The Physical Properties of Star-Forming Galaxies with Strong [O III] Lines at z=3.25

We present an analysis of physical properties of 34 [O III] emission-line galaxies (ELGs) at z=3.254$\pm$0.029 in the Extended Chandra Deep Field South (ECDFS). These ELGs are selected from deep narrow H2S(1) and broad Ks imaging of 383 arcmin$^{2}$ obtained with CFHT/WIRCam. We construct spectral energy distributions (SEDs) from U to Ks to derive the physical properties of ELGs. These [O III] ELGs are identified as starburst galaxies with strong [O III] lines of L([O III]) ~ 10$^{42.6}$ - 10$^{44.2}$ erg s$^{-1}$, and have stellar masses of M* ~ 10$^{9.0}$-10$^{10.6}$ M$_\odot$ and star formation rates of ~ 10-210 M$_\odot$ yr$^{-1}$. Our results show that 24% of our sample galaxies are dusty with Av>1 mag and EW(OIII)$_{rest}$ ~ 70-500 $\AA$, which are often missed in optically selected [O III] ELG samples. Their rest-frame UV and optical morphologies from HST/ACS and HST/WFC3 deep imaging reveal that these [O III] ELGs are mostly multiple-component systems (likely mergers) or compact. And 20% of them are nearly invisible in the rest-frame UV owing to heavy dust attenuation. Interestingly, we find that our samples reside in an overdensity consisting of two components: one southeast (SE) with an overdensity factor of $\delta_{gal}$ ~ 41 over a volume of 13$^{3}$ cMpc$^{3}$ and the other northwest (NW) with $\delta_{gal}$ ~ 38 over a volume of 10$^{3}$ cMpc$^{3}$. The two overdense substructures are expected to be virialized at z=0 with a total mass of ~ 1.1 x 10$^{15}$ M$_\odot$ and ~ 4.8 x 10$^{14}$ M$_\odot$, and probably merge into a Coma-like galaxy cluster.


INTRODUCTION
The past two decades have witnessed a wealth of progress in mapping galaxy formation and evolution. The current generation of multiwavelength deep surveys have revealed the detailed properties of galaxy populations out to z ∼ 2-3, where the cosmic star formation rate density (CSFRD) reaches its peak (Hopkins et al. 2006;Sobral et al. 2013;Madau et al. 2014;Khostovan et al. 2015). At z > 2-3 about one-quarter of the present-xzzheng@pmo.ac.cn day stars were formed in the progenitors of present-day massive galaxies (Madau et al. 2014), preferentially in the overdense environments (Thomas et al. 2005;Chiang et al. 2017). Characterizing the properties of galaxies at z > 3 is thus essential to understanding the early formation of massive galaxies and large-scale structures, as well as how the star formation activities are activated to reach the peak of CSFRD (Suzuki et al. 2015;Onodera et al. 2016Onodera et al. , 2020. The emission lines in the rest-frame optical spectra of galaxies (e.g., [O II] Kewley et al. 2019, for a review). Moreover, studies of emission-line galaxies (ELGs) at z > 3 provide insights into understanding the cosmic reionization (de Barros et al. 2016). The universe is fully ionized by z ∼ 6 (e.g., Fan et al. 2006;de Barros et al. 2014). Star-forming galaxies (SFGs) at z > 6 are thought to be the main contributors to the ionizing field in the era of reionization (e.g., Nakajima et al. 2014;Robertson et al. 2015). Owing to the opaque intergalactic medium (e.g., Worseck et al. 2014), the nature of ionizing sources in the reionization era is still not well understood.
These ionizing sources usually have prominent [O III]+Hβ emission (De Barros et al. 2019;). The strong [O III] emission lines may reveal the extreme conditions of the interstellar medium in a galaxy, and likely are associated with low metallicity and high ionizing parameters (McLinden et al. 2011;Nakajima et al. 2014;Onodera et al. 2020;Tang et al. 2021b). The extreme [O III] ELGs are often seen as analogs of galaxies in the reionization era Du et al. 2020;Tang et al. 2021aTang et al. , 2022. And galaxies with large [O III] equivalent widths (EWs; from 200Å to 800Å) are widely used to address the Lyα continuum escape fraction in the high-z universe (Fletcher et al. 2019; Barrow et al. 2020;Katz et al. 2020;. Their analogs at low z refer to the so-called "Green Pea" galaxies (Cardamone et al. 2009 (Jaskot et al. 2013;Yang et al. 2017;Yuma et al. 2019;Lumbreras-Calle et al. 2021;Liu et al. 2022). The [O III] lines redshift into the near-infrared (NIR) and mid-infrared (MIR) bands for z > 3 objects. Deep IR photometric and spectroscopic observations are thus crucial to identifying and studying SFGs at z > 3 (e.g., Bunker et al. 1995;Geach et al. 2008;Nakajima et al. 2013;Sobral et al. 2013;Khostovan et al. 2016).
