CLEAR: Boosted Lyα Transmission of the Intergalactic Medium in UV-bright Galaxies

The CLEAR Experiment is a cycle 23 HST observing program (Program GO-14227, PI: C. Papovich), which observed 12 fields in the CANDELS GOODS fields to 10 to 12 orbit depth with the G102 grism in the HST/WFC3 camera. Each field was observed at three position angles (separated by >10 degrees) to properly correct galaxy spectra from contamination. The goal of the CLEAR survey is to measure the distribution of Lyα emission in galaxies during the epoch of reionization at 6.0 < z < 8.2. The feasibility of detecting Lyα emission from HST/WFC3 grism observations has been proved in previous studies, such as Schmidt et al. (2016) from the Grism Lens-Amplified Survey from Space (Treu et al. 2015), and Tilvi et al. (2016) and Larson et al. (2018) from the Faint Infrared Grism Survey (Pirzkal et al. 2017). The data set also provides constraints on the stellar populations of 1 < z < 2 galaxies. Previous work on the metallicities, ages, and formation histories of massive galaxies at 1 < z < 2 has been published in Estrada-Carpenter et al. (2019, 2020), and the gas-phase metallicity gradients of star-forming galaxies are investigated in Simons et al. (2021). Also, Cleri et al. (2020) studied Paschen-β as a star-formation-rate indicator in low-redshift galaxies, using the CLEAR data set.

The data set has been extended to include all publicly available HST/WFC3 G102 and G141 grism observations in the CLEAR fields. For processing the grism observations, the grism redshift and line analysis software Grizli (Brammer 2019) have been utilized; the software retrieves the raw observations and performs astrometric alignment, contamination modeling, extracting spectra, and fitting continuum and emission-line models. The full details of the grism data reduction and spectral extractions are described in Estrada-Carpenter et al. (2019, 2020) as well as Simons et al. (2021).

While the primary CLEAR spectral extractions were made for galaxies based on the 3D-HST GOODS catalog (Skelton et al. 2014), we extracted the CLEAR spectra of high-z galaxies based on the updated HST CANDELS photometry and its segmentation maps from Finkelstein et al. (2022). The photometric selection of high-z candidate galaxies was done following the criteria described in Section 3.2 in Finkelstein et al. (2015). In brief, the selection is based on the full photometric redshift probability distribution functions of P(z) calculated by EAZY (Brammer et al. 2008) rather than simply using the best-fit redshifts. This includes, for example, the integral of P(z) under the primary redshift peak must be >70% of the total P(z) (for the full details, refer to Finkelstein et al. 2015). Additionally, visual inspection was performed for removing any artifacts (e.g., diffraction spikes, nearby bright sources) and checking the quality of photometry. This resulted in 180, 6.0 < z_phot < 8.2 galaxies in the CLEAR fields. However, checking the quality of the extracted spectra, in total 148 CLEAR spectra were collected after removing 32 spectra due to technical issues in treating the removal and corrections from contaminating spectrum from nearby galaxies using Grizli (e.g., we removed objects where residuals from contaminating spectra from nearby bright galaxies were too severe compared to the expected continuum or line flux for our objects).

In order to search Lyα emission lines or continuum objects detected from these high-z targets, we utilized the grism spectroscopy from the full sample of the CLEAR spectra of 148, 6.0 < z_phot < 8.2 galaxies. In this subsection, we describe the process we used to to determine if any emission lines or continuum spectra are present in the spectra of each galaxy.

2.2.1. Emission-line Search

2.2.2. Continuum-detection Search

The Texas Spectroscopic Search for Lyα Emission at the End of Reionization Survey (Jung et al. 2018, 2019, 2020) obtained deep spectroscopic data of over 120 z > 6 galaxies in the GOODS-N fields (65 galaxies from DEIMOS and 62 galaxies from MOSFIRE observations). These data provided the largest number of confirmed Lyα emitters at z > 7 with which to constrain the IGM H i fraction at z ∼ 7.6 (Jung et al. 2020). Using these data, Jung et al. (2020) noted tentative evidence that the Lyα EW distribution at z > 7 depends on the galaxies’ UV luminosities. Here we use the CLEAR sample to explore this finding, as the CLEAR data provides constraints on Lyα emission for a larger sample of faint galaxies.

