Open Access
Issue
A&A
Volume 676, August 2023
Article Number A53
Number of page(s) 30
Section Extragalactic astronomy
DOI https://doi.org/10.1051/0004-6361/202346232
Published online 07 August 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

Understanding the last phase transition of the Universe, known as the epoch of reionization (EoR), is one of the longstanding goals of extragalactic astronomy. The massive stars residing in the high-redshift(z) star-forming galaxies (SFGs) are thought to be the dominant ionizing agents that drive reionization (e.g., Robertson et al. 2015; Finkelstein et al. 2019). However, some studies suggest that low-luminosity active galactic nuclei (AGNs) may play a significant role in reionization (e.g., Madau & Haardt 2015; Dayal et al. 2020). In recent high-resolution cosmological simulations of Lyman-continuum (LyC, < 912 Å) emitting sources in the EoR, z > 5 simulated galaxies are studied with a detailed treatment of the multiphase interstellar medium (ISM) and stellar feedback. For example, within the project called Feedback in Realistic Environments (FIRE; Hopkins et al. 2018), it has been found that a majority of LyC escape comes from the very young (< 10 Myr) kiloparsec-scale star-forming regions of a galaxy, and the contribution from an older (> 10 Myr) stellar population is negligible. These ISM structures clear out the neutral gas column to allow the escape of ionizing photons at very low escape fractions (fesc) that can reach 10–20% only for a short time since the ionizing photon production from massive stars begins to decline after 3 Myr (e.g., Ma et al. 2015, 2020). At the same time, most of the ionizing photons are consumed by surrounding neutral gas clouds (e.g., Ma et al. 2015; Kakiichi & Gronke 2021). Moreover, the reionization phase is estimated to be a fast epoch where the Universe evolves from 90% neutral at z ∼ 8.22 to 10% neutral at z ∼ 6.25 in ∼300 Myr by massive starburst galaxies (Naidu et al. 2020). Understanding the mechanisms that facilitate the escape remains crucial.

On the other hand, deep rest-frame ultraviolet (UV) spectra (λ  ∼  1200 − 2000 Å) of several high-z galaxies (z  ∼  5 − 7; e.g., Stark et al. 2015; Mainali et al. 2018; Hutchison et al. 2019; Castellano et al. 2022; Tang et al. 2023) reveal prominent high-ionization nebular emission lines, such as He IIλ1640, O III]λλ1661,66, C III]λλ1907,09 (C III] hereafter), and C IVλλ1548,51 (C IV hereafter). To unveil their nature, the rest-frame UV is crucial, and analogs to these reionization sources at lower redshifts are the key to this. The rest-UV spectra can foster our understanding of SFGs in terms of the stellar populations hosting massive stars and their impact on the physical conditions of their ISM, their chemical evolution, feedback processes, and cosmic reionization. This is important because James Webb Space Telescope (JWST) instruments such as NIRSpec (Jakobsen et al. 2022) will cover blueward of ∼4500 Å only in objects z < 10. Understanding the ISM properties from UV spectra will therefore be essential for characterizing and interpreting the spectroscopic observations of high-z systems. Large and deep surveys such as the Lyman Break Galaxy (LBG) survey of z ∼ 3 (Steidel et al. 2003), the VIMOS Ultra Deep Survey (VUDS; Le Fèvre et al. 2015; Tasca et al. 2017), the MUSE Hubble Ultra Deep Survey (Bacon et al. 2017), and VANDELS (McLure et al. 2018; Pentericci et al. 2018; Garilli et al. 2021) have targeted SFGs at mainly z ∼ 2 − 4 to study the physical properties of SFGs with intense rest-UV emission lines. Significant improvements have also been made in the local Universe with relatively large samples of metal-poor SFGs in the COS Legacy Spectroscopic SurveY (CLASSY; Berg et al. 2022). From an observational point of view, the escape fraction at z ∼ 3 depends on galaxy properties such as the equivalent width (EW) of Lyα, the stellar mass, and the color excess by dust extinction (e.g., Steidel et al. 2018; Saxena et al. 2022a; Begley et al. 2022; Pahl et al. 2023). Moreover, the profile shape of Lyα is an essential predictor of the LyC escape fraction as this gives information about the covering fraction of neutral gas at the systemic velocity (Izotov et al. 2020; Flury et al. 2022a). These indirect probes are essential because ionizing fluxes cannot be measured at z > 6 because the opacity of the intergalactic medium (IGM) is too high (Inoue et al. 2014), and the measurement of the escape fraction of ionizing photons will heavily rely on indirect probes at high redshift (e.g., Xu et al. 2022a; Naidu et al. 2022) or on work on analogs at lower redshift (e.g., Schaerer et al. 2016; Flury et al. 2022b).

Galaxy-scale outflows are thought to be one of the processes that aid the escape of ionizing photons (e.g., Chisholm et al. 2017; Kim et al. 2020; Hogarth et al. 2020, and references therein). These outflows are expected to remove surrounding neutral gas and thus clear a pathway on which ionizing photons can escape efficiently. However, it is often challenging to reconcile the timescales of these strong galactic outflows with the timescale of the production and escape of ionizing photons in a galaxy. Outflowing gas of SFGs at z ∼ 2 has been reported both from optical emission lines such as [OIII]λ5007 (hereafter [OIII]) or Hα (e.g., Förster Schreiber et al. 2019; Übler et al. 2022) and from UV absorption lines (e.g., Steidel et al. 2010; Jones et al. 2018; Calabrò et al. 2022). Optical rest-frame emission lines are able to trace denser outflowing gas, providing an instantaneous snapshot of the ongoing ejective feedback. They are therefore in principle less contaminated by tenuous gas around galaxies (Concas et al. 2022). These emission lines are typically modeled with one narrow and one broad Gaussian component. The latter is considered the outflow component. The properties of the outflows are probed to depend on the properties of the host galaxy. For example, it is observed that the velocity of outflows increases as a function of star formation rate (SFR) and galaxy stellar mass (M; e.g., Weiner et al. 2009), which is also seen in simulations (e.g., Muratov et al. 2015). Freeman et al. (2019) reported based on the MOSFIRE Deep Evolution Field survey (MOSDEF; Kriek et al. 2015) that the broad-to-narrow flux ratio increases with stellar mass. Understanding how highly ionized gas outflows affect the properties of host galaxies gives insights into the vital role of the stellar feedback from young stars in the escape of ionizing photons.

Finally, analytic models explain that scaling relations such as the mass-metallicity relation (MZR; Tremonti et al. 2004) and the fundamental metallicity relation (FMR; Mannucci et al. 2010; Curti et al. 2020; Sanders et al. 2021) arise from an interplay between star formation, the infall of metal-poor IGM gas, and the ejection of metal-rich ISM gas. The FMR is a signature of the smooth long-lasting equilibrium between gas flows and secular evolution (e.g., Bouché et al. 2010; Lilly et al. 2013). Based on the FMR, less massive galaxies (M < 1011M) with a lower SFR produce less heavy elements that are more efficiently ejected because their potential wells are shallow. As result, for a given M, the gas metallicity decreases with SFR (Dayal et al. 2013). On the other hand, the lack of any correlation between the ratio of a secondary element, such as carbon (C) or nitrogen (N), to a primary element, such as oxygen (O), with SFR (e.g., Pérez-Montero & Contini 2009) could indicate that stellar winds eject metals, leaving their proportions unaffected in the remaining gas. The role of outflows in shaping the chemical properties of galaxies is another critical feature that needs to be to understood in galaxy evolution.

In this paper, we present an analysis of the physical properties of a sample of 35 SFGs at z ∼ 3 in which we search for the effects of the stellar feedback on the chemical properties of the SFGs based on the kinematics of the ionized gas. In Sect. 2 we present the sample selection and the optical and near-infrared (NIR) spectroscopy. In Sect. 3 we present the physical properties of the sample and the narrow+broad Gaussian modeling of the [OIII]λλ4959,5007 profiles. Then, we present the results of our study in Sect. 4. We describe the main source of ionization of our sample, their chemical abundances (the gas-phase metallicity (O/H) and the carbon-to-oxygen (C/O) abundance), and the properties of ionized gas kinematics based on the outflow component. Then, we discuss our results in Sect. 5 based on how outflows and star formation histories (SFHs) may affect the chemical properties of their host galaxy. Finally, we present our conclusions in Sect. 6.

Throughout this paper, we adopt a Λ-dominated flat universe with ΩΛ = 0.7, ΩM = 0.3, and H0 = 70 km s−1 Mpc−1. All magnitudes are quoted in the AB system. Equivalent widths are quoted in the rest frame and are positive for emission lines. We adopt a Chabrier (2003) initial mass function (IMF). We consider log(O/H) = 8.69 and log(C/O) = −0.26 (Asplund et al. 2009).

2. Data and sample selection

Our strategy is based on the combined analysis of the rest-UV+optical spectra of a diverse sample of SFGs at z = 2 − 4. Our main selection criterion (the detection of CIII]) depends on the rest-UV emission lines that are obtained by optical spectrographs. Our parent sample is a combination of two large surveys carried out with the VIMOS spectrograph (Le Fèvre et al. 2003): the VUDS (Le Fèvre et al. 2015; Tasca et al. 2017) and the VANDELS (McLure et al. 2018; Pentericci et al. 2018) surveys. From this parent sample, we selected galaxies with NIR spectroscopy targeting their rest-optical emission lines.

Our final sample contains 17 VANDELS galaxies (hereafter, C3-VANDELS) and 18 VUDS galaxies (hereafter, C3-VUDS). We describe the selection of the final sample and the NIR observations in the next sections.

2.1. Sample selection from rest-frame UV spectroscopy

2.1.1. VANDELS parent sample

We used spectroscopic data from VANDELS (McLure et al. 2018; Pentericci et al. 2018), a deep VIMOS survey of the CANDELS fields, which is a completed ESO public spectroscopic survey carried out using the Very Large Telescope (VLT). VANDELS covers two well-studied extragalactic fields: the UKIDSS Ultra Deep Survey (UDS) and the Chandra Deep Field South (CDFS). The final VANDELS data release, DR4, contains spectra of ∼2100 galaxies in the redshift range 1.0 < z < 7.0, with on-source integration times ranging from 20 to 80 h, where > 70% of the targets have at least 40 h of integration time (Garilli et al. 2021). The spectral resolution of VANDELS spectra is R ∼ 600 in the wavelength range 480 < λobs < 980 nm. At the redshift range 2 < z < 4 of our interest for the CIII] selection, 887 galaxies were observed with reliable redshift. From them, we selected a parent sample of 280 SFGs that show a signal-to-noise ratio (S/N) > 3 in CIII] with EW(CIII])  >  0. Most of them (74%) show EWs  <  5 Å, and only ∼5% show EWs  >  10 Å. To select this parent sample, we followed the method presented in Llerena et al. (2022) for the VANDELS DR3.

2.1.2. VUDS parent sample

We also used observations from the VUDS survey (Le Fèvre et al. 2015; Tasca et al. 2017), a massive 640-h (∼80 nights) spectroscopic campaign reaching extreme depths (i′< 25 mag) over three well-studied extragalactic fields: COSMOS, ECDFS, and VVDS-02h. Spectroscopic observations consisted of approximately 50 400 s of integration across the wavelength range 365 < λobs < 935 nm at a spectral resolution of R = 230. At 2 < z < 3.8, where the instrumental setup allows us to follow the CIII] line reliably, 3899 SFGs were observed. Our parent sample is described in Le Fèvre et al. (2019) and is selected with 2 < z < 3.8 and S/N > 3 in CIII]. The authors selected 1763 SFGs with EW(CIII])  >  0, most of which (75%) show EWs  <  5 Å, and only ∼8% show EWs  >  10 Å.

2.2. Rest-frame optical spectroscopy

We describe the NIR spectroscopy that we used to select the final sample to be analyzed in this paper. In summary, the rest-optical spectra for VANDELS targets were obtained from the NIRVANDELS survey (Cullen et al. 2021) using Keck/MOSFIRE (McLean et al. 2012). For the VUDS targets, we obtained the rest-optical spectra from different instruments. First, from the public MOSDEF survey (Kriek et al. 2015) using Keck/MOSFIRE. We also considered NIR spectroscopy observations with X-shooter (Vernet et al. 2011) for a subsample selected from Amorín et al. (2017). Finally, we also considered NIR spectroscopy observations with Magellan/FIRE (Simcoe et al. 2010). More details of the selection are presented in the following subsections.

2.2.1. MOSFIRE spectroscopy

From the parent sample of CIII] emitters selected in the VANDELS survey with reliable spectroscopic redshift, we cross-matched entries with the catalog of sources in the NIRVANDELS survey (Cullen et al. 2021). This is a Keck/MOSFIRE VANDELS follow-up survey for 35 sources at 2.95 ≤z≤ 3.8 and 10 sources at 2.09 ≤z≤ 2.61. The details of the observations and the data reduction are described in Cullen et al. (2021). The slit width was 0.7″, yielding a spectral resolution of ∼3650 in H and ∼3600 in K band, respectively. Nineteen galaxies in NIRVANDELS show S/N > 3 in CIII]. Two objects were discarded because they did not show optical emission lines, particularly, [OIII]. Our final C3-VANDELS sample contains 17 galaxies in the CDFS and UDS fields.

To build our sample of VUDS galaxies, we used the publicly available MOSDEF survey (Kriek et al. 2015), which is also a Keck/MOSFIRE survey that comprises NIR spectra of ∼1500 K-band selected galaxies targeted to lie within three distinct redshift intervals, 1.37 ≤ z ≤ 1.70, 2.09 ≤ z ≤ 2.61, and 2.95 ≤ z ≤ 3.80. The details of the observations and the data reduction are described in Kriek et al. (2015). The 0.7″ width slit results in a spectral resolution of R = 3000 for the J band. For the H and K bands, the resolution is the same as in the NIRVANDELS survey. The MOSDEF survey covers well-studied HST extragalactic legacy fields by the CANDELS and 3D-HST surveys: AEGIS, COSMOS, GOODS-N, GOODS-S, and UDS. We cross-matched the MOSDEF and our VUDS parent sample catalogs to increase the number of galaxies. We prioritized galaxies with S/N CIII]  >  3 and with [OIII] detected. Our subsample contains 11 galaxies in the COSMOS and ECDFS fields.

2.2.2. X-shooter spectroscopy and data reduction

Four additional VUDS galaxies were observed in a follow-up program (Program: 0101.B-0779, PI: Amorín R.) with VLT/X-shooter (Vernet et al. 2011), which is a wide-band echelle spectrograph where two dichroics split the light into three arms (UVB, VIS, and NIR), and exposures simultaneously. We only used the spectra from the NIR arm. These galaxies were selected from Amorín et al. (2017) for their intense CIII] emission in deep VUDS spectra. Due to their redshift (at z ∼ 2.4 and ∼3.4), bright optical emission lines fall within good transmission windows. Observations were made in echelle mode from 2018 May to 2019 March with 900 s integrations using 1.0, 0.9, and 0.9″ (UVB, VIS, and NIR) slits for a resolution of R = 5400, 8900, and 5600, respectively, in seeing conditions of ∼0.6 arcsec. Each observing block (OB) is ∼3600 s of integration time, and each galaxy has two to three OBs. One of the galaxies (VUDS 5101421970) has an additional observing block of 3600 s from the program 0103.B-0446(A) (PI: Nakajima, K.) to complete a total of 3 h of exposure. The NIR region of the X-shooter spans the combined Y, J, H, and K regions from 1024–2048 nm. The reduction of each OB was performed using the EsoReflex (Freudling et al. 2013) X-shooter pipeline (Modigliani et al. 2010), which provides merged 2D NIR, visible, and UVB spectra. With the pipeline, we performed dark subtraction, flat-fielding, flexure correction, and 2D mapping, wavelength calibration, and flux calibration with standard stars. This method was also used in Matthee et al. (2021), who analyzed three out of four of the galaxies in our X-shooter sample. To combine the OBs, we used the IRAF (Tody 1986) task imcombine with median and σ clipping. To extract the 1D spectrum, we used the trace by [OIII]λλ4959,5007+Hβ, which is clearly detected, but the continuum is not detected in any galaxy.

2.2.3. FIRE spectroscopy and data reduction

A follow-up with Magellan/Folded Port Infrared Echellette (FIRE; Simcoe et al. 2010) was carried out for three galaxies selected from the VUDS parent sample to have a CIII] detection and at z < 3 so that bright emission lines did not overlap with strong skylines. Observations were conducted in 2022 April and May. FIRE was used in the high-resolution echelle mode. The observations were conducted as follows: the J-band acquisition camera was used to locate a nearby star from which a blind offset was applied to position the science target in the slit. The slits were either 0.75″ or 1.0″ in width, depending on the seeing (< 1″), yielding a spectral resolution of R ∼ 5200. The slits were oriented at the parallactic angle to minimize differential atmospheric refraction. Exposure times of 900 s were used for ABBA dither sequences with total integrations ranging from 3 to 4 h. The readouts were performed with the sample-up-the-ramp mode to minimize overheads. For each science target, one A0V star was observed at a similar airmass for telluric correction. The data were reduced using the publicly available pipeline1 developed by the instrument team. Unfortunately, the bright emission lines in this subsample were affected by sky emission lines, which precludes the kinematic analysis explained in the following sections. We used this subsample to estimate the flux of emission lines, which are included in the chemical analysis, and to estimate EWs of bright observed lines. For the VUDS targets, our final C3-VUDS sample consists of 18 galaxies.

