A JWST/NIRSpec First Census of Broad-Line AGNs at z = 4 − 7: Detection of 10 Faint AGNs with M BH ∼ 10 6 − 10 8 M ⊙ and Their Host Galaxy Properties

We present a first statistical sample of faint type-1 AGNs at z > 4 identified by JWST/NIRSpec deep spectroscopy. Among the 185 galaxies at z spec = 3 . 8 − 8 . 9 confirmed with NIRSpec, our systematic search for broad-line emission reveals 10 type-1 AGNs at z = 4 . 015 − 6 . 936 whose broad component is only seen in the permitted H α line and not in the forbidden [Oiii] λ 5007 line that is detected with greater significance than H α . The broad H α line widths of FWHM ≃ 1000 − 6000 km s − 1 suggest that the AGNs have low-mass black holes with M BH ∼ 10 6 − 10 8 M ⊙ , remarkably lower than those of low-luminosity quasars previously identified at z > 4 with ground-based telescopes. JWST and HST high-resolution images reveal that the majority of them show extended morphologies indicating significant contribution to the total lights from their host galaxies, except for three compact objects two of which show red SEDs, probably in a transition phase from faint AGNs to low luminosity quasars. Careful AGN-host decomposition analyses show that their host’s stellar masses are systematically lower than the local relation between the black hole mass and the stellar mass, implying a fast black hole growth consistent with predictions from theoretical simulations. A high fraction of the broad-line AGNs ( ∼ 5%), higher than z ∼ 0, indicates that a number density of such faint AGNs is higher than an extrapolation of the quasar luminosity function, implying a large population of AGNs in the early universe. Such faint AGNs contribute to cosmic reionization, while the total contribution is not large, up to ∼ 50% at z ∼ 6, because of their faint nature.


INTRODUCTION
It has been known for over two decades that the mass of supermassive black holes (SMBHs) at z ∼ 0 are tightly correlated with the bulge properties of their host galaxies, such as velocity dispersion and bulge mass (e.g., Magorrian et al. 1998;Gebhardt et al. 2000;Kormendy & Ho 2013;Reines & Volonteri 2015).This tight correlation indicates the strong connection between the growth of central SMBHs and their host galaxies, known hari@icrr.u-tokyo.ac.jp as the galaxy-SMBH coevolution.Although the underlying physical mechanisms are still under debate, theoretical models suggest that feedback of active galactic nuclei (AGNs) connected to galaxy's merger histories play an important role (e.g., Granato et al. 2004;Di Matteo et al. 2005;Hopkins et al. 2006;Li et al. 2007).Since theoretical models usually make predictions for the time evolution of the systems (e.g., Anglés-Alcázar et al. 2017;Toyouchi et al. 2021;Valentini et al. 2021;Zhu et al. 2022;Habouzit et al. 2022;Trinca et al. 2022;Inayoshi et al. 2022;Hu et al. 2022;Zhang et al. 2023a,b), observations of both SMBHs and their host galaxies over cosmic time are essential to test and/or refine our current understandings of their build-up (e.g., Gallerani et al. 2017;Valiante et al. 2017).
Observations of AGNs at the high-redshift universe are thus crucial to understanding the evolution of the SMBH growth in cosmic history, but previous observations have been limited to bright quasars identified in surveys with ground-based telescopes (e.g., Bañados et al. 2016;Shen et al. 2019, see Inayoshi et al. 2020 for a review) including the most distant quasars at z > 7.5 (Bañados et al. 2018;Yang et al. 2020;Wang et al. 2021).Recent surveys using 4-8 m-class telescopes including the Subaru/Hyper Suprime-Cam survey (Aihara et al. 2018) are finding low-luminosity quasars at z ∼ 4 − 7 (e.g., Willott et al. 2010b;Kashikawa et al. 2015;Matsuoka et al. 2016;Onoue et al. 2017;Akiyama et al. 2018;Niida et al. 2020), but these low-luminosity quasars have moderately massive black hole masses with M BH ∼ 10 8 − 10 9 M ⊙ compared to AGNs found in the local universe (e.g., Liu et al. 2019).Although intensive spectroscopic observations targeting high redshift galaxies at z ≳ 7 have identified high ionization emission lines indicative of AGN activity (e.g., Tilvi et al. 2016;Laporte et al. 2017;Mainali et al. 2018), the physical properties of their central SMBHs are unclear.
The James Webb Space Telescope (JWST) was launched at the end of 2021 and started its operation in early 2022.JWST observations are now beginning to improve our understanding of the connection between high redshift AGNs and their host galaxies, especially in the low-luminosity and low-mass regimes.JWST/NIRCam deep and high-resolution images allow us to detect stellar lights from host galaxies of lowluminosity quasars at z ∼ 6 (Ding et al. 2022), and to identify the least-massive black hole candidate at z ∼ 5 (Onoue et al. 2023) and a triply-lensed red-quasar candidate at z ∼ 8 (Furtak et al. 2022).JWST/NIRSpec deep-spectroscopic observations have confirmed the candidate in Onoue et al. (2023) at z spec = 5.2 by detecting a broad Hα emission line that indicates a black hole mass of M BH ∼ 10 7 M ⊙ , with an additional finding of a red AGN at z spec = 5.6 (Kocevski et al. 2023).NIRSpec spectroscopy has also identified a supermassive black hole in a quiescent galaxy at z = 4.658 (Carnall et al. 2023).NIRSpec IFU observations have revealed broad Hα and Hβ emission lines in a galaxy at z = 5.55 ( Übler et al. 2023), which is interpreted as a type 1.8 AGN with a black hole mass of M BH ∼ 10 8 M ⊙ .Very recently, Larson et al. (2023) have reported a broad Hβ emission line in a galaxy at z spec = 8.7, suggesting a SMBH whose mass is M BH ∼ 10 7 M ⊙ 570 Myrs after the Big Bang (see also Bogdan et al. 2023;Goulding et al. 2023;Greene et al. 2023;Furtak et al. 2023;Fuji-moto et al. 2023;Kokorev et al. 2023;Labbe et al. 2023;Maiolino et al. 2023;Matthee et al. 2023 for studies that appeared after our initial submission of this paper).
Motivated by these recent JWST findings of high redshift AGNs, we systematically search for broad-line AGNs at z > 4 using the JWST/NIRSpec spectroscopic data presented in Nakajima et al. (2023).Our sample of low-luminosity AGNs with low-mass SMBHs allows us to obtain the first statistical view on the physical properties of faint AGNs and their host galaxies at z > 4 such as black hole masses, Eddington ratios, number densities, host's stellar mass, and morphologies, crucial to understand the galaxy-SMBH co-evolution in the early universe and implication for cosmic reionization.Moreover, the black hole masses of our low-luminosity AGNs can be directly compared with AGNs at z ∼ 0, allowing us to understand the redshift evolution over cosmic time.
This paper is organized as follows.Section 2 presents the JWST and HST observational data sets used in this study.In Section 3 we explain our systematic sample selection and the final AGN sample.We show our main results in Section 4 and discuss the contribution to cosmic reionization and the nature of our AGNs in Section 5. Section 6 summarizes our findings.Throughout this paper, we use the Planck cosmological parameter sets of the TT, TE, EE+lowP+lensing+BAO result (Planck Collaboration et al. 2020): Ω m = 0.3111, Ω Λ = 0.6899, Ω b = 0.0489, h = 0.6766, and σ 8 = 0.8102.All magnitudes are in the AB system (Oke & Gunn 1983).