CLEAR aims to measure the Lyα EW distribution during the reionization era at z > 6. Also, CLEAR is a blind survey with slitless spectroscopy, providing galaxy spectra for a wide dynamic range of UV magnitudes. Even nondetections in CLEAR are still highly constraining, placing upper limits on the Lyα visibility (Treu et al. 2012; Jung et al. 2018; Mason et al. 2019). The combination of the CLEAR data and Keck data then provided important, complimentary data. This allows us to explore the evolution of the Lyα EW distribution and its dependence on the UV absolute magnitude (M_UV) at higher significance than possible with either data set separately.

We derived the properties of our 148 sources in CLEAR by performing spectral energy distribution (SED) fittings with the Bruzual & Charlot (2003) stellar population model, following Jung et al. (2019, 2020). We utilized the HST ACS and WFC3 broadband photometry (B₄₃₅, V₆₀₆, i₇₇₅, I₈₁₄, z₈₅₀, Y₁₀₅, J₁₂₅, JH₁₄₀, and H₁₆₀) in addition to Spitzer Infrared Array Camera (IRAC) 3.6 μm and 4.5 μm band fluxes. Here, we use photometric redshift measurements that have been updated in Finkelstein et al. (2022) based on updated CANDELS photometry, including deeper I₈₁₄ and Spitzer/IRAC data (where those authors utilized a deblending technique to measure IRAC photometry more accurately using the HST images as priors). A Salpeter (1955) initial mass function was assumed with a stellar mass range of 0.1–100 M_⊙, and we allowed a metallicity range of 0.005–1.0 Z_⊙. We explored a range of exponential models of star formation histories with exponentially varying timescales (τ = 10 Myr, 100 Myr, 1 Gyr, 10 Gyr, 100 Gyr, −300 Myr, −1 Gyr. −10 Gyr). We use the Calzetti (2001) dust attenuation curve with E(B − V) values ranging from 0 to 0.8 for modeling galaxy spectra and add nebular emission lines, adopting the Inoue (2011) emission-line ratio, as done in Salmon et al. (2015). The IGM attenuation was applied to the SED models based on Madau (1995).

Physical properties were obtained from the best-fit models, minimizing χ². The uncertainties of physical parameters were estimated from SED fitting with 1000 Monte Carlo (MC) realizations of the perturbed photometric fluxes, according to their photometric errors for individual galaxies. UV magnitudes (M_UV) of galaxies were calculated from the averaged fluxes over a 100 Å-bandpass (at 1450–1550 Å) from the best-fit models, which were not dust-corrected. In our SED fitting, we fixed redshifts with the best-fit photometric redshifts calculated by EAZY.

In the left panel of Figure 1 we present the M_UV distribution of our CLEAR targets as a function of the redshift with the existing Keck DEIMOS and MOSFIRE targets from Jung et al. (2018, 2020). The black open circles show the CLEAR sample, and the blue and red filled circles denote targets from DEIMOS and MOSFIRE observations, respectively. The CLEAR galaxy sample does not show any noticeable bias, randomly distributed over a range of − 18 ≳ M_UV ≳ − 22, giving more UV-faint (M_UV > − 19) targets than the Keck observations. However, the Keck observations are much deeper than CLEAR in general, providing significantly lower EW limits for those targeted galaxies. Figure 1 also shows the stellar mass (M_star)–M_UV distribution of our CLEAR galaxies in the right panel, compared to the fiducial relation derived at z ∼ 7 by Song et al. (2016b). Similar to the Keck sample, most of the CLEAR targets are placed on the fiducial M_star–M_UV relation at z ∼ 7 of Song et al. (2016b) with no significant selection bias from the typical high-z galaxy population, although there are a few UV-faint and massive galaxies located above the fiducial relation.