In summary, our final sample combining C3-VANDELS and C3-VUDS samples with different NIR instruments contains 35 CIII] emitters, whose physical properties are described in the following section. In Table 1 we list the coordinates and spectroscopic redshifts based on CIII] of the individual galaxies in our sample, with notes on which NIR instrument was used.

Table 1.

Coordinates and spectroscopic redshift of the final sample.

3. Physical parameters of the final sample

3.1. Spectral energy distribution modeling

We performed a spectral energy distribution (SED) fitting of the entire final sample using BAGPIPES (Carnall et al. 2018). The photometric catalog for the C3-VANDELS sample comes from the CANDELS team (Galametz et al. 2013; Guo et al. 2013) and is the same as was used for the VANDELS team for targets with HST imaging (described in McLure et al. 2018; Garilli et al. 2021). For the galaxies in the CDFS field, we used the following bands: U-VIMOS, HST: F435W, F606W, F775W, F814W, F850LP, F098M, F105W, F125W, F160W, Ks-ISAAC, Ks-HAWKI, 3.6 μm-IRAC, and 4.5 μm-IRAC. For the galaxies in the UDS field, we used the bands U-CFHT, Subaru: B, V, R, i, z, HST: F606W, F125W, F160W, HAWKI: Y, Ks, WFCAM: J, H, K, 3.6 μm-IRAC, and 4.5 μm-IRAC.

For the C3-VUDS sample, we used the photometric catalog from COSMOS2015 (Laigle et al. 2016) for targets in the COSMOS field, and for galaxies in the ECDFS field, we used the photometric catalog used by the VUDS team (e.g., Calabrò et al. 2017; Ribeiro et al. 2017; Lemaux et al. 2022) that is obtained from Cardamone et al. (2010) and Grogin et al. (2011). For the galaxies in the COSMOS field, we considered the following bands: u-CFHT, VISTA: Y, J, H, and Ks; and Subaru: B, V, R, i, z, Y, 3.6 μm-IRAC, and 4.5 μm-IRAC. For the galaxies in the ECDFS field, we considered the following bands: MUSYC: U38, U, B, V, R, I, z, J, H, and K; and Subaru: IA427, IA464, IA484, IA505, IA527, IA574, IA624, IA679, IA709, IA738, IA767, IA827, 3.6 μm-IRAC, and 4.5 μm-IRAC.

We assumed an exponentially declining τ-model for the SFH. We obtained values for the timescale of τ > 5 Gyr, which implies a constant SFH at the redshift range covered by our sample. We allowed the metallicity to vary up to 0.25 Z, in agreement with the stellar metallicities observed in SFGs at z ∼ 3 (e.g., Cullen et al. 2019; Calabrò et al. 2021; Llerena et al. 2022). We included a nebular component that includes emission lines and nebular continuum emission in the model given the emission lines that are observed in these galaxies. The nebular component depends on the ionization parameter (log U) and can vary between −3 and −2 in our model. We used the Calzetti law to determine the dust reddening (Calzetti et al. 2000), with the total extinction AV free to vary between 0 and 2 mag. We constrained the ages from 100 Myr up to 1 Gyr. Our SED fitting leads to differences of ∼0.3 dex toward lower stellar mass and ∼0.1 dex toward higher SFR compared with a model with a similar SFH but a fixed solar metallicity and without a nebular model. The stronger effect on the SFR offset is the assumption of subsolar metallicity, while the stronger effect on the stellar mass offset is due to the inclusion of the nebular model. The stellar masses and SFRs obtained with the SED fitting are displayed in Fig. 1 and reported in Table 2. We highlight that the SFR is calculated over a timescale of 100 Myr. Our sample ranges across ∼2.4 dex in stellar mass from 107.9 to 1010.3M and ∼1.4 dex in SFR from 6.9 to 185 M yr−1. The physical parameters are reported in Table 2, including the color excess (E(B − V)SED) from the SED model, which ranges from 0.04 to 0.25 mag. The photometry and the resulting SED model are displayed in Figs. B.1 and B.2 for the C3-VANDELS and C3-VUDS samples, respectively. We performed a SED fitting following the same constraints to our VANDELS parent sample (2D histogram in Fig. 1), and we note that they follow the main sequence (MS) at z ∼ 3 according to Santini et al. (2017). The stellar mass was corrected for by a factor of 0.6 (Madau & Dickinson 2014) by adopting the same IMF as was assumed in this paper.

thumbnail Fig. 1.

Sample distributed along the star-forming main sequence at z ∼ 3 color-coded by EW([OIII]). The 2D histogram corresponds to the VANDELS parent sample at the same redshift range, and the dashed black line is the main sequence according to Santini et al. (2017). The magenta crosses and cyan triangles are reference samples at low (Berg et al. 2022) and intermediate redshifts (Maseda et al. 2017; Tang et al. 2021), respectively (see Sect. 3.1).

Table 2.

Main physical properties of the sample based on SED fitting.

We note that most galaxies of our combined samples are distributed along the MS. Seven (five of which are VUDS galaxies) show slightly higher SFRs than three times the observed 0.37 dex scatter of the relation. We compared our sample with other works at similar intermediate redshifts (Maseda et al. 2017; Tang et al. 2021) and with local metal-poor SFGs (Berg et al. 2022) in the following sections.

3.2. Emission line fluxes and EWs

The line flux and EW were measured individually in each galaxy. We measured the fluxes of the UV lines using single Gaussians and a linear component to include the local continuum. We considered Lyα, CIV, HeIIλ1640, OIII]λ1666, and CIII] in the set of lines to be measured. We considered the CIII] width as the maximum width of the lines, which is in the range of ∼250 − 300 km s−1 (FWHM ∼ 15 − 18 Å) and is different for each galaxy. The systemic redshift is based on the CIII] peak. Because the CIII] doublet is not resolved in our observations, we assumed an air wavelength of 1907.05 Å for the average peak. The observed fluxes and rest-frame EW of the observed lines are reported in Table 3. The uncertainties were estimated directly from the Gaussian fitting based on the covariance matrix using nonlinear least squares. For galaxies in the C3-VANDELS sample at z < 3, Lyα is not measured because it is not included in the VANDELS spectral range.

Table 3.

Observed fluxes and rest-frame EWs of the rest-UV emission lines.

On the other hand, the observed fluxes of the rest-optical lines are reported in Table 4. The fluxes are computed from a single Gaussian component (except for [OIII]λλ4959,5007, which is explained below). The local continuum was set to zero because it is not detected above 1 σ. We considered [OII]λ3727, [OII]λ3729, Hβ, [OIII]λλ4959,5007, Hα, and [NII]λ6583 to be measured. The spectra of the galaxies in the C3-VANDELS and C3-VUDS samples are displayed in Figs. A.1 and A.2, respectively.

Table 4.

Observed fluxes and rest-frame EWs of the rest-optical emission lines of our sample.

For [OIII]λλ4959,5007, we considered a more detailed modeling. We fit the [OIII]λλ4959,5007 profiles with two Gaussian components. In the other rest-optical lines, the S/N is unfortunately lower, and no similar analysis can be performed. We fit the doublet [OIII]λλ4959, 5007 simultaneously using LMFIT (Newville et al. 2016), considering a wavelength range from 4939 to 5027 Å to include only the doublet. We fixed the ratio of the two components to 1:2.98 (Storey & Zeippen 2000). We masked the region in the spectra between the two lines and the regions with strong sky residuals. As an initial guess, we included one narrow (100 km s−1) and one broad component (150 km s−1), and both widths were free to vary. We constrained the kinematics, assuming that the width of the components in each line is the same, as well as their peak velocities and ratios. We also performed a single Gaussian model to compare our results with and to test the improvement of the model with two Gaussians based on the Bayesian information criterion (BIC), which is a statistical measure used to compare models based on their fit to the data and complexity. According to this criterion, the model with the lowest BIC is the preferred model. When the difference between the BIC values of the two models is greater than 2, it is rated as positive evidence against the significance of the model with the higher BIC (Fabozzi et al. 2014). We chose the BIC criterion because it prevents the selection of an overfit model because it penalizes models with more free parameters. To compare, we estimated the ΔBIC = BICsingle − gaussian − BICdouble − gaussian and considered that the second component is statistically needed if ΔBIC > 2 (Fabozzi et al. 2014). A similar criterion was adopted for the complex line profiles of strong Lyα emitters (LAEs, Matthee et al. 2021) and Green Pea (GP) galaxies (Bosch et al. 2019; Hogarth et al. 2020) based on a χ2 minimization. We find that 23 (65%) out of the 35 galaxies in the sample show two kinematic components in their [OIII]λλ4959,5007 profiles. In the remaining 12 galaxies, only one single component is statistically justified. We only considered the second component to be broad for a difference of one spectral resolution element between the widths of the two Gaussian components. Otherwise, we considered that there are two narrow components in the double Gaussian model. The results from the fitting are shown in Figs. 2 and 3 for the subsample with two components in the C3-VANDELS and C3-VUDS samples, respectively. For [OIII], we report in Table 4 the global flux (double-Gaussian model) or the flux of the single-Gaussian model if the second component is not justified. The kinematic information of the fits is reported in Table 5. We considered the intensity peak of the narrower component as the systemic velocity.

thumbnail Fig. 2.

Best-fit of the [OIII] profiles from the C3-VANDELS sample with ΔBIC  >  2. The panels show single galaxies ordered by EW(CIII]) from left to right and top to bottom. At the top in each panel, we show the 2D spectrum with the detected lines. In the middle, we plot the models for [OIII]λ4959 (left) [OIII]λ5007 (right). The blue line shows the observed spectrum, and the red line shows the error spectrum. The Gaussian lines are normalized to the intensity peak of [OIII]λ5007. The dashed black lines are the narrow and broad components, and the magenta line is the global fit considering both components. The magenta-shaded region is the 3σ uncertainty of the fit. The green line is the single-Gaussian fit. The vertical gray line marks the systemic velocity traced by the peak intensity of the narrow component. The vertical blue line marks the peak intensity of the broad component. Bottom: The residuals (Δσ) for each model are shown with the same colors. The gray-shaded regions are masked regions due to sky residuals. The galaxies in the black square show two narrow components.

thumbnail Fig. 3.

Same as in Fig. 2, but for the C3-VUDS sample with ΔBIC > 2.

Table 5.

Kinematic properties of the ionized gas based on [OIII] modeling.

For the rest-EW, we used the SED model obtained from photometry to determine the continuum of the rest-optical emission lines of our sample. We are interested in qualitative trends with the EWs. We therefore did not perform an additional aperture correction for the continuum. For a particular line with an intensity peak at λpeak, the continuum was obtained as the average between the continuum at λpeak − 20 [Å] and λpeak + 20 [Å]. The measured EWs for Hβ and [OIII] are reported in Table 4. For the rest-UV lines, we used the local continuum measured directly from the spectra, which is detected in all galaxies in the final sample. We verified that the UV continuum from the SED fitting model was consistent with the continuum measured in the spectra. For example, for the C3-VANDELS sample, the difference for the continuum for CIII] between the SED model and the local continuum measured directly from the spectra is ∼14% on average. This difference implies a difference in the EW of ∼0.6 Å on average, which is smaller than the typical error of 0.93 Å in the EWs.

Most of the targets (20 out of 35) in our sample are at z > 3, which means that Hα is not included in the observed NIR spectral range. For this reason, we cannot determine the nebular extinction (E(B − V)g) based on the Balmer decrement for the entire sample. For the galaxies (12 out of 35) with a detected Hα and Hβ (S/N > 2), we estimated the nebular attenuation assuming Hα/Hβ = 2.79 under case B approximation for Te = 15 000 K and ne = 100 cm−3 (Pérez-Montero 2017) and considering the Cardelli law (Cardelli et al. 1989). The obtained E(B − V)g values are reported in Table 2. On the other hand, we obtained an estimate of the stellar extinction from the SED fitting. Most of our galaxies have a low dust content with E(B − V)SED < 0.25 mag (see Table 2). From this subsample, we extrapolated the nebular extinction for the entire sample based on their SED extinction. The best linear fit leads to E(B − V)g = 0.75 × E(B − V)SED + 0.19. We used this extrapolation only when E(B − V)g was not determined directly by the Balmer decrement. We corrected the observed fluxes (reported in Tables 3 and 4) assuming the Reddy et al. (2015) law with RV = 2.505, based on other works with UV emission lines (e.g., Mingozzi et al. 2022).

3.3. SFR surface density

We also estimate the instantaneous SFR(Hα) using the relation logSFR(Hα) = log(L(Hα)[erg s−1]) − 41.27, which assumes the Chabrier (2003) IMF (Kennicutt & Evans 2012), and the luminosities were corrected for dust reddening as explained in Sect. 3.2. When Hα was not available, we used Hβ assuming the same theoretical ratio Hα/Hβ = 2.79. Because Hα is only available for a small subsample and Hβ is fainter and has a lower S/N than [OIII], a multiple Gaussian components analysis is not possible for the entire sample. For this reason, we assumed the same broad-to-narrow flux (fB) ratio from [OIII] to take only the flux from the narrow Hα (or Hβ) into account when we estimated the SFR, and we excluded the contribution from the broad component.

We used the above values to derive the instantaneous SFR surface density (ΣSFR) defined as , where rH is the effective radius of the galaxy measured in the H band. We obtained the effective radius directly from the literature based on HST imaging. For the C3-VANDELS sample, they were obtained directly from the CANDELS catalog (van der Wel et al. 2012), while for the VUDS sample, they were obtained from Ribeiro et al. (2016). In both cases, they used GALFIT (Peng et al. 2002, 2010) and HST/F160W images to fit Sérsic profiles to obtain the effective radius. For the C3-VUDS subsample, no H-band HST image is available for 6 out of 18 galaxies. They are identified in Table 5. In these cases, we considered the effective radius using HST/F814W images from the same Ribeiro et al. (2016) catalog following a consistent method.

We note that at the redshift range covered for our sample, the F814W images trace the rest-frame UV-continuum by mostly massive stars, and the F160W images trace the rest-frame optical, which includes older stars and extended gas. Based on the C3-VUDS galaxies that have two images (12 galaxies), we find that the effective radius from the H-band image is typically a factor 1.4 greater than in the i band. In a few cases (3 out of 12), the differences can be up to a factor of ∼3, however. When only HST/F814W was available, we corrected the effective radius by a factor 1.4.

In Fig. 4 we display the HST/F160W images of the C3-VANDELS sample with two components in their [OIII] profile. In Fig. 5 we show the same, but for the C3-VANDELS sample with a single-Gaussian model. The images for the C3-VUDS sample are shown in Figs. C.1 and C.2 for the galaxies with and without a broad component, respectively. We note that in both cases, most of the galaxies (25 out of 35) tend to be compact, that is, they are smaller than the size-mass relation at z ∼ 2.75 of late-type galaxies (van der Wel et al. 2014). They show no clear evidence of mergers.

thumbnail Fig. 4.

HST/F160W images (Koekemoer et al. 2011) of the C3-VANDELS sample with ΔBIC> 2, i.e., the subsample with a broad component in the [OIII] profile. The images trace the rest-optical. The white contour is the 3σ level. The physical scale of 0.5 arcsec at their redshift is shown the left of each image, and the effective radius is shown on the right. The galaxy marked with a black square shows two narrow components in its [OIII] profile.

thumbnail Fig. 5.

Same as in Fig. 4, but for the C3-VANDELS sample without the features of a broad component in their [OIII] profile.

4. Results

4.1. Ionizing sources: Diagnostic diagrams

In this section, we analyze the ionization source of the galaxies in our sample. First, we use the UV diagnostic diagram based on the EW(CIII]) versus CIII]/HeIIλ1640 flux ratio and EW(CIV]) versus CIV/HeIIλ1640 flux ratio. The results are shown in Fig. 6. Most of the galaxies in the sample are consistent with being ionized by massive stars according to the demarcation lines in Nakajima et al. (2018). For some galaxies, these ratios are lower limits because HeIIλ1640 (or CIV) is not detected in their spectra, and then we determined 2σ upper limits. The more extreme [OIII] emitters tend to fall in the upper right corner of the diagrams, suggesting high EW(CIII]) and EW(CIV), but not very strong HeIIλ1640, which might be an indication of AGN contribution.

thumbnail Fig. 6.

UV diagnostic diagrams for our sample based on the EWs of CIII] (left) and CIV (right). In both panels, our sample is color-coded by EW([OIII]), and the dashed black lines are the demarcation between AGN (on the left) and SF (on the right) according to Nakajima et al. (2018). In the right panel, the symbols with magenta edges are galaxies classified as AGN according to the EW(CIII])–CIII]/HeII diagram.

