EDP Sciences logo
Subscriber Authentication Point
Search
Menu
Home All issues Volume 672 (April 2023) A&A, 672 (2023) A186 Full HTML
Open Access
Issue
A&A
Volume 672, April 2023
Article Number A186
Number of page(s) 10
Section Extragalactic astronomy
DOI https://doi.org/10.1051/0004-6361/202245532
Published online 20 April 2023

© The Authors 2023

Licence Creative CommonsOpen Access article, published by EDP Sciences, under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This article is published in open access under the Subscribe to Open model. Subscribe to A&A to support open access publication.

1. Introduction

In recent years, we have obtained increasing evidence that the reionization of hydrogen happened fairly late, approximately one billion years after the Big Bang (z ∼ 5.5 − 10), with a mid-point around z ∼ 7 − 8 (e.g., Fan et al. 2006; Stark et al. 2010; McGreer et al. 2015; Mason et al. 2018; Davies et al. 2018; Qin et al. 2021; Planck Collaboration VI 2020; Bolan et al. 2022). However, there is evidence for significant star formation before this time (e.g., Oesch et al. 2018; Hashimoto et al. 2018; McLeod et al. 2021), and thus it appears that reionization lags behind galaxy formation. The reason for this lag is unknown: we are still lacking a full physical understanding of the reionization process. In particular, we still do not know which types of galaxies drive the process, that is to say, which physical mechanisms mediate the production and escape of ionizing photons from galaxies. In order to produce such a late and fairly rapid reionization, the ionizing population could have been dominated by low mass, UV-faint galaxies with a low average escape fraction (∼5%; e.g., Mason et al. 2019; Qin et al. 2021). Alternatively, rarer, more massive galaxies with higher escape fractions could have been responsible (e.g., Sharma et al. 2017; Naidu et al. 2020). With only measurements of the timing of reionization, these scenarios are degenerate, and thus physical priors on the ionizing properties of galaxies across cosmic time are necessary to pinpoint the sources of reionization.

The total ionizing output of galaxies can be simply parameterized (e.g., Madau et al. 1999; Robertson et al. 2010) as the product of the production rate of ionizing photons relative to non-ionizing UV photons, ξion (determined by the stellar populations; e.g., Stanway et al. 2016) and the fraction of ionizing photons that escape the interstellar medium (ISM) into the intergalactic medium, fesc (determined by the structure and ionization state of the ISM, which is likely shaped by star formation and feedback; e.g., Trebitsch et al. 2017; Ma et al. 2020). Both of these quantities are also expected to vary with time in an individual galaxy, for example due to the lifetime and properties of young stellar populations, and depending on the effects of feedback and bursty star formation on the ISM.

While we can easily observe the non-ionizing UV photons from galaxies, the high optical depth of the intergalactic medium to ionizing photons makes direct measurements of the escaping ionizing spectrum statistically unlikely at z ≳ 3 (Inoue et al. 2014; Becker et al. 2021; Vanzella et al. 2018). Alternatively, fluxes of nonresonant recombination lines, emitted by gas that was ionized in HII regions around massive stars, can crucially measure the flux of ionizing photons that do not escape galaxies. In particular, Hα emission can be used to directly estimate (1 − fesc)ξion (e.g., Leitherer & Heckman 1995; Bouwens et al. 2016; Shivaei et al. 2018; Emami et al. 2020). As fesc is inferred to be low (≲10%) on average for Lyman-break galaxies at z ∼ 2 − 4 (Steidel et al. 2018; Begley et al. 2022; Pahl et al. 2023), measurements of Hα should trace the intrinsic production of ionizing photons reasonably well. The value of ξion can also be inferred from the strength of [OIII]+Hβ emission; however, due to the dependence of [OIII] emission on metallicity and the ionization parameter, the correlation is not as tight as with Hα (e.g., Chevallard et al. 2018).

Previous work at z ≲ 2.5, where direct Hα spectroscopy has been possible from the ground, has found a mean log10ξion [erg Hz−1]  ≈  25.3, with a scatter of ∼0.3 dex, likely dominated by variations in stellar populations between galaxies (e.g., Shivaei et al. 2018; Tang et al. 2019). At higher redshifts, where Hα redshifts into the infrared, broadband photometry with Spitzer has been used extensively to estimate Hα line fluxes (e.g., Schaerer & de Barros 2009; Shim et al. 2011; Stark et al. 2013; Smit et al. 2015; Bouwens et al. 2016; Lam et al. 2019; Maseda et al. 2020; Stefanon et al. 2022).

However, due to the limited spatial resolution and sensitivity of Spitzer, previous works were limited to studying ξion in isolated, bright (> L*) galaxies, where de-blending Infrared Array Camera (IRAC) photometry was possible (e.g., Bouwens et al. 2016), and using stacks for fainter galaxies (e.g., Lam et al. 2019; Maseda et al. 2020). With James Webb Space Telescope (JWST) it is finally possible to extend these studies to individual UV-faint galaxies (Endsley et al. 2022) and obtain rest-frame optical spectroscopy at z > 3 (e.g., Sun et al. 2022; Williams et al. 2022).

Results from previous analyses have been intriguing but require further investigation. Using stacked IRAC photometry, Lam et al. (2019) find no significant evidence for a strong correlation of ξion with MUV. However, Maseda et al. (2020) find a population of extremely UV-faint galaxies (MUV > −16) selected as Lyα emitters in deep Multi Unit Spectroscopic Explorer (MUSE) observations, which have very elevated ξion compared to higher luminosity galaxies and at fixed Hα equivalent widths (EWs), which implies that these efficient ionizing galaxies are particularly young and of low metallicity. It is thus important to examine the distribution of ξion at low UV luminosities, and to compare galaxies with and without Lyα emission to better understand the demographics of the ionizing population.

Furthermore, using early JWST NIRCam data, Endsley et al. (2022) discovered a population of UV-faint galaxies (MUV ∼ −19) at z ∼ 6.5 − 8 with high specific star formation rates (sSFRs) but low EW [OIII]+Hβ inferred from photometry. The high sSFR would imply high ξion due to the increased abundance of O and B stars. To explain the low [OIII]+Hβ EW, Endsley et al. (2022) suggest that either these galaxies have extremely low metallicities (reducing oxygen abundance) or, alternatively, that all nebular lines are reduced. A reduction in all nebular lines could be due to either them being produced in density-bounded HII regions with a very high ionizing escape fraction (e.g. Zackrisson et al. 2013; Marques-Chaves & Schaerer 2022) or a recent cessation of star formation. At z ∼ 3 − 7, both [OIII]+Hβ and Hα are visible in NIRCam photometry, enabling us to test these scenarios.

In this paper we make use of deep multiband Hubble Space Telescope (HST)/ACS, WFC3, and JWST/NIRCam imaging with overlapping MUSE observations, which enables us to blindly detect a spectroscopic sample with precision rest-frame UV-to-optical photometry. We measure the distribution of ξion over a broader luminosity range (−23 ≲ MUV ≲ −15.5) than previously possible in individual galaxies thanks to the excellent resolution and sensitivity of NIRCam at rest-optical wavelengths compared to Spitzer/IRAC as well as the power of gravitational lensing. We explore correlations of ξion with empirical galaxy properties. We find significant trends of increasing ξion with decreasing UV luminosity, decreasing UV β slope, and increasing Hα EW, all of which implies that the strongest ionizers are young sources with expected low metallicities. We also explore whether our sample shows evidence for very low metallicities or an extremely high escape fraction.