Zoom In Zoom Out Reset image size

The Lyα EW distribution is commonly described by a declining exponential form,

where EW is the Lyα rest-frame EW, and W₀ is a e-folding scale width (e.g., Cowie et al. 2010). Jung et al. (2018) introduced an improved methodology of measuring the Lyα EW distribution, motivated by earlier studies (e.g., Treu et al. 2012, 2013), which utilizes galaxies without detected emission lines as well as Lyα detected objects. Following Jung et al., we constructed a template of an expected number of Lyα emitters as a function of the detection significance, by accounting for all types of data incompleteness including instrumental wavelength coverage, the wavelength-dependent Lyα detection limit, the UV-continuum flux, and the photometric redshift probability distribution function, P(z), based on the combined data set of the CLEAR spectra. We then applied this to the CLEAR data set and combined these results with those from our previous work using the Keck/DEIMOS+MOSFIRE observations in Jung et al. (2018, 2019, 2020) to measure the Lyα EW distribution.

To calculate the expected number of Lyα emitters for a given model we did the following. We first estimated the detection sensitivity for an unresolved emission line in the grism spectrum for each object. The detection limit was inferred through MC simulations with the 1D spectra by assigning a mock Lyα emission line to the reduced 1D spectra for all objects. We recovered the line flux of this mock emission line with its error from the Gaussian line fitting with the IDL MPFIT package (Markwardt 2009). This mock emission line was created to have an intrinsic line profile equal to the best-fit line profile from one of the highest-S/N Lyα emission detected in z7_GND_42912 from the MOSFIRE observations of Jung et al. (2019, 2020), and we smoothed the line profile to match the G102 resolution (∼45 Å). We note that the choice of the emission-line profile has a negligible effect given the relatively low resolution of the G102 grism data (R ∼ 200). We consider our targets to be point-like sources in our 1D mock emission simulation. This assumption is appropriate as our target galaxies at this redshift are small, with a typical effective radius (R_eff) of R_eff < 1 kpc in general (or ≲3 pixels at the resolution of HST/WFC3) and there is evidence that these kinds of galaxies are even smaller at higher redshifts (Shibuya et al. 2019). Additionally, in Grizli, our 1D spectra are extracted using the source segmentation maps from the photometric catalog, thus the extracted 1D spectra include the contribution estimated from the (rest-frame UV) morphology as measured by the Y₁₀₅ band. We may not capture the contribution of Lyα from extended halos in 1D spectra (e.g., Leclercq et al. 2017; Wisotzki et al. 2018). However, an appropriate modeling of the extended component of Lyα at this redshift is beyond the scope of this study. In future, for example, JWST NIRSpec integral field unit observations will resolve the extended component of Lyα.

From these simulation emission lines we can derive emission-line flux limits for the CLEAR G102 data. The top panel of Figure 2 shows the 3σ detection limits over the wavelength range of the G102 grism, which corresponds to a redshift range of for Lyα emission at 6.0 ≲ z ≲ 8.2. In the figure, the colored dots show the detection limits for all the 148 individual galaxies in our sample; the red curve shows the median detection limit for all spectra. The bottom panel of Figure 2 shows the corresponding (median) 3σ rest-frame EW limits for galaxies as a function of the apparent UV magnitude. The EW limits were estimated by dividing the median 3σ emission-line flux limits (f_3σ−limit) by the continuum level (f_cont) and (1 + z_Lyα ) at each corresponding wavelength as below.

Zoom In Zoom Out Reset image size

After the detection limits were estimated, we constructed a template of an expected number of detected Lyα emission lines as a function of the detection significance. The template construction was done in a MC fashion, by simulating mock Lyα emission lines for entire target galaxies, following Jung et al. (2018, 2020). The MC simulation of individual mock emission lines has three main steps: (i) we allocate the wavelength of the simulated Lyα emission, randomly taken based on an object’s P(z); (ii) at each redshift, we simulate a Lyα line flux given the galaxy’s magnitude and EW, where the latter is drawn from the distribution in Equation (1) over a range for the assumed Lyα EW scale width, W₀ = 5–200 Å; (iii) finally, we measure the detection significance for this simulated emission line based on the precomputed emission-line detection limits taken from above. For each choice of W₀, we carried out 1000 sets of MC simulations of the mock emission lines of all individual target galaxies, producing a distribution of the expected number of Lyα detections as a function of the detection significance.