All the galaxies in the C3-VANDELS sample are consistent with pure stellar photoionization, but a few (five galaxies in each diagram) C3-VUDS galaxies lie in the AGN region or near the boundary between the two regions (symbols with magenta edges in the right panel in Fig. 6). In particular, two of these galaxies (5100998761 and 5101421970) also show unusually high Te > 20 000 K (see Sect. 4.4), and some nonthermal contribution cannot be ruled out according to these diagnostics. Such high Te are now observed in young SFGs in the EoR (z = 5.33 − 6.93) with intense EW([OIII]λλ4959,5007+Hβ)∼1000 Å (Matthee et al. 2023) and have previously been reported in extremely metal-poor starbursts in the local Universe (Kehrig et al. 2016).

Recently, the C IV/C III] ratio has been proposed as a potential indirect indicator to constrain the LyC escape fraction (Schaerer et al. 2022a; Saxena et al. 2022b). We also find that most of the galaxies are consistent within the errors with CIV/CIII]  <  0.7, which indicates that they are not strong LyC leakers and are likely to show fesc < 0.1. Only one VUDS galaxy (5101421970) shows CIV/CIII]  >  0.7, which in the UV diagnostic diagrams is classified as an AGN.

We also explored the Baldwin, Phillips & Terlevich (BPT; Baldwin et al. 1981) optical diagnostics to verify the SF nature of these galaxies. In Fig. 7 (left panel) we show the classical BPT diagram ([NII]λ6583/Hα versus [OIII]/Hβ) for the subsample of galaxies at z < 3 for which Hα is included in the K band. This subsample falls within the demarcation lines for SF regions and is broadly offset from the typical excitation of local SFGs. The subsample at z < 3 belongs to the fainter CIII] emitters in the entire sample with EW(CIII])  <  3 Å, in particular, those for which [NII]λ6584 is detected. The more extreme CIII] emitters (with EWs  >  4 Å) tend to show upper limits in [NII] and high [OIII]/Hβ ratios.

thumbnail Fig. 7.

Optical diagnostic diagrams for our sample. Left: Classical BPT diagram (Baldwin et al. 1981) for our subsample at z < 3, color-coded by EW(CIII]). The dashed black lines are the typical local AGN/SF demarcation lines (Kewley et al. 2001; Kauffmann et al. 2003). The dashed red line is the demarcation at z ∼ 3 according to Kewley et al. (2013). Right: Mass-excitation diagram for our entire sample color-coded by EW(CIII]). The dashed black and red lines are the AGN/SF demarcation at low-z (Juneau et al. 2014) and z ∼ 2.3 (Coil et al. 2015), respectively. As a reference, we include the stacks from Cullen et al. (2021). Our VANDELS subsample is a subset of these stacks. In both panels, the symbols with magenta edges are galaxies classified as AGN according to the UV diagnostic diagrams in Fig. 6.

Finally, in Fig. 7 (right panel), we explore the mass-excitation (MEx; Juneau et al. 2014) diagram for the entire sample (except for the galaxies without Hβ in the spectral range). Similarly to the BPT diagram, they are also consistent with SF. The more extreme CIII] emitters tend to fall in the low-mass region (roughly < 109.5M) and high [OIII]/Hβ > 4, which indicates a highly excited ISM. The two particularly extreme galaxies with possible AGN contribution lie in the SF region in this diagram (one of them shows a lower stellar mass in the sample). The above analysis suggests that a combination of UV and optical lines may give us more clues about their nature, particularly for extremely metal-poor SF galaxies, which may display very high electron temperatures. Hereafter, we consider that the galaxies in our sample are dominated by an SF with a highly excited ISM and show no clear evidence of an AGN contribution based on their emission lines.

4.2. Ionization parameter

We also studied the ionization properties of our sample of galaxies by means of a tracer of the ionization parameter log U. Because the wavelength coverage of our spectra is limited, we used the ratio as a proxy for log U (Kewley et al. 2019). We find [OIII]/[OII] values within 1 and 10, which imply a high ionization parameter between roughly log U = −3 and log U = −2 (Reddy et al. 2023a).

In Fig. 8 we present the relation for [OIII]/[OII] and ΣSFR found by Reddy et al. (2023a,b) for SFGs in the MOSDEF survey and others at higher redshift, suggesting that galaxies with more compact and violent star formation show more extreme ionization conditions. We find that our VANDELS and VUDS galaxies follow the same trend and lie closer to the z > 3 galaxies than those at z < 2.6 in the above works. Even though our sample galaxies mostly lies in the SF main sequence at z ∼ 3 (Fig. 1), their ionization properties and emission line EWs appear to be more extreme than those of their counterparts at lower redshift.

thumbnail Fig. 8.

Variation in [OIII]/[OII] with ΣSFR. Individual galaxies with detected [OII] are shown with red and blue symbols for the C3-VUDS and C3-VANDELS samples, respectively. We also include lower limits based on the upper limits on [OII]. The small green and magenta circles are galaxies from Reddy et al. (2023a) at z = 1.6 − 2.6 and Reddy et al. (2023b) at z = 2.7 − 6.3, respectively. The solid black line is the relation presented in Reddy et al. (2023a), and its extrapolation up to ΣSFR = 100 M yr−1 kpc−2 is plotted as the dashed black line.

4.3. EW relations

For the UV lines, our sample covers a range of EW(CIII]) from ∼1 Å to ∼15 Å, with a mean value of 5.6 Å (σ = 3.7 Å). Six galaxies (17%) show EW(CIII])  >  10 Å, similar to the EW values observed in galaxies at z > 6 (e.g., Stark et al. 2017; Hutchison et al. 2019). Our sample includes 14 galaxies at z ≳ 3 for which Lyα is included in our spectral range and with EW(Lyα)  >  20 Å. Five of them reach EW(Lyα)  >  100 Å.

For the optical lines, we find strong [OIII]λλ4959,5007 emission, with EW([OIII]) spanning from 102 Å to 1715 Å with a mean value of 563 Å (σ = 420 Å), which is well within the typical EWs defined for extreme emission line galaxies (EELG) at lower redshifts (e.g., Amorín et al. 2014, 2015; Calabrò et al. 2017; Pérez-Montero et al. 2021). Only five galaxies (14%) of the sample show EW([OIII])  >  1000 Å, which are typical values found for z > 6 EoR galaxies from photometric data (e.g., Endsley et al. 2021) and more recently with JWST spectroscopy (Matthee et al. 2023).

We explore the correlation between stellar mass and [OIII]+Hβ EWs in Fig. 9. Low-mass galaxies tend to show higher EWs, following a trend similar to that found in literature, both at z ∼ 2 − 3 (e.g., Maseda et al. 2017; Tang et al. 2021) and in reionization galaxies (Endsley et al. 2021). The most extreme EWs in our sample correspond to galaxies with EW(CIII]) > 5 Å and EW([OIII]λλ4959,5007+Hβ) > 500 Å, as shown in Fig. 10, which are still rare at z ∼ 3, but become the norm toward reionization (e.g., Smit et al. 2014; De Barros et al. 2019; Endsley et al. 2021; Sun et al. 2022a,b; Matthee et al. 2023). The trend in Fig. 10 also shows that the strong [OIII] and CIII] emitters at z > 3 are typically those with larger EWs in Lyα (≳20 Å). However, a few galaxies with weak Lyα emission are found among the strong [OIII] emitters, in agreement with previous works (Le Fèvre et al. 2019; Du et al. 2020).

thumbnail Fig. 9.

Relation between EW([OIII]λλ4959,5007+Hβ) and stellar mass. Our sample is color-coded by EW(CIII]). The small red and magenta circles are literature samples at intermediate redshifts from Maseda et al. (2017) and Tang et al. (2021), respectively. The dashed black line is the best fit (slope −0.34) with our data, and the gray shaded region is the 1σ observed scatter of 0.33 dex.

thumbnail Fig. 10.

Relation between EW([OIII]λλ4959,5007+Hβ) and EW(CIII]). Our sample is color-coded by EW(Lyα). The square symbols with cyan edges are galaxies in the C3-VANDELS at z < 3, for which Lyα is not in the spectral range. The small red and magenta circles are literature samples at intermediate redshifts from Maseda et al. (2017) and Tang et al. (2021), respectively. The dashed black line is the best fit (slope 0.68) with our data, and the gray shaded region is the 1σ observed scatter of 0.33 dex.

These results can be interpreted in the context of the galaxy ionizing photon production efficiency, which correlates with the Hα and [OIII] EWs (Tang et al. 2019). Even for extreme [OIII]+Hβ emitters that efficiently produce ionizing photons, the emerging Lyα line will not be necessarily intense due to its resonant nature, which makes it sensitive to dust content, to the neutral hydrogen column density, and to their spatial distribution (e.g., Hayes 2015). Our subsample of strong [OIII] emitters with lower EW(Lyα) shows a higher dust extinction. These objects are likely found in a very early phase after the onset of star formation, where the young (< 2–3 Myr) massive stars are still embedded in their dense and dusty birth clouds and did not yet have enough time to clear channels through the ISM via feedback (e.g., winds or supernovae), and then Lyα photons are trapped, while LyC photons are absorbed (Naidu et al. 2022).

4.4. Electron densities and temperatures

The two components of the doublet [OII]λλ3727, 3729 are detected and are not affected by strong sky residuals in 15 galaxies of our entire sample. For this subsample, we estimated the electron density using the getTemDen task in PyNeb assuming Te = 15 000 K and performing 100 Monte Carlo simulations in order to include the uncertainties of the observed fluxes. We find a wide range of electron densities from ∼47 to ∼1261 cm−3, with a mean value of 560 cm−3. These values are within the range of values observed at similar redshifts (e.g., Sanders et al. 2016; Reddy et al. 2023a). Given the uncertainties (we obtained a mean value of ∼270 cm−3) in the estimated ne values due to the low S/N of [OII]λλ3727, 3729, we report the densities in Table 6, but in the following sections, we use only the mean value of the entire sample.

Table 6.

Electron density, temperature, and chemical abundances estimated for our sample.

In order to estimate the electron temperature, we used the OIII]λ1666/[OIII]λ5007 ratio where available. OIII]λ1666 is detected with an S/N > 2 in 21 galaxies in our sample. We used the getTemDen task in PyNeb assuming ne = 560 cm−3. Because of the offset between the Te([OIII]) found using this ratio and the commonly used ratio [OIII]λ4363/[OIII]λ5007 found in local galaxies, we corrected the obtained temperature down by the typical difference of −0.025 dex obtained for CLASSY SFGs in Mingozzi et al. (2022), which leads to differences of ∼ − 1000 K. We report these temperatures in Table 6. As discussed in Mingozzi et al. (2022), one of the intrinsic reasons that may explain the offset between the two methods is that the ISM is patchy. The UV light is visible only through the less dense (and/or less reddened) regions along the line of sight, while the optical light may also arise from denser (and/or more reddened regions). We find Te([OIII])  >  1.3 × 104 K. Some objects have very high temperatures but do not exceed 2.5 × 104 K, as reported in Table 6. We find a mean value of 1.77 × 104 K for the entire sample with typical errors of 0.2 × 104 K, which were calculated with 100 Monte Carlo simulations taking into account a normal distribution of the fluxes of the lines used to estimate Te([OIII]). Our results are consistent with the early results from the JWST/NIRSpec, based on the first direct detection of the faint auroral line [OIII]λ4363 at z ∼ 8, where Te from 1.2 × 104 and up to 2.8 × 104 K are estimated. The Te-based metallicities range from extremely metal poor (12+log(O/H) < 7) to about one-third solar (Schaerer et al. 2022b; Trump et al. 2023; Curti et al. 2023; Nakajima et al. 2022).

We note that the OIII]λ1666/[OIII]λ5007 ratio was obtained from two different instruments, which may cause possible systematic errors due to flux calibration and dust attenuation corrections because the wavelength separation of the lines is large. First, we checked that the spectra were consistent with the photometry based on the SED model to avoid flux-matching issues. For example, the difference between the mean continuum between 1450 and 1500 Å in the SED model and the observed spectrum is ∼10% of the observed flux on average, which is lower than the scatter in this spectral region, which is 25% of the observed flux on average. Additionally, we verified that changing the dust attenuation law does not affect our results. For instance, when we consider the Calzetti et al. (2000) law, the mean Te changes by only 76 K, which is negligible compared with the typical uncertainties.

4.5. Oxygen and carbon abundance

The study of the abundances of heavy elements in the ISM of galaxies provides valuable insights into the physical processes that cause their formation and how the relative importance of these processes has changed across cosmic time (see the review by Maiolino & Mannucci 2019). Due to the line production mechanisms, nebular UV emission can be used to also directly calculate the physical and chemical conditions under which emission lines are produced. For instance, while the oxygen abundance (O/H) is the standard measure of a gas-phase metallicity in galaxies, CIII] provides a path to estimating the C abundance, which is a non-α element. Because C is primarily produced in stars with lower masses than stars that produce O, the injection of C and O into the ISM occurs on different timescales. This provides a probe of the duration, history, and burstiness of the star formation (e.g., Berg et al. 2019). The relative abundance of C and O can also be relatively unaffected by hydrodynamical processes, such as outflows of enriched gas (e.g., Edmunds 1990). In particular, the line intensity ratio C III]/O III]1666 has been used to estimate the relative C/O abundances at different redshifts (e.g., Garnett et al. 1995; Shapley et al. 2003; Erb et al. 2010; Steidel et al. 2016; Berg et al. 2016, 2019; Pérez-Montero & Amorín 2017; Amorín et al. 2017; Llerena et al. 2022). Moreover, the combination of UV+optical emission lines has been explored to constrain Te using the OIII]λ1666/[OIII]λ5007 ratio (Pérez-Montero & Amorín 2017), and then to estimate the gas-phase metallicity, in particular when the optical auroral line [OIII]λ4363 is not available because it is too weak or because the spectra do not cover it.

We derived the C/O abundance based on photoionization models using the public version 4.232 of HCM-UV (Pérez-Montero & Amorín 2017), using the POPSTAR synthetic SEDs (Mollá et al. 2009) as an ionizing source for the models and assuming the relations between metallicity and excitation for EELGs assumed by the code when no auroral emission line is provided. The code HCM-UV performs a Bayesian-like calculation that compares extinction-corrected UV emission line fluxes and their uncertainties with the prediction of a grid of models covering a wide range of values in O/H, C/O, and log U. The code calculates C/O as the average of the χ2-weighted distribution of the C/O values in the models. Then, C/O is fixed in the grid of models, and both O/H and log U are calculated as the mean of the model input values in the χ2-weighted distribution. The χ2 values for each model are derived from the quadratic relative differences between the observed and predicted emission line ratios. The uncertainties of the derived parameters are calculated as the quadratic addition of the weighted standard deviation and the dispersion of the results after a Monte Carlo simulation.

As input, we used CIII], OIII]λλ1661,1666, Hβ, and [OIII]. We did not include CIV in the input because CIV is only detected in 42% of our sample. We considered the ratio OIII]λ1661/OIII]λ1666 = 0.44, based on photoionization models (Gutkin et al. 2016). Because of the spectral resolution of ∼7 Å at 1666 Å for the C3-VUDS sample, the OIII]λλ1661,1666 is blended, and then the measured flux corresponds to the doublet. On the other hand, for the C3-VANDELS sample, the spectral resolution is ∼3 Å at 1666 Å. We multiplied the measured OIII]λ1666 flux by a factor 1.44 to account for the doublet. We included the flux of the doublet OIII]λλ1661,1666 in the input when the line was detected with S/N > 2. In the case of an upper limit, the 2σ limit was considered as input in the code for the other lines, with an error of 1σ, following the method described in Pérez-Montero et al. (2023). The obtained C/O values are ∼0.11 dex higher on average than the value that is obtained when the C3O3 calibration proposed by Pérez-Montero & Amorín (2017) is used.

We find a mean value of log(C/O) = − 0.52 (∼54% solar) with a typical error of 0.15 dex, which is consistent with the typical value of SFGs at z ∼ 3 based on stacking (Shapley et al. 2003; Llerena et al. 2022). Our results of log(C/O) range from −0.90 (23% solar) to −0.15 (128% solar) and are consistent with the wide range of C/O values observed in local blue compact dwarf (BCD) galaxies (Garnett et al. 1995, 1997; Kobulnicky et al. 1997; Kobulnicky & Skillman 1998; Izotov et al. 1999; Thuan et al. 1999; Berg et al. 2016; Senchyna et al. 2021) and giant HII regions (Garnett et al. 1995, 1999; Kurt et al. 1995; Mattsson 2010; Senchyna et al. 2021). The lowest values in our sample are also within the range of values reported of log(C/O) = −0.83±0.38 at a metallicity < 2% solar for EoR galaxies at z ∼ 8 from JWST spectra and are consistent with the large dispersion in log(C/O) observed in z ∼ 0 − 2 low-metallicity dwarf galaxies without evidence of an evolution in the C/O versus O/H relation (Arellano-Córdova et al. 2022). Even a lower value log(C/O) = −1.01 ± 0.22 has been reported at z = 6.23, which is consistent with the expected yield from core-collapse supernovae, indicating negligible carbon enrichment from intermediate-mass stars (Jones et al. 2023).

