First Results from the Lyman Alpha Galaxies in the Epoch of Reionization (LAGER) Survey: Cosmological Reionization at z ∼ 7

The LAGER survey is the largest narrowband survey yet for LAEs at z ≳ 7. It is currently ongoing, using the Dark Energy Camera (DECam, with FOV ∼ 3 deg²) on the NOAO-CTIO 4 m Blanco telescope together with an optimally designed custom narrowband filter, NB964¹² (central wavelength ∼9642 Å, FWHM ∼ 90 Å). LAGER is designed to select more than a hundred z ∼ 7 LAEs over an area of 12 deg² in four fields (8 × 10⁶ cMpc³). Currently, LAGER has collected 47 hr NB964 narrowband imaging in three fields (CDF-S, COSMOS, and DLS-f5) taken during nine nights in 2015 December, 2016 February, March, and November.

The deepest NB964 imaging is done in COSMOS, where we have obtained 34 hr NB964 exposure in a 3 deg² field. The exposure time per NB964 frame is 900 s. Consecutive exposures were dithered by ∼100'' so that chip gaps and bad pixels do not lead to blank areas in the final stacks. We also took ∼0.5–1 hr z and Y  band exposure per field to exclude possible transients. There are deep archival DECam and Subaru broadband images in COSMOS.

We downloaded the reduced and calibrated DECam resampled images from the NOAO Science Archive, which were processed through the NOAO Community Pipeline (version 3.7.0; Valdes et al. 2014). The individual frames were stacked following the weighting method in Annis et al. (2014) with SWarp (version 2.38.0; Bertin 2010). The seeing of the final stacked LAGER-COSMOS narrowband image is 093.

The zero-magnitudes of the DECam images were calibrated to the Subaru Suprime-Cam broadband magnitudes of the stars in the public COSMOS/UltraVISTA K_S-selected catalog (Muzzin et al. 2013). We estimated the image depth by measuring the root mean square (rms) of the background in blank places where detected (>3σ) signals were masked out. In a 2' diameter aperture, the 3σ limiting AB magnitudes of the DECam images are [u, g, r, i, z, Y, NB964]^3σ = [26.2, 27.4, 26.9, 26.6, 26.1, 24.3, 25.6]. Deep Subaru Suprime-Cam broadband and narrowband images are available in the central 2-deg² area. We downloaded the raw images from the archival server SMOKA (Baba et al. 2002), and produced our own broadband and narrowband stacked images (follow-up work of Jiang et al. 2013). Their 3σ limiting magnitudes within a 2'' diameter aperture are [B, g', V, r', i', z', NB711, NB816, NB921]_3σ ∼ [27.9, 27.6, 27.2, 27.4, 27.3, 25.9, 26.0, 26.1, 26.2]. These Subaru Suprime-Cam images are deeper and have better seeing than the DECam images on average.

We calculated the narrowband completeness via Monte Carlo simulations with the BALROG software (Suchyta et al. 2016). The completeness fraction is defined as the SExtractor recovery percentage of the randomly distributed artificial sources in the NB964 image as a function of narrowband magnitude. For point sources, the narrowband NB964 aperture magnitudes corresponding to the 80%, 50%, and 30% completeness fractions are 24.2, 25.0, and 25.3, respectively. For pseudo LAEs, we choose fake sources similar to those used in Konno et al. (2017), which have a Sérsic profile with the Sérsic index of n = 1.5 and the half-light radius of r_e ∼ 0.9 kpc (0.17 arcsec at z = 6.9). The narrowband NB964 aperture magnitudes corresponding to the 80%, 50%, and 30% completeness fractions are 24.3, 24.7, and 25.0 for pesudo LAEs, respectively.