Lastly, we fit the data for each galaxy to the suite of simulated templates over the range of redshift and emission-line EW. We then combined these results to construct a probability distribution function (PDF) for the e-folding scale width (W₀) of the Lyα EW distribution. We utilized a Markov Chain Monte Carlo (MCMC) algorithm for this fitting. As counting the number of Lyα emission-line detections is a general Poisson problem, we used a Poisson likelihood. We used the Cash (1979) statistic to describe the Poisson likelihood. We then used a Metropolis–Hastings MCMC sampler (Metropolis et al. 1953; Hastings 1970) to construct the PDF of W₀, generating 10⁵ MCMC chains. For more information on this methodology, see Jung et al. (2019, 2020).

Jung et al. (2018) constrained the characteristic e-folding scale width W₀ of the Lyα EW distribution at 6 < z < 7 using data from Keck/DEIMOS for a sample of galaxies. They measured the 1σ and 2σ upper limits of W₀ < 36 Å and <125 Å, respectively. Similarly, Jung et al. (2020) measured Å for galaxies at 7.0 < z < 8.2 using data from Keck/MOSFIRE observations. These values are shown in Figure 3.

Zoom In Zoom Out Reset image size

Here, we report new measurements on the evolution of the Lyα EW from the combined analysis of the CLEAR and Keck data sets. The CLEAR data provide important constraints on the evolution of Lyα as the lack of detections implies strong evolution in the number of sources with high Lyα EWs at these redshifts. The updated constraints improve the constraint on the EW e-folding scale width, W₀, to be <17 Å (1σ) and <28 Å (2σ) at 6.0 < z < 7.0 and Å at 7.0 < z < 8.2 (these are shown as red, large symbols in Figure 3). These measurements confirm the decline in W₀ already seen at 6.0 < z < 7.0 compared to z < 6.

Observations at lower redshifts (2 ≲ z ≲ 6) show the Lyα EW e-folding scale width (W₀) decreases with increasing UV luminosity (e.g., Ando et al. 2006; Stark et al. 2010; Schaerer et al. 2011; Cassata et al. 2015; Furusawa et al. 2016; Wold et al. 2017; Hashimoto et al. 2017; Oyarzún et al. 2017; Santos et al. 2020). In contrast, Jung et al. (2020) reported tentative evidence that W₀ increases with increasing UV luminosity at z > 6 from the brightest galaxies (M_UV < −21). Although the initial evidence was weak due to the large measurement errors, this finding is consistent with a picture of reionization where the IGM around brighter galaxies becomes ionized earlier than that of fainter objects, allowing for more efficient Lyα-photon escape (e.g., Zheng et al. 2017; Mason et al. 2018a; Endsley et al. 2021a).

CLEAR provides a means of improving the constraint on the evolution of W₀ as a function of the absolute UV magnitude as the CLEAR data cover many more lower-luminosity objects. We therefore recalculated W₀ from the combined data set of CLEAR and Keck/DEIMOS+MOSFIRE observations in different bins of the UV absolutemagnitude (M_UV). These W₀ measurements are listed in the third column in Table 1 and shown in the left panel of Figure 4 as filled red dots. From this analysis, the CLEAR data improve the evidence that the e-folding scale width (W₀) of the Lyα EW distribution at 6.0 < z < 8.2 shows a possible upturn of W₀ in the brightest UV magnitude bin, −22 < M_UV < −21, although it is consistent with no upturn at the 1σ level. The PDF of W₀ is constructed from 10⁵ MCMC chains as described in Section 3.1. We then inspect what fraction of the MCMC chains have an upturn of W₀ in the brightest UV magnitude bin. We find that 76% of the MCMC chains show such an upturn (or a greater W₀). In addition, we find that 91% of the chains show an increasing trend between the Lyα IGM transmission and the (increasing) UV luminosity as shown in Figure 4. We discuss this further in the next subsection. This is in strong contrast with the measurements at 2 < z < 6 that show W₀ decreases with increasing UV luminosity (Santos et al. 2020, shown as black open circles), indicating that at z > 6 the Lyα emission for UV-brighter galaxies is less affected by the IGM compared to UV-fainter galaxies.