For the gas-phase metallicity derivation, we first used the HCM-UV code with the same input as described before. We then compared these results with the results obtained from an alternative method using PyNeb (Luridiana et al. 2012, 2015). To derive the total oxygen abundance, we used the approximation . To estimate O2+/H+, we used the getIonAbundance task in PyNeb, assuming the corrected electron temperature, the global [OIII] flux, and ne = 560 cm−3. To estimate O+/H+, we used the same task with [OII]λ3727 where available (otherwise, we used [OII]λ3729 if available). To estimate Te[OII], we assumed the relation Te([OII]) = based on photoionization models to infer Te([OII]) from Te([OIII]) (Pérez-Montero 2017). The total oxygen abundance log(O/H)PyNeb with this method is reported in Table 6.

A comparison of these two methods finds an offset of ∼0.4 dex (a factor of 2.5) toward a lower metallicity for the results based on HCm-UV. For the galaxies in which the lacking detected lines prevent us from using the PyNeb method, we corrected the gas-phase metallicity from HCM-UV upward by a factor 0.4 dex. These values are reported in Col. 4 in Table 6. We find a mean value of log(O/H)HC M−UV + 12 = 7.91 (17% solar), with values ranging from 7 to 73% solar. Finally, we compared our metallicity derivation with the expected values using the EW(CIII])-metallicity calibration proposed by Mingozzi et al. (2022) for local analogs, which is displayed in Fig. 11. For most galaxies in our sample, our results are consistent within 2σ with the local relation (dashed blue line), which has an observed scatter of 0.18 dex. Our best fit (dashed red line) shows an offset toward lower values compared to the local relation, and the observed scatter is 0.24 dex.

thumbnail Fig. 11.

Relation between gas-phase metallicity and EW(CIII]) for our sample. The dashed blue line and shaded blue region correspond to the relation found in Mingozzi et al. (2022) at low z and their observed 2σ scatter. The dashed red line is our best fit, and the observed 2σ scatter is the shaded red region. The sample is color-coded by OIII] flux when this line is detected at an S/N > 2. In the other cases, only error bars are shown.

We compared our estimates with the results using stron-line calibration (SLC) from pure optical and UV lines. For example, for the O32 = log calibration from Pérez-Montero et al. (2021), we find that the obtained metallicities tend to be ∼0.19 dex offset toward higher metallicities. The values found using this calibrator are reported in Col. 2 in Table 6. A slightly higher mean offset of ∼0.30 dex is found using the Bian et al. (2018) O32 calibration. Moreover, for the calibrations based on UV lines alone, for instance, He2-O3C3 (Byler et al. 2020) based on the ratios and , we find metallicities that tend to be ∼0.42 dex lower than our estimated metallicites. This mean offset toward lower metallicities is reduced to ∼0.30 dex for the proposed corrected He2-O3C3 calibration in Mingozzi et al. (2022), where offsets between UV and optical methods have been observed in local galaxies (Mingozzi et al. 2022). The values found using this calibrator are reported in Col. 3 in Table 6.

Hereafter, we consider the metallicities derived combining UV and optical emission lines in Table 6 Col. 4 for our discussion. This method allowed us to estimate gas-phase metallicities for the entire sample of galaxies.

4.6. Ionized gas kinematics

We interpret the narrow component in the [OIII] profile as the warm gas tracing virial motions within SF regions, while the broad component is interpreted as the turbulent outflowing gas with a velocity in reference to the systemic velocity (see, e.g., Amorín et al. 2012; Arribas et al. 2014; Freeman et al. 2019; Hogarth et al. 2020). The outflowing gas can be SF or AGN driven and depends on the source of energy. From observations, SF-driven winds are closely coupled to SF properties such as ΣSFR, whereas AGN-driven winds are strongly correlated with stellar mass and are rare in low-mass galaxies (Förster Schreiber et al. 2019). In the literature, broad components have been observed in AGN with a large velocity dispersion > 200 km s−1 (Rodríguez del Pino et al. 2019; Förster Schreiber et al. 2019). In order to obtain the intrinsic velocity dispersion of the [OIII] lines, we subtracted the instrumental (σins) and thermal (σther) widths in quadrature. For the former, we considered the resolution in the H band of R = 3600 (for MOSFIRE) and R = 5600 (for X-shooter), which corresponds to σins ∼ 35 km s−1 and ∼23 km s−1, respectively. For the latter, we assumed the same formula as in Hogarth et al. (2020) and assumed the mean Te value obtained for the sample of 1.77 × 104 K, which leads to σther ∼ 0.8 km s−1. As described in Sect. 3.2, we assumed that the second component is broad when the difference was greater than 35 km s−1 with respect to the narrower component, which coincides with the global peak of the profile. Three of the 23 galaxies with two kinematic components do not show a broad component, but they show two narrow components instead. These galaxies are marked in Table 5. In these cases, the complex profile more likely traces a merger instead of a turbulent outflowing gas, but it is not totally clear from HST images where obvious companions are not observed (see the images marked by the black square in Figs. 4 and C.1). We excluded these three galaxies from the interpretation of outflowing gas.

The intrinsic velocity dispersion (σvel) of the narrow component shows values ranging from ∼19 to ∼102 km s−1, with a mean value of 57 km s−1 (typical error of 3.7 km s−1) for the entire sample. When we consider only the galaxies that show broad components, the σvel of the narrow component are up to 70 km s−1, with a mean value of 48 km s−1. The broad components range from 73 to 210 km s−1, with a mean value of 121 km s−1 (typical error of 34 km s−1). The uncertainties are estimated directly from the Gaussian fitting based on the covariance matrix using nonlinear least squares. We note that the continuum nondetection implies that the true intensity and width of the broader component fit in [OIII] should be considered as possible lower limits.

The broad component is blueshifted from the systemic velocity in 14 of these galaxies, and it is redshifted in 6 galaxies, with a mean velocity shift (ΔvB) of −43 and 34 km s−1, respectively, as shown in Fig. 12. In both cases, the broad component shows a mean velocity dispersion σB = 118 − 124 km s−1, which is in the range of typical widths for ionized outflows from SFGs at similar redshifts with comparable EWs (e.g., Matthee et al. 2021) and in local analogs (e.g., Amorín et al. 2012; Bosch et al. 2019; Hogarth et al. 2020). The velocity shift shows typical errors of 7 km s−1. The velocity shift is < 10 km s−1 in 5 of these galaxies, which is lower than the NIR resolution. They may show outflows very close to the systemic velocity. We highlight that none of the galaxies in our sample shows the extremely broad wings (σ ∼ 255 km s−1) observed in local giant HII regions that are dominated by supernovae feedback (Castaneda et al. 1990). Our sample covers values from 24% to 63% for the broad-to-narrow flux ratio (fB), with a mean value of 43% (σ = 0.13).

thumbnail Fig. 12.

Distribution of the velocity shift ΔvB of the broad component in our sample. The dashed black and red lines mark the mean values when the broad component is blue- and redshifted, respectively.

For the three galaxies of the C3-VUDS sample that were observed with X-shooter, we find consistent results with the kinematic analysis presented in Matthee et al. (2021), in which the authors considered only our three galaxies with z < 3. Additionally, the resolved Lyα profiles of these galaxies were analyzed in Naidu et al. (2022). For example, in the Lyα profile of 510583858, an intense blue peak is reported, which suggests an inflow of gas (Yang et al. 2014). This interpretation is in line with that from the kinematics of the ionized gas, where we find a redshifted broad component for the optical lines. The other two galaxies included in Naidu et al. (2022) show Lyα profiles with weaker blue peaks compared with the dominant red peak, which is indicative of outflowing material.

4.6.1. Outflow interpretation for the broad emission

Here we note that the blueshifted broad component can be interpreted as an outflow on the near side of the galaxy or as an inflow on the far side, and vice versa for the redshifted components. Moreover, the dust distribution may also play a role in the interpretation because dust may be blocking the flow (Arribas et al. 2014). Based on the low dust content in our sample galaxies, we assume that in either case, blue- or redshifted, the broad component primarily traces outflowing material, and the shift depends on the geometry and on the particular line of sight.

In order to derive the maximum velocity (vmax) of the unresolved outflow, we followed the same approach as in Hogarth et al. (2020), Avery et al. (2021), Concas et al. (2022) and references therein. We considered vmax = |Δv|+2 × σB in order to include the blue- or redshifted cases as outflowing gas, where the maximum velocity will be in the line profile wings dominated by the outflow (see, e.g., Rodríguez del Pino et al. 2019; Lutz et al. 2020). We find a mean maximum velocity of 283 km s−1, ranging from 183 to 460 km s−1. For the C3-VANDELS sample, we also performed a qualitative comparison with the kinematics analysis based on rest-UV absorption lines described in Calabrò et al. (2022). They found that the bulk ISM velocity along the line of sight is 60 ± 10 km s−1 for low-ionization gas, consistent with the mean value found in our sample. However, for the maximum velocity, they find a mean value of 500 km s−1. The higher value compared with our results may be explained by the low resolution of the VANDELS spectra, which leads to higher unresolved line widths. Another possible explanation is that nebular emission lines may trace denser gas than ISM absorption lines (e.g., Marasco et al. 2023). In this case, the emission lines would trace smaller-scale outflows near the star cluster and with lower velocities, while the UV absorption lines would trace large-scale outflows with a galactic extent and with higher velocities (Chisholm et al. 2016, and references therein).

In a galaxy-by-galaxy comparison using the bulk velocity traced by the combined fit of low-ionization lines (SiIIλ1260, SiIIλ1526, CIIλ1334, and AlIIλ1670), we find that in 8 out of the 11 C3-VANDELS galaxies with detected broad component, the bulk velocity and the offset of the broad component agrees in the direction of the flow (i.e., if they are blue- or redshifted). In 2 cases, we find a broad component that is blueshifted while the bulk velocity is redshifted, but they are consistent within the uncertainties. Only in one case did we find a redshifted broad component and a negative bulk velocity. Our results suggest that the broad component of [OIII] indeed traces flowing material that roughly agrees with the assumption of being blue- or redshifted in both absorption and emission lines.

4.6.2. Outflow velocities and star formation properties

We related the maximum velocity of the outflow with the global properties of our sample. We find a weak positive correlation between the maximum velocity and the galaxy stellar mass (Pearson correlation coefficients ρ: 0.22, 0.91σ significance). This trend suggests that more massive galaxies can power faster outflows than less massive galaxies. We find no correlation for the relation with SED-derived SFR (ρ ∼ 0, 0.36σ significance). The correlation is stronger (ρ = 0.16, 0.66σ significance) when we consider the instantaneous SFR traced by Hβ (or Hα). The slope of the best fit is 0.06, which agrees with the trend found in Arribas et al. (2014) for a sample of local luminous and ultraluminous infrared galaxies (U/LIRGs) at galactic and subgalactic (i.e., star-forming clump) scales. This suggests that the longer-timescale SFR does not trace the actual outflow, while the instantaneous SFR is a more real tracer of the outflow and the maximum velocity that can be reached. This might be the reason why only a marginal (at 2σ level) correlation was found between the maximum velocity and SFR in Calabrò et al. (2022).

In Fig. 13 we explore the relation between the maximum velocity of the outflow and the ΣSFR. A relation between the two properties is expected, as explained by simple models (e.g., Heckman et al. 2015; Xu et al. 2022b). The warm ionized outflow is described by a collection of clouds or filaments driven outward by the momentum transferred by the very hot gas of the stellar ejecta from the starburst, which creates a fast-moving wind that accelerates the ambient gas. We find a weak positive correlation (ρ = 0.40, 1.69σ significance), which is consistent with what was also found from absorption lines (e.g., Calabrò et al. 2022). This suggests that even though they might trace different regions, they might be triggered by the exact mechanisms and with a similar origin. Interestingly, we find that our best fit shows a shallower slope of 0.06 ± 0.03 (dashed green line) compared to the slope of 0.13 (scaled dashed black line) found in Arribas et al. (2014). Here, we used the instantaneous SFR traced by the narrow component of Hβ (or Hα), assuming the same ratio as observed in [OIII]. Based on the UV absorption lines, a similar trend is observed with a steeper slope of 0.18 (dashed red line) found in the CLASSY survey with local metal-poor galaxies (Xu et al. 2022b).

thumbnail Fig. 13.

Relation between outflow velocity and ΣSFR. Our sample is color-coded by the SFR(Hβ)/SFRSED ratio. Our sample is divided into galaxies with blue- (crosses) and redshifted (pentagons) broad components. The dashed black line corresponds to the best slope from Arribas et al. (2014), and the black squares show their observed data. The dashed green line is our best fit. The dashed red line is the best fit from Xu et al. (2022b). Two galaxies with broad components (CDFS023527 and 511245444) are excluded from this plot because Hα and Hβ are not included in the observed spectral range.

Additionally, we used the SFR(Hβ)/SFRSED ratio as a proxy of the burstiness of the galaxy because the two SFR estimators trace different timescales. The SFR(Hβ) is more sensitive to younger ages. We find a mean SFR(Hβ)/SFRSED ratio of 1.4 with values ranging from 0.4 to 5.1. Half of the sample shows SFR(Hβ) higher than SFRSED. Galaxies with a higher burstiness tend to also show high ΣSFR, which is displayed in the color-codes in Fig. 13.

5. Discussions

5.1. Mass-metallicity and fundamental metallicity relations

The gas-phase metallicity of galaxies encodes information about the physical processes that drive the evolution of galaxies. The emergence of scaling relations, such as the MZR, plays a fundamental role in understanding galaxy formation processes and constraining physical models of galaxy evolution. As galaxies grow, feedback processes shape the MZR because they promote the dispersion and removal of a significant fraction of metals from the star-forming regions into the CGM. The selective loss of newly synthesized heavy elements could be particularly important in low-mass galaxies because their potential wells are shallow (e.g., Tremonti et al. 2004; Andrews & Martini 2013). In addition, the slope and normalization of the MZR may evolve with redshift, with high-z galaxies showing lower metallicity for a given stellar mass (e.g., Sanders et al. 2021). Simulations explain this evolution by invoking the higher gas fractions of galaxies at higher redshifts (Torrey et al. 2019).

In Fig. 14 we show that our subset of galaxies with reliable Hβ luminosities follows the MZR built using MOSDEF main-sequence galaxies at similar redshift by Sanders et al. (2021) and the z ∼ 0 relation for local analogs of similar stellar mass in the CLASSY survey (Berg et al. 2022). Our sample is mostly within 3σ, the intrinsic scatter of these relations. A few galaxies show a lower metallicity. We interpret this apparent offset as a likely selection effect, that is, our sample includes strong emission line galaxies that tend to have a higher SFR, more extreme ionization conditions, and a lower metallicity than the sample we used to establish the MZR that we considered for reference. The high log([OIII]/Hβ)  >  0.6 of our sample is roughly the higher excitation in the lower-mass bin used for the MZR in Sanders et al. (2021). This is also evident in Fig. 8, showing that our sample has higher [OIII]/[OII] ratios than the MOSDEF galaxies that were used to determine the MZR at z ∼ 3. Compared to strong [OIII] emitters at higher redshifts, our galaxies follow a similar trend to that found for a more extreme subset of EoR galaxies (Matthee et al. 2023; Heintz et al. 2022).

thumbnail Fig. 14.

Relation between stellar mass, gas-phase metallicity, and SFR. Top: Mass-metallicity relation: The dashed red line is the MZR of SFGs at z ∼ 3.3 from Sanders et al. (2021) and the shaded red region is the observed 3σ scatter. For comparison, metal-poor local galaxies from Berg et al. (2022) are included as black circles, and the corresponding MZR is shown as the dashed black line. The red squares are SFGs at z > 6.25 from Matthee et al. (2023; stack) and Heintz et al. (2022; single-lensed galaxies). Middle (bottom): Differences between the metallicity from FMR and our metallicity estimates as a function of sSFR (metallicity). The shaded gray region is the 2σ observed scatter of 0.22 dex for the FMR at z ∼ 3.3 (Sanders et al. 2021). In all panels, our sample is color-coded by ΣSFR, and the triangles are the galaxies without broad components. Pentagons and crosses are galaxies in our sample that show red- and blueshifted broad components, respectively.

The scatter of galaxies around the median MZR correlates with the SFR, that is, the FMR (Mannucci et al. 2010; Curti et al. 2020), which implies that at fixed stellar mass, galaxies with higher SFR show lower metallicities. Observationally, the FMR appears to be invariant with redshift up to z ∼ 3.5 (e.g., Curti et al. 2020; Sanders et al. 2021), although low-mass starbursting galaxies may deviate clearly (Amorín et al. 2014; Calabrò et al. 2017). The shape of the FMR is found to be modeled by the age of the stellar populations, with younger (< 150 Myr) and more bursty SF showing lower metallicity (Duarte Puertas et al. 2022). Models and simulations find that the FMR results from the smooth evolution of galaxies in a quasi-equilibrium state, which is regulated by inflows and outflows over time (Lilly et al. 2013; Nelson et al. 2019a). The strength of the implied SFR-metallicity relation is found to depend on feedback via the shape of the star formation history, particularly for low-mass starbursting galaxies (Torrey et al. 2018).