JWST/NIRSpec Spectra
We use JWST/NIRSpec datasets reduced in Nakajima et al. (2023).Here we briefly describe the observations and the data reduction.Please see Nakajima et al. (2023) for details.
The data sets used in this study were obtained in the Early Release Observations (EROs; Pontoppidan et al. 2022) targeting the SMACS 0723 lensing cluster field (ERO-2736) and the Early Release Science (ERS) observations of GLASS (ERS-1324, PI: T. Treu;Treu et al. 2022) and the Cosmic Evolution Early Release Science (CEERS; ERS-1345, PI: S. Finkelstein;Finkelstein et al. 2022b).The ERO data were taken in the medium resolution (R ∼ 1000) filter-grating pairs F170LP-G235M and F290LP-G395M covering the wavelength ranges of 1.7 − 3.1 and 2.9 − 5.1 µm, respectively.The total exposure time of the ERO data is 4.86 hours for each filtergrating pair.The GLASS data were taken with high resolution (R ∼ 2700) filter-grating pairs of F100LP-G140H, F170LP-G235H, and F290LP-G395H covering the wavelength ranges of 1.0 − 1.6, 1.7 − 3.1 and 2.9 − 5.1 µm, respectively.The total exposure time of the GLASS data is 4.9 hours for each filter-grating pair.CEERS data were taken with the Prism (R ∼ 100) that covers 0.6 − 5.3 and medium-resolution filter-grating pairs of F100LP-G140M, F170LP-G235M, and F290LP-G395M covering the wavelength ranges of 1.0 − 1.6, 1.7 − 3.1 and 2.9 − 5.1 µm, respectively.The total exposure time of the CEERS data is 0.86 hours for each filtergrating pair.These data were reduced in Nakajima et al. (2023) with the JWST pipeline version 1.8.5 with the Calibration Reference Data System (CRDS) context file of jwst 1028.pmap or jwst 1027.pmap with additional processes improving the flux calibration, noise estimate, and the composition.Reduced spectra are corrected for slit loss (see Nakajima et al. 2023 for details).Finally, we obtain NIRSpec spectra of a total of 185 galaxies at z spec = 3.8 − 8.9.

JWST/NIRCam and HST/ACS&WFC3 Images
For the GLASS field, we use the JWST/NIRCam and HST/ACS&WFC3 images produced by the Ultradeep NIRSpec and NIRCam ObserVations before the Epoch of Reionization (UNCOVER) team.1In the GLASS NIRSpec field around the center of the Abell 2744 lensing cluster, the JWST/NIRCam images were taken in the F115W, F150W, F200W, F277W, F356W, and F444W bands in the UNCOVER program (GO-2561, PIs: I. Labbe and R. Bezanson;Bezanson et al. 2022), and the HST/ACS and WCS3 images were taken in the F435W, F606W, F814W, F105W, F125W, F140W, and F160W-bands in the Hubble Frontier Field program (PI: J. Lotz; Lotz et al. 2017) and the Beyond Ultra-deep Frontier Fields And Legacy Observations (BUFFALO) program (PI: C. Steinhardt;Steinhardt et al. 2020).
For the CEERS field, we use the JWST/NIRCam and HST/ACS&WFC3 official images produced by the CEERS team (version 0.5; Bagley et al. 2022).2In the CEERS field, the JWST/NIRCam images were taken in the F115W, F150W, F200W, F277W, F356W, F410M, and F444W bands in the CEERS program, and the HST/ACS and WCS3 images were taken in the F606W, F814W, F105W, F125W, F140W, and F160W-bands in the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS; Grogin et al. 2011;Koekemoer et al. 2011, see also Finkelstein et al. 2022a).For part of the CEERS field where the official images are not available (i.e., fields whose data were taken in December 2022), we reduced the NIRCam data in the same manner as Harikane et al. (2023) using the JWST pipeline version 1.8.5 with the CRDS context file of jwst 1027.pmap.

Emission Line Fitting
From 185 galaxies at z spec = 3.8 − 8.9 in Nakajima et al. (2023), we search for broad-line AGNs that show broad permitted lines such as Hα and Hβ.We fit rest-frame optical emission lines such as Hβ, [Oiii]λλ4959,5007, Hα, and [Nii]λλ6548,6584.We use the line spread functions made from spectra of a planetary nebula in Isobe et al. ( 2023) for spectra taken with medium and high-resolution gratings.For spectra taken with the Prism, we use the FWHM measured in Isobe et al. ( 2023) from the planetary nebula's spectra assuming a Gaussian profile because Prism's line spread function made from the spectra is severely contaminated by nearby lines due to its low resolution.
First, we fit the Hβ and [OIII]λλ4959,5007 lines with an assumed line ratio of f [OIII]5007 /f [OIII]4959 = 2.98 that is taken from theoretical calculations (Storey & Zeippen 2000).Initially, we fit the emission lines with a single component.Most sources can be fitted with a single component whose line width is narrow (FWHM < 500 km s −1 ), but some sources show a weak moderately broad component in [Oiii]λ5007 that may be due to outflow based on visual inspection (see Zhang et al. 2023c for statistics).For such sources, we simultaneously fit the [OIII]λλ4959,5007 with the narrow (FWHM narrow < 500 km s −1 ) and outflow (FWHM outflow > FWHM narrow ) components.Then we fit the Hα and [Nii]λλ6548,6584 emission line with an assume line ratio of f [NII]6584 /f [NII]6548 = 2.94.We find that the [Nii] and most of the Hα lines can be fitted with a single narrow component (FWHM < 500 km s −1 ), but some sources show a broad component in Hα.For such sources, we simultaneously fit the narrow (FWHM narrow < 500 km s −1 ) and broad (FWHM broad > 500 km s −1 ) components for Hα, and the narrow components for [Nii]λλ6548,6584.For sources showing the outflow components in the [Oiii] line, we add an outflow component (FWHM narrow < FWHM outflow < FWHM broad ) in Hα.