The paper is structured as follows. In Sect. 2 we describe the photometric and spectroscopic data for our study. In Sect. 3 we describe how we infer the ionizing production rate, ξion, and in Sect. 4 we describe the correlations we find between ξion and other galaxy properties and present a comparison to the literature. We discuss our results and state our conclusions in Sect. 6.

We assume a flat Λ cold dark matter cosmology with Ωm = 0.3,  ΩΛ = 0.7 and h = 0.7. All magnitudes are in the AB system.

2. Data

For this work we selected fields with multiband HST/ACS and JWST/NIRCam imaging and overlapping MUSE spectroscopy. We selected sources detected with Lyα emission (z ∼ 2.9 − 6.7 in MUSE) and sources with a high probability of being in the same redshift range based on photometric redshift, and we used the HST + JWST photometry to extract optical emission line fluxes. Below we describe the data sets and the selection of our sample.

2.1. Imaging

We used JWST NIRCam imaging in parallel to and of the cluster Abell 2744 from the GLASS-JWST program ERS-1324 (PI Treu; Treu et al. 2022) and the UNCOVER1 program GO-2561 (co-PIs Labbé and Bezanson).

The GLASS-JWST NIRCam observations discussed in this paper were taken in parallel to NIRISS observations of the cluster Abell 2744 on June 28–29, 2022. They are centered at RA = 3.5017025 deg and Dec = −30.3375436 deg and consist of imaging in seven bands: F090W (total exposure time: 11 520 s), F115W (11520 s.), F150W (6120 s.), F200W (5400 s.), F277W (5400 s.), F356W (6120 s.), and F444W (23400 s.). The UNCOVER NIRCam observations of the Abell 2744 cluster were taken on November 2-15, 2022. They are centered at RA = 3.5760475 deg and Dec = −30.37946 deg and consist of imaging in seven bands: F115W (10823 s.), F150W (10823 s.), F200W (6700 s.), F277W (6700 s.), F356W (6700 s.), F410M (6700 s.), and F444W (8246 s.).

In our analysis, we also included new and archival HST imaging; the ACS imaging is particularly important for constraining photometric redshifts. This includes new HST/ACS data in F606W (59 530 s.), F775W (23 550 s.), and F814W (123 920 s) from HST-GO/DD program 172312 (PI Treu), as well as archival data acquired under the Hubble Frontier Fields program (HST-GO/DD-13495, PI Lotz; Lotz et al. 2017), BUFFALO (HST-GO-15117 PI Steinhardt; Steinhardt et al. 2020), and programs HST-GO-11689 (PI Dupke), HST-GO-11386 (PI Rodney), HST-GO-13389 (PI Siana), HST-GO-15940 (PI Ribeiro), and HST-SNAP-16729 (PI Kelly). Not all HST bands cover every object in our sample, and we only kept objects in our sample that have a well-constrained photometric redshift, usually meaning that there is ACS coverage (see Sect. 2.4). We also included HST/WFC3 imaging for completeness, but it is generally not as constraining as the NIRCam fluxes.

The image reduction and calibration, and the methods used to detect sources and measure multiband photometry in both fields, closely follow that of Brammer et al. (in prep.). Briefly, we pulled calibrated images from the Mikulski Archive for Space Telescopes (MAST) 3 and processed them with the grizli pipeline (Brammer et al. 2022). The pipeline first aligns the exposures to external catalogs and to one another and corrects for any distortion within the image. Following this, we subtracted a sky-level background, divided out flat-field structure using custom flat-field images, and corrected for 1/f noise. We also corrected for NIRCam image anomalies, which include persistence, any remaining cosmic rays, and “snowballs” (see Rigby et al. 2023). Finally, we applied zero-point corrections calculated by G. Brammer4 and drizzled all exposures to a common pixel grid.

For source detection, we used SEP, Source Extraction and Photometry (Barbary 2018), to perform aperture photometry on the F444W detection image in each field.

2.2. VLT/MUSE spectroscopy

MUSE spectroscopy of the Abell 2744 cluster was obtained through ESO program 094.A-0115 (PI Richard) and is described by Mahler et al. (2018) and Richard et al. (2021). We used their publicly available catalog to select Lyα-emitting galaxies. The data comprise a 4 sq. arcmin mosaic centered on the cluster core. Four 1 sq. arcmin quadrants were observed for a total of 3.5, 4, 4, and 5 h, respectively, and the center of the cluster was observed for an additional 2 h. The median line flux 1σ uncertainty in the MUSE data is 3.6 × 10−19 erg s−1 cm−2. This corresponds to a 5σ EW limit of ∼4 − 30 Å over z ∼ 3 − 7 for a galaxy with MUV = −19 (the median for our sample before accounting for magnification as EW is invariant under magnification).

Very Large Telescope (VLT)/MUSE spectroscopy in the GLASS-JWST NIRCam fields were obtained through a new ESO Director’s Discretionary Time program, 109.24EZ.001 (co-PIs Mason, Vanzella), on the nights of July 28 and August 20, 2022. The data comprise five pointings (four of which are over 4 sq. arcmin and overlap with NIRCam imaging), each with 1 h of exposure time. The raw data are publicly available on the ESO archive5. The reduction, calibration, and source detection methods used for this work are identical to techniques described in Caminha et al. (2017, 2019). A full assessment of the depth is ongoing, but, based on the ∼4 h depth of the Mahler et al. (2018) observations described above, we estimate a 5σ EW limit of ∼8 − 60 Å in these shallower data.

In this work we used 102 spectroscopic confirmations at z ∼ 2.9 − 6.7: 42 from the GLASS-JWST NIRCam fields and 60 from the Abell 2744 cluster field.

2.3. Gravitational lensing magnification

For the galaxies detected in the core of the Abell 2744 cluster, we corrected for gravitational lensing magnification using the model from Bergamini et al. (2023). The median magnification of the sample is μ = 3.54, and 90% of the galaxies have μ = 2 − 20. We removed sources with a magnification with μ > 50 (12 sources) due to high uncertainties in the model near the critical curves. The galaxies in the parallel fields are ∼3 − 10′ away from the cluster core, where the magnification is expected to be modest (μ ≈ 1). We do not account for the magnification of these sources.

2.4. Sample selection

For this work we focused on selecting a sample of galaxies at z ∼ 3 − 7 with high purity. We selected 102 MUSE Lyα-detected galaxies with overlapping HST/ACS and JWST/NIRCam data as described above. We also selected a comparison sample of galaxies based on peak photometric redshift, within the same footprint as the MUSE observations, which we expect to have slightly lower Hα EWs than the Lyα-selected sample.

We found the photometric redshift distribution of all sources detected as described in Sect. 2.1 using EAZY (Brammer et al. 2008) and using all available photometric bands. To build a photometric sample with high purity, following Bouwens et al. (2016), we selected sources with the peak of their photometric redshift between 2.9 < z < 6.7 and kept only sources that have 90% of the redshift probability density between Δz ∼ 1 of the peak of their distribution. The resulting high purity photometric sample consists of 268 galaxies.

The redshift and UV magnitude distribution of our sample is shown in Fig. 1. The median redshift of the full sample is 4.02, and the Lyα-selected sample has a median redshift of 3.95. The median MUV is −18.1, with a Kolmogorov-Smirnov (KS) test showing no significant difference between the Lyα- and photometrically selected samples.

thumbnail Fig. 1.

Galaxies studied in this work: Lyα-detected galaxies (in purple) and galaxies photometrically selected (with no Lyα-detected, in gray). Top: Distribution of redshifts for the spectroscopic and photometric samples. We show the spectroscopic redshift, where available, or the peak photometric redshift. Bottom: UV magnitude distribution for our sample. We find a median value of −18.14 ± 1.58, with no statistically significant difference between the two samples.