Since Subaru broadband images are deeper and have better seeings than those of the DECam broadband images, we only apply selection criteria for z ∼ 7 LAEs in the central 2-deg² area covered by both the Suprime-Cam and the DECam. The selection criteria for z ∼ 7 LAEs include narrowband NB964 with ≥5σ detection (NB964 ≤ NB964_5σ = 25.07), narrowband NB964 excess over broadband z (z - NB964 ≥ 1.0),¹³ and nondetections in both DECam ugri and Suprime-Cam Bg'Vr'i', N711, N816, N921 bands. We specially exclude broadband detections in the 108 (4 pixel) diameter aperture to exclude marginal signals in these bands. 156 sources passed the selection criteria, but a large fraction of them are fake sources (bleed trails, diffraction spikes, etc.) after our visual check on NB964 image. With our team's tripartite visual check on both NB964 and broadband images, 27 candidate z ∼ 7 galaxies are selected. No transient was identified among them using the available broadband images.

We estimate the Lyα line fluxes and EWs of the candidates using narrowband and broadband photometry.¹⁴ We further exclude 4 out of the 27 candidates with estimated line EW of <10Å.

In total, 23 candidate z ∼ 7 LAEs, which have Lyα line fluxes in the range of 0.8–7.8 × 10⁻¹⁷ ergs cm⁻² s⁻¹ and rest-frame equivalent widths EW_R ≥ 10 Å, are selected. The corresponding observed Lyα line luminosities are 5.4–43.4 × 10⁴² erg s⁻¹. Their positions and their multi-band thumbnail images are plotted in Figure 1.

Zoom In Zoom Out Reset image size

The four brightest LAEs, with L(Lyα) = 19.3–43.4 × 10⁴² erg s⁻¹, are the most luminous candidate LAEs known at z ≳ 7, thanks to the large survey volume of LAGER. We have confirmed three of the brightest LAEs at z ∼ 7 (Hu et al. 2017) with our recent Magellan/IMACS spectroscopic follow-up observations.

One of the candidates, J09:59:50.99+02:12:19.1, was previously selected and spectroscopically confirmed as a z = 6.944 LAE with Magellan/IMACS narrowband imaging and spectroscopy over a much smaller field (Rhoads et al. 2012). It is well recovered with our selection using new DECam narrowband imaging.

We use the V/V_max method to calculate the Lyα LF at z = 6.9. The formula below is adopted as

$$\begin{eqnarray}&&{\rm{\Phi }}(L){dL}=\displaystyle \sum _{{L}_{i}\in [L-{\rm{\Delta }}L/2,L+{\rm{\Delta }}L/2]}\displaystyle \frac{1}{{V}_{\max }{f}_{\mathrm{comp}}({{NB}}_{i})}.\end{eqnarray} \tag{ 1 }$$

Here V_max is simply a constant, the maximum survey volume for LAEs (V_max = 1.26 × 10⁶ cMpc³ calculated from sky coverage and NB964 central wavelength/FWHM), and f_comp(NB_i) is the detection completeness for sources with narrowband magnitude m(NB_i) interpolated in the narrowband detection completeness for fake LAEs (see Section 2.2).

The Lyα LF of LAEs at z ∼ 7 in the LAGER-COSMOS field is listed in Table 1, and plotted in Figure 2 together with LAE LFs at z ∼ 5.7–7.3 from the literature. The LF shows dramatic evolution not only in normalization, but also in the shape.

Zoom In Zoom Out Reset image size

Table 1.  Lyα Luminosity Function at z ∼ 7 from LAGER-COSMOS Field

log10(Li)	N ± Δ Na	$\langle {f}_{\mathrm{comp}.}\rangle$	log10Φ(Li)
(ΔL = 0.125)	$[{L}_{i}\pm \tfrac{{\rm{\Delta }}L}{2}]$		[(dlog10L)−1 Mpc−3]
42.71	${6}_{-2.4}^{+3.6}$	0.21	$-{3.74}_{-0.22}^{+0.20}$
42.84	${7}_{-2.6}^{+3.8}$	0.44	$-{4.00}_{-0.20}^{+0.19}$
42.96	${4}_{-1.9}^{+3.2}$	0.56	$-{4.34}_{-0.28}^{+0.25}$
43.09	${2}_{-1.3}^{+2.7}$	0.80	$-{4.80}_{-0.45}^{+0.37}$
43.21	<2.6	0.82	<−5.10
43.33	${1}_{-0.8}^{+2.3}$	0.85	$-{5.13}_{-0.77}^{+0.52}$
43.46	${2}_{-1.3}^{+2.7}$	0.88	$-{4.84}_{-0.45}^{+0.37}$
43.59	${1}_{-0.8}^{+2.3}$	0.90	$-{5.15}_{-0.77}^{+0.52}$

Note.