Selection Criteria
Based on these fitting results, we select type 1 AGNs showing broad emission only in a permitted line (i.e., Hα and/or Hβ).We select sources that show 1. broad (FWHM > 1000 km s −1 ) and significant (SNR > 5) permitted Hα and/or Hβ emission line, and 2. narrow (FWHM < 700 km s −1 ) forbidden [Oiii] and [Nii] emission lines, even if an outflow component is seen.

Our AGN Sample
Using the selection criteria above, we select a total of 10 galaxies that show broad Hα emission lines.Figures 1-3 show NIRSpec spectra of the selected galaxies, and Table 1 summarizes the physical properties derived in Nakajima et al. (2023).CEERS 00746 and CEERS 02782 were already reported as broad-line AGNs in Kocevski et al. (2023) as CEERS 3210 and CEERS 1670, respectively, and our measured broad line widths are consistent with those of Kocevski et al. (2023) within 1 − 2σ.These 10 galaxies show significant broad-line emission only in a permitted line (Hα), and their forbidden lines, especially [Oiii]λ5007, which is detected with a signal-to-noise ratio higher than Hα, are well fitted with narrow (FWHM < 500 km s −1 ) components.GLASS 160133 and GLASS 150029 show outflow component whose line width is FWHM = 540 and 140 km s −1 in [Oiii]λ5007, respectively.We evaluate the fittings using the Akaike Information Criterion (AIC; Akaike 1974), which is defined by AIC = −2logL + 2k, where L is the maximum likefood and k is the number of free parameters.As shown in Table 1, all of the selected objects are well fitted with the broad component rather than without the broad component with ∆AIC > 20.These properties are consistent with the fact that they are broad-line type-1 AGNs.
Figure 4 shows the UV magnitudes and redshifts of our selected AGNs.Our selected AGNs are typically faint with a UV magnitude of −22 ≲ M UV ≲ −17 mag, much fainter than the low-luminosity quasars found in ground-based observations at similar redshifts (e.g., Akiyama et al. 2018;Niida et al. 2020;Matsuoka et al. 2018).None of our AGNs is selected as X-ray AGNs in Giallongo et al. (2019), probably because their Xray emission is too faint to be detected with current X-ray observations.Our AGNs are widely distributed in the UV magnitude and redshift space and are not biased compared to the star-forming galaxy sample in Nakajima et al. (2023), except for a possible gap in −21 < M UV < −20 mag probably due to the small sample size.Since we have found 10 AGNs in the sample of 185 galaxies at z > 3.8, about 5% of the galaxies at z = 4−7 with −22 ≲ M UV ≲ −17 mag are broad-line type-1 AGNs, which is higher than z ∼ 0 galaxies, as discussed in Section 4.6.Note that this value, "5%", should be taken with caution, because the observed AGN fraction depends on the sensitivity (see discussion in Section 4.6) and the selection function for the parent spectroscopic sample that could be very complex.
There are two other galaxies showing a broad permitted line but not selected due to their inefficient signalto-noise ratio.CEERS 01465 at z spec = 5.269 shows a possible broad Hα emission line with an FWHM of 3603 +1136 −291 km s −1 , but the signal-to-noise ratio of the broad line is ∼ 4, lower than the threshold value of our selection criteria.The other galaxy is CEERS 01019 at z spec = 8.679, which was first confirmed spectroscopically with Lyα by Zitrin et al. (2015).This galaxy was reported to show a 4.3σ Nvλ1243 whose ionization potential is 77 eV (Mainali et al. 2018), and re-     For GLASS 160133 and GLASS 150029, we also show the outflow components with FWHM ≲ 500 km s −1 .The right panels show the spectra around Hα+[Nii]λλ6548,6584 with the logarithmic scale.The broad-line components only seen in Hα, which are detected with a higher signal-to-noise ratio than [Oiii]λ5007, indicates that these objects are type-1 AGNs.z>3.8 Galaxies Our analysis also shows a similar broad Hβ line with FWHM = 1664 +475 −302 km s −1 with ∼ 3σ, consistent with the results in Larson et al. (2023).Although these two galaxies are good candidates for broad-line AGNs, we do not include them in our final sample due to the moderate signal-to-noise ratio of the broad line.
Broad emission lines are sometimes interpreted as mergers or outflows, but the observed broad emission lines in our AGNs seen only in Hα are not likely made by mergers or outflows.Merging galaxies show broad emission lines both in the permitted and forbidden lines and the emission lines are decomposed with two or more narrow components, typically with FWHM < 500 km s −1 (e.g., Hashimoto et al. 2019;Romano et al. 2021), different from the broad (FWHM > 1000 km s −1 ) single component only seen in Hα in our AGNs.Galaxies with strong outflows also show broad emission lines, but such broad lines are usually seen in both permitted and forbidden lines (e.g., Freeman et al. 2019;Xu et al. 2022).Moreover, the line width of the broad line in starforming galaxies is typically FWHM ∼ 300−500 km s −1 , not larger than 1000 km s −1 .Type-2 AGNs sometimes show broad emission lines due to outflows with FWHM > 1000 km s −1 , but this broad component is again seen in both [Oiii] and Hα lines (e.g., Förster Schreiber et al. 2014;Genzel et al. 2014).One remaining possibility is a strong low-metallicity outflow whose metallicity and ionization parameter are tuned to show a broad component only in Hα, but such outflow is not observed even in low-metallicity galaxies (Xu et al. 2022).Given these comparisons, it is reasonable to interpret that our selected galaxies harbor broad-line type-1 AGNs.