In order to explore the position of our sample in the FMR, we compared the gas-phase metallicities obtained with the expected FMR values. We defined the parameter ΔFMR = log(O/H)−log(O/H)FMR, where log(O/H)FMR is the relation found in Sanders et al. (2021), which depends on the stellar mass and SFR. While most of the galaxies in our sample are consistent (within the uncertainties in the gas-phase metallicity) with the observed 2σ scatter of the relation (middle and bottom panel in Fig. 14), we find a weak negative trend between ΔFMR and specific SFR (sSFR = SFR/M) similar to that shown in galaxies at z > 6.25 with comparable sSFR (e.g., Matthee et al. 2023; Heintz et al. 2022). However, no dependence is found on ΣSFR. Although this trend appears to be independent of the method used to estimate the metallicity, we should take this result with caution as the uncertainties in metallicity are still large.

Some caveats may affect our comparisons. First, the stellar masses in our work were obtained assuming subsolar stellar metallicities, which could lead to differences of up to ∼0.3 dex compared to assuming solar values, as in Sanders et al. (2021), as mentioned in Sect. 3.1. Second, the different methods used to derive metallicities may lead to possible systematic differences. We used a Te-consistent method based on the comparison of UV and optical emission line ratios with predictions from photoionization models. As described in Sect. 4.5, metallicity differences of up to ∼0.3 dex can be found when a sole strong-line calibration is used instead. Acknowledging these caveats, we are interested in the emerging trends of these comparisons, not in absolute values.

In conclusion, our sample follows the MZR at z ∼ 3. A subset shows slightly lower metallicities. These galaxies also show lower metallicities than expected from the FMR, which could be due to the relatively extreme ionization properties of their young and intense SF regions. Recent gas accretion fueling a compact starburst and the intense stellar feedback it produces, that is, inflows and outflows, are physical mechanisms that could naturally drive the observed offset in the MZR and FMR toward lower metallicities. Probing relative abundances (e.g., C/O or N/O) that depend on the SF history of galaxies (e.g., Vincenzo & Kobayashi 2018; Berg et al. 2019) can be a useful tool for exploring this hypothesis.

5.2. The C/O-O/H relation

The C/O abundance may provide general trends in the evolutionary state of a galaxy and its ISM. Models (e.g., Henry et al. 2000; Mollá et al. 2015; Mattsson 2010) and observations of local galaxies (Garnett et al. 1995; Berg et al. 2016, 2019) showed that C/O increases with metallicity for galaxies with Z ≳ 20% solar. This trend can be explained because C is primarily produced by the triple-α process in both massive and low- to intermediate-mass stars, but in massive stars, carbon arises almost exclusively from metallicity-dependent stellar winds, mass loss, and ISM enrichment, which are higher at higher metallicities (Henry et al. 2000). Instead, in younger metal-poor systems, the delayed release of C (mostly produced by low- and intermediate-mass) relative to O (produced almost exclusively by massive stars) appears to be the driver of the observed trend (Garnett et al. 1995). However, the large dispersion of the C/O values over a wide range in metallicity for galaxies at z ∼ 0 − 2 (∼0.2 dex) suggests that the C/O abundance is highly sensitive to other factors, such as the detailed SFH, with longer burst durations and lower star formation efficiencies that correspond to low C/O ratios (Berg et al. 2019).

In Fig. 15 (left panel) we show our galaxy sample in the C/O–O/H plane. We find no apparent increase in C/O with metallicity, as models predict (e.g., Mattsson 2010; Nicholls et al. 2017), but a large scatter of C/O values around a metallicity of ∼10 − 20% solar. Despite the uncertainties in both C/O and O/H, part of this scatter could be physical, as discussed in previous works that reported similar findings at low and high redshifts (e.g., Amorín et al. 2017; Berg et al. 2019; Llerena et al. 2022). Figure 15 shows no clear correlation of the C/O scatter with ΣSFR or sSFR, which may suggest a more local effect on the C/O–O/H relation at low metallicity. This topic will be addressed in a future more specific study. Finally, these results also suggest some caution in using C/O as an indicator of metallicity because it may be subject to large uncertainty and possible selection effects.

thumbnail Fig. 15.

Relation between C/O, O/H and stellar mass. Left panel: C/O–O/H relation. Our sample is color-coded by ΣSFR. The symbols with cyan edges are galaxies with limits in OIII]λ1666. We compare our results with local BCD galaxies (small green circles; Garnett et al. 1995, 1997; Kobulnicky et al. 1997; Kobulnicky & Skillman 1998; Izotov et al. 1999; Thuan et al. 1999; Berg et al. 2016; Senchyna et al. 2021) and HII regions (small blue squares; Garnett et al. 1995, 1999; Kurt et al. 1995; Mattsson 2010; Senchyna et al. 2021). We also show chemical evolution models from the literature as dashed lines with the colors described in the legend (Garnett et al. 1995; Mattsson 2010; Nicholls et al. 2017). The dashed magenta line is the scaled N/O-mass relation in Andrews & Martini (2013) assuming the constant C/N factor based on Berg et al. (2019). The shaded magenta region is the 1σ uncertainty considering the observed scatter in the relation and the conversion factor. Right panel: Relation between C/O and stellar mass. The symbols are the same as in the left panel. In this panel, the dashed black line is the relation presented in Llerena et al. (2022) at z ∼ 3 based on stacking.

On the other hand, the right panel in Fig. 15 shows that our galaxies appear to be consistent with an increase in C/O with stellar mass. In order to compare our results with local galaxies, we re-scaled the N/O-stellar mass relation reported in Andrews & Martini (2013) and assumed a constant C/N conversion (Berg et al. 2019) to obtain the relation shown by the dashed magenta line. Our sample shows a large scatter in C/O for a given stellar mass that is roughly consistent with Llerena et al. (2022) for CIII] emitters at z ∼ 3 using stacking. However, they appear to follow the trend, expected for their stellar mass, suggesting that a fraction of their C/O may have a secondary origin. This is not seen in the O/H–C/O plane. While we do not find a correlation with outflow velocities and ΣSFR, one possible interpretation for these trends is that a recent metal-poor inflow may have produced a dilution of O/H while keeping the C/O as high as expected for their stellar mass. To explore this and other possible interpretations further, larger representative samples over a wide range of mass and metallicity are needed. JWST spectroscopy will certainly help us to reduce the uncertainties in the chemical abundance of high-z galaxies, thus providing a more robust interpretation of the C/O–O/H relation for samples at z≳ 4 (e.g., Arellano-Córdova et al. 2022).

In conclusion, the low metallicity and relatively large C/O abundance of our sample suggest that these galaxies are in a relatively active and early phase of chemical enrichment. The complex interplay between metal content and stellar feedback is discussed in the following sections.

5.3. Outflow properties and their relation with the star formation rate density

One important parameter in understanding the effects of outflows in the properties of galaxies is the mass-loading factor (η = out/SFR), which measures how efficiently outflows remove gas from the galaxy relative to the formation of stars. To be consistent with previous works, we followed the same models as in Concas et al. (2022) to calculate the mass-loading factor η. For a multiconical or spherical outflow and a constant outflow velocity, the mass outflow rate out from the Hα line is given by

(1)

where the factor C depends on the assumed outflow history, Rout is the radius of the outflow, vmax is the maximum outflow velocity defined in Sect. 4.6, and is the mass outflow, which is given by

(2)

where is the dust-corrected luminosity of the outflow (broad) component of Hα. To estimate , we used the dust-corrected Hβ luminosity and assumed the broad-to-narrow flux ratio from [OIII]. We did not use the model-based [OIII] because it depends on the metallicity (see Concas et al. 2022), and we preferred to keep the mass outflow rate independent of the method used to derive metallicity. However, we verified that the values obtained by using the two lines differed by 0.28 dex, with higher out values when the [OIII] line was used. This difference was also discussed in Concas et al. (2022). They reported agreement only when a higher metallicity of 50% solar was assumed.

From Eq. (2), we find the outflow mass ranging from 107.0 to 108.03 M, with a mean value of 107.58 M. Because our data do not allow us a reliable estimate of ne for the outflow component (i.e., using only the broad emission of [OII]λλ3727,3729), we assumed an outflow mean electron density of 380 cm−3, which is based on the stacking of 33 galaxies with SF-driven outflows (Förster Schreiber et al. 2019). This assumption makes our estimate consistent with other works using the same value (e.g., Davies et al. 2019; Concas et al. 2022; Gupta et al. 2023). When we use the mean (global) density of 560 cm−3 obtained for our sample, the resulting η values are 0.16 dex lower, which is lower than the mean uncertainties (0.2 dex). It does not affect our results and conclusions.

For the mass outflow rate, we assumed a constant outflow rate that started at −t = −Rout/vmax, which leads to C = 1, as in other works (Concas et al. 2022). We assumed that the outflow radius Rout is the effective radius measured in the F160W band in WFC3/HST, as described in Sect. 3.3, which has values ranging from 0.2 to 3.6 kpc, with a mean Rout = 1.6 kpc. We assumed that the outflow velocity is the maximum velocity (see Table 5, Sect. 4.6). The mass outflow rates range from 1.4 to 57 M yr−1, with a mean out = 13.3 M yr−1. This is nearly half the mean SFR derived for the sample. As a comparison, our sample agrees with the relation Mout-SFR reported in Avery et al. (2021) at the more extreme SFR values, which indicates that SF is the driver of the outflow. They found a linear slope of out ∝ SFR0.97 for integrated outflows on local MaNGA galaxies using Hα to measure outflows. This relation is also consistent with the values reported in Marasco et al. (2023) for a sample of 19 nearby systems above the local MS, which shows lower values of SFRs and out than our sample. We also note that some of our galaxies show higher values than those found in Xu et al. (2022b) for a sample of local dwarf galaxies using UV absorption lines with similar high SFRs.

In order to estimate η, we considered the instantaneous SFR using Balmer lines, as explained in Sect. 3.3. We report the obtained values in Table 5. We find a mean value of η = 0.54 with values ranging from 0.05 to 3.26. Our mean value is higher than the estimate of η ∼ 0.2 using stacking from a sample of SFGs at z = 3 − 4 with EW(Hβ+[OIII]λ5007) > 600 Å (Gupta et al. 2023) and for a local GP galaxy (Hogarth et al. 2020). On the other hand, our mean value is lower than the typical η of 1.5 for neutral and low-ionization gas of SFGs at z ∼ 3 derived from absorption lines (e.g., Calabrò et al. 2022).

In Fig. 16 (top panel) we display the relation between the mass loading factor and stellar mass for our sample galaxies and compare it with similar theoretical and observational results from the literature. Our galaxies show a very wide range of η values in a limited range of M. While this precludes a rigorous study of possible trends in η with M at z ∼ 3, we find that for a given stellar mass, galaxies show scatter of more than one order of magnitude in η values according to their star formation rate surface density. We find a clear trend in the scatter with galaxies with more compact star formation, that is, ΣSFR≳10 M yr−1 kpc−2, showing higher η for a given M. Instead, galaxies with ΣSFR ⪅ 10 M yr−1 kpc−2 show values that are ∼1 dex lower η on average. While the former appears roughly consistent with the trend predicted by FIRE-2 simulations (dot-dashed lines; Pandya et al. 2021), the latter appears to be more consistent with the scaled trend (by a factor ∼1/141 lower) predicted by the Illustris-TNG simulations (Nelson et al. 2019b).

thumbnail Fig. 16.

Relation between the mass-loading factor, stellar mass and ΣSFR. Top panel: Relation between the mass-loading factor and stellar mass. Our sample is color-coded by ΣSFR. The dashed black line is the rescaled relation found in simulations according to Nelson et al. (2019b). The dot-dashed line is the relation found in Pandya et al. (2021). The small black (magenta) circles are single simulated halos from Pandya et al. (2021) in the redshift range 2–4 (< 0.05). We also include observational results from McQuinn et al. (2019), Concas et al. (2022), and Marasco et al. (2023). Bottom panel: Relation between the mass-loading factor and ΣSFR. The symbols are the same as in the top panel, but our sample is color-coded by EW([OIII]). The red stars are stacks at z = 3 − 4 from Gupta et al. (2023). The dashed green line is the best fit for log(ΣSFR[M yr−1 kpc−2]) > −3 including our sample and the observed galaxies and stacks from literature.

Compared to recent observational studies using a similar two-Gaussian decomposition of strong emission lines, Fig. 16 typically shows higher η values than main-sequence galaxies at cosmic noon (e.g., Concas et al. 2022) and nearby dwarf galaxies (e.g., Marasco et al. 2023). The relative disagreement with simulation predictions (e.g., Nelson et al. 2019b), that is, a dropoff in η at M ≲ 1010M compared to the predicted increasing trend, was interpreted as evidence of possible inefficient feedback efficiency in low-mass galaxies.

However, our results suggest that the compactness of the SF regions in low-mass galaxies, that is, their ΣSFR, is key to determining the impact of outflows and stellar feedback in low-mass systems. Our results agree well with recent findings by McQuinn et al. (2019), who also reported a clear increasing trend for the outflow mass-loading factors of local dwarf galaxies with centrally concentrated star formation (green squares in Fig. 16). The few BCDs included in Marasco et al. (2023) also show comparably higher η values. This is also evident from Fig. 16 (bottom panel), in which we find a clear trend that galaxies with a higher ΣSFR show higher η (best fit in the green line). The resulting trend consistently includes local dwarfs and results from the stacking of EELGs at z ∼ 3 − 4 (Gupta et al. 2023; red stars in Fig. 16).

These results strongly suggest that galaxies of a given stellar mass may have different feedback effects according to their SF surface density, with low-mass galaxies with a higher ΣSFR experiencing more effective feedback. These results differ from simulations, where an opposite trend is found (e.g., Nelson et al. 2019a; Pahl et al. 2023). As discussed in Nelson et al. (2019a), this trend depends on the velocity threshold of outflow particles, and the difference with observations might be explained by small-scale relations that are not considered in simulations (see also McQuinn et al. 2019, for a detailed discussion). Within this context, we note that our sample of galaxies mainly consists of strong [OIII] emitters with high EWs, which push the relation with η toward higher ΣSFR, as shown in Fig. 16. Thus, it appears possible that low-mass galaxies with a moderate SFR and lower emission line EWs are more consistent with literature data and values observed in simulated halos at low-z (z < 0.05).

Interestingly, recent works suggest a relation between the SFR surface density and the escape of ionizing photons (i.e. ΣSFR ∝ fesc), which is intimately related to the ability of stellar feedback, that is, outflows, to clear out young star-forming regions from dust and neutral, thus creating optically thin channels from which ionizing photons may eventually escape (e.g., Naidu et al. 2020; Flury et al. 2022a). Within this context, our results suggest that young low-mass galaxies with strongly mass-loaded outflows, that is, showing broad emission line components, could be clear candidates for favorable conditions for Lyman photon escape.

Recently, the spatially resolved study of the Sunburst arc, a lensed metal-poor galaxy at z = 2.37 showing LyC escape, presented by Mainali et al. (2022), revealed a strong blueshifted broad emission component in [OIII]. Remarkably, the broad-to-narrow ratio in the leaking clump is 120%, whereas for the nonleaking regions, this ratio falls to 35%. When we compare these results with the fB values obtained for our sample and other LyC indirect diagnostics (e.g., Flury et al. 2022a), we find that only a few galaxies with the higher fB and ΣSFR, as well as high [OIII]/[OII], could be considered as our best candidates for LyC leakage.

5.4. Effects of stellar feedback on chemical abundances

We discuss the impact of stellar feedback traced by outflows in the chemical abundances of the host galaxies. We use Fig. 17, in which we plot the oxygen and carbon nebular abundances versus η and β = log(SFRHα/SFRSED), the ratio of the instantaneous SFR, traced by Hα, and the SED-based SFRSED, tracing longer-timescales. β is often considered a burstiness parameter (Scalo 1986; Guo et al. 2016).

thumbnail Fig. 17.

Relation of the chemical abundances (oxygen abundance in the top panels and C/O in the bottom panels) with the mass-loading factor (left panels) and with burstiness (right panels). Our sample is divided into galaxies in which the broad component is blueshifted (blue crosses) or redshifted (red pentagons) or for which no broad component was detected (black squares). The dashed cyan lines are the mean value, and the shaded region is the mean observed scatter. Our sample is color-coded by ΣSFR.

For our sample, the correlation (ρ = −0.15, 0.6σ significance) between metallicity and η is not significant. This suggests that the metallicity is insensitive to the strength of the outflow at global spatial scales. We do not find compelling evidence that younger starbursts (high β) show a lower metallicity (ρ = −0.04, 0.17σ significance). This also suggests that the metallicity is insensitive to the SF timescale at global spatial scales. No correlation is found with C/O neither (ρ ∼ 0, 0.1σ significance). These results are somewhat consistent with the position of our sample in the mass-metallicity-SFR relations and the large scatter they show in the relation between η and M.

However, galaxies with higher η appear to have a weak increasing trend with the C/O abundance, which is slightly higher for galaxies with stronger outflows and denser star formation. This implies a certain level of selective enrichment due to outflows that could be in place in these galaxies. Our best fit is consistent with a weak correlation (ρ = 0.42, 2σ significance) that is not far from the scatter we measure around the mean C/O value for the sample (cyan regions in Fig. 17). For this reason, a more complete analysis of the impact of stellar feedback on the chemical properties of the galaxies will be done in future works.