3. Inferring the ionizing photon production rate

3.1. Inferring nebular emission line strengths from photometry

To estimate nebular emission line fluxes from broadband photometry, we followed approaches in the literature and fit the spectral energy distribution (SED) to the full photometry, excluding bands we expected to contain strong nebular emission lines (e.g., Shim et al. 2011; Stark et al. 2013; Mármol-Queraltó et al. 2016; Bouwens et al. 2016). This provided us with a model for the continuum flux in those bands that we could subtract from the observed photometry to infer the line flux.

We used BAGPIPES to fit SEDs (Carnall et al. 2018). We adopted BC03 (Bruzual & Charlot 2003) templates and excluded any nebular emission contribution. We did not consider any broadbands where Hα or [OIII]+Hβ are observed according to each galaxy’s redshift. For ease of comparison to the literature (e.g., Maseda et al. 2020; Lam et al. 2019), we assumed a Chabrier (2003) initial mass function and a Small Magellanic Cloud (SMC; Prevot et al. 1984) dust attenuation law, allowing AV to vary from 0 − 3 mag. Because metallicity is not well known at the range of redshifts we explored, we allowed metallicity to vary from 0 − 2 Z⊙. And because star formation histories are notoriously difficult to constrain at high redshifts (Strait et al. 2021), we assumed an exponentially rising delayed τ star formation history, allowing τ to vary freely. For the spectroscopically confirmed Lyman α emitters, we fixed the redshift at the Lyα redshift. For our photometric sample, we used the photometric redshift obtained from EAZY with a uniform prior with Δz = 1 (see Sect. 2.4).

We then compared the SED model of the galaxy’s continuum to the broadbands where Hα or [OIII]+Hβ fall. We multiplied the non-nebular SED posteriors by the transmission of the aforementioned broadbands to obtain the contribution of the galaxy’s continuum to the observed flux. By subtracting this continuum flux contribution from the observed photometry, we were then able to recover the flux distributions of the Hα and [OIII]+Hβ emission lines for each galaxy. We compared our measurements with a sample of six galaxies with [OIII]+Hβ EW measurements from the GLASS-ERS program using JWST/NIRISS (Boyett et al. 2022a), finding that our method recovers the EW of these sources to within ∼20 − 40%. A full comparison of these photometric inference methods is left to future work.

There are some limitations to our method for obtaining line fluxes, such as contamination from the 4000 Å break in the broadband that contains [OIII]+Hβ, the chance that the line falls outside the effective width of any of our broadband filters, or Hα and [OIII]+Hβ falling on the same band. We considered a contribution of 6.8% from [NII] to the calculated Hα flux, and 9.5% from [SII] according to Anders (2003). We removed galaxies with a poor χ2 score (> 50) on their SED fit; we chose this value by ignoring all galaxies on the high end of the χ2 distribution.

The advantage of this approach, unlike estimating line fluxes directly from the SED fitting, is that it does not depend strongly on star formation history assumptions and allows us to make a mostly empirical measurement of the line fluxes. We obtain comparable results using the flux in the band redward of Hα as the continuum flux, assuming a flat optical continuum (see also, e.g., Maseda et al. 2020). Estimating other physical parameters, such as the star formation rate and stellar mass, from the SED fitting did not give reliable results. This is because the fitting was too dependent on the initial assumptions and needed extremely young ages (< 10 Myrs) and an instantaneous burst of star formation to recreate the observed nebular emissions.

The following results consist of 83 and 64 Lyα-emitting galaxies with Hα and [OIII]+Hβ emission line measurements, respectively, and a photometric sample of 220 and 203 galaxies with Hα and [OIII]+Hβ emission line measurements, respectively. We see both lines in 62 Lyα galaxies and 177 photometrically selected galaxies. We see no apparent biases in our MUV distribution after narrowing down the sample. Nebular emission flux errors are derived from the 68% confidence interval of the resulting distributions.

3.2. Measuring UV absolute magnitude and slope

To infer the UV absolute magnitude, MUV (magnitude at 1500 Å), and β slope, we fit the power law (e.g., Rogers et al. 2013) fλ ∝ λβ to the fluxes from the HST and JWST bands. We performed the fit using a Markov chain Monte Carlo sampling and the python module emcee (Foreman-Mackey et al. 2013). We assumed flat priors for β and MUV, with bounds −4 < β < 1 and −25 < MUV < −12, sufficient to explore the common value ranges for galaxies (e.g., Bouwens et al. 2014).

To obtain the photometric bands that are observing the UV rest frame of our galaxies, we excluded any bands that fall blueward of Lyman-α and might be affected by the Lyman break. For the same reason, we excluded bands redward of the 4000 Å break in the rest frame. After these requirements, we are left with three or four bands for each source. In the case of galaxies with Lyman-α detected in MUSE, we used the line’s redshift. For photometrically selected galaxies, in each call of the likelihood, we randomly drew a redshift from a Normal distribution, N(μ = zphot, σ = 0.5), and selected the appropriate photometric bands. For lensed sources, we considered magnifications and applied them following the same random draw method as for the redshift. We used the corresponding magnification and error obtained from the Bergamini et al. (2023) lensing model.

3.3. Determination of ξion

We defined the production rate of ionizing photons, ξion, as the ratio between the luminosity of observed ionizing photons and the intrinsic luminosity of the ionizing UV photons (e.g., Leitherer & Heckman 1995):

(1)

where LHα is the unattenuated Hα luminosity in erg s−1 and Lν, UV, intr is the intrinsic UV luminosity density at 1500 Å. The models from where the conversion factor is derived assume a young population of massive stars equivalent to a massive HII region. We assumed this type of environment to be similar to what we would find in young galaxies.

Because Hα is produced by the excitation of hydrogen gas from ionizing radiation that does not escape the galaxy, and because we cannot directly measure fesc in our sample, we note that the production rate we obtain is for ionizing photons that did not escape the galaxy, ξion(1 − fesc).

We first calculated LHα directly from the SED obtained in Sect. 3.1, after accounting for dust attenuation (Prevot et al. 1984). To obtain the intrinsic value of the UV luminosity, we took the dust attenuation into account following Lam et al. (2019), who defined the intrinsic UV luminosity as , where fesc, UV is the fraction of escaping UV photons not absorbed by the dust. For this, we used the SMC dust law defined by Prevot et al. (1984):

(2)

where β is the UV slope obtained in Sect. 3.2. Galaxies with slopes bluer than β < −2.23 were assumed to be dust-free and therefore not corrected for dust.

In the following, uncertainties on ξion are at the 68% confidence intervals and were obtained from propagating the uncertainty in the Hα flux from its resulting distribution, as described in Sect. 3.1. The posterior distributions for β and MUV were obtained as described in Sect. 3.2.

3.4. Correlation analysis

For the purpose of studying the correlations between galaxy properties, we used the python package linmix6 to perform Bayesian linear regression, including intrinsic scatter and accounting for two-dimensional errors (Kelly 2007). We fit for log10[(1 − fesc)ξion]=α + βX + ϵ, where ϵ is the intrinsic scatter and is assumed to be normally distributed with a variance of . We recovered the best-fit trend line from the posteriors as well the 68% confidence interval on the parameters. We report the results in Table 1 and show the best-fit line in the figure plots.

Table 1.

Linear fitting parameters for trends with log10(1 − fesc)ξion.