However, the NIR observations of high-z galaxies can be carried out only in the J, H and K s bands on the ground owing to the atmospheric transmission. And the high sky background leads such observations to be very time-consuming and available only for a limited sky area. MOSFIRE on board the Keck telescope is an efficient instrument in taking NIR spectroscopy of high-z galaxies (McLean et al. 2012). The NIR spectroscopic surveys with MOSFIRE, e.g., the Keck Baryonic Structure Survey (KBSS-MOSFIRE; Steidel et al. 2014) and the MOSFIRE Deep Evolution Field survey (MOS-DEF; Kriek et al. 2015;Shapley et al. 2015;Reddy et al. 2018), obtained rest-frame optical spectra of thousands of galaxies mostly at z ∼ 1.4-3 based on the H-band selection. In contrast, the NIR observations with space-borne facilities are free from the atmospheric emission, but constrained by the thermal emission from the facilities. The IR grism surveys using WFC3 on board the Hubble Space Telescope (HST ), e.g., the WFC3 Infrared Spectroscopic Parallel Survey (WISP; Atek et al. 2011), the MAMMOTH-Grism HST slitless spectroscopic survey ) and the 3D-HST survey (Brammer et al. 2012;Momcheva et al. 2016), provided lowresolution rest-frame optical spectra for a large number of galaxies out to z ∼ 2.5. These NIR surveys have built a more comprehensive view of the physical properties of galaxies at 1 < z < 3. The James Webb Space Telescope (JWST) will offer unprecedented sensitivities in the NIR and MIR to conduct imaging and spectroscopy of high-z galaxies and revolutionize our understanding of galaxy formation and evolution since the era of reionization.
Deep imaging through narrow-and broadband K s enables us to detect the emission lines [O III]λλ4959,5007 in galaxies over 3 < z < 3.7, and even determine the [O III] luminosity functions (Reddy et al. 2008;Kashikawa et al. 2011;Khostovan et al. 2015;Sobral et al. 2015;Gong et al. 2017;Khostovan et al. 2020). Such observations are often used to identify Hα and other emission lines at lower redshifts (Khostovan et al. 2015(Khostovan et al. , 2016. It has been verified that the approach with NIR narrowband imaging is effective in probing ELGs within a narrow redshift range of δz/(1 + z) =1%-2% over a large sky coverage (Sobral et al. 2013). On the other hand, the presence of strong [O III] emission lines may cause an excess of the observed K s flux relative to the continuum flux derived from broadband spectral energy distributions (SEDs) and be used to identify [O III] ELGs at 3 < z < 3.7 (Onodera et al. 2020). Pilot studies of [O III] ELGs with spectroscopic observations have been contributed to addressing the kinematic and structural evolution of [O III] SFGs (Steidel et al. 2010;McLinden et al. 2013;Schenker et al. 2013;Gillman et al. 2019;Price et al. 2020;Tran et al. 2020;Yates et al. 2020), as well as metal enrichments (e.g., Kewley et al. 2008;Mannucci et al. 2010;Sommariva et al. 2012;Nakajima et al. 2014;Troncoso et al. 2014;Nakajima et al. 2016).
In the Extended Chandra Deep Field South (ECDFS), a deep narrowband imaging survey has been carried out with CFHT/WIRCam, detecting a sample of 34 [O III] ELGs at z ∼ 3.25 (An et al. 2014, hereafter A14). Here we conduct a detailed analysis of the physical properties of these [O III] ELGs using the publicly available multiwavelength data. This paper is organized as follows: In Section 2, we briefly introduce our narrowband imaging observations and multiwavelength data used in our analysis. Section 3 displays the input parameters of SED fitting and gives the results. We present the physi-
The H 2 S(1) and K s imaging data of ECDFS are used to identify ELGs with the K s − H 2 S(1) color excess (Bunker et al. 1995) following where Σ is the significant factor and σ H2S(1) and σ Ks are background noises in the two bands. The Σ is introduced to quantify the significance of a narrowband excess relative to the combined 1σ photometric error from both the narrow-and broadband. Here f H2S(1) refers to H 2 S(1) flux as f H2S(1) = 0.3631 × 10 0.4 (25−H2S(1)) . The noises and fluxes are given in units of µJy. Using the color excess criteria, in total 8720 sources were detected with S/N> 5 in both H 2 S(1) and K s . Their fluxes were measured from the corresponding images using the software tool SExtractor (Bertin et al. 1996). With the selection criteria of Σ = 3 and EW> 50Å, in total 140 objects were securely selected as emission-line candidates.

Public Data
We utilize optical U -, B-, V -, R-and I-band photometric catalog and imaging data from the Multiwavelength Survey by Yale-Chile (MUSYC; Gawiser et al. 2006;Cardamone et al. 2010); HST /ACS F606W (V 606 ) and F850LP (z 850 ) imaging from the GEMS survey (Rix et al. 2004;Caldwell et al. 2008); HST /WFC3 F125W (J 125 ) and F160W (H 160 ) imaging from the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS; Grogin et al. 2011;Koekemoer et al. 2011); and CFHT/WIRCam J and K s imaging data (Hsieh et al. 2012), in conjunction with our H 2 S(1) imaging data. Fluxes in these 12 bands are obtained for the sample of 140 emission-line candidates. Note that 16 targets are optically too faint to be included in the MUSYC public catalog, and 72 sources out of the 140 candidates have J 125 and H 160 data since the CANDELS survey only covers a part of ECDFS.