Finally, we identify in Fig. 17 galaxies with blue- and redshifted broad components to explore potential differences for in(out)flows, but there is no clear distinction among them. We compared them to galaxies without broad emission. While this may suggest that we did not detect these components because of the geometry of the gas flow or the depth of our spectra rather than because of different nebular physical conditions, it may also indicate that a global parameter such as β is not efficient in identifying chemo-dynamical differences in these unresolved galaxies.

Therefore, we conclude that unresolved spectroscopy is likely insufficient to discern between these two possible scenarios in our sample. Spatially resolved spectroscopy is needed to consider the geometry of the ISM, compare it with models, and further explore the connection between gas flows, chemical abundances, and the star formation of galaxies at z ∼ 3, which is now possible using the NIRSpec IFU on board the JWST.

6. Conclusions

We presented a detailed analysis of the chemical abundances and kinematics of the ionized gas of low-mass (107.9–1010.3M) SFGs at z ∼ 3. We used new follow-up NIR spectroscopy for a sample of 35 SFGs selected on the basis of their rest-UV emission line properties (from Lyα to CIII]) from two previous works using ultradeep optical spectra of the VANDELS (Llerena et al. 2022) and VUDS (Amorín et al. 2017) surveys. For VANDELS targets, our sample was assembled from Keck/MOSFIRE spectra of the NIRVANDELS survey (Cullen et al. 2021). For VUDS targets, our sample was assembled from MOSDEF spectra (Kriek et al. 2015) and from new spectra obtained with VLT/X-shooter and Magellan/FIRE.

We focused our analysis of the NIR spectra on strong emission lines in the rest-optical, from [OII]λ3727 to Hα. We characterized the main properties of the sample based on the UV and optical datasets. We discussed scaling relations involving the gas metallicity and C/O abundances of the galaxies, which are derived using Te-consistent methods based on photoionization models and the observed UV and optical emission line ratios. In addition, using the available high-resolution spectra, we performed an analysis of the [OIII]λλ4959,5007 emission line profiles with a multi-Gaussian fitting technique to investigate the ionized gas kinematics of the galaxies and discuss the connection between stellar feedback and chemical enrichment in these young low-mass SFGs. We summarize our main results and conclusions below.

  • According to diagnostic diagrams based on both UV and optical emission line ratios, the dominant source of ionization in our sample of SFGs is massive stars. While 14% of the sample show UV emission line ratios that are closer to those expected from AGN models, we find that their optical line ratios are instead consistent with pure stellar photoionization. Overall, our sample is characterized by high [OIII]/Hβ > 4 ratios, which suggests high-ionization conditions in the ISM.

  • We find that the rest-frame EW(CIII]) ranges from 1 Å to 15 Å and the EW([OIII]) ranges from 102 Å to 1715 Å. We derived positive correlations between the EWs of bright UV and optical emission lines. About 15% of our sample show EW([OIII]λλ4959,5007) > 1000 Å that closely resemble those measured in z > 6 EoR galaxies with photometric data (e.g., Endsley et al. 2021) and, more recently, with JWST spectroscopy (e.g., Matthee et al. 2023).

  • For galaxies with reliable measurements of the OIII]λ1666/[OIII]λ5007 ratio, we find mean electron temperatures Te = 1.8 × 104 K. Consequently, we used the code HCM-UV based on UV photoionization models to consistently derive low gas-phase metallicities and C/O abundances. We find a wide range of metallicity (12+log(O/H) ∼ 7.5–8.5) with a mean value of 12+log(O/H) = 7.91 or 17% solar. Using alternative methods, we find differences of up to ∼0.3 dex toward higher (lower) metallicities when using pure optical (UV) strong-line calibrations, which are larger than typical uncertainties. We also derived a wide range of C/O abundance ratios ranging from log(C/O)  =   − 0.9 to log(C/O) = –0.15 (23% and 128% solar, respectively) with a mean value of log(C/O)  =   − 0.52 (54% solar) that is consistent with previous results for SFGs at z ∼ 3 based on stacking spectra (Shapley et al. 2003; Llerena et al. 2022). The oxygen and carbon abundances for the highest EW galaxies in our sample agree very well with values obtained for galaxies at z > 6 with JWST spectra (Arellano-Córdova et al. 2022; Jones et al. 2023)

  • Our sample follows a mass-metallicity relation with a slope that is consistent with previous work at similar redshifts. The sample shows an offset of about 0.3 dex to lower metallicities, however, which appears to be consistent with the low-metallicity envelope of the MZR scatter (Curti et al. 2020; Sanders et al. 2021). While these differences could be explained by the different methods that were used to estimate the metallicities, we conclude that the high-ionization properties of our sample most likely drive these offsets. Furthermore, we find that for a given stellar mass, galaxies with lower metallicities tend to show stronger deviations from the FMR. These results suggest that our SFGs experience a rapid and active episode of massive star formation in which outflows from stellar feedback and accretion of fresh gas can act as significant regulators of their mass and metal content.

  • From the analysis of the CO-O/H relation, we find no apparent increase in C/O with metallicity, as models predict (e.g., Mattsson 2010; Nicholls et al. 2017), but a large scatter of C/O values around a metallicity of ∼10 − 20% solar. On the other hand, our galaxies appear to be consistent with an increase in C/O with stellar mass, suggesting that a fraction of their C/O may have a secondary origin. One possible interpretation for these trends is that a recent metal-poor inflow may dilute O/H while keeping the C/O as high as expected for their stellar mass. To explore this and other possible interpretations further, larger representative samples over a wide range of mass and metallicity are needed.

  • From a detailed multi-Gaussian component fitting of the [OIII]λλ4959,5007 line profiles, we find that 65% of our galaxies show two distinct kinematic components: a narrow component with an intrinsic velocity dispersion of σN ∼ 57 km s−1 that accounts for the core of the lines, and a broader component with σB ∼ 121 km s−1 that best fits the extended line wings. We find that the broad component is typically blue- or redshifted by ∼30 − 40 km s−1 with respect to the narrow component in most galaxies. Following the close similarities with local analogs, such as the Green Peas (Amorín et al. 2012; Hogarth et al. 2020), we interpret the narrow and broad kinematic components as gas tracing virial motions and turbulent outflowing ionized gas driven by strong star formation, respectively. From our kinematic analysis, we find typical outflow velocities of ∼280 km s−1, which are found to correlate weakly with stellar mass but strongly with the instantaneous SFR traced by Balmer lines and ΣSFR.

  • From our kinematic analysis, we find a mean mass-loading factor η = 0.54 (with a wide range of 0.05–3.26 and a typical uncertainty of 0.3) that is higher than the typical value observed in SFGs at similar redshift. We find galaxies with more compact star formation, that is, ΣSFR≳10 M yr−1 kpc−2, showing higher η for a given M at the stellar mass range covered by our sample (log(M/M) < 10.2). This suggests that for a given stellar mass, denser starbursts in low-mass low-metallicity galaxies produce stronger outflows. This indicates that stellar mass alone, as concluded by some studies at lower redshift, does not necessarily determine how effectively gas is removed by stellar feedback, and that the star formation and ISM densities can regulate this process in low-mass galaxies, as some simulations predict.

Overall, our results suggest a complex interplay between star formation, gas kinematics, and chemical enrichment in relatively young galaxies at z ∼ 3. When observed during a young burst of SF, the ionization properties are extreme, and their chemical abundances are strongly regulated by their significant gas accretion and stellar feedback, which make them outliers of key scaling relations. In this phase, SFGs may show broad emission line components imprinting the turbulent ionized gas that is outflowing from the starbursting regions. While outflows appear to be ubiquitous in the rapid star-forming episodes of low-mass galaxies at z ∼ 3, their role as a regulator of the gas metallicity could be significantly stronger in galaxies developing denser starbursts. Outflows are in turn suggested as an important mechanism to shape the ISM properties and to facilitate the escape of ionizing photons. Because a subsample of galaxies in this work closely resembles EELGs at z > 6, we conclude that the above results suggest that similar findings could be common in galaxies at higher redshifts that are observed with deep JWST high-resolution spectra.


Acknowledgments

We thank the anonymous referee for the detailed review and useful suggestions that helped to improve this paper. This work is based on data products from observations made with ESO Telescopes at La Silla Paranal Observatory under ESO program ID 194.A-2003 (PIs: Laura Pentericci and Ross McLure). This paper includes data gathered with the 6.5 meter Magellan Telescopes located at Las Campanas Observatory, Chile. We thank Ross McLure, Fergus Cullen, and James Dunlop as part of the NIRVANDELS team for providing us with part of the data used in this paper. M.Ll. acknowledges support from the National Agency for Research and Development ANID/Scholarship Program/Doctorado Nacional/2019-21191036. R.A. acknowledges support from ANID FONDECYT Regular Grant 1202007. This work has made extensive use of Python packages astropy (Astropy Collaboration 2018), numpy (Harris et al. 2020), and Matplotlib (Hunter 2007).

