Galaxy Luminosity Function Research Articles

Cosmic star-formation history and black hole accretion history inferred from the JWST mid-infrared source counts

ABSTRACT With the advent of the JWST, extragalactic source count studies were conducted down to sub-μJy in the mid-infrared (MIR), which is several tens of times fainter than what the previous-generation infrared (IR) telescopes achieved in the MIR. In this work, we aim to interpret the JWST source counts and constrain cosmic star-formation history (CSFH) and black hole accretion history (BHAH). We employ the backward evolution of local luminosity functions (LLFs) of galaxies to reproduce the observed source counts from sub-μJy to a few tens of mJy in the MIR bands of the JWST. The shapes of the LLFs at the MIR bands are determined using the model templates of the spectral energy distributions (SEDs) for five representative galaxy types (star-forming galaxies, starbursts, composite, AGN type 2 and 1). By simultaneously fitting our model to all the source counts in the six MIR bands, along with the previous results, we determine the best-fitting evolutions of MIR LFs for each of the five galaxy types, and subsequently estimate the CSFH and BHAH. Thanks to the JWST, our estimates are based on several tens of times fainter MIR sources, the existence of which was merely an extrapolation in previous studies.

JWST Constraints on the UV Luminosity Density at Cosmic Dawn: Implications for 21 cm Cosmology

An unprecedented array of new observational capabilities are starting to yield key constraints on models of the epoch of first light in the Universe. In this Letter we discuss the implications of the UV radiation background at cosmic dawn inferred by recent JWST observations for radio experiments aimed at detecting the redshifted 21 cm hyperfine transition of diffuse neutral hydrogen. Under the basic assumption that the 21 cm signal is activated by the Lyα photon field produced by metal-poor stellar systems, we show that a detection at the low frequencies of the EDGES and SARAS3 experiments may be expected from a simple extrapolation of the declining UV luminosity density inferred at z ≲ 14 from JWST early galaxy data. Accounting for an early radiation excess above the cosmic microwave background suggests a shallower or flat evolution to simultaneously reproduce low- and high-z current UV luminosity density constraints, which cannot be entirely ruled out, given the large uncertainties from cosmic variance and the faint-end slope of the galaxy luminosity function at cosmic dawn. Our findings raise the intriguing possibility that a high star formation efficiency at early times may trigger the onset of intense Lyα emission at redshift z ≲ 20 and produce a cosmic 21 cm absorption signal 200 Myr after the Big Bang.

A new tool to derive simultaneously exponent and extremes of power-law distributions

ABSTRACT Many experimental quantities show a power-law distribution p(x) ∝ x−α. In astrophysics, examples are: size distribution of dust grains or luminosity function of galaxies. Such distributions are characterized by the exponent α and by the extremes xminxmax where the distribution extends. There are no mathematical tools that derive the three unknowns at the same time. In general, one estimates a set of α corresponding to different guesses of xminxmax. Then, the best set of values describing the observed data is selected a posteriori. In this paper, we present a tool that finds contextually the three parameters based on simple assumptions on how the observed values xi populate the unknown range between xmin and xmax for a given α. Our tool, freely downloadable, finds the best values through a non-linear least-squares fit. We compare our technique with the maximum likelihood estimators for power-law distributions, both truncated and not. Through simulated data, we show for each method the reliability of the computed parameters as a function of the number N of data in the sample. We then apply our method to observed data to derive: (i) the slope of the core mass function in the Perseus star-forming region, finding two power-law distributions: α = 2.576 between $1.06\, \mathrm{M}_{\odot }$ and $3.35\, \mathrm{M}_{\odot }$, α = 3.39 between $3.48\, \mathrm{M}_{\odot }$ and $33.4\, \mathrm{M}_{\odot }$; (ii) the slope of the γ-ray spectrum of the blazar J0011.4+0057, extracted from the Fermi-LAT archive. For the latter case, we derive α = 2.89 between 1484 MeV and 28.7 GeV; then we derive the time-resolved slopes using subsets of 200 photons each.