Morphology
Figure 5 shows snapshots of our selected AGNs, and Figure 6 present their false color images.Remarkably, the selected AGNs have a variety of morphologies, including not only compact point sources but also extended sources.CEERS 00746, CEERS 00672, and CEERS 02782 show compact morphologies consistent with the point-spread function (PSF).Among these three sources, CEERS 00746 (CEERS 3210 in Kocevski et al. 2023) and CEERS 00672 show red spectral-energy distributions (SEDs) and large dust attenuation estimated from the Balmer decrement as presented in Section 4.2, indicating that these two sources are dusty red AGNs.The remaining seven AGNs show extended morphologies, different from the ones reported in Kocevski et al. (2023).Such a high fraction (70%) of extended morphologies indicates that the total lights of the faint AGNs with −22 ≲ M UV ≲ −17 mag are partly dominated by their host galaxies like Seyfert galaxies, which is discussed in Section 5.2 (see also Bowler et al. 2021).Interestingly, many AGNs show clumpy morphologies that may indicate merger activity, which is consistent with a scenario in which a merger triggers the AGN activity (Hopkins et al. 2006).

Dust Attenuation
We evaluate the dust attenuation of our selected AGNs using the color excess, E(B − V), estimated from the Balmer decrement in Nakajima et al. (2023)   False color stamps of our faint AGNs.We use JWST/NIRCam F115W, F200W, and F356W images, while HST/ACS F814W and HST/WFC3 F125W and F160W are used for the objects whose NIRCam images are not available.Each thumbnail is 2 ′′ ×2 ′′ in size.broad-line AGNs include both blue and red AGNs.The red AGNs are found only in a faint UV-magnitude range with −18 < M UV < −16 mag (Figure 4).

Emission Lines
In Figure 9, we plot the narrow-line ratios of our AGNs in the BPT diagram (Baldwin et al. 1981) that is commonly used to classify galaxies as dominated by emission from AGN or star formation.Our AGNs are located in the region with high [Oiii]/Hβ and low [Nii]/Hα, similar to other galaxies at z > 3.8 without broad line emission (Nakajima et al. 2023;Sanders et al. 2023), and above the sequence of z ∼ 0 galaxies, as seen in z = 2 − 3 star forming galaxies (e.g., Steidel et al. 2014;Shapley et al. 2015;Kashino et al. 2017).These results indicate that our galaxies harboring AGNs cannot be distinguished from normal star-forming galaxies at z > 4 with the BPT diagram, as discussed in Kocevski et al. (2023), because of the low metallicities in these sources.Indeed, low-metallicity (Z < Z ⊙ ) AGNs with moderately weak [Nii] emission are found at z ∼ 0 in Kawasaki et al. (2017).As shown in Table 1, metallicities of our galaxies having AGNs are typically sub-solar, resulting in the weak [Nii] emission in contrast to typical z ∼ 0 AGNs showing strong [Nii] emission.
We have investigated the presence of rest-frame UV emission lines and rest-optical high-ionization lines in these galaxies having AGNs.We do not find significant high-ionization lines such as NeVλ3426 and Heiiλ4686.Rest-frame UV Ciii]λλ1907,1909 emission lines are seen in four objects, CEERS 01244, GLASS 150029, CEERS 01665, and CEERS 00397, with rest-frame equivalent widths (EW) of EW 0 CIII] = 7 − 13 Å, comparable to both star-forming galaxies and AGNs (Nakajima et al. 2018).Among them, CEERS 00397 The BPT diagram.The red symbols show narrow emission line ratios of our galaxies having AGNs at z = 4 − 7, and the cyan symbols are other galaxies at z ≳ 4 in Nakajima et al. (2023).The black contours show SDSS galaxies at z ∼ 0 (Abazajian et al. 2009).The black solid and dashed curves are boundaries between AGNs and starforming galaxies obtained in Kewley et al. (2001) and Kauffmann et al. (2003), respectively.Our galaxies having AGNs cannot be distinguished from normal star-forming galaxies at z > 4 with the BPT diagram.also shows broad (FWHM ∼ 1000 km s −1 ) and redshifted (+800 km s −1 ) Civλλ1548,1551 and tentative Heiiλ1640 lines (the top panels in Figure 10).As shown in the bottom panels in Figure 10, rest-UV diagnostics The spectrum shows the broad (FWHM ∼ 1000 km s −1 ) and redshifted (+800 km s −1 ) Civ, tentative Heii, and clear Ciii] emission lines.(Bottom:) The rest-UV line diagnostics to separate AGNs and star-forming galaxies.The dashed lines are criteria to distinguish AGNs, composites, and star-forming galaxies proposed in Hirschmann et al. (2019Hirschmann et al. ( , 2022)), while the dotted line is to separate AGNs from star-forming galaxies proposed in Nakajima et al. (2018).Our faint AGN, CEERS 00397, is located in the composite region of Hirschmann et al. (2019Hirschmann et al. ( , 2022) ) and around the boundary in Nakajima et al. (2018), indicating that these rest-UV lines in CEERS 00397 are partly dominated by its AGN.
using Ciii], Civ, and Heii indicate that the AGN activity is partly contributing the these emission line fluxes but not dominating the total fluxes, which is consistent with its Seyfert-like nature.