^a1σ Poisson errors from Gehrels (1986).

Download table as:  ASCIITypeset image

The relatively fainter end of our LF (42.65 ≤ log₁₀L_Lyα ≤ 43.25) at z ∼ 7 can be fitted with a a single Schechter function in the form of

$$\begin{eqnarray}&&{\rm{\Phi }}(L){dL}={{\rm{\Phi }}}^{* }{\left(\displaystyle \frac{L}{{L}^{* }}\right)}^{\alpha }\exp \left(-\displaystyle \frac{L}{{L}^{* }}\right)d\left(\displaystyle \frac{L}{{L}^{* }}\right).\end{eqnarray} \tag{ 2 }$$

Because our survey does not go far below L*, we fix α = −2.5 in our fitting, which is suggested from recent studies on Lyα LFs at z = 5.7 and z = 6.6 by Konno et al. (2017). Konno et al. (2017) report the largest narrowband surveys to date for LAEs at z = 5.7 and z = 6.6 taken with Subaru HSC on the 14–21 deg² sky, and their best-fit Schechter function analyses also include the smaller but much deeper narrowband surveys taken with Subaru Suprime-Cam (Ouchi et al. 2008, 2010) at the corresponding redshifts. Considering that the Lyα LFs by Konno et al. (2017) are derived from the largest LAE samples at z = 5.7 and z = 6.6 with the widest luminosity range, and we have used consistent selection and analysis methods, we choose their Lyα LFs for comparison in the following sections. At z > 7, we choose the Lyα LF at z = 7.3 by Konno et al. (2014) for comparison. The best-fit Schechter function parameters of these LFs are listed in Table 2 and plotted in the upper panel of Figure 3. We find a large drop in Φ of Lyα LF, compared with that at z = 5.7 and z = 6.6. In the relatively fainter end, our Lyα LF at z ∼ 7 is in agreement within the 1σ measurement uncertainties of that at z = 7.3 .

Zoom In Zoom Out Reset image size

Table 2.  Best-fit Schechter Parameters and Lyα Luminosity Densities at z ∼ 5.7, 6.6, 6.9, and 7.3

z	Area	Volume	L Fitted Range	log10[${L}_{\mathrm{Ly}\alpha }^{* }$]	log10[Φ*]	log10[ρLyα]a	log10[${\rho }_{\mathrm{Ly}\alpha }^{\mathrm{Bleft}}$]b	Reference
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	deg2	[cMpc3]		[erg s−1]	[Mpc−3]	[erg s−1 Mpc−3]	[erg s−1 Mpc−3]
Two free parameters [${L}_{{Ly}\alpha }^{* }$, Φ]* with fixed α = −2.5
5.7	13.8	1.16 × 107	42.4–44.0	43.24 ± 0.05	$-{4.09}_{-0.11}^{+0.13}$	39.54 ± 0.01	38.24	Ouchi et al. (2008) + Konno et al. (2017)
6.6	21.2	1.91 × 107	42.4–44.0	${43.25}_{-0.05}^{+0.06}$	−4.32 ± 0.13	39.33 ± 0.02	38.05	Ouchi et al. (2010) + Konno et al. (2017)
6.9	2.0	1.26 × 106	42.65–43.25	${42.77}_{-0.16}^{+0.32}$	$-{3.77}_{-0.76}^{+0.54}$	c ${38.92}_{-0.04}^{+0.03}$	c37.90	This study
			42.65–43.65	${43.74}_{-0.38}^{+0.55}$	$-{5.76}_{-0.81}^{+0.72}$	38.77 ± 0.05	37.82	This study
7.3	0.45	2.5 × 105	42.4–43.0	${42.77}_{-0.34}^{+1.23}$	$-{4.09}_{-1.91}^{+1.09}$	${38.55}_{-0.07}^{+0.08}$	d⋯	Konno et al. (2014)

Notes. The best-fit parameters in column 5 and column 6 are derived by fitting the Schechter function with fixed α = −2.5 on the Lyα LFs listed in column 9 with the χ²-statistics. The errors in columns 5–7 are estimated from their 1σ confidence contours (e.g., Coe 2009) shown in the top panel of Figure 3.