Revisiting X-Ray-bright Optically Normal Galaxies with the Chandra Source Catalog

X-ray bright optically normal galaxies (XBONGs) are galaxies with X-ray luminosities consistent with those of active galactic nuclei (AGNs) but no evidence of AGN optical emission lines. Crossmatching the Chandra Source Catalog version 2 with the Sloan Digital Sky Survey sample of spectroscopically classified galaxies, we have identified 817 XBONG candidates with L X > 1042 erg s−1 and X-ray to optical flux ratio F XO > 0.1. Comparisons with Wide-field Infrared Survey Explorer colors and near-IR, optical, UV, and radio luminosities show that the loci of XBONGs are in-between those of control samples of normal galaxies and quasars and are consistent with low-luminosity quasars. We find that 43% of the XBONG sample have X-ray colors suggesting N H > 1022 cm−2, double the fraction in the QSO sample, suggesting that a large fraction of XBONG are highly obscured AGNs. However, ∼50% of the XBONGs are not obscured and have X-ray colors harder than those of normal galaxies. Some of these XBONGs have spatially extended X-ray emission. These characteristics suggest that they may be unidentified galaxy groups and clusters. Using the X-ray luminosity functions of QSOs, galaxies, groups and clusters, we estimate the approximate fraction of extended XBONGs to be <20%. We also assess the approximate fraction of XBONGs whose AGN signatures are diluted by stellar light of host galaxies to be ∼30%, based on their redshift and deviation from the extrapolation of the QSO L X–L r relation.

The impact of UV variability on the abundance of bright galaxies at z ≥ 9

ABSTRACT JWST observations have revealed a population of galaxies bright enough that potentially challenge standard galaxy formation models in the Λ cold dark matter (ΛCDM) cosmology. Using a minimal empirical framework, we investigate the influence of variability on the rest-frame ultra-violet (UV) luminosity function of galaxies at z ≥ 9. Our study differentiates between the median UV radiation yield and the variability of UV luminosities of galaxies at a fixed dark matter halo mass. We primarily focus on the latter effect, which depends on halo assembly and galaxy formation processes and can significantly increase the abundance of UV-bright galaxies due to the upscatter of galaxies in lower-mass haloes. We find that a relatively low level of variability, σUV ≈ 0.75 mag, matches the observational constraints at z ≈ 9. However, increasingly larger σUV is necessary when moving to higher redshifts, reaching $\sigma _{\rm UV} \approx 2.0\, (2.5)\, {\rm mag}$ at z ≈ 12 (16). This implied variability is consistent with expectations of physical processes in high-redshift galaxies such as bursty star formation and dust clearance during strong feedback cycles. Photometric constraints from JWST at z ≳ 9 therefore can be reconciled with a standard ΛCDM-based galaxy formation model calibrated at lower redshifts without the need for adjustments to the median UV radiation yield.

Cosmic variance and the inhomogeneous UV luminosity function of galaxies during reionization

ABSTRACT When the first galaxies formed and starlight escaped into the intergalactic medium to reionize it, galaxy formation and reionization were both highly inhomogeneous in time and space, and fully coupled by mutual feedback. To show how this imprinted the UV luminosity function (UVLF) of reionization-era galaxies, we use our large-scale, radiation-hydrodynamics simulation CoDa II to derive the time- and space-varying halo mass function and UVLF, from z ≃ 6–15. That UVLF correlates strongly with local reionization redshift: earlier-reionizing regions have UVLFs that are higher, more extended to brighter magnitudes, and flatter at the faint end than later-reionizing regions observed at the same z. In general, as a region reionizes, the faint-end slope of its local UVLF flattens, and, by z = 6 (when reionization ended), the global UVLF, too, exhibits a flattened faint-end slope, ‘rolling-over’ at MUV ≳ −17. CoDa II’s UVLF is broadly consistent with cluster-lensed galaxy observations of the Hubble Frontier Fields at z = 6–8, including the faint end, except for the faintest data point at z = 6, based on one galaxy at MUV = −12.5. According to CoDa II, the probability of observing the latter is $\sim 5~{{\ \rm per\ cent}}$. However, the effective volume searched at this magnitude is very small, and is thus subject to significant cosmic variance. We find that previous methods adopted to calculate the uncertainty due to cosmic variance underestimated it on such small scales by a factor of 2–4, primarily by underestimating the variance in halo abundance when the sample volume is small.