Zoom In Zoom Out Reset image size

Table 1. Lyα EW e-folding Scale Width (W₀) and the Ratio of Observed-to-Expected (intrinsic) EWs at z > 6 at Different M_UV

		W0		W0,z>6 / W0,z<6
MUV (LUV/L* a )	Ngalaxy	z ∼ 2−6	6.0 < z < 8.2	6.0 < z < 8.2
		Santos et al. (2020)	This study
−20 < MUV < −19 (0.15 < LUV/L* < 0.4) b	77	Å	Å
− 21 < MUV < − 20 (0.4 < LUV/L* < 1.0)	92	Å	Å
− 22 < MUV < − 21 (1.0 < LUV/L* < 2.5)	30	Å	Å

Notes.

^a , where is the characteristic magnitude of a typical galaxy with an approximate value of at z ∼ 7 (Finkelstein et al. 2015). ^b Our sample in this faintest UV magnitude group could be incomplete in the CANDELS GOODS Wide fields due to the ∼1 magnitude shallower detection limits in the Wide fields than those in the CANDELS GOODS Deep fields, as shown in Figure 1 (left). Thus, we also calculated W₀ in −20 < M_UV < −19 excluding objects in the Wide fields, which provides Å. Although it is slightly increased from our fiducial value listed in the table, the change is insignificant and does not alter our conclusion.

Download table as:  ASCIITypeset image

We note that we cannot completely rule out the possibility of the redshift evolution of galaxy physical properties (e.g., interstellar medium condition), which alters the Lyα properties significantly even without the IGM effect (Dayal et al. 2011; Hassan & Gronke 2021). However, such decreasing trend of W₀ has been reported from most lower redshifts studies at z < 6 (see references earlier in this section), and evolutionary features of Lyα-emitting galaxies are not expected to be extreme, particularly between z < 6 and z > 6.

Our W₀ measurements indicate different evolution of the Lyα EW distribution between bright and faint galaxies into the epoch of reionization, z ≳ 6. As shown in Figure 4, the W₀ values obtained for the UV-fainter objects (M_UV > −21) at z > 6 are substantially lower than those at lower redshift (2 < z < 6), whereas there is milder evolution in W₀ for the UV-brighter objects (M_UV < −21) from z < 6 to z > 6.

We use the change in the rest-frame Lyα EW as an estimate for the evolution of the Lyα transmission in the IGM transmission. We calculate the ratio of the observed-to-expected (intrinsic EW) at z > 6, EW_obs/EW_int); ¹² therefore,

where we take W_0,z<6 as the intrinsic EW after the IGM became completely ionized. For our analysis, we adopted W_0,z<6, the measurements of the EW distribution at 2 < z < 6 from Santos et al. (2020) as the intrinsic values, and our W₀ measurements as the observed values at z > 6.

The EW ratio, , has been used heavily in the literature to interpret the evolution in Lyα emission lines in the context of the changing IGM into the EoR (e.g., Konno et al. 2018; Weinberger et al. 2018; Jung et al. 2020; Goto et al. 2021). In practice, this ratio approximates the global average IGM transmission at z > 6 compared to z < 6. As such, it assumes that the Lyα photon production and escape from galaxies do not evolve between the two redshift ranges. It also does not take into account the detailed transmission of the IGM to Lyα photons along sight lines to individual galaxies (and the latter requires sophisticated simulations that include the radiative effects of Lyα in cosmological simulations of reionization, and utilizing such simulations is beyond the scope of this work). The details of Lyα transmission are very model dependent, with large systematics between studies (see discussion below). Therefore, we interpret the ratio in Equation (3) as an estimate for the Lyα IGM transmission because it allows us to contextualize our measurements and compare to previous observation studies (but we use a larger spectroscopic sample). However, below we do provide comparison to some theoretical models, particularly ones that predict an increasing Lyα transmission with bright UV magnitude (as we observe that here). This provides some physical interpretation for our results and provides guidance for future theoretical studies.