4. Results

In this section we present our results. In Sect. 4.1 we study the trends between ξion, Hα EW, MUV, and the β slope, and in Sect. 4.2 we investigate whether our sample shows evidence for galaxies with very high ionizing photon escape fractions and/or very low metallicities.

4.1. Behavior of ξion

Figure 2 shows the distribution of (1 − fesc)ξion for our Lyα-selected and photometric samples. We find median values of log10ξion 25.39 ± 0.64 and 25.31 ± 0.43, respectively, and 25.33 ± 0.47 [Hz erg−1] for the complete data set. We find an intrinsic scatter of 0.42 dex, obtained by subtracting in quadrature the average uncertainty in log10ξion (= 0.21 dex) from the standard deviation of the observed distribution. The recovered intrinsic scatter is broader by ∼0.1 dex than that found by Bouwens et al. (2016) and Shivaei et al. (2018) in MUV ≲ −20 galaxies. The broad distribution of ξion is likely an outcome of the broad range of stellar populations in these galaxies, that is to say, due to a range of star formation histories (and thus ages) and stellar metallicities (see e.g., Shivaei et al. 2018).

thumbnail Fig. 2.

Distribution of (1 − fesc)ξion. The purple histogram includes galaxies with Lyα emission detection and the gray, galaxies without. Overall, Lyα-emitting galaxies show stronger ionizing photon production than galaxies with no Lyα emission, with median values 25.39 ± 0.64 and 25.31 ± 0.43, respectively. We show the median relation from the literature at z ∼ 2 − 5 as a dashed black line (e.g., Shivaei et al. 2018; Bouwens et al. 2016; Lam et al. 2019).

We performed a two-sample KS test to determine whether the Lyα-selected and photometric samples are drawn from the same distribution. We recover a p-value of 0.03, meaning it is likely that the underlying distributions are different; this is consistent with the results from Saldana-Lopez et al. (2022), where a statistically significant difference is found between the ξion distributions of Lyα emitters and non-Lyα emitters at z ∼ 3 − 5. Given that galaxies with strong Lyα emission also likely have high ionizing photon escape fractions (e.g., Verhamme et al. 2015; Dijkstra et al. 2016), it is likely that the intrinsic ionizing photon production efficiency of these galaxies is even higher than what we can infer based on Hα emission.

Figure 3 shows (1 − fesc)ξion versus UV magnitude and demonstrates the revolutionary capabilities of MUSE and JWST/NIRCam: we are able to spectroscopically confirm extremely UV-faint galaxies via their high Lyα EW, and we are able to infer Hα, and therefore ξion, from much fainter individual galaxies than was previously possible with Spitzer, where stacking was necessary at MUV ≳ −20 (e.g., Lam et al. 2019; Maseda et al. 2020). We reach ∼1 dex lower than any previous studies at similar redshifts and without needing to use stacking methods. We can reach individual detections of very faint galaxies, MUV < −17. We also find results consistent with those at z ∼ 2 (Shivaei et al. 2018) and at z ∼ 4 − 5 for > L* galaxies (Bouwens et al. 2016) and < L* galaxies (Lam et al. 2019, where a stacking analysis was used), as shown in Fig. 2. We note that our observations demonstrate the large scatter in (1 − fesc)ξion at fixed MUV, which was not possible to observe in previous analyses that used the stacking of Spitzer photometry for UV-faint galaxies.

thumbnail Fig. 3.

MUV vs. (1 − fesc)ξion. Lyα-detected galaxies are shown with purple stars and photometrically selected sample with no Lyα-detected with gray circles. We show data from Maseda et al. (2020), Harikane et al. (2018), Lam et al. (2019), and Bouwens et al. (2016) as colored boxes for comparison. We find evidence for an increase in log10(1 − fesc)ξion toward fainter UV magnitudes, with a slope of 0.03 ± 0.02, but only when considering the range where our sample is MUV complete (MUV < −18.1). We show literature constraints at similar redshifts as colored shapes (Bouwens et al. 2016; Harikane et al. 2018; Lam et al. 2019; Maseda et al. 2020), noting that all constraints fainter than MUV ≳ −20 were obtained by stacking IRAC photometry.

As described in Sect. 3.4, we performed a linear regression to assess correlations in our data. In contrast to Lam et al. (2019), we find significant evidence for a weak trend between ξion and MUV, where the highest ξion tends to come from the faintest galaxies. Since our sample is not MUV complete, we only study the correlation up to the peak of our MUV distribution (=−18.14) in Fig. 1. We find log10[(1−fesc)ξion] = (0.03 ± 0.02)(MUV + 20) + 25.36 ± 0.03, but with a large scatter (see Table 1).

Figure 4 shows that ξion follows a strong trend with Hα EW, as found in previous work (Harikane et al. 2018; Lam et al. 2019; Tang et al. 2019). Such works were limited to the highest Hα EW values, while we reach ∼0.75 dex lower due to the sensitivity of NIRCam. This trend is consistent with a picture where ξion is elevated in the youngest, most highly star-forming galaxies (e.g., Tang et al. 2019). We find log10[(1 − fesc)ξion]=(0.73 ± 0.04)(log10EWHα − 2.5)+25.15 ± 0.02. The measurement by Maseda et al. (2020), obtained from a stack of extremely UV-faint galaxies with high Lyα EWs, lies significantly above our sample and values from the rest of the literature, with higher (1 − fesc)ξion at fixed Hα EWs. As discussed by Maseda et al. (2020), this likely implies their sources have a much lower gas-phase metallicity than other samples.

thumbnail Fig. 4.

Comparison of the Hα EW with the ionizing photon production that does not escape the galaxy. Lyα-detected galaxies are shown as stars and photometrically selected galaxies with no Lyα as circles. As above, error bars are only shown for 30% of the sources for clarity. We color-code these two samples by UV β slope. In the top panel we show the distribution of Hα EWs for the same two samples compared to the values found by Tang et al. (2019). We add data from Harikane et al. (2018) and Lam et al. (2019), which are at the high end of our observed Hα EW distribution, for comparison. We see that a higher ξion correlates very strongly with a higher Hα EW. Galaxies with detected Lyα emission have an Hα EW distribution with higher values, median 732 ± 187 Å compared to 457 ± 161 with a Kolmogorov-Smirnov test p-value ≪0.01. The sources with the reddest UV slopes systematically lie below the best-fit relation at fixed Hα EW.

We also find that Lyα-selected galaxies have a higher Hα EW than the photometrically selected sample (median EW = 732 ± 187 Å compared to 457 ± 161 Å for the photometric sample). A two-sample KS test establishes that the EW distributions of the two samples are different (p-value ≪0.01). This is likely the primary driver of the increased ξion distribution for the Lyα-selected sample (Fig. 2).

At a fixed Hα EW, we see a clear tendency for galaxies with very blue β UV slopes to have elevated ξion (Fig. 4). This trend is also seen in the full sample (Fig. 5), where we find high ξion is weakly correlated to a blue β slope, but with a large scatter. We find log10(1 − fesc)ξion = ( − 0.20 ± 0.04)(β + 2)+25.41 ± 0.01 (see Table 1). Similar correlations have been seen at z ∼ 6 (e.g., Ning et al. 2023). Using a KS test, we find no significant difference in the β distributions for the Lyα and photometric samples. Our sample has a median β = −2.1.

thumbnail Fig. 5.

UV β slope vs. ξion(1 − fesc). Lyα-detected galaxies are shown in purple and the photometrically selected sample with no Lyα-detected in gray. As above, error bars are only shown for 30% of the sources for clarity. We add the stacked measurements from Lam et al. (2019) for comparison. We find a very weak trend of increasing ξion with decreasing β, with a linear slope of −0.10 ± 0.06.