References

  1. Amorín, R., Vílchez, J. M., Hägele, G. F., et al. 2012, ApJ, 754, L22 [CrossRef] [Google Scholar]
  2. Amorín, R., Sommariva, V., Castellano, M., et al. 2014, A&A, 568, L8 [CrossRef] [EDP Sciences] [Google Scholar]
  3. Amorín, R., Pérez-Montero, E., Contini, T., et al. 2015, A&A, 578, A105 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  4. Amorín, R., Fontana, A., Pérez-Montero, E., et al. 2017, Nat. Astron., 1, 0052 [Google Scholar]
  5. Andrews, B. H., & Martini, P. 2013, ApJ, 765, 140 [NASA ADS] [CrossRef] [Google Scholar]
  6. Arellano-Córdova, K. Z., Berg, D. A., Chisholm, J., et al. 2022, ApJ, 940, L23 [CrossRef] [Google Scholar]
  7. Arribas, S., Colina, L., Bellocchi, E., Maiolino, R., & Villar-Martín, M. 2014, A&A, 568, A14 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  8. Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481 [NASA ADS] [CrossRef] [Google Scholar]
  9. Astropy Collaboration (Price-Whelan, A. M., et al.) 2018, AJ, 156, 123 [Google Scholar]
  10. Avery, C. R., Wuyts, S., Förster Schreiber, N. M., et al. 2021, MNRAS, 503, 5134 [NASA ADS] [CrossRef] [Google Scholar]
  11. Bacon, R., Conseil, S., Mary, D., et al. 2017, A&A, 608, A1 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  12. Baldwin, J. A., Phillips, M. M., & Terlevich, R. 1981, PASP, 93, 5 [Google Scholar]
  13. Begley, R., Cullen, F., McLure, R. J., et al. 2022, MNRAS, 513, 3510 [NASA ADS] [CrossRef] [Google Scholar]
  14. Berg, D. A., Skillman, E. D., Henry, R. B. C., Erb, D. K., & Carigi, L. 2016, ApJ, 827, 126 [NASA ADS] [CrossRef] [Google Scholar]
  15. Berg, D. A., Erb, D. K., Henry, R. B. C., Skillman, E. D., & McQuinn, K. B. W. 2019, ApJ, 874, 93 [NASA ADS] [CrossRef] [Google Scholar]
  16. Berg, D. A., James, B. L., King, T., et al. 2022, ApJS, 261, 31 [NASA ADS] [CrossRef] [Google Scholar]
  17. Bian, F., Kewley, L. J., & Dopita, M. A. 2018, ApJ, 859, 175 [CrossRef] [Google Scholar]
  18. Bosch, G., Hägele, G. F., Amorín, R., et al. 2019, MNRAS, 489, 1787 [NASA ADS] [CrossRef] [Google Scholar]
  19. Bouché, N., Dekel, A., Genzel, R., et al. 2010, ApJ, 718, 1001 [Google Scholar]
  20. Byler, N., Kewley, L. J., Rigby, J. R., et al. 2020, ApJ, 893, 1 [NASA ADS] [CrossRef] [Google Scholar]
  21. Calabrò, A., Amorín, R., Fontana, A., et al. 2017, A&A, 601, A95 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  22. Calabrò, A., Castellano, M., Pentericci, L., et al. 2021, A&A, 646, A39 [EDP Sciences] [Google Scholar]
  23. Calabrò, A., Pentericci, L., Talia, M., et al. 2022, A&A, 667, A117 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  24. Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682 [NASA ADS] [CrossRef] [Google Scholar]
  25. Cardamone, C. N., van Dokkum, P. G., Urry, C. M., et al. 2010, ApJS, 189, 270 [Google Scholar]
  26. Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245 [Google Scholar]
  27. Carnall, A. C., McLure, R. J., Dunlop, J. S., & Davé, R. 2018, MNRAS, 480, 4379 [Google Scholar]
  28. Castaneda, H. O., Vilchez, J. M., & Copetti, M. V. F. 1990, ApJ, 365, 164 [NASA ADS] [CrossRef] [Google Scholar]
  29. Castellano, M., Pentericci, L., Cupani, G., et al. 2022, A&A, 662, A115 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  30. Chabrier, G. 2003, PASP, 115, 763 [Google Scholar]
  31. Chisholm, J., Tremonti, C. A., Leitherer, C., Chen, Y., & Wofford, A. 2016, MNRAS, 457, 3133 [NASA ADS] [CrossRef] [Google Scholar]
  32. Chisholm, J., Tremonti, C. A., Leitherer, C., & Chen, Y. 2017, MNRAS, 469, 4831 [NASA ADS] [CrossRef] [Google Scholar]
  33. Coil, A. L., Aird, J., Reddy, N., et al. 2015, ApJ, 801, 35 [Google Scholar]
  34. Concas, A., Maiolino, R., Curti, M., et al. 2022, MNRAS, 513, 2535 [NASA ADS] [CrossRef] [Google Scholar]
  35. Cullen, F., McLure, R. J., Dunlop, J. S., et al. 2019, MNRAS, 487, 2038 [Google Scholar]
  36. Cullen, F., Shapley, A. E., McLure, R. J., et al. 2021, MNRAS, 505, 903 [CrossRef] [Google Scholar]
  37. Curti, M., Mannucci, F., Cresci, G., & Maiolino, R. 2020, MNRAS, 491, 944 [Google Scholar]
  38. Curti, M., D’Eugenio, F., Carniani, S., et al. 2023, MNRAS, 518, 425 [Google Scholar]
  39. Davies, R. L., Förster Schreiber, N. M., Übler, H., et al. 2019, ApJ, 873, 122 [Google Scholar]
  40. Dayal, P., Ferrara, A., & Dunlop, J. S. 2013, MNRAS, 430, 2891 [Google Scholar]
  41. Dayal, P., Volonteri, M., Choudhury, T. R., et al. 2020, MNRAS, 495, 3065 [Google Scholar]
  42. De Barros, S., Oesch, P. A., Labbé, I., et al. 2019, MNRAS, 489, 2355 [NASA ADS] [CrossRef] [Google Scholar]
  43. Du, X., Shapley, A. E., Tang, M., et al. 2020, ApJ, 890, 65 [NASA ADS] [CrossRef] [Google Scholar]
  44. Duarte Puertas, S., Vilchez, J. M., Iglesias-Páramo, J., et al. 2022, A&A, 666, A186 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  45. Edmunds, M. G. 1990, MNRAS, 246, 678 [NASA ADS] [Google Scholar]
  46. Endsley, R., Stark, D. P., Chevallard, J., & Charlot, S. 2021, MNRAS, 500, 5229 [Google Scholar]
  47. Erb, D. K., Pettini, M., Shapley, A. E., et al. 2010, ApJ, 719, 1168 [Google Scholar]
  48. Fabozzi, F. J., Focardi, S. M., Rachev, S. T., & Bala, G. 2014, The Basics of Financial 1594 Econometrics: Tools, Concepts, and Asset Management Applications (Wiley) [CrossRef] [Google Scholar]
  49. Finkelstein, S. L., D’Aloisio, A., Paardekooper, J.-P., et al. 2019, ApJ, 879, 36 [Google Scholar]
  50. Flury, S. R., Jaskot, A. E., Ferguson, H. C., et al. 2022a, ApJ, 930, 126 [NASA ADS] [CrossRef] [Google Scholar]
  51. Flury, S. R., Jaskot, A. E., Ferguson, H. C., et al. 2022b, ApJS, 260, 1 [NASA ADS] [CrossRef] [Google Scholar]
  52. Förster Schreiber, N. M., Übler, H., Davies, R. L., et al. 2019, ApJ, 875, 21 [Google Scholar]
  53. Freeman, W. R., Siana, B., Kriek, M., et al. 2019, ApJ, 873, 102 [Google Scholar]
  54. Freudling, W., Romaniello, M., Bramich, D. M., et al. 2013, A&A, 559, A96 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  55. Galametz, A., Grazian, A., Fontana, A., et al. 2013, ApJS, 206, 10 [Google Scholar]
  56. Garilli, B., McLure, R., Pentericci, L., et al. 2021, A&A, 647, A150 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  57. Garnett, D. R., Skillman, E. D., Dufour, R. J., et al. 1995, ApJ, 443, 64 [Google Scholar]
  58. Garnett, D. R., Skillman, E. D., Dufour, R. J., & Shields, G. A. 1997, ApJ, 481, 174 [CrossRef] [Google Scholar]
  59. Garnett, D. R., Shields, G. A., Peimbert, M., et al. 1999, ApJ, 513, 168 [NASA ADS] [CrossRef] [Google Scholar]
  60. Grogin, N. A., Kocevski, D. D., Faber, S. M., et al. 2011, ApJS, 197, 35 [NASA ADS] [CrossRef] [Google Scholar]
  61. Guo, Y., Ferguson, H. C., Giavalisco, M., et al. 2013, ApJS, 207, 24 [NASA ADS] [CrossRef] [Google Scholar]
  62. Guo, Y., Rafelski, M., Faber, S. M., et al. 2016, ApJ, 833, 37 [CrossRef] [Google Scholar]
  63. Gupta, A., Tran, K.-V., Mendel, T., et al. 2023, MNRAS, 519, 980 [Google Scholar]
  64. Gutkin, J., Charlot, S., & Bruzual, G. 2016, MNRAS, 462, 1757 [Google Scholar]
  65. Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357 [Google Scholar]
  66. Hayes, M. 2015, PASA, 32, e027 [NASA ADS] [CrossRef] [Google Scholar]
  67. Heckman, T. M., Alexandroff, R. M., Borthakur, S., Overzier, R., & Leitherer, C. 2015, ApJ, 809, 147 [Google Scholar]
  68. Heintz, K. E., Brammer, G. B., Giménez-Arteaga, C., et al. 2022, arXiv e-prints [arXiv:2212.02890] [Google Scholar]
  69. Henry, R. B. C., Edmunds, M. G., & Köppen, J. 2000, ApJ, 541, 660 [NASA ADS] [CrossRef] [Google Scholar]
  70. Hogarth, L., Amorín, R., Vílchez, J. M., et al. 2020, MNRAS, 494, 3541 [NASA ADS] [CrossRef] [Google Scholar]
  71. Hopkins, P. F., Wetzel, A., Kereš, D., et al. 2018, MNRAS, 480, 800 [NASA ADS] [CrossRef] [Google Scholar]
  72. Hunter, J. D. 2007, Comput. Sci. Eng., 9, 90 [Google Scholar]
  73. Hutchison, T. A., Papovich, C., Finkelstein, S. L., et al. 2019, ApJ, 879, 70 [NASA ADS] [CrossRef] [Google Scholar]
  74. Inoue, A. K., Shimizu, I., Iwata, I., & Tanaka, M. 2014, MNRAS, 442, 1805 [NASA ADS] [CrossRef] [Google Scholar]
  75. Izotov, Y. I., Chaffee, F. H., Foltz, C. B., et al. 1999, ApJ, 527, 757 [NASA ADS] [CrossRef] [Google Scholar]
  76. Izotov, Y. I., Schaerer, D., Worseck, G., et al. 2020, MNRAS, 491, 468 [NASA ADS] [CrossRef] [Google Scholar]
  77. Jakobsen, P., Ferruit, P., Alves de Oliveira, C., et al. 2022, A&A, 661, A80 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  78. Jones, T., Stark, D. P., & Ellis, R. S. 2018, ApJ, 863, 191 [NASA ADS] [CrossRef] [Google Scholar]
  79. Jones, T., Sanders, R., Chen, Y., et al. 2023, ApJ, 951, L17 [CrossRef] [Google Scholar]
  80. Juneau, S., Bournaud, F., Charlot, S., et al. 2014, ApJ, 788, 88 [NASA ADS] [CrossRef] [Google Scholar]
  81. Kakiichi, K., & Gronke, M. 2021, ApJ, 908, 30 [CrossRef] [Google Scholar]
  82. Kauffmann, G., Heckman, T. M., Tremonti, C., et al. 2003, MNRAS, 346, 1055 [Google Scholar]
  83. Kehrig, C., Vílchez, J. M., Pérez-Montero, E., et al. 2016, MNRAS, 459, 2992 [NASA ADS] [CrossRef] [Google Scholar]
  84. Kennicutt, R. C., & Evans, N. J. 2012, ARA&A, 50, 531 [NASA ADS] [CrossRef] [Google Scholar]
  85. Kewley, L. J., Dopita, M. A., Sutherland, R. S., Heisler, C. A., & Trevena, J. 2001, ApJ, 556, 121 [Google Scholar]
  86. Kewley, L. J., Dopita, M. A., Leitherer, C., et al. 2013, ApJ, 774, 100 [NASA ADS] [CrossRef] [Google Scholar]
  87. Kewley, L. J., Nicholls, D. C., & Sutherland, R. S. 2019, ARA&A, 57, 511 [Google Scholar]
  88. Kim, C.-G., Ostriker, E. C., Somerville, R. S., et al. 2020, ApJ, 900, 61 [Google Scholar]
  89. Kobulnicky, H. A., & Skillman, E. D. 1998, ApJ, 497, 601 [NASA ADS] [CrossRef] [Google Scholar]
  90. Kobulnicky, H. A., Skillman, E. D., Roy, J.-R., Walsh, J. R., & Rosa, M. R. 1997, ApJ, 477, 679 [NASA ADS] [CrossRef] [Google Scholar]
  91. Koekemoer, A. M., Faber, S. M., Ferguson, H. C., et al. 2011, ApJS, 197, 36 [NASA ADS] [CrossRef] [Google Scholar]
  92. Kriek, M., Shapley, A. E., Reddy, N. A., et al. 2015, ApJS, 218, 15 [NASA ADS] [CrossRef] [Google Scholar]
  93. Kurt, C. M., Dufour, R. J., Garnett, D. R., et al. 1995, Rev. Mex. Astron. Astrofis. Conf. Ser., 3, 223 [NASA ADS] [Google Scholar]
  94. Laigle, C., McCracken, H. J., Ilbert, O., et al. 2016, ApJS, 224, 24 [Google Scholar]
  95. Le Fèvre, O., Saisse, M., Mancini, D., et al. 2003, SPIE Conf. Ser., 4841, 1670 [Google Scholar]
  96. Le Fèvre, O., Tasca, L. A. M., Cassata, P., et al. 2015, A&A, 576, A79 [Google Scholar]
  97. Le Fèvre, O., Lemaux, B. C., Nakajima, K., et al. 2019, A&A, 625, A51 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  98. Lemaux, B. C., Cucciati, O., Le Fèvre, O., et al. 2022, A&A, 662, A33 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  99. Lilly, S. J., Carollo, C. M., Pipino, A., Renzini, A., & Peng, Y. 2013, ApJ, 772, 119 [NASA ADS] [CrossRef] [Google Scholar]
  100. Llerena, M., Amorín, R., Cullen, F., et al. 2022, A&A, 659, A16 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  101. Luridiana, V., Morisset, C., & Shaw, R. A. 2012, IAU Symp., 283, 422 [NASA ADS] [Google Scholar]
  102. Luridiana, V., Morisset, C., & Shaw, R. A. 2015, A&A, 573, A42 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  103. Lutz, D., Sturm, E., Janssen, A., et al. 2020, A&A, 633, A134 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  104. Ma, X., Kasen, D., Hopkins, P. F., et al. 2015, MNRAS, 453, 960 [NASA ADS] [CrossRef] [Google Scholar]
  105. Ma, X., Quataert, E., Wetzel, A., et al. 2020, MNRAS, 498, 2001 [Google Scholar]
  106. Madau, P., & Dickinson, M. 2014, ARA&A, 52, 415 [Google Scholar]
  107. Madau, P., & Haardt, F. 2015, ApJ, 813, L8 [Google Scholar]
  108. Mainali, R., Zitrin, A., Stark, D. P., et al. 2018, MNRAS, 479, 1180 [NASA ADS] [Google Scholar]
  109. Mainali, R., Rigby, J. R., Chisholm, J., et al. 2022, ApJ, 940, 160 [NASA ADS] [CrossRef] [Google Scholar]
  110. Maiolino, R., & Mannucci, F. 2019, A&ARv, 27, 3 [Google Scholar]
  111. Mannucci, F., Cresci, G., Maiolino, R., Marconi, A., & Gnerucci, A. 2010, MNRAS, 408, 2115 [NASA ADS] [CrossRef] [Google Scholar]
  112. Marasco, A., Belfiore, F., Cresci, G., et al. 2023, A&A, 670, A92 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  113. Maseda, M. V., Brinchmann, J., Franx, M., et al. 2017, A&A, 608, A4 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  114. Matthee, J., Sobral, D., Hayes, M., et al. 2021, MNRAS, 505, 1382 [NASA ADS] [CrossRef] [Google Scholar]
  115. Matthee, J., Mackenzie, R., Simcoe, R. A., et al. 2023, ApJ, 950, 67 [NASA ADS] [CrossRef] [Google Scholar]
  116. Mattsson, L. 2010, A&A, 515, A68 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  117. McLean, I. S., Steidel, C. C., Epps, H. W., et al. 2012, SPIE Conf. Ser., 8446, 84460J [NASA ADS] [Google Scholar]
  118. McLure, R. J., Pentericci, L., Cimatti, A., et al. 2018, MNRAS, 479, 25 [NASA ADS] [Google Scholar]
  119. McQuinn, K. B. W., van Zee, L., & Skillman, E. D. 2019, ApJ, 886, 74 [NASA ADS] [CrossRef] [Google Scholar]
  120. Mingozzi, M., James, B. L., Arellano-Córdova, K. Z., et al. 2022, ApJ, 939, 110 [NASA ADS] [CrossRef] [Google Scholar]
  121. Modigliani, A., Goldoni, P., Royer, F., et al. 2010, SPIE Conf. Ser., 7737, 773728 [Google Scholar]
  122. Mollá, M., García-Vargas, M. L., & Bressan, A. 2009, MNRAS, 398, 451 [CrossRef] [Google Scholar]
  123. Mollá, M., Cavichia, O., Gavilán, M., & Gibson, B. K. 2015, MNRAS, 451, 3693 [CrossRef] [Google Scholar]
  124. Muratov, A. L., Kereš, D., Faucher-Giguère, C.-A., et al. 2015, MNRAS, 454, 2691 [NASA ADS] [CrossRef] [Google Scholar]
  125. Naidu, R. P., Tacchella, S., Mason, C. A., et al. 2020, ApJ, 892, 109 [NASA ADS] [CrossRef] [Google Scholar]
  126. Naidu, R. P., Matthee, J., Oesch, P. A., et al. 2022, MNRAS, 510, 4582 [CrossRef] [Google Scholar]
  127. Nakajima, K., Schaerer, D., Le Fèvre, O., et al. 2018, A&A, 612, A94 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  128. Nakajima, K., Ouchi, M., Xu, Y., et al. 2022, ApJS, 262, 3 [CrossRef] [Google Scholar]
  129. Nelson, D., Springel, V., Pillepich, A., et al. 2019a, Comput. Astrophys. Cosmol., 6, 2 [Google Scholar]
  130. Nelson, D., Pillepich, A., Springel, V., et al. 2019b, MNRAS, 490, 3234 [Google Scholar]
  131. Newville, M., Stensitzki, T., Allen, D. B., et al. 2016, Astrophysics Source Code Library [record ascl:1606.014] [Google Scholar]
  132. Nicholls, D. C., Sutherland, R. S., Dopita, M. A., Kewley, L. J., & Groves, B. A. 2017, MNRAS, 466, 4403 [Google Scholar]
  133. Pahl, A. J., Shapley, A., Steidel, C. C., et al. 2023, MNRAS, 521, 3247 [NASA ADS] [CrossRef] [Google Scholar]
  134. Pandya, V., Fielding, D. B., Anglés-Alcázar, D., et al. 2021, MNRAS, 508, 2979 [NASA ADS] [CrossRef] [Google Scholar]
  135. Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ, 124, 266 [Google Scholar]
  136. Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2010, AJ, 139, 2097 [Google Scholar]
  137. Pentericci, L., McLure, R. J., Garilli, B., et al. 2018, A&A, 616, A174 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  138. Pérez-Montero, E. 2017, PASP, 129, 043001 [CrossRef] [Google Scholar]
  139. Pérez-Montero, E., & Amorín, R. 2017, MNRAS, 467, 1287 [Google Scholar]
  140. Pérez-Montero, E., & Contini, T. 2009, MNRAS, 398, 949 [CrossRef] [Google Scholar]
  141. Pérez-Montero, E., Amorín, R., Sánchez Almeida, J., et al. 2021, MNRAS, 504, 1237 [CrossRef] [Google Scholar]
  142. Pérez-Montero, E., Amorín, R., Pérez-Díaz, B., Vílchez, J. M., & García-Benito, R. 2023, MNRAS, 521, 1556 [CrossRef] [Google Scholar]
  143. Reddy, N. A., Kriek, M., Shapley, A. E., et al. 2015, ApJ, 806, 259 [NASA ADS] [CrossRef] [Google Scholar]
  144. Reddy, N. A., Sanders, R. L., Shapley, A. E., et al. 2023a, ApJ, 951, 56 [NASA ADS] [CrossRef] [Google Scholar]
  145. Reddy, N. A., Topping, M. W., Sanders, R. L., Shapley, A. E., & Brammer, G. 2023b, ApJ, 952, 167 [CrossRef] [Google Scholar]
  146. Ribeiro, B., Le Fèvre, O., Tasca, L. A. M., et al. 2016, A&A, 593, A22 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  147. Ribeiro, B., Le Fèvre, O., Cassata, P., et al. 2017, A&A, 608, A16 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  148. Robertson, B. E., Ellis, R. S., Furlanetto, S. R., & Dunlop, J. S. 2015, ApJ, 802, L19 [Google Scholar]
  149. Rodríguez del Pino, B., Arribas, S., Piqueras López, J., Villar-Martín, M., & Colina, L. 2019, MNRAS, 486, 344 [CrossRef] [Google Scholar]
  150. Sanders, R. L., Shapley, A. E., Kriek, M., et al. 2016, ApJ, 816, 23 [Google Scholar]
  151. Sanders, R. L., Shapley, A. E., Jones, T., et al. 2021, ApJ, 914, 19 [CrossRef] [Google Scholar]
  152. Santini, P., Fontana, A., Castellano, M., et al. 2017, ApJ, 847, 76 [NASA ADS] [CrossRef] [Google Scholar]
  153. Saxena, A., Pentericci, L., Ellis, R. S., et al. 2022a, MNRAS, 511, 120 [NASA ADS] [CrossRef] [Google Scholar]
  154. Saxena, A., Cryer, E., Ellis, R. S., et al. 2022b, MNRAS, 517, 1098 [Google Scholar]
  155. Scalo, J. M. 1986, Fund Cosmic Phys., 11, 1 [NASA ADS] [Google Scholar]
  156. Schaerer, D., Izotov, Y. I., Verhamme, A., et al. 2016, A&A, 591, L8 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  157. Schaerer, D., Izotov, Y. I., Worseck, G., et al. 2022a, A&A, 658, L11 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  158. Schaerer, D., Marques-Chaves, R., Barrufet, L., et al. 2022b, A&A, 665, L4 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  159. Senchyna, P., Stark, D. P., Charlot, S., et al. 2021, MNRAS, 503, 6112 [NASA ADS] [CrossRef] [Google Scholar]
  160. Shapley, A. E., Steidel, C. C., Pettini, M., & Adelberger, K. L. 2003, ApJ, 588, 65 [Google Scholar]
  161. Simcoe, R. A., Burgasser, A. J., Bochanski, J. J., et al. 2010, SPIE Conf. Ser., 7735, 773514 [NASA ADS] [Google Scholar]
  162. Smit, R., Bouwens, R. J., Labbé, I., et al. 2014, ApJ, 784, 58 [NASA ADS] [CrossRef] [Google Scholar]
  163. Stark, D. P., Walth, G., Charlot, S., et al. 2015, MNRAS, 454, 1393 [Google Scholar]
  164. Stark, D. P., Ellis, R. S., Charlot, S., et al. 2017, MNRAS, 464, 469 [NASA ADS] [CrossRef] [Google Scholar]
  165. Steidel, C. C., Adelberger, K. L., Shapley, A. E., et al. 2003, ApJ, 592, 728 [Google Scholar]
  166. Steidel, C. C., Erb, D. K., Shapley, A. E., et al. 2010, ApJ, 717, 289 [Google Scholar]
  167. Steidel, C. C., Strom, A. L., Pettini, M., et al. 2016, ApJ, 826, 159 [NASA ADS] [CrossRef] [Google Scholar]
  168. Steidel, C. C., Bogosavljević, M., Shapley, A. E., et al. 2018, ApJ, 869, 123 [Google Scholar]
  169. Storey, P. J., & Zeippen, C. J. 2000, MNRAS, 312, 813 [NASA ADS] [CrossRef] [Google Scholar]
  170. Sun, F., Egami, E., Pirzkal, N., et al. 2022a, ApJ, accepted [arXiv:2209.03374] [Google Scholar]
  171. Sun, F., Egami, E., Pirzkal, N., et al. 2022b, ApJ, 936, L8 [NASA ADS] [CrossRef] [Google Scholar]
  172. Tang, M., Stark, D. P., Chevallard, J., & Charlot, S. 2019, MNRAS, 489, 2572 [NASA ADS] [CrossRef] [Google Scholar]
  173. Tang, M., Stark, D. P., Chevallard, J., et al. 2021, MNRAS, 501, 3238 [NASA ADS] [CrossRef] [Google Scholar]
  174. Tang, M., Stark, D. P., Chen, Z., et al. 2023, MNRAS, submitted [arXiv:2301.07072] [Google Scholar]
  175. Tasca, L. A. M., Le Fèvre, O., Ribeiro, B., et al. 2017, A&A, 600, A110 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  176. Thuan, T. X., Izotov, Y. I., & Foltz, C. B. 1999, ApJ, 525, 105 [NASA ADS] [CrossRef] [Google Scholar]
  177. Tody, D. 1986, SPIE Conf. Ser., 627, 733 [Google Scholar]
  178. Torrey, P., Vogelsberger, M., Hernquist, L., et al. 2018, MNRAS, 477, L16 [Google Scholar]
  179. Torrey, P., Vogelsberger, M., Marinacci, F., et al. 2019, MNRAS, 484, 5587 [NASA ADS] [Google Scholar]
  180. Tremonti, C. A., Heckman, T. M., Kauffmann, G., et al. 2004, ApJ, 613, 898 [Google Scholar]
  181. Trump, J. R., Arrabal Haro, P., Simons, R. C., et al. 2023, ApJ, 945, 35 [NASA ADS] [CrossRef] [Google Scholar]
  182. Übler, H., Förster Schreiber, N. M., van der Wel, A., et al. 2022, MNRAS, submitted [arXiv:2210.03106] [Google Scholar]
  183. van der Wel, A., Bell, E. F., Häussler, B., et al. 2012, ApJS, 203, 24 [NASA ADS] [CrossRef] [Google Scholar]
  184. van der Wel, A., Franx, M., van Dokkum, P. G., et al. 2014, ApJ, 788, 28 [Google Scholar]
  185. Vernet, J., Dekker, H., D’Odorico, S., et al. 2011, A&A, 536, A105 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  186. Vincenzo, F., & Kobayashi, C. 2018, MNRAS, 478, 155 [NASA ADS] [CrossRef] [Google Scholar]
  187. Weiner, B. J., Coil, A. L., Prochaska, J. X., et al. 2009, ApJ, 692, 187 [Google Scholar]
  188. Xu, X., Henry, A., Heckman, T., et al. 2022a, ApJ, 933, 202 [NASA ADS] [CrossRef] [Google Scholar]
  189. Xu, X., Heckman, T., Henry, A., et al. 2022b, ApJ, 933, 222 [NASA ADS] [CrossRef] [Google Scholar]
  190. Yang, Y., Zabludoff, A., Jahnke, K., & Davé, R. 2014, ApJ, 793, 114 [NASA ADS] [CrossRef] [Google Scholar]