By fitting their SEDs composed of 12-band data points with the software tool EAZY (Easy and Accurate Redshifts from Yale; Brammer et al. 2008), photometric redshifts (photo-z) were derived for the 140 emissionline candidates (the modeling will be introduced in Section 3). Of them, 34 ELGs with 2.8 < z phot < 3.3 are identified as [O III] emitters at z ∼ 3.25. With the public catalogs available in the literature, we also identify that 8 of these 34 ELGs have spectroscopic redshifts (specz). Except for one source that has a spec-z at 3.083, the spec-z of the remaining seven sources is in range of 3.208-3.248, confirming that these ELGs are located at z ∼ 3.25.   · · · · · · · · · · · · · · · 0.09±0.02 0.57±0.06 · · · · · · 0.62±0.08 6.92±0.36 3.65±0.14 34 53.087608 Note that all fluxes are given in units of µJy.

Emitters of Our Samples
The redshift distribution of the 34 [O III] ELGs and H 2 S(1) width are shown in Figure 1. Because the narrowband filter H 2 S(1) has a relatively small band width ∆λ, we attribute the emission line detected in this band to [O III]λ5007 (this will be discussed later in Section 6.1). More details about emission-line source selection and the EAZY SED fittings can ben found in An et al. (2013) and A14.

SED FITTING
In order to maximize signal-to-noise ratio (S/N) for aperture-matched colors between the 12 bands mentioned above, A14 firstly determined colors for MUSYC, HST and CFHT bands, respectively, and then matched three sets of colors to establish SED input fluxes from U to K s . Finally, the SED was scaled up to meet the total flux of K s derived from aperture photometry within a diameter of 2 corrected for the missing flux out of the aperture (see A14, for more details). The photometric fluxes in the 12 bands from U to K s for our sample of 34 [O III] ELGs are listed in Table 1.
Photometric redshifts (photo-z) were derived from these SEDs using EAZY with relatively small errors partially because of the narrowband data points linked to given redshifts traced by emission lines. The galaxy SED templates were generated from a library of six independent templates in EAZY. This gives a fast determination on photo-z but sacrifices the accuracy in modeling the details of SEDs (e.g., line fluxes). Therefore, we utilize the Code Investigating GALaxy Emission (CIGALE; Boquien et al. 2019) to analyze the SEDs of our [O III] ELGs with the improved galaxy templates. CIGALE produces millions of models to fit the observational data and estimates their physical properties such as stellar mass, star formation rate (SFR), and dust attenuation, while applying a Bayesian statistical analysis approach to estimate the results. The photo-z obtained with EAZY are used in the CIGALE fitting as the input redshift since CIGALE is not optimized to measure photometric redshift.
The Chabrier (2003) IMF and the stellar population synthesis model from Bruzual&Charlot (2003) are adopted for the fitting. We set three values of metallicity as 0.0004, 0.008 and 0.02 (for Z =0.02) in stellar models. A delayed form of star formation history (SFH), SFR ∝ t/τ 2 main exp(−t/τ main ), is adopted in our fitting too, where t is the time of the star-formation onset and τ main is the e-folding time of the main stellar population. Such a functional form is more physical than a simple exponential SFH because it removes the discontinuity in SFR at t = 0 and is able to produce an increasing SFR when τ is large (Carnall et al. 2019). A starburst component f burst can be added at a given mass fraction as well. This SFH can fit the high-z SFGs well since they usually have a relatively strong star formation activity. It also avoids the systematic biases caused by the degeneracy between the slope of the dust attenuation curve, effective dust attenuation, and the intrinsic UV slope of model templates (Yuan et al. 2019;Villa-Vélez et al. 2021;Qin et al. 2022). The degeneracy is indeed smaller for blue SFGs. However, high-z SFGs may have nonnegligible dust attenuation, and this could increase the degeneracy for these dusty objects.
For nebular emission, the initial parameters include the dimensionless ionization parameter log U , the escape fraction of Lyman continuum photons f esc and the fraction of Lyman continuum photons absorbed by dust f dust . The radiation strength U is set to be −2.0, while f esc and f dust have values of 0.0, 0.1, and 0.2.
We adopt the modified Calzetti law (Calzetti et al. 2000;Noll et al. 2009b) to describe the dust attenuation in our fitting. The attenuation for nebular emission is higher than that for stellar emission. We set E(B − V ) factor , a ratio of E(B − V ) star to the E(B − V ) gas , to be 0.44 (Calzetti et al. 2000). Meanwhile, the color excess of the nebular lines E(B − V ) gas and the slope of the power law modifying the attenuation curve δ are chosen to change freely, in the range of 0.05-0.8 and −1.0 to 0.2, respectively. We also take the amplitude of UV bump to be 0, 1, 2, 3, or 4, where 3 corresponds to the value of the Milky Way. And the ratio of total to selective extinction R v is fixed to a standard value of 3.1.