M BH − L bol Relation
We estimate the black hole mass, M BH , and bolometric luminosity, L bol , of our AGNs.The black hole mass is estimated with the following equation calibrated at z ∼ 0 in Greene & Ho (2005): where the L Hα,broad is the extinction-correlated broadline Hα luminosity, and FWHM Hα,broad is the FWHM of the broad Hα emission line.We use the broad-line luminosity as an input of Equation ( 16) following Kocevski et al. (2023), because it is not clear whether the narrowline emission line seen in our AGNs originates from the AGN or Hii regions in its host galaxy.The estimated black hole masses are presented in Table 2.If we include the narrow line component in the Hα luminosity, the black hole mass increase by 0.1 dex on average (0.4 dex at the maximum).
Estimating the bolometric luminosity of our AGNs is not straightforward, because the continuum luminosities observed with the HST and JWST images are possibly significantly contributed by the lights from their host galaxies.Therefore, we instead estimate the bolometric luminosity from the Hα luminosity.Our best estimates of the bolometric luminosity, L bol , comes from the following equation between the luminosity at the rest- frame 5100 Å, L 5100 , and the Hα luminosity including both the broad and narrow components (Greene & Ho 2005): (17) and the relation between L 5100 and L bol with the bolometric correction in Richards et al. (2006), L bol = 10.33 × L 5100 .Since it is not clear whether the narrowline emission in our AGNs originates from the AGN or star formation in host galaxies, we set the lower limit of the bolometric luminosity using Equation ( 17) with the broad-line Hα luminosity as an input assuming that the narrow component comes from the host galaxy and the bolometric correction of L bol = 9.8 × L 5100 in McLure & Dunlop (2004).As the upper limit, we use the following equation in Netzer (2009)   where the Hβ luminosity is estimated from the extinction-corrected narrow-line Hα line assuming Case B recombination.The estimated bolometric luminosities are presented in Table 2.
Figure 11 shows the estimated M BH and L bol of our AGNs.The black hole masses and bolometric luminosities of our AGNs are M BH ∼ 10 6 − 10 8 M ⊙ and L bol ∼ 10 44 − 10 46 erg s −1 , respectively, indicating that our AGNs have lower black hole masses and lower bolometric luminosities than quasars at z ∼ 4 − 7 identified in the ground-based observations (Trakhtenbrot et al. 2011;Matsuoka et al. 2019;Shen et al. 2019;Onoue et al. 2019).The two red AGNs show relatively higher M BH and L bol , indicating that these AGNs might be in the transition phase between the faint AGNs with low M BH and low-luminosity quasars.
The black hole masses of our AGNs are comparable to those of z ∼ 0 AGNs in Liu et al. (2019), but our AGNs show bolometric luminosities higher than those of z ∼ 0 AGNs in Liu et al. (2019) on average, resulting in higher Eddington ratios (λ Edd = L bol /L Edd ).This distribution of higher λ Edd of our AGNs compared to those in z ∼ 0 may be due to selection bias, because AGNs with faint broad-line Hα emission may not be selected in our selection criteria with the signal-to-noise ratio threshold.Deeper NIRSpec spectroscopy is needed to investigate whether the AGNs at z > 4 with M BH ∼ 10 6 −10 8 M ⊙ have systematically higher λ Edd than z ∼ 0 AGNs or not.

AGN-Host Decomposition
To estimate the stellar mass of AGN's host galaxies, we conduct the AGN-host decomposition analysis using the high-resolution JWST and HST images, in the same manner as Zhang et al. (2023d).We conduct the two-dimensional decomposition by fitting the images of our AGNs in all bands with 1) a PSF profile only, 2) a PSF and a Sérsic profile, and 3) two PSFs and one Sérsic profile.For each band, we generate the PSF by selecting and stacking bright stars in the same fields.For the Sérsic profile, we restrict the Sérsic index n to the range of n = 1 − 4. We find that CEERS 00746 and CEERS 00672 are well-fitted with a PSF only model, and their host galaxies are not seen.CEERS 02782 is also well fitted with a PSF, but a tentative residual is seen, possibly indicative of its host galaxy.The other sources are fitted with a PSF and a Sérsic profile, except for CEERS 01236, which is well fitted with a model of two PSFs and a Sérsic profile, indicating that this source could be a dual AGN.Examples of our results are shown in Figure 12, and fitted models are summarized in Table 3.For GLASS 160133, we cannot obtain a reasonable fitting solution with either PSF, PSF+Sérsic, or 2×PSF+Sérsic models.

SED fitting
To estimate the stellar mass, we conduct SED fittings for host galaxy components using prospector (Johnson et al. 2021).The procedure of the SED fitting is the same as that of Harikane et al. (2023), except for the fixed redshifts based on the NIRSpec emission line measurements.We use the stellar population synthesis package, Flexible Stellar Population Synthesis (FSPS; Conroy et al. 2009;Conroy & Gunn 2010) for stellar SEDs, and include nebular emission from the photoionization models of Cloudy (Byler et al. 2017).We assume the Chabrier (2003) stellar initial mass function (IMF) of 0.1 − 100 M ⊙ , the intergalactic medium (IGM) attenuation model of Madau (1995), the Calzetti et al.  for GLASS 160133.Thus we use the stellar mass estimated from the SED fitting to total lights as the upper limit for this object.
(2000) dust attenuation law, and a fixed metallicity of 0.2 Z ⊙ .We choose a flexible star formation history as adopted in Harikane et al. (2023) with a continuity prior.The estimated stellar masses are presented in Table 3.The systematic uncertainty on the stellar mass due to the fixed prior is typically ∼ 0.2 dex (Leja et al. 2019), which is smaller than the statistical uncer-   tainty.For sources whose host galaxies are not seen (i.e., CEERS 00746, CEERS 00672, and CEERS 02782) and GLASS 160133, we estimate the stellar mass by fitting the total lights of that source with the galaxy templates using prospector, and use the derived value as the upper limit.

Result
Figure 13 shows the stellar masses and black hole masses of our AGNs.Compared to the AGNs at z ∼ 0 (Reines & Volonteri 2015), our AGNs at z ∼ 4 − 7 have similar black hole masses but systematically lower stellar masses.Similar results are obtained in previous studies with a smaller number of AGNs (Kocevski et al. 2023;Übler et al. 2023;Kocevski et al. 2023), but we confirm this trend of lower M * (higher M BH ) with a sample of 10 AGNs at z = 4 − 7. Our results at z = 4 − 7 showing lower M * (higher M BH ) than z ∼ 0 AGNs indicate that the black hole grows faster than its host galaxy at high redshift.The fast black hole growth is also suggested by previous studies at z ∼ 2 (e.g., Zhang et al. 2023d).Such over-massive black holes compared to their host stellar masses are indeed predicted in some theoretical simulations (e.g., Toyouchi et al. 2021;Trinca et al. 2022;Inayoshi et al. 2022;Hu et al. 2022;Zhang et al. 2023b).