4.2. A search for high-escape-fraction and extremely low-metallicity galaxies

As well as being a tracer of the ionizing photon production of galaxies, nebular emission lines are also sensitive to the escape fraction. Zackrisson et al. (2013) proposed that in galaxies with a very high ionizing escape fraction, one would expect a reduction in nebular emission line strength (Hβ EW ≲ 30 Å) and extremely blue UV slopes (β < −2.5) due to the lack of nebular continuum. Early JWST observations have discovered potentially very blue galaxies (Topping et al. 2022, though cf. Cullen et al. 2023) and galaxies with weak nebular line emission yet high sSFRs (via [OIII]+Hβ; Endsley et al. 2022), potentially indicating a population with a high ionizing escape fraction. However, the observation of low [OIII]+Hβ line strengths could also be caused by very low gas-phase metallicity (decreasing the strength of [OIII] emission) or a recent turnoff in star formation (which would also decrease all nebular emission lines). Given the redshift range of our sample, we can infer both Hα and [OIII]+Hβ line strengths for 241 galaxies, allowing us to test these scenarios and to search for galaxies with a high escape fraction. We obtained the [OIII]+Hβ nebular line fluxes as described in Sect. 3.1.

In Fig. 6 we show UV β slopes as a function of intrinsic Hα EW for our sample (where we correct for dust attenuation as described in Sect. 3.1). We compare our sample to the region proposed by Zackrisson et al. (2017) to have fesc > 0.5. While several sources fall into this region, and also have low [OIII]+Hβ EWs (≲100 Å), the uncertainties are too large to make them robust candidates. We discuss this further in Sect. 5.2.

thumbnail Fig. 6.

Comparison between the intrinsic (unattenuated) EW of Hα and the UV β slope, color-coded by [OIII]+Hβ EW. Lyα galaxies are shown with star-shaped markers, and the photometric sample as circles. Galaxies shown in gray do not have [OIII]+Hβ EW measurements. We show the region predicted by Zackrisson et al. (2017) to show fesc > 0.5. We rescale from Hβ EW to intrinsic Hα with a case B recombination scenario of factor 2.89, assuming a flat optical continuum in fλ, which we confirm from the SED fitting done in Sect. 3.1.

Figure 7 shows the Hα EW as a function of [OIII]+Hβ for our sample. We see the expected positive correlation between both nebular emission lines, as these lines are all generated by the effects of stellar ionizing radiation. We see a very large scatter (with a range of ∼1.5 dex) as expected due to variations in metallicity, temperature, and the ionization parameter, all of which affect the strength of individual [OIII] galaxies (e.g., Maiolino et al. 2008; Steidel et al. 2014; Sanders et al. 2021). We find log10EW(Hα) = 0.97 ± 0.06(log10EW([OIII]+Hβ)−2.5)+2.52 ± 0.03.

thumbnail Fig. 7.

Comparison between the EW of Hα and [OIII]+Hβ. Lyα-detected galaxies are shown as stars and the photometrically selected sample with no Lyα-detected as circles. In the top panel we show the distribution of [OIII]+Hβ EW for both of our samples. We find a very strong correlation between Hα EW and the [OIII]+Hβ EW, though with large scatter. The dashed lines are the correlation trends found for this work (red) and Tang et al. (2019, green). The Hα EW/[OIII]+Hβ EW is higher than the z ∼ 2 sample from Tang et al. (2019), which was selected to have a strong [OIII] EW, implying that we might be observing lower-metallicity galaxies.

Galaxies with detected Lyα emission tend to occupy the top right of the plot, with strong nebular emission lines, suggesting they are young star-forming galaxies with low metallicities and large ionization parameters that produce the copious amounts of ionizing photons needed to power these emission lines (see e.g., Yang et al. 2017; Du et al. 2020; Tang et al. 2021, for more detailed studies). We find the Lyα-selected galaxies have stronger [OIII]+Hβ EWs compared to the photometric population, following the trend with Hα EW in Fig. 4. However, as discussed by Tang et al. (2021), not all galaxies with strong nebular emission are detected in Lyα, indicating that Lyα transmission is reduced due to a high column density of neutral gas in these systems and/or inclination effects. We compare our data to a z ∼ 2 sample by Tang et al. (2019), which was selected based on strong [OIII] emission. We find a similar correlation, but overall our ratio of Hα EW/[OIII]+Hβ EW is higher by ∼0.1 dex. Given that the Tang et al. (2019) sample has a significantly subsolar gas-phase metallicity, Z < 0.3Z⊙ (Tang et al. 2021), the decrease we observe in [OIII] at fixed Hα EW would likely imply an overall lower metallicity due to a lower number of metal atoms in our sample.

5. Discussion

5.1. The profile of a strong ionizer

Thanks to the depth of JWST/NIRCam, we have been able to assess trends of ξion at z > 3 across the broadest range of galaxy properties to date. From these results, we corroborate previous work at lower redshifts and high luminosities and push the measurement of ξion to a large sample of individual UV-faint galaxies for the first time.

We find that galaxies with strong ionizing photon emission tend to have high Hα EWs, low UV luminosities, blue UV β slopes, and Lyα emission – all implying that these galaxies are young and likely have a low dust content, low metallicity, and a high O/B star population that is capable of producing hard ionizing photons (e.g., Tang et al. 2019; Boyett et al. 2022b). This picture of the integrated emission from galaxies is complemented by high spatial resolution observations of highly magnified arcs with JWST. They have revealed extremely young star clusters (≲10 Myr) with [OIII]+Hβ EW > 1000 Å. They dominate the ionizing photon production in their galaxy (Vanzella et al. 2022, 2023), indicating that there can be large variations in ξion in individual galaxies if they contain multiple stellar populations, but also that the variation is primarily driven by the age of the stellar populations. We also find that, overall, our Lyα galaxy sample has higher ξion than the photometrically selected one; the primary reason for this difference is that the former has higher Hα EWs (Fig. 4). The enhanced prevalence of Lyα emission in strong Hα emitters is likely a combination of an increased production of Lyα photons due to the young stellar population implied by the strong Hα and (potentially) an increase in the Lyα escape fraction in the ISM (Tang et al. 2021; Naidu et al. 2022). In these rapidly star-forming galaxies, the hard ionizing radiation may be ionizing the ISM and/or feedback may disrupt the ISM gas, leading to a reduced HI column density and dust cover. We note that the galaxies with the highest (1 − fesc)ξion are not necessarily all Lyα emitters, likely due to variance in the geometry and column density of neutral gas and dust in these sources. Ning et al. (2023) show this same correlation between ξion and Lyα for a broad range of luminosities and EWs.

5.2. The ionizing photon escape fraction

In Sect. 4.2 we explore whether our sample shows signs of high ionizing photon escape fraction, fesc, using the low Hβ EW–blue UV β slope region defined by Zackrisson et al. (2017) for fesc > 0.5. While several sources fell into this region, with both low Hα EW and [OIII]+Hβ EW (≲100 Å), the uncertainties on the line flux measurements are too large for them to be robust candidates. More precise emission line measurements with JWST spectroscopy will be vital for identifying such candidates and their relative abundance in the galaxy population.

The lack of high fesc candidates amongst the Lyα-selected galaxies is also surprising. As the same conditions (a low neutral gas covering fraction) facilitate both Lyα escape and Lyman continuum escape, a correlation between the two is expected (e.g., Verhamme et al. 2015; Dijkstra et al. 2016; Reddy et al. 2016).