Due to the lack of detections in the far-IR, the SED of dust emission is not well constrained in our work. The dust emission is modeled with templates from Dale et al. (2014), which refines the PAH emission and also adds an optional active galactic nucleus (AGN) fraction in the modeling. The star-forming component is parameterized by a single parameter α defined as dM d (U ) ∝ U −α dU , where M d is the dust mass and U is the radiation field intensity. We set the AGN fraction to be zero, and the slope is fixed at α = 2.
As pointed out by Lambrides et al. (2020), the Xray selection might miss faint AGNs at z > 2 because their host galaxies usually have strong star formation activities. As a check, we add detailed AGN models from Fritz et al. (2006) in CIGALE to test our results. The ratio of the maximum to minimum radii of the dust torus is set to be 30 and 100, while the optical depth at 9.7 µm is 0.3 and 2. And the AGN fraction has a value of 0, 0.01, and 0.1. All the input parameters of CIGALE fitting are given in Table 2    The best-fit model from CIGALE is chosen using the least-squares method, and the reduced χ 2 (χ 2 r = χ 2 /(N − 1)) is used as a global indicator to quantify the quality of the fitting. Figure 2 shows the results of CIGALE fitting of 34 [O III] ELGs; their ID, photo-z (spec-z if available), and parameter χ 2 r of the fittings are shown in each panel. The first figure consists of 20 [O III] ELGs with J 125 and H 160 imaging data, while the second panel consists of the remaining 14 [O III] ELGs without these two bands. Only one object (ID=10) shows a large χ 2 r > 5 owing to the detection upper limits in five MUSYC bands. About 65% of the objects have χ 2 r < 1 and 85% have χ 2 r < 3, suggesting that most of our bestfit SEDs are obtained in good quality.
Note that the 2175Å bump is present in the best-fit SEDs for some [O III] ELGs in our sample (e.g., ID=5, 13). We will further examine in Section 4.5 whether such a feature is caused by measurement uncertainties or a solid detection of the 2175Å bump in z ∼ 3. ELGs, only one object is identified as an X-ray source (XID=760) in the 7 Ms catalog. Its absorption-corrected intrinsic 0.5−7.0 keV luminosity is 1.12 × 10 45 erg s −1 . Moreover, this object has the highest [O III] luminosity L [O III] = 10 44.2 erg s −1 , as well as highest stellar mass M * = 10 11.4 M (see Table 3). No other X-ray counterparts are found in the 7 Ms catalog for the remaining [O III] ELGs in our sample. The CIGALE results also show that the majority of our [O III] ELGs contain no or a negligible AGN component. Including the X-ray source, six sample galaxies in our sample have an AGN fraction of 0.1, while about half have no AGN fraction in the fitting.

Morphologies
We use HST /ACS V 606 and z 850 imaging data from the GOODS and GEMS surveys to investigate the morphologies of our sample galaxies. For z ∼ 3.25, these two bands correspond to the rest-frame far-UV (FUV: 1402Å) and near-UV (NUV: 2130Å). Note that three [O III] ELGs do not have V 606 imaging data owing to the incomplete coverage of the GEMS observations. We adopt the HST /ACS F814W (I 814 ) imaging data to replace V 606 for making color images of these three objects. Figure 3 shows the color images made with V 606 and z 850 for our [O III] ELGs. A variety of morphologies can be seen. Here we define five morphological types in terms of the compactness and numbers of components of a galaxy. We assign the types of UV faint, compact, diffuse (including clumpy and tidal ones), merging, and multiple component with type ID 1-5, respectively. The morphological types of these 34 [O III] ELGs are visually classified by three of us (R.W., J.R. and S.L.). The median of three classifications is adopted for each galaxy, and the results are presented in Table 3.
Our results of morphological classifications show that about 6% (2/34) are too faint to be securely resolved in both V 606 and z 850 ; 35% (12/34) appear to be compact with R e < 0. 3 and the majority of these compact ones are relatively bright; 38% (13/34) of our sample [O III] ELGs have diffuse emission out to R = 1 − 2 (7-15 kpc) or have apparent tidal/clumpy features within 1 ; 9% (3/34) look like mergers, with two obvious galactic nuclei connected by tidal bridges; and 12% (4/34) are pairs of two or three components with comparable colors and sizes and are separated by < 2 without clear tidal bridges between them.
There are 20 sample ELGs having the HST /WFC3 J 125 and H 160 images from CANDELS. For z ∼ 3.25, these two bands correspond to the rest-frame NUV (2939Å) and U (3624Å). Figure 4 presents their color images made with the two-band data. We carry out morphological classification with the J 125 and H 160 images. The results are very similar to those based on the V 606 + z 850 images. Note that one object (ID=20) is invisible in V 606 and z 850 and appears as a compact galaxy in J 125 and H 160 . We take the morphological classifications based on V 606 and z 850 as our main morphology properties for these 34 [O III] ELGs.