UV Luminosity Function
We estimate the number density of our broad-line AGNs detected with NIRSpec.Since the selection function including the target selection completeness with  2021) (see also Izumi et al. 2018Izumi et al. , 2019)).The gray crosses and circles are z ∼ 0 AGNs in Kormendy & Ho (2013) and Reines & Volonteri (2015), respectively.The black solid line with the shaded region is the relation at z ∼ 0 in Reines & Volonteri (2015), and the dashed lines correspond to MBH/M * = 0.1, 0.01, and 0.001.Our z = 4 − 7 AGNs show systematically higher MBH (lower M * ) compared to z ∼ 0 galaxies.
NIRSpec/MSA is not known, we cannot precisely calculate the number density of our AGNs.Instead, we present rough estimates based on the spectroscopic results.We divide our AGN sample and the entire galaxy sample in Nakajima et al. (2023) into magnitude and redshift bins and calculate the fraction of the AGNs to the entire sources in each bin.Table 4 summarizes our calculated AGN fraction at each redshift and magnitude bin.Using this AGN fraction, f AGN (z, M UV ), we estimate the number density of the AGNs, Φ AGN , from the galaxy UV luminosity function fitted with the double-power law form, Φ, as where the fitting parameters of the galaxy luminosity function are taken from Harikane et al. (2022), which are consistent with Adams et al. (2020Adams et al. ( , 2022)).The error of the number density includes both the Poisson error (Gehrels 1986) and the cosmic variance.We estimate the cosmic variance following the procedures in Somerville et al. (2004), using the bias value of faint quasars at z ∼ 4 (He et al. 2018) and the effective vol- ume assuming the survey area of 72 arcmin 2 (= 8 NIR-Spec pointings).Using the effective volume calculated here, we can also evaluate the lower limit of the number density based on the observed number of AGNs in each bin.If the 1σ lower limit calculated with f AGN is higher than the volume-based value, we replace the 1σ lower limit with the volume-based lower limit.
Figure 14 shows the calculated number densities of our AGNs at z ∼ 4 − 7 with previous measurements for AGNs and galaxies.Tables 4 and 5 The number density of our broad-line type-1 AGNs detected with NIRSpec is higher than an extrapolation of the quasar luminosity functions (Akiyama et al. 2018;Niida et al. 2020;Matsuoka et al. 2018), because these quasar studies only select compact objects.The best estimates of the number densities are higher than those of the X-ray selected AGNs (Parsa et al. 2018;Giallongo et al. 2019), probably due to the limited sensitivity in X-ray observations.Our best estimates are higher than Matthee et al. (2023) and lower than Maiolino et al. (2023), because of the difference in sensitivities and selection functions.As shown in Figure 15, the typical sensitivity of NIRCam Grism spectra in Matthee et al. (2023) is lower than those of the NIRSpec data in this study and Maiolino et al. (2023), resulting in the low AGN fraction in Matthee et al. (2023).The slightly higher AGN fraction in Maiolino et al. (2023) compared to this study may be due to the slightly higher typical sensitivity in the JADES data, or a biased selection to AGN candidates in the NIRSpec MSA target selection as discussed in Maiolino et al. (2023).

Contribution to Cosmic Reionization
To discuss the contribution of our AGNs to cosmic reionization, we estimate the cosmic ionizing emissivity density, ϵ 912 , and the cosmic photoionization rate, Γ, following Giallongo et al. (2019).First, we estimate the rest-frame UV emissivity at 1450 Å, ϵ 1450 , by integrating the UV luminosity function at each redshift down to M UV = −18 mag, the same limit as Giallongo et al. (2019).Since an AGN does not dominate all of the restframe UV light as discussed in Section 4.1, we multiply the UV luminosity density with the fraction of light from an AGN to all light including both an AGN and its host galaxy, which is estimated to be ∼ 50% in rest-frame UV (1450 Å) based on the decomposition analysis in Section 4.5.1.Then we calculate ϵ 912 from ϵ 1450 using the SED presented in Lusso et al. (2015) and an ionizing photon escape fraction.The escape fraction of faint AGNs is highly uncertain, with a variety of the values reported (f esc = 0.3−0.8)by previous studies (e.g., Micheva et al. 2017;Grazian et al. 2018;Romano et al. 2019;Iwata et al. 2022).In this calculation, we assume f esc = 0.5.The photoionization rate, Γ, is estimated from ϵ 912 using an equation presented in Lusso et al. (2015).We increase the AGN emissivity by a factor of 1.2 to include the contribution by radiative recombination in the IGM following D' Aloisio et al. (2018).Calculated emissivities and photoionization rates are summarized in Table 5.
Figure 16 presents the cosmic ionizing emissivity density and the photoionization rate from our AGNs with previous measurements.The emissivities of our AGNs at z ∼ 4 − 6 are comparable to those of X-ray selected AGNs in Giallongo et al. (2019), although our AGNs selected with broad emission lines may not be the same population from X-ray selected AGNs.The cosmic photoionization rates from our AGNs are also comparable to those of X-ray-selected AGNs.At z ∼ 6, the photoionization rate of our AGNs is lower than the measurement Figure 14.UV luminosity functions at z ∼ 4, 5, 6, and 7.The red diamonds show the number densities of our broad-line (BL) AGNs identified with NIRSpec (Table 4), and the red dashed lines with the shaded regions are the best-fit functions (Table 5).The cyan triangles and circles are X-ray selected AGNs in Parsa et al. (2018) and Giallongo et al. (2019), respectively, and the blue squares show the number densities of quasars at z ∼ 4, 5, and 6 (Akiyama et al. 2018;Niida et al. 2020;Matsuoka et al. 2018).In the z ∼ 5 panel, the red open star and circles are number density estimates of z ∼ 5 faint BL AGNs detected with NIRSpec in Kocevski et al. (2023) and Maiolino et al. (2023), respectively, while the red crosses are estimates for BL AGNs found in NIRCam Grism observations in Matthee et al. (2023).The orange open circles at z ∼ 4 and 6 are number densities of red and compact AGN candidates in Labbe et al. (2023).The results of Matthee et al. (2023) and Labbe et al. (2023) are horizontally shifted by −0.1 dex for visualization purposes.The blue pentagon and circle at z ∼ 7 are lower limits for a red quasar (Fujimoto et al. 2022) and the radio-selected AGN (Endsley et al. 2022), respectively, whose number densities are estimated in Fujimoto et al. (2022).The gray symbols show the number densities of galaxies (diamond: Harikane et al. 2022, square: Bouwens et al. 2021, triangle: Bowler et al. 2015, 2017, circle: Finkelstein et al. 2015).The cyan solid and dashed lines are the best-fit double power-law functions based on the X-ray selected AGNs in Parsa et al. (2018) and Giallongo et al. (2019), respectively, and the blue lines are the best-fit functions and their extrapolations based on the quasars (Akiyama et al. 2018;Niida et al. 2020;Matsuoka et al. 2018).The number densities of our AGNs newly identified in the NIRSpec spectra are higher than the extrapolations of the quasar luminosity function.
Broad-line AGN fraction and 3σ sensitivity for an Hα broad emission line in each study.The sensitivity is for a broad-line width of FWHM = 2000 km s −1 .The purple, blue, and green shaded regions indicate AGN fractions and the 16th−84th percentile of the broad-line sensitivities for reported AGNs in JADES (Maiolino et al. 2023), CEERS+GLASS (this study), and EIGER+FRESCO (Matthee et al. 2023), respectively.The low AGN fraction from NIRCam Grism observations in Matthee et al. (2023) is due to the shallower depth compared to NIRSpec observations in this study and Maiolino et al. (2023).
from the Lyα forest analysis (e.g., Bosman et al. 2022;Davies et al. 2018;Wyithe & Bolton 2011).This indicates that such faint AGNs contribute to cosmic reionization, while the total contribution is not large, up to ∼ 50% at z ∼ 6, because of their faint nature.