^aLyα luminosity densities integrated over the best-fit Schechter functions at 42.4 < log₁₀L_Lyα < 44.0 (column 7). Note the errors of Lyα luminosity densities are likely underestimated because we do not consider the uncertainties of the faint-end slope in that calculation. ^bLyα luminosity densities integrated at the bright end (43.3 < log₁₀L_Lyα < 43.7; column 8). ^cThe Lyα luminosity densities in this row are calculated by integrating the truncated Schechter function plotted as the red solid curve in Figure 2 and discussed in Section 3.2. We use the Lyα density in column 7 of this row to estimate the neutral IGM fraction at z ∼ 7 in Section 4.2. ^dThe survey for z = 7.3 LAEs does not have large enough volume to cover the bright end.

Download table as:  ASCIITypeset image

More strikingly, we see a clear bump, i.e., significant excess to the Schechter function in the bright end of our z ∼ 7 LF (log₁₀L_Lyα ≳ 43.25). It demonstrates that, while the space density of faint LAEs drops tremendously from z = 5.7 and 6.6 to z = 6.9, the luminous LAEs show no significant change. Similarly but much less prominently, the LF evolution between z = 5.7 and z = 6.6 may also be differential, with a significant decline at the faint end, but not clear evolution at the bright end (Matthee et al. 2015; Santos et al. 2016; Konno et al. 2017). In Figure 2 we also plot a scaled-down and truncated version (red solid line at high luminosity end) of Lyα LF at z = 5.7 from Konno et al. (2017), normalized by the UV luminosity density evolution¹⁵ between z = 5.7 and z = 6.9, which appears consistent with our LF at the bright end within statistical uncertainties.

Cosmological reionization was well under way by z ∼ 9 (Planck Collaboration et al. 2016) and ended by z ∼ 6 (Fan et al. 2006). Lyα galaxies at z ≳ 6, e.g., from LAE surveys at z = 5.7 (Ouchi et al. 2008; Santos et al. 2016; Konno et al. 2017), at z = 6.6 (Ouchi et al. 2010; Matthee et al. 2015; Konno et al. 2017), at z = 6.9 (our LAGER-COSMOS sample in this work and Ota et al. 2017), and at z = 7.3 (Konno et al. 2014), are unique samples useful to probe both galaxy evolution and reionization.

In this work, we detect for the first time a significant bump at the bright end of Lyα LF at z ≳ 7. Previous surveys for LAEs at z ≳ 7 failed to reveal a bump as they covered much smaller volumes.

The existence of such a bright-end bump is consistent with the scenario proposed by Haiman & Cen (2005), that bright LAEs are less attenuated by a neutral IGM than faint LAEs, as the larger Strömgren sphere surrounding luminous LAEs alleviates the neutral IGM absorption (also see Santos et al. 2016; Konno et al. 2017). Therefore, the evolution in the bright end of the LF better reflects the intrinsic evolution of LAEs, while the faint end is controlled by the evolution of both galaxies and the IGM.

The bottom panel of Figure 3 shows the evolution of luminosity densities both for Lyα and UV photons. Compared to LBGs (yellow shaded region), an accelerated evolution of LAE Lyα densities from z = 6.6 (green diamonds) to z = 7.3 (purple cross), as reported by Konno et al. (2014), is clearly seen in Figure 3. Seeing that UV photons detected in LBGs are insensitive to the neutral hydrogen in EoR, Konno et al. (2014) concluded that such rapid evolution in LAE LFs can be attributed to a large neutral IGM fraction at z ∼ 7.3. Our LAGER survey shows rapid evolution from z = 6.6 to 6.9, and a smaller difference between z = 6.9 and 7.3 (see Figure 3). It is possible that the evolution in the neutral fraction accelerated after z = 6.9, though a smooth, monotonic neutral fraction evolution at 6.6 ≤ z ≤ 7.3 also fits the LAGER results and the measurements of Konno et al. (2014) within their uncertainties. Note that the cosmic time from z ∼ 6.6 to 6.9 and from z ∼ 6.9 to 7.3 are approximately equal, which is about 50 Myr.