We point out that there are two UV-faint [O III] ELGs (ID=20, 33) being heavily attenuated by dust in restframe UV. As shown in Figure 2, these two objects' SEDs show clear dust reddening, while several other objects (e.g., ID=10, 12, 29) also exhibit reddening from their SEDs. We find that these galaxies are all compact, indicating that the compactness helps to maintain a dustier star-forming environment, although the rest-UV morphologies are very sensitive to young stellar populations with large uncertainties due to dust attenuation. We stress that our sample selection based on the H 2 S(1) and K s observations is able to pick these dusty [O III] ELGs, which were missed by the previous studies based on the optical sample selections.

Colors and [O III] EWs
The U V J diagram is widely used to separate SFGs and quiescent galaxies (Williams et al. 2009;Brammer et al. 2011;Whitaker et al. 2011). Similarly, the U and V filters given in Maíz Apellániz (2006), and the J filter from the Two Micron All Sky Survey (2MASS) are adopted to derive the rest-frame U − V and V − J colors from the best-fit model SEDs of our [O III] ELGs. The calculation is done with CIGALE . Figure 5 shows the distribution of our [O III] ELGs in the rest-frame U −V and V −J diagrams. The stellar masses derived from CIGALE are used to color-code the data points. The selection criteria to distinguish star-forming and quiescent galaxies are adopted from Williams et al. (2009). For a comparison, we also show a sample of galaxies with 9 < log (M * /M ) < 11 at 2.8 < z < 3.7 from the 3D-HST GOODS-South catalog (Skelton et al. 2014).
It is clear from Figure 5  As described in Section 2.1 (see also A14), the restframe EWs of our [O III] samples are estimated using the formula from Geach et al. (2008) as (2) where ∆λ H2S(1) and ∆λ Ks are the widths of the narrowand broadband filters and f H2S(1) and f Ks are the flux densities in these two bands. The estimated EWs are listed in Table 3  The [O III] EWs of our sample ELGs vary over a wide range from 70Å to 500Å, with a median value of ∼200Å and 15% of them are larger than 300Å. The relation between A V derived from CIGALE fitting and EW is shown in Figure 6. There are eight sample galaxies having  including [O III] emitters at z ∼ 3.2 from Suzuki et al. (2015), and at z ∼ 3.3 from Onodera et al. (2016Onodera et al. ( , 2020. We perform an orthogonal distance regression (ODR) fit to the data points of our sample and estimate the dispersion. In addition, compact SFGs tend to be found at the upper envelope of the SFMS (e.g., Barro et al. 2017;Gómez-Guijarro et al. 2019). Our [O III] ELGs are mainly young and blue SFGs with strong star formation activities, and more than one-third of our sample galaxies are classified as compact ones in terms of their morphologies in the rest-frame UV. We thus argue that the compact [O III] ELGs in our sample resemble compact SFGs at z ∼ 3.25 as young starburst galaxies with high SFRs.

Dust Attenuation
The empirical dust attenuation curve of Calzetti et al. (2000) is widely used for starburst galaxies. Noll et al. (2009b)  ELGs color-coded with stellar mass. The gray density map shows the distribution of galaxies selected with 9 < log (M * /M ) < 11 at 2.8 < z < 3.7 from the 3D-HST GOODS-South catalog. The selection box to separate star-forming and quiescent galaxies is from Williams et al. (2009 power-law function with a slope of δ and adding a 2175Å bump that is described by a Lorentzian-like Drude profile. We take the modified Calzetti law as the dust attenuation curve in our analysis and derive dust attenuation A V from the SED fitting with CIGALE. In practice, the stellar color excess E(B − V ) star and attenuation curve slope δ are set as free parameters in the fitting. From the best-fit results with CIGALE, we obtain E(B − V ) star , δ, and the strength of the 2175Å bump for each sample galaxy. We are able to estimate the attenuation at a given band using the global attenuation formula from Boquien et al. (2019) as  Table 3.
It can be clearly seen from Figure 2 that the 2175Å bump is present in the best-fit SEDs of seven sample galaxies (e.g., ID=5 and 13). The 2175Å bump is redshifted to around 9250Å at z ∼ 3.25, corresponding to I and z 850 . These seven galaxies have relatively high SFRs and A [O III] . Moreover, half of them are compact, and the other half are extended in morphology. We do not see a connection between the presence of the 2175Å bump and galaxy morphology among these z ∼ 3.25 Previous studies reported that the 2175Å bump is commonly seen in SFGs up to z ∼ 2.6 (Buat et al. 2011;Wild et al. 2011;Shivaei et al. 2020;Kashino et al. 2021). Noll et al. (2009a) pointed out that at least 30% of SFGs at 1 < z < 2.5 exhibit a significant 2175Å bump. In our SED fitting, the 2175Å bump is introduced when the z 850 flux is lower than the I flux in an SED. We caution that either an overestimate of the I flux or an underestimate of the z 850 flux might demand a stronger 2175Å bump in the models for a better fit. We examine the 2175Å bump of those best-fit SEDs, finding that most of the seven sample galaxies have a higher I flux than the z 850 . Due to the lack of spectroscopic and more photometric data, it is difficult to securely confirm the 2175Å bump. More efforts are necessary to investigate the origin of the 2175Å bump in z ∼ 3.25 [O III] ELGs.