The estimated EW ratios using Equation (3) are listed in the last column in Table 1 and shown in the right panel in Figure 4. Under the assumptions that go into Equation (3) (see above), the fact that the IGM transmission for Lyα photons appears to be higher for the UV-brightest galaxies can be interpreted as evidence that the IGM is highly ionized and more transparent around the brightest galaxies at z > 6. This result will require further confirmation with larger and more sensitive data sets from future telescopes, such as JWST and NGRST. Under the same assumptions, the decrease in IGM transmission for fainter galaxies (M_UV > −21) can be interpreted as evidence that the IGM is more neutral where Lyα photons suffer significantly more attenuation. This adds to the growing observational evidence and findings of simulations that reionization is a spatially inhomogeneous process, where large ionized bubbles were often created by bright galaxies (e.g., Zheng et al. 2017; Castellano et al. 2018; Jung et al. 2019, 2020; Hu et al. 2021; Endsley et al. 2021b; Park et al. 2021; Qin et al. 2022). We expand on this interpretation in the next section.

While the interpretation of the EW ratio in Equation (3) can be used as an estimate of the Lyα IGM transmission, it is simplistic as it does not include variations in the Lyα IGM transmission depending on the local reionization history with no assumptions on the (either smooth or patchy) IGM evolution (e.g., the sizes and kinematics of individual ionized regions) nor the detailed Lyα emission properties of individual galaxies. However, the line-of-sight variations in the Lyα IGM transmission have been examined in previous studies, and a large number of these predict that the Lyα IGM transmission should increase with increasing galaxy UV luminosity (e.g., Mason & Gronke 2020; Gronke et al. 2021; Park et al. 2021; Smith et al. 2022; Qin et al. 2022), which corresponds to what our results support. Regarding possible effects of the interstellar medium (ISM), our measurements assume that we do not expect strong ISM evolution in galaxies between the two redshift ranges: similar ISM conditions in galaxies with similar UV brightness. The M_UV–Δv_Lyα (Lyα velocity offset) relation may evolve with the redshift, but the M_UV dependency of Δv_Lyα becomes even smaller at higher redshifts (e.g., Equation (3) in Mason et al. 2018a), and relatively small velocity offsets in higher-z UV-bright galaxies were reported (e.g., Matthee et al. 2020). Thus, if this evolutionary effect is considered, the additional boost on the Lyα IGM transmission from UV-brighter galaxies due to their relatively larger Δv_Lyα becomes even weaker. Thus, this means that the ISM effects are unlikely to be a factor nullifying our results and conclusions in this study. Although other important physical conditions in the ISM, such as dust geometry, nebular metallicity, galaxy morphology, and line-of-sight variation in the ISM level, need to be considered as well, our current knowledge on these factors is largely limited and left for future study.

To make more quantitative interpretation of the evolution of the IGM requires more sophisticated simulations that include both the detailed production and escape of Lyα from individual galaxies, in addition to the radiative transfer of those photons through an expanding, inhomogeneously ionized IGM. This has been the focus of multiple studies (e.g., Hutter et al. 2014; Mesinger et al. 2015; Mason et al. 2018a; Weinberger et al. 2018). While these simulations are enlightening, they also depend on other assumptions (e.g., the reionization history, the evolution of ionized bubble sizes, and M_UV—halo mass relation). Here, our intent is to provide results from the latest sets of Lyα observations combined with our empirical approach. In future work, it will be very useful to extend these investigations to study how the Lyα IGM transmission varies depending on the reionization history and the characteristic ionized bubble sizes around galaxies (e.g., Garel et al. 2021; Park et al. 2021; Smith et al. 2022) as well as the intrinsic Lyα emission properties of galaxies during the epoch of reionization such as Lyα velocity offset (e.g., Stark et al. 2017; Hashimoto et al. 2018; Hutchison et al. 2019; Matthee et al. 2020).