Furthermore, compared to the evolution of Lyα densities ρ_Lyα between redshifts of 5.7 and 6.9, the evolution in the bright end only ${\rho }_{\mathrm{Ly}\alpha }^{\mathrm{Bright}}$ follows a much smoother trend, somehow similar to that of the UV densities of LBGs (bottom panel of Figure 3). This demonstrates that either the intrinsic LAE LF evolves moderately between z = 5.7 and 6.9, or that IGM attenuation plays a role even at the highest luminosity bin. Consequently, the dramatic and rapid evolution between z = 6.6 and 6.9 in the faint end can be mainly attributed to the change of neutral IGM fraction. This evolution trend is in agreement with the Lyα fraction test which shows the fraction of LBGs with visible Lyα lines drops significantly at z ≳ 7 (e.g., Schenker et al. 2014).

Following Ouchi et al. (2010) and Konno et al. (2014), we compare the Lyα luminosity densities at z = 6.9 (in EoR) and at z = 5.7 (when reionization is completed) to estimate the effective IGM transmission factor ${T}_{{Ly}\alpha ,z}^{\mathrm{IGM}}$. Assuming the stellar population, interstellar medium (ISM), and dust are similar at z = 6.9 and z = 5.7, we obtain that

$$\begin{eqnarray}&&{T}_{z=6.9}^{\prime }=\displaystyle \frac{{T}_{{Ly}\alpha ,z=6.9}^{\mathrm{IGM}}}{{T}_{{Ly}\alpha ,z=5.7}^{\mathrm{IGM}}}=\displaystyle \frac{{\rho }_{z=6.9}^{{Ly}\alpha }/{\rho }_{z=5.7}^{{Ly}\alpha }}{{\rho }_{z=6.9}^{\mathrm{UV}}/{\rho }_{z=5.7}^{\mathrm{UV}}}.\end{eqnarray} \tag{ 3 }$$

Here ρ^UV and ρ^Lyα are UV and Lyα luminosity densities, respectively. The ratio of UV luminosity densities ${\rho }_{z=6.9}^{\mathrm{UV}}/{\rho }_{z=5.7}^{\mathrm{UV}}=0.63\pm 0.09$ is obtained from Finkelstein et al. (2015). We obtain the ratio of observed Lyα luminosity densities (column 7 in Table 2) ${\rho }_{z=6.9}^{{Ly}\alpha }/{\rho }_{z=5.7}^{{Ly}\alpha }=0.24\pm 0.03$. Thus, we estimate ${T}_{z=6.9}^{\prime }=0.38\pm 0.07$ ¹⁶ (cf., ${T}_{z=7.3}^{\prime }=0.29$ from Konno et al. 2014 and ${T}_{z=6.6}^{\prime }=0.70\pm 0.15$ from Konno et al. 2017).

Converting the Lyα emission line transmission factor to neutral IGM fraction x_{H i} is model dependent (e.g., Santos 2004; Dijkstra et al. 2007; McQuinn et al. 2007). An important factor is the shift of the Lyα line with respect to the systemic velocity, which is widely observed and may be explained by Lyα radiative transfer in a galactic wind or outflow. The Lyα velocity shift measurements of z ∼ 6–7 galaxies are very limited and are mostly around 100–200 km s⁻¹ (Stark et al. 2015, 2017; Pentericci et al. 2016; but see the two UV luminous galaxies with 400–500 km s⁻¹ reported by Willott et al. 2015). With the analytic model of Santos (2004), assuming a Lyα velocity shift of 0 and 360 km s⁻¹, the value of ${T}_{z=6.9}^{\prime }=0.38$ corresponds to x_{H i} ∼ 0.0 and 0.6, respectively.