We estimate A FUV at 1500Å with Equation 3 and show the correlation between A FUV and UV slope β in Figure 8. There are eight sample galaxies that are compact (R < 0.1 ), faint, or nearly invisible in the restframe UV (see Figure 3), indicating a highly obscured environment around these galaxies. In addition, these sources also show upper limits in both U and B in their SEDs (see Figure 2); therefore, the UV slope β becomes meaningless for such a case. Thus, these eight sources are not included in Figure 8. The UV slope β is often used as a measure of dust obscuration in the sense that a redder UV slope (a higher β) is linked to a higher dust attenuation (Meurer et al. 1999). From Figure 8, however, we show that our sample galaxies do not exhibit a clear correlation between β and A FUV . As can be seen, some of our sample galaxies are barely visible in the rest-frame FUV owing to the heavy dust attenuation, and the estimates of β and A FUV could be significantly biased by the leaking UV radiation.

[O III] Luminosity
Following Ly et al. (2011), we estimate the line flux of [O III]λ5007 from the narrowband excess using  Table 3. Figure 9 shows the relation between SFR and [O III] luminosity. A strong correlation is seen. Such a correlation has been reported before (e.g. Straughn et al. 2009;Villa-Vélez et al. 2021). It is also clear that most of our [O III] ELGs have dust-corrected luminosities in the range of 10 42.6 − 10 43.6 . In order to show the representativeness of luminosity for our [O III] ELGs, the luminosity functions of [O III] ELGs at z ∼ 3.24 from Khostovan et al. (2015) are adopted for comparison. Note that the luminosity function parameter results in their work are uncorrected for dust and AGN contribution owing to the undeveloped roles of dust on emission lines at the high-z universe. The characteristic luminosity of [O III] ELGs at z ∼ 3.24 in Khostovan et al. (2015)  The scatter is large owing to the lack of UV photometric data.  .04 a Morphology Type: 1-UV faint; 2-compact; 3-diffuse/clumpy/tidal; 4-merger; 5-multiple components. b A typical error of U − V is 0.12. c A typical error of V − J is 0.24. d X-ray source (with XID=760 in the Chandra 7 Ms catalog). Figure 10 shows the spatial distribution of our sample of 34 [O III] ELGs. It is obvious that they are strongly clustered and form an overdensity. We estimate its overdensity factor as δ gal = N group /N field − 1. The effective detection area of our sample is 383 arcmin 2 . The narrowband filter H 2 S(1) (λ c = 2.130 µm, ∆λ = 0.0293 µm) covers a redshift span of z = 3.254 ± 0.029 for [O III]λ5007, corresponding to a radial scale of 50.9 comoving Mpc (cMpc). The total comoving volume of ECDFS is then estimated to be 41.5 3 cMpc 3 .

General Field Number Density
In order to estimate the overdensity factor of [O III] ELGs in ECDFS, we firstly estimate the number density in general fields. The High-redshift(Z) Emission Line Survey (HiZELS) provides a large sample of ELGs identified by narrowband excesses in COSMOS and UDS (Sobral et al. 2013). The NB K filter (λ c = 2.1210 µm, ∆λ = 0.0210 µm) probes [O III] ELGs at z ∼ 3.24, which is similar to our sample galaxies. Given that the improved photometric catalog is available in COSMOS, we cross the NB K -excess object catalog from Sobral et al. (2013) with the COSMOS2020 catalog (Weaver et al. 2021) to identify the [O III] ELGs in COSMOS. Following the selection criteria used in Khostovan et al. (2015), we pick in total 159 NB K -excess objects with 2.8 ≤ z phot ≤ 4 to be [O III] ELGs. We use this sample to estimate the number density of [O III] ELGs at z ∼ 3.24 in general fields.  ELGs are located in a smaller area than the detection area. We construct a density map for our z ∼ 3.25 [O III] ELGs and estimate the coverage area of the overdensity in ECDFS. Following Zheng et al. (2021), the detection area is divided into a grid of 1 × 1 cells, and the number of ELGs is counted in each cell to obtain the number density. A Gaussian kernel of σ = 1 (i.e., 1.9 cMpc at z ∼ 3.25) is used to convolve the grid and yield the density map as shown in Figure 10. The contour levels are drawn at the 4, 8, 12, 16, and 20 × the surface [O III] emitter number density of general fields.
Clearly, the majority of our sample galaxies are located in the central region of ECDFS. The red solid rectangle in Figure 10  ELGs are also located in the SE component, confirming this overdensity region is located at z ∼ 3.25.