The UV luminosity functions of bright z &amp;gt; 8 galaxies: determination from ∼0.41 deg2 of HST observations along ∼300 independent sightlines

ABSTRACT We determine the bright end of the rest-frame UV luminosity function (UVLF) at z = 8–10 by selecting bright z ≳ 8 photometric candidates from the largest systematic compilation of HST (pure-)parallel observations to date, the Super-Brightest-of-Reionizing-Galaxies (SuperBoRG) data set. The data set includes ∼300 independent sightlines from WFC3 observations, totalling 800–1300 arcmin2 (depending on redshift). We identify 31 z ≳ 8 candidates via colour selection and photo-z analysis with observed magnitude (24.1 &amp;lt; H160 &amp;lt; 26.6) and 1σ range probability of $58\!-\!95~{{\ \rm per\ cent}}$ of being true high-z galaxies. Following detailed completeness and source recovery simulations, as well as modelling of interloper contamination, we derive rest-frame UVLFs at z = 8–10 down to MUV ≃ −23. We find that the bright end of the galaxy luminosity function can be described both by a Schechter and by a double power-law function, with our space-based large area determination showing some tentative discrepancies with the luminosity functions derived from ground-based observations at the same redshifts. We identify two possible explanations. The first is that the UVLF includes a substantial contribution from AGNs at magnitudes MUV &amp;lt; −22. The UVLF we derive at z = 8 is consistent with no evolution of the bright end from z = 6–7 UVLFs that include AGNs. An alternative (and non-exclusive) possibility is that the dust content of the observed galaxies decreases with redshift, thereby countering the naturally expected decrease in the UV luminosity functions because of a decreased star-formation rate with increasing redshift. Both scenarios raise interesting prospects to further understand galaxy formation in extreme objects during the epoch of reionization.

A New Sample of Transient Ultraluminous X-Ray Sources Serendipitously Discovered by Swift/XRT

Ultraluminous X-ray sources (ULXs) are our best laboratories for studying extreme super-Eddington accretion. Most studies of these objects are of relatively persistent sources; however, there is growing evidence to suggest a large fraction of these sources are transient. Here we present a sample of five newly reported transient ULXs in the galaxies NGC 4945, NGC 7793, and M81 serendipitously discovered in Swift/XRT observations. Swift monitoring of these sources have provided well-sampled lightcurves, allowing for us to model the lightcurves with the disk-instability model of Hameury & Lasota, which implies durations of 60–400 days and that the mass-accretion rate through the disk is close to or greater than the Eddington rate. Of the three source regions with prior Hubble Space Telescope imaging, color–magnitude diagrams of the potential stellar counterparts show varying ages of the possible stellar counterparts. Our estimation of the rates of these sources in these three galaxies is 0.4–1.3 yr−1. We find that, while persistent ULXs dominate the high end of galaxy luminosity functions, the number of systems that produce ULX luminosities are likely dominated by transient sources.

The total rest-frame UV luminosity function from 3 &amp;lt; z &amp;lt; 5: a simultaneous study of AGN and galaxies from −28 &amp;lt; MUV &amp;lt; −16

ABSTRACT We present measurements of the rest-frame ultraviolet luminosity function (LF) at redshifts z = 3, z = 4, and z = 5, using 96894, 38655, and 7571 sources, respectively, to map the transition between active galactic nuclei (AGN) and galaxy-dominated ultraviolet emission shortly after the epoch of reionization (EoR). Sources are selected using a comprehensive photometric redshift approach, using 10 $\, {\rm deg}^2$ of deep extragalactic legacy fields covered by both HSC and VISTA. The use of template fitting spanning a wavelength range of 0.3–2.4 μm achieves 80–90 per cent completeness, much higher than the classical colour–colour cut methodology. The measured LF encompasses −26 &amp;lt; MUV &amp;lt; −19.25. This is further extended to −28.5 &amp;lt; MUV &amp;lt; −16 using complementary results from other studies, allowing for the simultaneous fitting of the combined AGN and galaxy LF. We find that there are fewer UV luminous galaxies (MUV &amp;lt; −22) at z ∼ 3 than z ∼ 4, indicative of an onset of widespread quenching alongside dust obscuration, and that the evolution of the AGN LF is very rapid, with their number density rising by around two orders of magnitude from 3 &amp;lt; z &amp;lt; 6. It remains difficult to determine if a double power law functional form is preferred over the Schechter function to describe the galaxy UV LF. Estimating the hydrogen ionizing photon budget from our UV LFs, we find that AGN can contribute to, but cannot solely maintain, the reionization of the Universe at z = 3–5. However, the rapidly evolving AGN LF strongly disfavours a significant contribution within the EoR.