By comparing our observed Lyα LF at z = 6.9 to that predicted with the radiative transfer simulations by McQuinn et al. (2007; their Figure 4), we obtain x_{H i} ∼ 0.40–0.60 at z = 6.9. Similar results are obtained in the analytic studies considering ionization bubbles (Furlanetto et al. 2006; Dijkstra et al. 2007). We should note that these studies predicted a suppression of the LF that is rather uniform across a wide range of luminosities (e.g., Section 4 of McQuinn et al. 2007). This suggests that the true distribution of ionized region sizes may differ appreciably from those used in the literature (Furlanetto et al. 2006; Dijkstra et al. 2007; McQuinn et al. 2007). We conclude that the neutral hydrogen fraction is x_{H i} ∼ 0.4–0.6 at z = 6.9, where both the uncertainties in the IGM transmission factor calculation and the theoretical model predictions are considered. The LBG Lyα fraction test by Schenker et al. (2014) yields a similar neutral hydrogen fraction (${x}_{{\rm{H}}{\rm{I}}}={0.39}_{-0.09}^{+0.08}$) at z ∼ 7. Note that smaller x_{H i} = 0.3 ± 0.2 at z = 6.6 is obtained by Konno et al. (2017) with Subaru HSC and Suprime-Cam surveys.

The bump at the bright end of the Lyα LF is an indicator of large enough (>1 pMpc radius) ionized bubbles, where the Hubble flow can bring Lyα photons out of resonance, thus leading to a different evolution of Lyα LF at the bright end and the faint end. The uneven distribution of these LAEs may indicate ionized bubbles in a patchy reionization at z ∼ 7. Larger LAGER samples in future will more definitely establish whether the degree of clustering in Figure 1 requires patchy reionization.

The origin of the ionized bubbles could be extrinsic, intrinsic, or both. From Malhotra & Rhoads (2006), which gave the proper radius of a Strömgren sphere and the Gunn–Peterson effect optical depth of a LAE with outflow velocity ΔV in EoR, we know that even the brightest LAE can not produce a large enough ionized bubble to effectively reduce the optical depth to τ ∼ 1. We thus would expect extrinsic contribution from additional satellite galaxies associated with the luminous LAEs. From Figure 1, we see such fainter nearby sources (e.g., the projected separations between LAE-1, 3, and 11 are ≲ 3.4 arcmin, which corresponds to 1 pMpc at z = 6.93). Further, deeper narrowband imaging would enable us to probe the bubbles by detecting more fainter galaxies associated with the luminous LAEs.

The intrinsic explanation is that these luminous LAEs may be physically unusual objects, perhaps of a type not commonly found at lower redshift. For example, they could represent an AGN population that dominates over the ordinary star-forming galaxies at lg₁₀L_Lyα > 43.3 (e.g., Matsuoka et al. 2016). Alternatively, they could be star-forming galaxies where metallicities and/or dust abundances are low enough to enable considerably larger Lyα production or escape than is commonly seen at z < 6 (e.g., Stark et al. 2015, 2017; Mainali et al. 2017). At z ∼ 7, Bowler et al. (2014) find an excess of luminous LBGs where there should be an exponential cutoff at the bright end of their UV LFs. NIR and FIR spectroscopic follow-up of these luminous LBGs show very large Lyα velocity offsets (>300 km s⁻¹; Willott et al. 2015; Stark et al. 2017), which are significantly larger than the predicted velocity offsets for galaxies in EoR by Choudhury et al. (2015). Besides that, unusually strong carbon lines reported in two other LBGs at z ∼ 7 (CIVλ1548 Å in A1703-zd6 at z = 7.045 and C iii] 1909 Å in EGS-zs8-1 at z = 7.73) indicate unusual harder and more intense UV radiation than that of z ∼ 3 LBGs (Stark et al. 2015, 2017). In addition, a strong He ii line has been reported in the brightest Lyα emitter at z = 6.6 (CR7), suggesting  that it could be powered by either Pop. III stars (Sobral et al. 2015) or perhaps powered by accretion onto a direct collapse black hole (Pallottini et al. 2015).

Future deeper NB964 imaging and IR (NIR and/or FIR) spectroscopic observation will help us to determine the nature of these luminous LAEs, and probe the ionized bubbles in a patchy reionization at z ∼ 7.