We then estimate the volume number density of [O III] ELGs in these components. We note that the redshift span of z = 3.254 ± 0.029 used to calculate the radial scale may be too large for calculating the comoving volumes because our sample [O III] ELGs unlikely fulfill the band width of H 2 S(1) and the actual radial size of the [O III] overdensity is smaller than 50.9 cMpc. Given that the available spec-z of our sample [O III] ELGs are all at z < 3.255, the half of the redshift span covered by the H 2 S(1) filter (see Figure 1), we thus take half of the radial comoving distance (25.5 cMpc) to be the upper limit for the radial size of the [O III] overdensity. On the other hand, the radial size of the overdensity is unlikely smaller than the the minor axis of each component area, giving a lower limit of the radial comoving distance. Therefore, we take the the lower and upper limits of the radial size to estimate the comoving volume for these three components.

Present-day Mass
The typical size of a protocluster at z = 3 is about 20 cMpc (Chiang et al. 2013). We estimate the expected total mass at z ∼ 0 for the overdensity components at z = 3.25 in ECDFS using from Steidel et al. (1998). Hereρ is the mean comoving matter density of the universe, which equals to 3H 2 0 8πG = 4.1×10 10 M cMpc −3 , δ m is the matter overdensity; and V true = V obs /C. From Steidel et al. (1998), V obs is the observed comoving volume and C is a correction factor estimated using C = 1 + f − f (1 + δ m ) 1/3 , where f = Ω m z 4/7 , and f = 0.98 at z = 3.25. And δ m is linked to the galaxy overdensity by 1 + b δ m = C (1 + δ gal ), where b is the [O III] emitter bias factor. We adopt the linear bias b = 3.43 for [O III] ELGs in the redshift range of 2-3 from Zhai et al. (2021) as the bias for [O III] ELGs at z = 3.25.
We calculate the correction factor C and matter overdensity δ m for two overdensity components. For the SE substructure, we obtain δ m = 4.33 and C = 0.27 for the lower limits and δ m = 2.67 and C = 0.47 for the upper limits. For the NW substructure, we obtain δ m = 4.26 and C = 0.27 for the lower limits and δ m = 2.39 and C = 0.51 for the upper limits. And for the entire structure, we get δ m = 1.83, 1.40 and C = 0.59, 0.67 for the lower and upper limits, respectively. So the present-day mass is then estimated to be ∼ 1.1×10 15 M for the SE substructure and ∼ 4.8 × 10 14 M for the NW substructure. And the present-day mass of the entire structure is ∼ 2.3 × 10 15 M .
Based on these estimates, we conclude that the overdensity traced by our [O III] ELGs is indeed a massive protocluster of galaxies at z ∼ 3.25 in ECDFS. These two substructures are expected to become virialized at z = 0, with the SE substructure probably being a high-mass "Coma-type" cluster of ∼ 10 15 M and the NW substructure forming an intermediate-mass "Virgo-type" cluster of (3-9) × 10 14 M . Moreover, the two substructures are separated by 21.8 cMpc, which is the characteristic size of a massive protocluster at z ∼ 3, and they probably merge into a more massive Coma-like galaxy cluster in the present day.  (Suzuki et al. 2016). We note that seven targets in our sample have spectroscopic redshifts in the range of 3.20 < z < 3.25 and are spatially mixed with other sample galaxies, implying that our sample galaxies are most likely distributed at 3.225 < z < 3.25 and the contribution by [ The vast majority of z > 3 galaxies are star-forming, and those with strong emission lines are very common at high redshifts. Our sample [O III] ELGs at z ∼ 3.25 are about 1-2 Gyr before the cosmic star formation peak and are expected to provide clues for understanding how galaxies grow and enhance star formation. The strong [O III] emission lines can be generated from the ionized regions around the hot young massive stars in a galaxy. Galaxies with extremely strong [O III] emission lines at this epoch are found to have preferentially lower metallicity and higher ionization parameters powered by intense star formation activities (Nakajima et al. 2014 (Forrest et al. 2017;Cohn et al. 2018;Forrest et al. 2018;Onodera et al. 2020;Tran et al. 2020;Tang et al. 2021a,b). They are typically small with M * ∼ 10 8 − 10 9 M and SFR ∼ 20-50 M yr −1 (Maseda et al. 2014;Tran et al. 2020). At increasing stellar mass, EELGs tend to have higher metallicity and stronger continuum emission from evolved stellar populations. Tran et al. (2020)  Protoclusters are considered as ideal laboratories to study galaxy properties in the dense environments, as well as the environmental effects on galaxy formation and evolution. Previous studies on z > 3 protoclusters mainly identify them with Lyα emitters (LAEs), Lyman break galaxies (LBGs), and submillimeter galaxies (SMGs). Up to date, there are more than 30 protoclusters reported at z > 3 with spectroscopically confirmed galaxies (see Harikane et al. 2019, for a review). These protoclusters are likely to form "Virgo-type" galaxy clusters at z = 0 with a total mass of (3-9) × 10 14 M .