Gravitational lensing modification of the high-redshift galaxy luminosity function

ABSTRACT The bright end of the rest-frame ultraviolet luminosity function (UVLF) of high-redshift galaxies is modified by gravitational lensing magnification bias. Motivated by recent discoveries of very high-z galaxies with JWST, we study the dependence of magnification bias on the finite size of sources at 6 &amp;lt; z &amp;lt; 14. We calculate the magnification probability distributions and use these to calculate the magnification bias assuming a rest-frame Schechter UVLF for galaxies at redshift 6 &amp;lt; z &amp;lt; 14. We find that the finite size of bright high-redshift galaxies together with lens ellipticity significantly suppresses magnification bias, producing an observed bright end which declines more sharply than the power law resulting from assumption of point sources. By assuming a luminosity–size relation for the source population and comparing with the observed z = 6 galaxy luminosity function from Harikane et al. (2022), we show that the UVLF can be used to set mild constraints on the galaxies intrinsic size, favouring smaller galaxies compared to the fiducial luminosity–size relation. In the future, wide surveys using Euclid and Roman Space Telescope will place stronger constraints. We also tabulate the maximum magnification possible as a function of source size and lens ellipticity.

Measurement of the evolving galaxy luminosity and mass function using clustering-based redshift inference

ABSTRACT We develop a framework for using clustering-based redshift inference (cluster-$z$ ) to measure the evolving galaxy luminosity function (GLF) and galaxy stellar mass function (GSMF) using Wide-field Infrared Survey Explorer W1 (3.4 μm) mid-infrared photometry and positions. We use multiple reference sets from the Galaxy And Mass Assembly survey, Sloan Digital Sky Survey and Baryon Oscillation Spectroscopic Survey. Combining the resulting cluster-$z$ s allows us to enlarge the study area, and by accounting for the specific properties of each reference set, making best use of each reference set to produce the best overall result. Thus we are able to measure the GLF and GSMF over ∼7500 deg2 of the Northern Galactic Cap up to $z$ &amp;lt; 0.6. Our method can easily be adapted for new studies with fainter magnitudes, which pose difficulties for the derivation of photo-$z$ s. With better statistics in future surveys this technique is a strong candidate for studies with new emerging data from, e.g. the Vera C Rubin Observatory, the Euclid mission or the Nancy Grace Roman Space Telescope.

The Hα Luminosity Function of Galaxies at z ∼ 4.5

We present the Hα luminosity function (LF) derived from a large sample of Lyman break galaxies at z ∼ 4.5 over the GOODS-South and North fields. This study makes use of the new, full-depth Spitzer/IRAC [3.6] and [4.5] imaging from the GOODS Re-ionization Era wide-Area Treasury from the Spitzer program. The Hα flux is derived from the offset between the continuum flux estimated from the best-fit spectral energy distribution, and the observed photometry in IRAC [3.6]. From these measurements, we build the Hα LF and study its evolution providing the best constraints of this property at high redshift, where spectroscopy of Hα is not yet available. Schechter parameterizations of the Hα LF show a decreasing evolution of Φ∗ with redshift, increasing evolution in L ∗, and no significant evolution in the faint-end slope at high z. We find that star formation rates (SFRs) derived from Hα are higher than those derived from the rest-frame UV for low SFR galaxies but the opposite happens for the highest SFRs. This can be explained by lower mass galaxies (also lower SFR) having, on average, rising star formation histories (SFHs), while at the highest masses the SFHs may be declining. The SFR function is steeper, and because of the excess SFR(Hα) compared to SFR(UV) at low SFRs, the SFR density estimated from Hα is higher than the previous estimates based on UV luminosities.

JWST high-redshift galaxy constraints on warm and cold dark matter models

Context. Warm dark matter is a possible alternative to cold dark matter to explain cosmological structure formation. Aims. We study the implications of the latest JWST data on the nature of dark matter. Methods. We compare properties of high-redshift galaxies observed by JWST with hydrodynamical simulations, in the standard cold dark matter model and in warm dark matter models with a suppressed linear matter power spectrum Results. We find that current data are neither in tension with cold dark matter nor with warm dark matter models with mWDM &gt; 2 keV, since they probe bright and rare objects whose physical properties are similar in the different scenarios. Conclusions. We also show how two observables, the galaxy luminosity functions and the galaxy correlation function at small scales of faint objects, can be promising tools for discriminating between the different dark-matter scenarios. Further hints may come from early stellar-mass statistics and galaxy CO emission.