As discussed by Topping et al. (2022), however, it is possible for galaxies with high fesc but very young ages to still have high nebular emission due to high ionizing photon production. It is likely that the criteria proposed by Zackrisson et al. (2017) can only find high fesc systems within the bounds of the assumptions made for their model, such as galaxy star formation histories, ages, metallicities, and dust levels, but also the stellar models used. Our results suggest the low luminosity galaxies with high sSFRs but low [OIII]+Hβ EWs observed by Endsley et al. (2022) may be more likely due to variations in metallicity than due to the high fesc.

6. Conclusions

We have inferred the hydrogen ionizing photon production rate, modulo the escape fraction, in the largest sample of individual sub-L*z > 3 galaxies to date, spanning −23 ≲ MUV ≲ −15.5 with a median MUV = −18.1, thanks to deep JWST/NIRCam imaging. This has enabled us to track the demographics of the ionizing population. Our conclusions are as follows:

  1. The median log10(1 − fesc)ξion of our sample is 25.33 ± 0.47 with an intrinsic scatter of 0.42 dex. The inferred ξion distribution of our sample has values in a range of ∼1.5 dex, implying a wide range of galaxy properties and ages.

  2. We find significant trends of increasing (1 − fesc)ξion with increasing Hα EW, decreasing UV luminosity, and decreasing UV slope, all suggesting that the galaxies most efficient at producing ionizing photons are young, highly star-forming, and normally expected to have low metallicities and be dust-poor.

  3. We find galaxies selected with strong Lyα emission to have higher ξion than photometrically selected galaxies, with median log10(1 − fesc)ξion values of 25.39 ± 0.64 and 25.31 ± 0.43, respectively. We find the Lyα-detected galaxies have an elevated Hα EW distribution, and thus the increased ξion is likely driven by the selection based on Lyα selecting a younger population. As strong Lyα emitters also likely have high ionizing photon escape fractions, this implies the intrinsic production rate of ionizing photons in these galaxies could be significantly higher than what we can infer from Hα luminosities.

  4. We examine our sample for signs of very high fesc by comparing the inferred strengths of nebular emission lines ([OIII]+Hβ and Hα) and the strength of the nebular continuum via the UV β slope. We find no significant evidence for sources with high-escape-fraction galaxies with low nebular emission line strengths and very blue UV β slopes. The reduced strength of the [OIII]+Hβ EWs in our z > 3 sample compared to a sample at z ∼ 2 from Tang et al. (2019) implies our sample likely has a lower gas-phase metallicity and/or ionization parameter.

We have demonstrated the power of JWST/NIRCam photometry to more precisely constrain the rest-frame optical emission of UV-faint high redshift galaxies than previously possible with Spitzer/IRAC. These observations allow us to constrain the production rate of ionizing photons from early galaxies, corroborating the picture obtained from previous stacking analyses, that ξion is elevated in young, highly star-forming galaxies but that there is a broad distribution of ξion, likely driven by variations in galaxy properties and ages.

With JWST spectroscopy it is becoming possible to obtain direct measurements of optical emission lines in large samples (e.g., Sun et al. 2022; Williams et al. 2022; Matthee et al. 2022). Deriving a census of the ionizing photon production rate across the full galaxy population will be necessary to fully understand reionization. Here we have shown that ξion is elevated in UV-faint galaxies with strong nebular emission lines, likely due to young ages. While a thorough analysis of the implications of our results for reionization is beyond the scope of this work, it becomes more prominent at high redshift (e.g., Boyett et al. 2022b; Endsley et al. 2022), implying that it would be possible to complete reionization with modest fesc. Considering the full distributions of ξion and fesc across galaxy properties will be required to assess the primary drivers of reionization.

Acknowledgments

We thank the co-PIs Ivo Labbé and Rachel Bezanson for the conception and public availability of the UNCOVER JWST Program (GO-2561), which made much of this work possible. We thank Mengtao Tang for sharing data the emission line catalog from Tang et al. (2019). This work is based on observations collected at the European Southern Observatory under ESO programmes 109.24EZ.001 and 094.A-0115. This work is based on NASA/ESA HST and JWST data which were obtained from the Mikulski Archive for Space Telescopes at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST. The HST observations are associated with programs GO/DD-17231, GO/DD-13495, GO-15117, GO-11689, GO-11386, GO-13389, GO-15940 and SNAP-16729. The JWST observations are associated with programs JWST-ERS-1324 and GO-2561. C.M. and G.P. acknowledge support by the VILLUM FONDEN under grant 37459. The Cosmic Dawn Center (DAWN) is funded by the Danish National Research Foundation under grant DNRF140. We acknowledge financial support from NASA through grant JWST-ERS-1324. We acknowledge support from the INAF Large Grant 2022 “Extragalactic Surveys with JWST” (PI Pentericci), and support through grants PRIN-MIUR 2017WSCC32, 2020SKSTHZ. This research is supported in part by the Australian Research Council Centre of Excellence for All Sky Astrophysics in 3 Dimensions (ASTRO 3D), through project number CE170100013.