Nature of the NIRSpec-Detected AGNs and Redshift Evolution
Our broad-line type-1 AGNs have low-mass black holes with M BH ∼ 10 6 − 10 8 M ⊙ .Given that more than half of the AGNs show extended morphologies, the total lights are partly dominated by lights from their host galaxies rather than those of the AGNs, suggesting that these AGNs are type-1 Seyfert galaxies.Some of them show weak broad Hα emission lines with a broad-tonarrow-line luminosity ratio of L broad /L narrow ∼ 0.2 − 0.6, suggesting that they are intermediate-type AGNs such as type 1.5, 1.8, and 1.9 Seyfert galaxies (Osterbrock 1981;Blandford 1990;Whittle 1992).Indeed, such a weak broad component observed in our AGNs is similarly seen in some Seyfert 1.5-1.9galaxies (e.g., Stern & Laor 2012, their Figure 1).The moderate line width of the broad component (< 2000 km s −1 ) in some of our AGNs are comparable to those of narrow-line Seyfert 1 galaxies.
Although our AGNs are similar to intermediate-type AGNs seen in local universe, the fraction of such AGNs may increase toward higher redshifts.Our analysis indicates that about 5% of the galaxies at z ∼ 4 − 7 harbor faint type-1 AGNs, while studies of local AGNs implies that only 1-2% of galaxies with similar bolometric luminosities (L bol ∼ 10 44 − 10 46 erg s −1 , or L 6166 ∼ 10 43 − 10 44 erg s −1 ) are type-1 AGNs (Stern & Laor 2012).Spectroscopic studies for z ∼ 3 Lyman break galaxies indicate that ∼ 1% of them are type-1 AGNs.Although a larger sample of broad-line AGNs at z > 4 is needed, our study finding the high fraction (∼ 5%) implies a possible redshift evolution of the type-1 AGN fraction.Some of our AGNs show weak (but statistically significant) broad Hα emission lines compared to their narrow components with low broad-to-narrow line flux ratios, which are detected in the deep NIRspec spectra but will be missed in shallow spectroscopic observations.Although the broad emission lines are detected, the low broad-to-narrow line flux ratios suggest that these AGNs may be significantly contributed from Hii regions in their host galaxies, or partly obscured, implying a high obscured fraction in the universe at z > 4. Indeed, several X-ray observations indicate a higher fraction of obscured AGNs in the higher redshift universe at z ∼ 0 − 4 (e.g., Hasinger 2008;Ueda et al. 2014;Liu et al. 2017;Zappacosta et al. 2018;Iwasawa et al. 2020;Peca et al. 2023).From recent ALMA observations, Gilli et al. (2022) predict that AGNs at z > 3 are significantly obscured by their host galaxies' interstellar medium, which may reasonably explain the observed high obscured fraction at high redshifts.This indicates there may be many obscured AGNs (type-2 AGNs) in the z > 4 universe possibly more than the type-1 AGNs we have identified, consistent with recent findings of heavily obscured AGNs at z ∼ 7 (Fujimoto et al. 2022;Endsley et al. 2022).

SUMMARY
In this paper, we conduct a systematic search for broad-line emission in the JWST/NIRSpec deep spectra of a total of 185 galaxies at z spec = 3.8 − 8.9.We identify 10 faint type-1 AGNs at z spec = 4.015 − 6.936, two of which are reported in Kocevski et al. (2023), allowing us to conduct the first statistical census of such faint AGNs in the early universe.Our major findings are summarized below.mitted Hα line (Figures 1-3).Their forbidden [Oiii]λ5007 line is detected with a higher signalratio than Hα, but show only narrow emission with FWHM ≲ 500 km s −1 .These spectral features are consistent with the fact that these objects are type-1 AGNs.The broad emission in some objects is very weak with a broad line-to-narrow line ratio of L Hα,broad /L Hα,narrow = 0.2 − comparable to those seen in local intermediate-type AGNs.
2. High-resolution JWST/NIRCam and HST/ACS and WFC3 images reveal that seven AGNs in our sample show extended morphologies, indicating significant contributions to the total lights from their host galaxies (Figure 5).The remaining three objects are dominated by compact emission consistent with the PSF, and two of them show red SEDs implying that they are dusty compact AGNs (Figures 7 and 8).
3. Our selected AGNs show narrow emission line ratios of high [Oiii]/Hβ and low [Nii]/Hα, similar to star-forming galaxies at z ≳ 4.These AGNs cannot be distinguished from normal star-forming with the classical BPT diagram (Figure 9), probably due to their low metallicities.