Appendix A: Spectra of the sample

In Fig. A.1 and A.2 we show the spectra of the galaxies for the C3-VANDELS and C3-VUDS samples, respectively. We highlight the spectral regions including the emission lines Lyα, CIVλλ1548,51, HeIIλ1640, OIII]λλ1661,66, CIII]λλ1907,09, [OII]λλ3727,29, Hβ, [OIII]λλ4959,5007, and Hα.

thumbnail Fig. A.1.

Rest-frame spectra of the galaxies in the C3-VANDELS subsample. From the left to right panels, we highlight the following emission lines (which are marked with the dashed black line): Lyα, CIVλλ1548,51, HeIIλ1640, OIII]λλ1661,66, CIII]λλ1907,09, [OII]λλ3727,3729, Hβ, [OIII]λλ4959,5007, and Hα. The flux density is in arbitrary units. Each galaxy is shown with the same color in each panel, and its ID is plotted in the [OIII]λλ4959,5007 panel.

thumbnail Fig. A.2.

Same as in Fig. A.1, but for the galaxies in the C3-VUDS sample.

Appendix B: Plot of the SED fitting

In Fig. B.1 and B.2 we display the photometry used and the SED model for each galaxy in the C3-VANDELS and C3-VUDS samples, respectively.

thumbnail Fig. B.1.

SED model for the galaxies in the C3-VANDELS sample. The red squares are the photometric points, and the solid black line is the resulting spectrum from the SED fitting using BAGPIPES, as described in Sec. 3.1.

thumbnail Fig. B.2.

Same as in Fig. B.1, but for the galaxies in the C3-VUDS sample.

Appendix C: HST imaging of the C3-VUDS sample

In Fig. C.1 and C.2 we show the HST images of the C3-VUDS sample for the galaxies with ΔBIC> 2 and with ΔBIC< 2, respectively.

thumbnail Fig. C.1.

HST/F160W images (Koekemoer et al. 2011) of the C3-VUDS sample with ΔBIC> 2, i.e., the subsample of galaxies with a broad component in their [OIII] profile. The images trace the rest-optical. The white contour is the 3σ level. The physical scale of 0.5 arcsec at their redshift is shown left of each image, and on the right, the effective radius is shown. The galaxies with only i-band HST/F814W are labeled. The galaxies marked with a black square show two narrow components in their [OIII] profile.

thumbnail Fig. C.2.

Same as in Fig. C.1, but for the C3-VUDS sample without features of a broad component in their [OIII] profile.

All Tables

Table 1.

Coordinates and spectroscopic redshift of the final sample.

Table 2.

Main physical properties of the sample based on SED fitting.

Table 3.

Observed fluxes and rest-frame EWs of the rest-UV emission lines.

Table 4.

Observed fluxes and rest-frame EWs of the rest-optical emission lines of our sample.

Table 5.

Kinematic properties of the ionized gas based on [OIII] modeling.

Table 6.

Electron density, temperature, and chemical abundances estimated for our sample.

All Figures

thumbnail Fig. 1.

Sample distributed along the star-forming main sequence at z ∼ 3 color-coded by EW([OIII]). The 2D histogram corresponds to the VANDELS parent sample at the same redshift range, and the dashed black line is the main sequence according to Santini et al. (2017). The magenta crosses and cyan triangles are reference samples at low (Berg et al. 2022) and intermediate redshifts (Maseda et al. 2017; Tang et al. 2021), respectively (see Sect. 3.1).

In the text
thumbnail Fig. 2.

Best-fit of the [OIII] profiles from the C3-VANDELS sample with ΔBIC  >  2. The panels show single galaxies ordered by EW(CIII]) from left to right and top to bottom. At the top in each panel, we show the 2D spectrum with the detected lines. In the middle, we plot the models for [OIII]λ4959 (left) [OIII]λ5007 (right). The blue line shows the observed spectrum, and the red line shows the error spectrum. The Gaussian lines are normalized to the intensity peak of [OIII]λ5007. The dashed black lines are the narrow and broad components, and the magenta line is the global fit considering both components. The magenta-shaded region is the 3σ uncertainty of the fit. The green line is the single-Gaussian fit. The vertical gray line marks the systemic velocity traced by the peak intensity of the narrow component. The vertical blue line marks the peak intensity of the broad component. Bottom: The residuals (Δσ) for each model are shown with the same colors. The gray-shaded regions are masked regions due to sky residuals. The galaxies in the black square show two narrow components.

In the text
thumbnail Fig. 3.

Same as in Fig. 2, but for the C3-VUDS sample with ΔBIC > 2.

In the text
thumbnail Fig. 4.

HST/F160W images (Koekemoer et al. 2011) of the C3-VANDELS sample with ΔBIC> 2, i.e., the subsample with a broad component in the [OIII] profile. The images trace the rest-optical. The white contour is the 3σ level. The physical scale of 0.5 arcsec at their redshift is shown the left of each image, and the effective radius is shown on the right. The galaxy marked with a black square shows two narrow components in its [OIII] profile.

In the text
thumbnail Fig. 5.

Same as in Fig. 4, but for the C3-VANDELS sample without the features of a broad component in their [OIII] profile.

In the text
thumbnail Fig. 6.

UV diagnostic diagrams for our sample based on the EWs of CIII] (left) and CIV (right). In both panels, our sample is color-coded by EW([OIII]), and the dashed black lines are the demarcation between AGN (on the left) and SF (on the right) according to Nakajima et al. (2018). In the right panel, the symbols with magenta edges are galaxies classified as AGN according to the EW(CIII])–CIII]/HeII diagram.

In the text
thumbnail Fig. 7.

Optical diagnostic diagrams for our sample. Left: Classical BPT diagram (Baldwin et al. 1981) for our subsample at z < 3, color-coded by EW(CIII]). The dashed black lines are the typical local AGN/SF demarcation lines (Kewley et al. 2001; Kauffmann et al. 2003). The dashed red line is the demarcation at z ∼ 3 according to Kewley et al. (2013). Right: Mass-excitation diagram for our entire sample color-coded by EW(CIII]). The dashed black and red lines are the AGN/SF demarcation at low-z (Juneau et al. 2014) and z ∼ 2.3 (Coil et al. 2015), respectively. As a reference, we include the stacks from Cullen et al. (2021). Our VANDELS subsample is a subset of these stacks. In both panels, the symbols with magenta edges are galaxies classified as AGN according to the UV diagnostic diagrams in Fig. 6.

In the text
thumbnail Fig. 8.

Variation in [OIII]/[OII] with ΣSFR. Individual galaxies with detected [OII] are shown with red and blue symbols for the C3-VUDS and C3-VANDELS samples, respectively. We also include lower limits based on the upper limits on [OII]. The small green and magenta circles are galaxies from Reddy et al. (2023a) at z = 1.6 − 2.6 and Reddy et al. (2023b) at z = 2.7 − 6.3, respectively. The solid black line is the relation presented in Reddy et al. (2023a), and its extrapolation up to ΣSFR = 100 M yr−1 kpc−2 is plotted as the dashed black line.

In the text
thumbnail Fig. 9.

Relation between EW([OIII]λλ4959,5007+Hβ) and stellar mass. Our sample is color-coded by EW(CIII]). The small red and magenta circles are literature samples at intermediate redshifts from Maseda et al. (2017) and Tang et al. (2021), respectively. The dashed black line is the best fit (slope −0.34) with our data, and the gray shaded region is the 1σ observed scatter of 0.33 dex.

In the text
thumbnail Fig. 10.

Relation between EW([OIII]λλ4959,5007+Hβ) and EW(CIII]). Our sample is color-coded by EW(Lyα). The square symbols with cyan edges are galaxies in the C3-VANDELS at z < 3, for which Lyα is not in the spectral range. The small red and magenta circles are literature samples at intermediate redshifts from Maseda et al. (2017) and Tang et al. (2021), respectively. The dashed black line is the best fit (slope 0.68) with our data, and the gray shaded region is the 1σ observed scatter of 0.33 dex.

In the text
thumbnail Fig. 11.

Relation between gas-phase metallicity and EW(CIII]) for our sample. The dashed blue line and shaded blue region correspond to the relation found in Mingozzi et al. (2022) at low z and their observed 2σ scatter. The dashed red line is our best fit, and the observed 2σ scatter is the shaded red region. The sample is color-coded by OIII] flux when this line is detected at an S/N > 2. In the other cases, only error bars are shown.

In the text
thumbnail Fig. 12.

Distribution of the velocity shift ΔvB of the broad component in our sample. The dashed black and red lines mark the mean values when the broad component is blue- and redshifted, respectively.

In the text
thumbnail Fig. 13.

Relation between outflow velocity and ΣSFR. Our sample is color-coded by the SFR(Hβ)/SFRSED ratio. Our sample is divided into galaxies with blue- (crosses) and redshifted (pentagons) broad components. The dashed black line corresponds to the best slope from Arribas et al. (2014), and the black squares show their observed data. The dashed green line is our best fit. The dashed red line is the best fit from Xu et al. (2022b). Two galaxies with broad components (CDFS023527 and 511245444) are excluded from this plot because Hα and Hβ are not included in the observed spectral range.

In the text
thumbnail Fig. 14.

Relation between stellar mass, gas-phase metallicity, and SFR. Top: Mass-metallicity relation: The dashed red line is the MZR of SFGs at z ∼ 3.3 from Sanders et al. (2021) and the shaded red region is the observed 3σ scatter. For comparison, metal-poor local galaxies from Berg et al. (2022) are included as black circles, and the corresponding MZR is shown as the dashed black line. The red squares are SFGs at z > 6.25 from Matthee et al. (2023; stack) and Heintz et al. (2022; single-lensed galaxies). Middle (bottom): Differences between the metallicity from FMR and our metallicity estimates as a function of sSFR (metallicity). The shaded gray region is the 2σ observed scatter of 0.22 dex for the FMR at z ∼ 3.3 (Sanders et al. 2021). In all panels, our sample is color-coded by ΣSFR, and the triangles are the galaxies without broad components. Pentagons and crosses are galaxies in our sample that show red- and blueshifted broad components, respectively.

In the text
thumbnail Fig. 15.

Relation between C/O, O/H and stellar mass. Left panel: C/O–O/H relation. Our sample is color-coded by ΣSFR. The symbols with cyan edges are galaxies with limits in OIII]λ1666. We compare our results with local BCD galaxies (small green circles; Garnett et al. 1995, 1997; Kobulnicky et al. 1997; Kobulnicky & Skillman 1998; Izotov et al. 1999; Thuan et al. 1999; Berg et al. 2016; Senchyna et al. 2021) and HII regions (small blue squares; Garnett et al. 1995, 1999; Kurt et al. 1995; Mattsson 2010; Senchyna et al. 2021). We also show chemical evolution models from the literature as dashed lines with the colors described in the legend (Garnett et al. 1995; Mattsson 2010; Nicholls et al. 2017). The dashed magenta line is the scaled N/O-mass relation in Andrews & Martini (2013) assuming the constant C/N factor based on Berg et al. (2019). The shaded magenta region is the 1σ uncertainty considering the observed scatter in the relation and the conversion factor. Right panel: Relation between C/O and stellar mass. The symbols are the same as in the left panel. In this panel, the dashed black line is the relation presented in Llerena et al. (2022) at z ∼ 3 based on stacking.

In the text
thumbnail Fig. 16.

Relation between the mass-loading factor, stellar mass and ΣSFR. Top panel: Relation between the mass-loading factor and stellar mass. Our sample is color-coded by ΣSFR. The dashed black line is the rescaled relation found in simulations according to Nelson et al. (2019b). The dot-dashed line is the relation found in Pandya et al. (2021). The small black (magenta) circles are single simulated halos from Pandya et al. (2021) in the redshift range 2–4 (< 0.05). We also include observational results from McQuinn et al. (2019), Concas et al. (2022), and Marasco et al. (2023). Bottom panel: Relation between the mass-loading factor and ΣSFR. The symbols are the same as in the top panel, but our sample is color-coded by EW([OIII]). The red stars are stacks at z = 3 − 4 from Gupta et al. (2023). The dashed green line is the best fit for log(ΣSFR[M yr−1 kpc−2]) > −3 including our sample and the observed galaxies and stacks from literature.

In the text
thumbnail Fig. 17.

Relation of the chemical abundances (oxygen abundance in the top panels and C/O in the bottom panels) with the mass-loading factor (left panels) and with burstiness (right panels). Our sample is divided into galaxies in which the broad component is blueshifted (blue crosses) or redshifted (red pentagons) or for which no broad component was detected (black squares). The dashed cyan lines are the mean value, and the shaded region is the mean observed scatter. Our sample is color-coded by ΣSFR.

In the text
thumbnail Fig. A.1.

Rest-frame spectra of the galaxies in the C3-VANDELS subsample. From the left to right panels, we highlight the following emission lines (which are marked with the dashed black line): Lyα, CIVλλ1548,51, HeIIλ1640, OIII]λλ1661,66, CIII]λλ1907,09, [OII]λλ3727,3729, Hβ, [OIII]λλ4959,5007, and Hα. The flux density is in arbitrary units. Each galaxy is shown with the same color in each panel, and its ID is plotted in the [OIII]λλ4959,5007 panel.

In the text
thumbnail Fig. A.2.

Same as in Fig. A.1, but for the galaxies in the C3-VUDS sample.

In the text
thumbnail Fig. B.1.

SED model for the galaxies in the C3-VANDELS sample. The red squares are the photometric points, and the solid black line is the resulting spectrum from the SED fitting using BAGPIPES, as described in Sec. 3.1.

In the text
thumbnail Fig. B.2.

Same as in Fig. B.1, but for the galaxies in the C3-VUDS sample.

In the text
thumbnail Fig. C.1.

HST/F160W images (Koekemoer et al. 2011) of the C3-VUDS sample with ΔBIC> 2, i.e., the subsample of galaxies with a broad component in their [OIII] profile. The images trace the rest-optical. The white contour is the 3σ level. The physical scale of 0.5 arcsec at their redshift is shown left of each image, and on the right, the effective radius is shown. The galaxies with only i-band HST/F814W are labeled. The galaxies marked with a black square show two narrow components in their [OIII] profile.

In the text
thumbnail Fig. C.2.

Same as in Fig. C.1, but for the C3-VUDS sample without features of a broad component in their [OIII] profile.

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.