Identifying Active Galactic Nuclei at z ∼ 3 from the HETDEX Survey Using Machine Learning

We used data from the Hobby–Eberly Telescope Dark Energy Experiment (HETDEX) to study the incidence of AGN in continuum-selected galaxies at z ∼ 3. From optical and infrared imaging in the 24 deg2 Spitzer HETDEX Exploratory Large Area survey, we constructed a sample of photometric-redshift selected z ∼ 3 galaxies. We extracted HETDEX spectra at the position of 716 of these sources and used machine-learning methods to identify those which exhibited AGN-like features. The dimensionality of the spectra was reduced using an autoencoder, and the latent space was visualized through t-distributed stochastic neighbor embedding. Gaussian mixture models were employed to cluster the encoded data and a labeled data set was used to label each cluster as either AGN, stars, high-redshift galaxies, or low-redshift galaxies. Our photometric redshift (photoz) sample was labeled with an estimated 92% overall accuracy, an AGN accuracy of 83%, and an AGN contamination of 5%. The number of identified AGN was used to measure an AGN fraction for different magnitude bins. The ultraviolet (UV) absolute magnitude where the AGN fraction reaches 50% is M UV = −23.8. When combined with results in the literature, our measurements of AGN fraction imply that the bright end of the galaxy luminosity function exhibits a power law rather than exponential decline, with a relatively shallow faint-end slope for the z ∼ 3 AGN luminosity function.

Dust-free starburst galaxies at redshiftsz&amp;gt; 10

ABSTRACTOne of the most distant galaxies GN-z11 was formed when the Universe was ≤400 Myr old, and it displays a burst-like star formation rate ∼25 M⊙ yr−1 with a metallicity Z ∼ 0.2 ± 0.1 Z⊙. It resembles z = 2–3 galaxies (at ‘cosmic noon’) except for the fact that the measured reddening E(B − V) = 0.01 ± 0.01 indicates the presence of little or no dust. This marked absence of dust hints towards violent dynamical events that destroy or evacuate dust along with gas out of the galaxy on a relatively short time-scale and make it transparent. We apply a 3D numerical model to infer possible physical characteristics of these events. We demonstrate that the energetics of the observed star formation rate is sufficient to tear apart the dusty veil on time-scales of 20–25 Myr. This can explain the apparent lack of evolution of ultraviolet luminosity function of galaxies between z ≥ 10 and z ∼ 7, by compensating for the lower galaxy masses at higher redshift by the absence of dust. We show, however, that this is a temporary phenomenon and soon after the last of the supernovae explosions have taken place, the expanding shell shrinks and obscures the galaxy on time-scales of ≈5–8 Myr.

The miniJPAS survey quasar selection – II. Machine learning classification with photometric measurements and uncertainties

ABSTRACTAstrophysical surveys rely heavily on the classification of sources as stars, galaxies, or quasars from multiband photometry. Surveys in narrow-band filters allow for greater discriminatory power, but the variety of different types and redshifts of the objects present a challenge to standard template-based methods. In this work, which is part of a larger effort that aims at building a catalogue of quasars from the miniJPAS survey, we present a machine learning-based method that employs convolutional neural networks (CNNs) to classify point-like sources including the information in the measurement errors. We validate our methods using data from the miniJPAS survey, a proof-of-concept project of the Javalambre Physics of the Accelerating Universe Astrophysical Survey (J-PAS) collaboration covering ∼1 deg2 of the northern sky using the 56 narrow-band filters of the J-PAS survey. Due to the scarcity of real data, we trained our algorithms using mocks that were purpose-built to reproduce the distributions of different types of objects that we expect to find in the miniJPAS survey, as well as the properties of the real observations in terms of signal and noise. We compare the performance of the CNNs with other well-established machine learning classification methods based on decision trees, finding that the CNNs improve the classification when the measurement errors are provided as inputs. The predicted distribution of objects in miniJPAS is consistent with the putative luminosity functions of stars, quasars, and unresolved galaxies. Our results are a proof of concept for the idea that the J-PAS survey will be able to detect unprecedented numbers of quasars with high confidence.

Are Brightest Cluster Galaxies Special?

A longstanding question when studying Brightest Cluster Galaxies (BCGs) and their evolution is whether their luminosities are statistical extremes of the luminosity distribution of other cluster member galaxies, or whether their luminosities follow a different,“special” distribution. To explore this question, we use the two Tremaine–Richstone (TR) statistics that sometimes lead to different conclusions about BCG specialness. We run Monte Carlo simulations of galaxy clusters with varying galaxy luminosity functions and BCG properties, and explore the sensitivity of the TR statistics to the adjusted BCG properties. We identify cases where the two statistics lead to different conclusions. The first of the TR statistics, which relies on the spread of BCG luminosities, is more reliable for determining specialness than the second, which utilizes the spread in the luminosity gap between the BCG and second ranked galaxy.