References

  1. Anders, P., & Fritze-v Alvensleben, U., 2003, A&A, 401, 1063 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  2. Barbary, K. 2018, Astrophysics Source Code Library [record ascl:1811.004] [Google Scholar]
  3. Becker, G. D., D’Aloisio, A., Christenson, H. M., et al. 2021, MNRAS, 508, 1853 [NASA ADS] [CrossRef] [Google Scholar]
  4. Begley, R., Cullen, F., McLure, R. J., et al. 2022, MNRAS, 513, 3510 [NASA ADS] [CrossRef] [Google Scholar]
  5. Bergamini, P., Acebron, A., Grillo, C., et al. 2023, A&A, 670, A60 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  6. Bolan, P., Lemaux, B. C., Mason, C., et al. 2022, MNRAS, 517, 3263 [Google Scholar]
  7. Bouwens, R. J., Illingworth, G. D., Oesch, P. A., et al. 2014, ApJ, 793, 115 [Google Scholar]
  8. Bouwens, R. J., Smit, R., Labbé, I., et al. 2016, ApJ, 831, 176 [Google Scholar]
  9. Boyett, K., Mascia, S., Pentericci, L., et al. 2022a, ApJ, 940, L52 [NASA ADS] [CrossRef] [Google Scholar]
  10. Boyett, K. N. K., Stark, D. P., Bunker, A. J., Tang, M., & Maseda, M. V. 2022b, MNRAS, 513, 4451 [NASA ADS] [CrossRef] [Google Scholar]
  11. Brammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, ApJ, 686, 1503 [Google Scholar]
  12. Brammer, G., Strait, V., Matharu, J., & Momcheva, I. 2022, https://zenodo.org/record/6672538 [Google Scholar]
  13. Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000 [NASA ADS] [CrossRef] [Google Scholar]
  14. Caminha, G. B., Grillo, C., Rosati, P., et al. 2017, A&A, 600, A90 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  15. Caminha, G. B., Rosati, P., Grillo, C., et al. 2019, A&A, 632, A36 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  16. Carnall, A. C., McLure, R. J., Dunlop, J. S., & Davé, R. 2018, MNRAS, 480, 4379 [Google Scholar]
  17. Chabrier, G. 2003, PASP, 115, 763 [Google Scholar]
  18. Chevallard, J., Charlot, S., Senchyna, P., et al. 2018, MNRAS, 479, 3264 [Google Scholar]
  19. Cullen, F., McLure, R. J., McLeod, D. J., et al. 2023, MNRAS, 520, 14 [NASA ADS] [CrossRef] [Google Scholar]
  20. Davies, F. B., Hennawi, J. F., Bañados, E., et al. 2018, ApJ, 864, 142 [NASA ADS] [CrossRef] [Google Scholar]
  21. Dijkstra, M., Gronke, M., & Venkatesan, A. 2016, ApJ, 828, 71 [Google Scholar]
  22. Du, X., Shapley, A. E., Tang, M., et al. 2020, ApJ, 890, 65 [NASA ADS] [CrossRef] [Google Scholar]
  23. Emami, N., Siana, B., Alavi, A., et al. 2020, ApJ, 895, 116 [NASA ADS] [CrossRef] [Google Scholar]
  24. Endsley, R., Stark, D. P., Whitler, L., et al. 2022, ArXiv e-prints [arXiv:2208.14999] [Google Scholar]
  25. Fan, X., Strauss, M. A., Becker, R. H., et al. 2006, AJ, 132, 117 [NASA ADS] [CrossRef] [Google Scholar]
  26. Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125, 306 [Google Scholar]
  27. Harikane, Y., Ouchi, M., Shibuya, T., et al. 2018, ApJ, 859, 84 [NASA ADS] [CrossRef] [Google Scholar]
  28. Hashimoto, T., Laporte, N., Mawatari, K., et al. 2018, Nature, 557, 392 [NASA ADS] [CrossRef] [Google Scholar]
  29. Inoue, A. K., Shimizu, I., Iwata, I., & Tanaka, M. 2014, MNRAS, 442, 1805 [NASA ADS] [CrossRef] [Google Scholar]
  30. Kelly, B. C. 2007, ApJ, 665, 1489 [Google Scholar]
  31. Lam, D., Bouwens, R. J., Labbé, I., et al. 2019, A&A, 627, A164 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  32. Leitherer, C., & Heckman, T. M. 1995, ApJS, 96, 9 [NASA ADS] [CrossRef] [Google Scholar]
  33. Lotz, J. M., Koekemoer, A., Coe, D., et al. 2017, ApJ, 837, 97 [Google Scholar]
  34. Ma, X., Quataert, E., Wetzel, A., et al. 2020, MNRAS, 498, 2001 [Google Scholar]
  35. Madau, P., Haardt, F., & Rees, M. J. 1999, ApJ, 514, 648 [CrossRef] [Google Scholar]
  36. Mahler, G., Richard, J., Clément, B., et al. 2018, MNRAS, 473, 663 [Google Scholar]
  37. Maiolino, R., Nagao, T., Grazian, A., et al. 2008, A&A, 488, 463 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  38. Mármol-Queraltó, E., McLure, R. J., Cullen, F., et al. 2016, MNRAS, 460, 3587 [Google Scholar]
  39. Marques-Chaves, R., Schaerer, D., Á lvarez-Márquez, J., et al. 2022, MNRAS, 517, 2972 [CrossRef] [Google Scholar]
  40. Maseda, M. V., Bacon, R., Lam, D., et al. 2020, MNRAS, 493, 5120 [Google Scholar]
  41. Mason, C. A., Treu, T., Dijkstra, M., et al. 2018, ApJ, 856, 2 [Google Scholar]
  42. Mason, C. A., Fontana, A., Treu, T., et al. 2019, MNRAS, 485, 3947 [NASA ADS] [CrossRef] [Google Scholar]
  43. Matthee, J., Mackenzie, R., Simcoe, R. A., et al. 2022, ApJ, submitted [arXiv:2211.08255] [Google Scholar]
  44. McGreer, I. D., Mesinger, A., & D’Odorico, V. 2015, MNRAS, 447, 499 [NASA ADS] [CrossRef] [Google Scholar]
  45. McLeod, D. J., McLure, R. J., Dunlop, J. S., et al. 2021, MNRAS, 503, 4413 [NASA ADS] [CrossRef] [Google Scholar]
  46. Naidu, R. P., Tacchella, S., Mason, C. A., et al. 2020, ApJ, 892, 109 [NASA ADS] [CrossRef] [Google Scholar]
  47. Naidu, R. P., Matthee, J., Oesch, P. A., et al. 2022, MNRAS, 510, 4582 [CrossRef] [Google Scholar]
  48. Ning, Y., Cai, Z., Jiang, L., et al. 2023, ApJ, 944, L1 [NASA ADS] [CrossRef] [Google Scholar]
  49. Oesch, P. A., Bouwens, R. J., Illingworth, G. D., Labbe, I., & Stefanon, M. 2018, ApJ, 855, 105 [NASA ADS] [CrossRef] [Google Scholar]
  50. Pahl, A. J., Shapley, A., Steidel, C. C., et al. 2023, MNRAS, 521, 3247 [NASA ADS] [CrossRef] [Google Scholar]
  51. Planck Collaboration VI. 2020, A&A, 641, A6 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  52. Prevot, M. L., Lequeux, J., Maurice, E., Prevot, L., & Rocca-Volmerange, B. 1984, A&A, 132, 389 [Google Scholar]
  53. Qin, Y., Mesinger, A., Bosman, S. E. I., & Viel, M. 2021, MNRAS, 506, 2390 [NASA ADS] [CrossRef] [Google Scholar]
  54. Reddy, N. A., Steidel, C. C., Pettini, M., Bogosavljević, M., & Shapley, A. E. 2016, ApJ, 828, 108 [Google Scholar]
  55. Richard, J., Claeyssens, A., Lagattuta, D., et al. 2021, A&A, 646, A83 [EDP Sciences] [Google Scholar]
  56. Rigby, J., Perrin, M., McElwain, M., et al. 2023, PASP, 135, 048001 [NASA ADS] [CrossRef] [Google Scholar]
  57. Robertson, B. E., Ellis, R. S., Dunlop, J. S., McLure, R. J., & Stark, D. P. 2010, Nature, 468, 49 [Google Scholar]
  58. Rogers, A. B., McLure, R. J., & Dunlop, J. S. 2013, MNRAS, 429, 2456 [Google Scholar]
  59. Saldana-Lopez, A., Schaerer, D., Chisholm, J., et al. 2022, MNRAS, submitted [arXiv:2211.01351] [Google Scholar]
  60. Sanders, R. L., Shapley, A. E., Jones, T., et al. 2021, ApJ, 914, 19 [CrossRef] [Google Scholar]
  61. Schaerer, D., & de Barros, S. 2009, A&A, 502, 423 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  62. Sharma, M., Theuns, T., Frenk, C., et al. 2017, MNRAS, 468, 2176 [NASA ADS] [CrossRef] [Google Scholar]
  63. Shim, H., Chary, R.-R., Dickinson, M., et al. 2011, ApJ, 738, 69 [NASA ADS] [CrossRef] [Google Scholar]
  64. Shivaei, I., Reddy, N. A., Siana, B., et al. 2018, ApJ, 855, 42 [Google Scholar]
  65. Smit, R., Bouwens, R. J., Franx, M., et al. 2015, ApJ, 801, 122 [Google Scholar]
  66. Stanway, E. R., Eldridge, J. J., & Becker, G. D. 2016, MNRAS, 456, 485 [NASA ADS] [CrossRef] [Google Scholar]
  67. Stark, D. P., Ellis, R. S., Chiu, K., Ouchi, M., & Bunker, A. 2010, MNRAS, 408, 1628 [Google Scholar]
  68. Stark, D. P., Schenker, M. A., Ellis, R., et al. 2013, ApJ, 763, 129 [Google Scholar]
  69. Stefanon, M., Bouwens, R. J., Illingworth, G. D., et al. 2022, ApJ, 935, 94 [NASA ADS] [CrossRef] [Google Scholar]
  70. Steidel, C. C., Rudie, G. C., Strom, A. L., et al. 2014, ApJ, 795, 165 [Google Scholar]
  71. Steidel, C. C., Bogosavljevic, M., Shapley, A. E., et al. 2018, ApJ, 869, 123 [NASA ADS] [CrossRef] [Google Scholar]
  72. Steinhardt, C. L., Jauzac, M., Acebron, A., et al. 2020, ApJS, 247, 64 [Google Scholar]
  73. Strait, V., Bradač, M., Coe, D., et al. 2021, ApJ, 910, 135 [NASA ADS] [CrossRef] [Google Scholar]
  74. Sun, F., Egami, E., Pirzkal, N., et al. 2022, ApJ, submitted [arXiv:2209.03374] [Google Scholar]
  75. Tang, M., Stark, D. P., Chevallard, J., & Charlot, S. 2019, MNRAS, 489, 2572 [NASA ADS] [CrossRef] [Google Scholar]
  76. Tang, M., Stark, D. P., Chevallard, J., et al. 2021, MNRAS, 503, 4105 [NASA ADS] [CrossRef] [Google Scholar]
  77. Topping, M. W., Stark, D. P., Endsley, R., et al. 2022, ApJ, 941, 153 [NASA ADS] [CrossRef] [Google Scholar]
  78. Trebitsch, M., Blaizot, J., Rosdahl, J., Devriendt, J., & Slyz, A. 2017, MNRAS, 478, 5607 [Google Scholar]
  79. Treu, T., Roberts-Borsani, G., Bradac, M., et al. 2022, ApJ, 935, 110 [NASA ADS] [CrossRef] [Google Scholar]
  80. Vanzella, E., Nonino, M., Cupani, G., et al. 2018, MNRAS, 476, L15 [Google Scholar]
  81. Vanzella, E., Castellano, M., Bergamini, P., et al. 2022, ApJ, 940, L53 [NASA ADS] [CrossRef] [Google Scholar]
  82. Vanzella, E., Claeyssens, A., Welch, B., et al. 2023, ApJ, 945, 53 [NASA ADS] [CrossRef] [Google Scholar]
  83. Verhamme, A., Orlitová, I., Schaerer, D., & Hayes, M. 2015, A&A, 578, A7 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  84. Williams, H., Kelly, P. L., Chen, W., et al. 2022, ArXiv e-prints [arXiv:2210.15699] [Google Scholar]
  85. Yang, H., Malhotra, S., Rhoads, J. E., & Wang, J. 2017, ApJ, 847, 38 [NASA ADS] [CrossRef] [Google Scholar]
  86. Zackrisson, E., Inoue, A. K., & Jensen, H. 2013, ApJ, 777, 39 [NASA ADS] [CrossRef] [Google Scholar]
  87. Zackrisson, E., Binggeli, C., Finlator, K., et al. 2017, ApJ, 836, 78 [NASA ADS] [CrossRef] [Google Scholar]