Our faint AGNs have low-mass black holes with
M BH ∼ 10 6 −10 8 M ⊙ , remarkably lower than those of low-luminosity quasars previously identified at z > 4 ground-based telescopes 11).Our AGNs at z = 4 − 7 show higher bolometric luminosities than AGNs at z ∼ 0 with black hole masses on due to the selection bias.Deeper NIRSpec spectroscopy is needed to confirm this trend.
5. We estimate the stellar masses of AGN's host galaxies with careful AGN-host analyses.The estimated stellar masses are systematically lower than the black hole mass-stellar mass relation at z ∼ 0 (Figure 13).stellar mass (higher black hole mass) at high redshifts implies fast hole growth, which is consistent with from theoretical simulations.
6.We estimate that ∼ 5% of galaxies at z = 4 − 7 are type-1 AGNs on average, whose fraction is higher than that in local galaxies with similar luminosities.The estimated number density of our faint AGNs is higher than an extrapolation of the quasar UV luminosity functions and is comparable to those of X-ray selected AGNs (Figure 14).Estimates of the ionizing emissivity and photoionization rate by the faint AGNs indicate that such faint AGNs contribute to cosmic reionization, while the total contribution is not large, up to ∼ 50% at z ∼ 6, because of their faint nature (Figure 16).
This study demonstrates that NIRSpec spectroscopy is an efficient and powerful tool to find faint type-1 AGNs with low mass black holes that are embedded in their host galaxies.Future large NIRSpec observations will uncover a large number of high redshift faint AGNs, allowing us to investigate the galaxy-SMBH co-evolution in the early universe.

Figure 1 .
Figure 1.NIRSpec spectra of CEERS 01244, GLASS 160133, GLASS 150029, and CEERS 00746.For each object, the left and middle panels show spectra around Hβ+[Oiii]λλ4959,5007 and Hα+[Nii]λλ6548,6584, respectively.The 2D and 1D spectra are shown in the top and bottom panels, respectively.The red dashed line with the shaded region shows the best-fit broad-line component (FWHM > 1000 km s −1 ) and other red dashed lines show the best-fit narrow components (FWHM < 500 km s −1 ).For GLASS 160133 and GLASS 150029, we also show the outflow components with FWHM ≲ 500 km s −1 .The right panels show the spectra around Hα+[Nii]λλ6548,6584 with the logarithmic scale.The broad-line components only seen in Hα, which are detected with a higher signal-to-noise ratio than [Oiii]λ5007, indicates that these objects are type-1 AGNs.

Figure 4 .
Figure4.UV magnitude and spectroscopic redshifts of our selected broad-line AGNs (the diamonds).The magenta diamonds are red AGNs (E(B − V) > 0.5) and the red diamonds are other AGNs identified in this study.The gray circles show star-forming galaxies inNakajima et al. (2023).

Figure 5
Figure 5. 3 ′′ × 3 ′′ images of our faint type-1 AGNs.The HST/ACS F606W, F814W, and the JWST/NIRCam F115W, F150W, F200W, F277W, F356W, and F444W images are shown.For objects whose NIRCam images are not available, we instead show the HST/WFC3 F125W and F160W images.The NIRSpec MSA aperture is shown with the blue rectangle in the longest wavelength band.More than half of the sources show extended morphologies, indicating that the total lights are significantly contributed from their host galaxies.
Figure 6.False color stamps of our faint AGNs.We use JWST/NIRCam F115W, F200W, and F356W images, while HST/ACS F814W and HST/WFC3 F125W and F160W are used for the objects whose NIRCam images are not available.Each thumbnail is 2 ′′ ×2 ′′ in size.
Figure 7.Dust attenuation estimated from the Balmer decrement as a function of spectroscopic redshift (left) and UV magnitude (right).The magenta diamonds are red AGNs (E(B − V) > 0.5) and the red diamonds are other AGNs identified in this study.The gray circles show star-forming galaxies inNakajima et al. (2023).

Figure 8 .
Figure8.SEDs of two Red AGNs, CEERS 00746 (left) and CEERS 00672 (right).CEERS 00746 is also reported inKocevski et al. (2023).The red circles and arrows show the measured magnitudes and 2σ upper limits, respectively.The blue curve is the best-fit SEDs if we use galaxy templates.The inset panel shows the false color image in Figure6.
Figure9.The BPT diagram.The red symbols show narrow emission line ratios of our galaxies having AGNs at z = 4 − 7, and the cyan symbols are other galaxies at z ≳ 4 inNakajima et al. (2023).The black contours show SDSS galaxies at z ∼ 0(Abazajian et al. 2009).The black solid and dashed curves are boundaries between AGNs and starforming galaxies obtained inKewley et al. (2001) andKauffmann et al. (2003), respectively.Our galaxies having AGNs cannot be distinguished from normal star-forming galaxies at z > 4 with the BPT diagram.
present our number density estimates and parameters of the best-fit double power-law function,

Figure 16 .
Figure 16.(Left:) Cosmic ionizing emissivity density, ϵ912, contributed by AGNs as a function of redshift.The red diamonds show estimates for our AGNs identified with the NIRSpec spectra assuming the escape fraction of 50%.The cyan circles and blue are measurements from X-ray (Giallongo et al. 2019) and quasars (Palanque-Delabrouille et al. 2013), respectively.The gray squares are measurements in Becker & Bolton (2013), and the black curve is the model in Madau & Haardt (2015).(Right:) Cosmic photoionization rate, Γ, a function of redshift.The red show estimates for our AGNs, and the cyan circles are measurements in Giallongo et al. (2019).The gray open symbols show measurements from the Lyα forest analysis in quasar spectra (circle: et al. 2022, diamond: Davies et al. 2018, square: Becker & Bolton 2013, pentagon: Wyithe & Bolton 2011, triangle: Faucher-Giguère et al. 2008).

Table 1 .
Physical Larson et al. (2023)int Type-1 AGNs Name R.A. Decl.zspecMUV 12 + log(O/H) E(B − V) SNR Hα,broad ∆AIC # * CEERS 01019 is reported as CEERS 1019 inLarson et al. (2023).The signal-to-noise ratio of the broad Hβ line instead of Hα is presented.# Difference between AIC values for the fittings without the broad component and with the broad component.

Table 3 .
Stellar Mass of Host Galaxies