Galaxy Populations in Groups and Clusters: Evidence for a Characteristic Stellar Mass Scale at M ∗ ∼ 109.5 M ⊙

We use the DR9 of the DESI legacy imaging survey and SDSS galaxy groups to measure the conditional luminosity function (CLF) for groups with halo mass M h ≥ 1012 M ⊙ and redshift 0.01 ≤ z ≤ 0.08, down to a limiting r-band magnitude of M r = −10 to −12. For given halo masses we measure the CLF for the total populations and for the red and blue populations classified using the (g − z) color. We find a clear faint-end upturn in the CLF of red satellites, with a slope α ≈ −1.8, which is almost independent of halo mass. This faint-end upturn is not seen for the blue and total populations. Our stellar population synthesis modeling shows that (g − z) provides a clean red/blue division and that red group galaxies defined by (g − z) are all dominated by old stellar populations. The fraction of old galaxies as a function of galaxy luminosity shows a minimum at M r ∼ −18, corresponding to M * ∼ 109.5 M ⊙. This scale is independent of halo mass and is comparable to the characteristic luminosity at which galaxies show a dichotomy in surface brightness and size, suggesting that the dichotomy in the old fraction and in galaxy structure may have a common origin. The rising of the old fraction at the faint end for Milky Way (MW)−sized halos is in good agreement with the quenched fraction measured for the MW/M31 system and from the ELVES survey. We discuss the implications of our results for the formation and evolution of low-mass galaxies and for the stellar mass functions of low-mass galaxies to be observed at high redshift.

The extent of intergalactic metal enrichment from galactic winds during the Cosmic Dawn

ABSTRACT One of the key processes driving galaxy evolution during the Cosmic Dawn is supernova feedback. This likely helps regulate star formation inside of galaxies, but it can also drive winds that influence the large-scale intergalactic medium. Here, we present a simple semi-analytic model of supernova-driven galactic winds and explore the contributions of different phases of galaxy evolution to cosmic metal enrichment in the high-redshift (z ≳ 6) Universe. We show that models calibrated to the observed galaxy luminosity function at z ∼ 6–8 have filling factors $\sim 1{{\%}}$ at z ∼ 6 and $\sim 0.1{{\%}}$ at z ∼ 12, with different star formation prescriptions providing about an order of magnitude uncertainty. Despite the small fraction of space filled by winds, these scenarios predict an upper limit to the abundance of metal-line absorbers in quasar spectra at $z \gtrsim 5$ which is comfortably above that currently observed. We also consider enrichment through winds driven by Pop III star formation in minihalos. We find that these can dominate the total filling factor at $z \gtrsim 10$ and even compete with winds from normal galaxies at z ∼ 6, at least in terms of the total enriched volume. But these regions have much lower overall metallicities, because each one is generated by a small burst of star formation. Finally, we show that Compton cooling of these supernova-driven winds at $z \gtrsim 6$ has only a small effect on the cosmic microwave background.

An Ultra-deep Multiband Very Large Array (VLA) Survey of the Faint Radio Sky (COSMOS-XS): New Constraints on the Cosmic Star Formation History

We make use of ultra-deep 3 GHz Karl G. Jansky Very Large Array (VLA) observations of the COSMOS field from the multiband COSMOS-XS survey to infer radio luminosity functions (LFs) of star-forming galaxies (SFGs). Using ∼1300 SFGs with redshifts out to z ∼ 4.6, and fixing the faint and bright end shape of the radio LF to the local values, we find a strong redshift trend that can be fitted by pure luminosity evolution with the luminosity parameter given by α L ∝ (3.40 ± 0.11) − (0.48 ± 0.06)z. We then combine the ultra-deep COSMOS-XS data set with the shallower VLA-COSMOS 3 GHz large project data set over the wider COSMOS field in order to fit for joint density+luminosity evolution, finding evidence for significant density evolution. By comparing the radio LFs to the observed far-infrared and ultraviolet (UV) LFs, we find evidence of a significant underestimation of the UV LF by 22% ± 14% at high redshift (3.3 < z < 4.6, integrated down to ). We derive the cosmic star formation rate density (SFRD) by integrating the fitted radio LFs and find that the SFRD rises up to z ∼ 1.8 and then declines more rapidly than previous radio-based estimates. A direct comparison between the radio SFRD and a recent UV-based SFRD, where we integrate both LFs down to a consistent limit (), reveals that the discrepancy between the radio and UV LFs translates to a significant (∼1 dex) discrepancy in the derived SFRD at z > 3, even assuming the latest dust corrections and without accounting for optically dark sources.