All Tables

Table 1.

Linear fitting parameters for trends with log10(1 − fesc)ξion.

In the text

All Figures

thumbnail Fig. 1.

Galaxies studied in this work: Lyα-detected galaxies (in purple) and galaxies photometrically selected (with no Lyα-detected, in gray). Top: Distribution of redshifts for the spectroscopic and photometric samples. We show the spectroscopic redshift, where available, or the peak photometric redshift. Bottom: UV magnitude distribution for our sample. We find a median value of −18.14 ± 1.58, with no statistically significant difference between the two samples.
In the text
thumbnail Fig. 2.

Distribution of (1 − fesc)ξion. The purple histogram includes galaxies with Lyα emission detection and the gray, galaxies without. Overall, Lyα-emitting galaxies show stronger ionizing photon production than galaxies with no Lyα emission, with median values 25.39 ± 0.64 and 25.31 ± 0.43, respectively. We show the median relation from the literature at z ∼ 2 − 5 as a dashed black line (e.g., Shivaei et al. 2018; Bouwens et al. 2016; Lam et al. 2019).
In the text
thumbnail Fig. 3.

MUV vs. (1 − fesc)ξion. Lyα-detected galaxies are shown with purple stars and photometrically selected sample with no Lyα-detected with gray circles. We show data from Maseda et al. (2020), Harikane et al. (2018), Lam et al. (2019), and Bouwens et al. (2016) as colored boxes for comparison. We find evidence for an increase in log10(1 − fesc)ξion toward fainter UV magnitudes, with a slope of 0.03 ± 0.02, but only when considering the range where our sample is MUV complete (MUV < −18.1). We show literature constraints at similar redshifts as colored shapes (Bouwens et al. 2016; Harikane et al. 2018; Lam et al. 2019; Maseda et al. 2020), noting that all constraints fainter than MUV ≳ −20 were obtained by stacking IRAC photometry.
In the text
thumbnail Fig. 4.

Comparison of the Hα EW with the ionizing photon production that does not escape the galaxy. Lyα-detected galaxies are shown as stars and photometrically selected galaxies with no Lyα as circles. As above, error bars are only shown for 30% of the sources for clarity. We color-code these two samples by UV β slope. In the top panel we show the distribution of Hα EWs for the same two samples compared to the values found by Tang et al. (2019). We add data from Harikane et al. (2018) and Lam et al. (2019), which are at the high end of our observed Hα EW distribution, for comparison. We see that a higher ξion correlates very strongly with a higher Hα EW. Galaxies with detected Lyα emission have an Hα EW distribution with higher values, median 732 ± 187 Å compared to 457 ± 161 with a Kolmogorov-Smirnov test p-value ≪0.01. The sources with the reddest UV slopes systematically lie below the best-fit relation at fixed Hα EW.
In the text
thumbnail Fig. 5.

UV β slope vs. ξion(1 − fesc). Lyα-detected galaxies are shown in purple and the photometrically selected sample with no Lyα-detected in gray. As above, error bars are only shown for 30% of the sources for clarity. We add the stacked measurements from Lam et al. (2019) for comparison. We find a very weak trend of increasing ξion with decreasing β, with a linear slope of −0.10 ± 0.06.
In the text
thumbnail Fig. 6.

Comparison between the intrinsic (unattenuated) EW of Hα and the UV β slope, color-coded by [OIII]+Hβ EW. Lyα galaxies are shown with star-shaped markers, and the photometric sample as circles. Galaxies shown in gray do not have [OIII]+Hβ EW measurements. We show the region predicted by Zackrisson et al. (2017) to show fesc > 0.5. We rescale from Hβ EW to intrinsic Hα with a case B recombination scenario of factor 2.89, assuming a flat optical continuum in fλ, which we confirm from the SED fitting done in Sect. 3.1.
In the text
thumbnail Fig. 7.

Comparison between the EW of Hα and [OIII]+Hβ. Lyα-detected galaxies are shown as stars and the photometrically selected sample with no Lyα-detected as circles. In the top panel we show the distribution of [OIII]+Hβ EW for both of our samples. We find a very strong correlation between Hα EW and the [OIII]+Hβ EW, though with large scatter. The dashed lines are the correlation trends found for this work (red) and Tang et al. (2019, green). The Hα EW/[OIII]+Hβ EW is higher than the z ∼ 2 sample from Tang et al. (2019), which was selected to have a strong [OIII] EW, implying that we might be observing lower-metallicity galaxies.
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.