Only a few overdensity structures have been reported at z > 3 traced with [O III] emitters (Maschietto et al. 2008;Forrest et al. 2017). We show that our sample [O III] ELGs reside in a massive overdense structure in ECDFS. The SE substructure spreads over an area of 7 × 5 while the NW substructure covers an area of 4. 4 × 4. 4. This overdensity of [O III] ELGs at z ∼ 3.25 is a new structure discovered in ECDFS. The SE and NW components have an overdensity factor about 20-60 over different comoving volumes owing to the the limits of radial comoving distance. These two substructures are expected to be virialized at z = 0 and probably form a massive cluster with ∼ 1.1 × 10 15 M for SE and ∼ 4.8 × 10 14 M for NW. And the two substructures probably merge into a more massive single Coma-like galaxy cluster with ∼ 2.3 × 10 15 M .
In ECDFS, there is one overdensity traced by extreme Hβ + [O III] emitters at z ∼ 3.5 discovered by the ZFOURGE and GOODS-ALMA surveys (Forrest et al. 2017;Zhou et al. 2020). Forrest et al. (2017) found a redshift peak at z = 3.5 with EELGs and SELGs in the ZFOURGE catalog. The seventh nearest-neighbor measure is used to build the overdensities projected on the sky, revealing the densest region of extreme [O III] emitters in ECDFS, with 53 member galaxies over a scale of 8.1 cMpc. Maschietto et al. (2008) reported 13 [O III] emitters around the radio galaxy MRC 0316−257 at z = 3.13. This radio-selected protocluster consists of 32 LAEs over a 7 × 7 region (Venemans et al. 2005). These 13 [O III] emitters form an overdensity with δ gal ∼ 2.5. Kuiper et al. (2012) found that MRC 0316−257 has a foreground structure at z = 3.10 traced by three spectroscopically confirmed [O III] emitters. They pointed out that the two structures are unlikely part of a larger protocluster based on a two-dimensional Kolmogorov-Smirnov test. We identify two substructures in our z = 3.25 overdense structure traced by 34 [O III] ELGs in ECDFS. We lack spectroscopic redshifts to see whether the NW substructure is located at the same redshift as the SE substructure. From the spatial distribution of our 34 [O III] ELGs, the two substructures likely belong to the same large-scale structure.
A recent work from the VANDELS survey presented several overdensities traced by Lyman-α emitters at 2 < z < 4 in CDFS and UDS (Guaita et al. 2020). We take their overdensities near z = 3.25 for a comparison. As shown in Figure 10, the central locations of three overdensities at z=3.17, 3.23, and 3.29 are overplotted in ECDFS. However, the spatial locations of the VANDELS Lyα overdensities are not exactly coincident with our SE and NW components. The spatial offsets between them could be explained by the systematic offsets between different populations from the density tracers, say, the [O III] and Lyα emitters. [O III] emitters are more massive and metal rich, while the Lyα emitters are mostly low-mass and metal poor. The typical present-day mass of the VANDELS overdensities is about 0.3 × 10 13 M , which is about a factor of two and three lower than our NW and SE components, respectively.
In addition, some of our sample [O III] ELGs exhibit significant dust attenuation and high SFR, compared to normal SFGs at the same redshifts. This hints that star formation and metal enrichment in this overdensity are enhanced. No detection of extreme [O III] ELGs (EW([O III]) rest > 500Å) in this overdensity also supported the acceleration of galaxy evolution in the overdense environment, in which low-mass and lowmetallicity starburst galaxies are deficient. However, protoclusters at similar redshifts have been found to have quiescent galaxies largely concentrated in the overdense region, probably due to the environmental quenching McConachie et al. 2022). These hint that the evolutionary states of protoclusters largely decide the environmental impacts on the member galaxies at z > 3.

SUMMARY AND CONCLUSIONS
Using the deep narrowband H 2 S(1) and broadband K s imaging of ECDFS, we identify a sample of 34 [O III] ELGs at z ∼ 3.25 and carry out an analysis of their physical properties. Using preexisting multiwavelength data, we construct SEDs from U to K s and perform SED fitting with CIGALE to obtain rest-frame U V J colors, stellar mass, SFR, dust attenuation, [ And the large fraction of compact sources in our sample also are tend to be located above the star formation main sequence of z ∼ 3 SFGs.
3. We find that our [O III] ELGs trace an overdense structure at z = 3.25. This structure is composed of two substructures of scales of 5 × 7 and 4. 4 × 4. 4, separated by 21.8 cMpc. We take the half of narrowband filter redshift span as the upper limit and the minor axis of the three different overdensity rectangular areas as the lower limit for the line-of-sight comoving distance to estimate the number density per comoving volume. Our estimate suggests that this structure has an overdensity factor δ gal ∼ 9-12 over a comoving volume of 25 3 -23 3 cMpc 3 . The SE and NW substructures are denser with δ gal in the range of 22-60 over a volume of 15 3 -11 3 cMpc 3 , and 18-57 over a volume of 12 3 -8 3 cMpc 3 , respectively. We estimate their present-day mass to be ∼ 1.1 × 10 15 M for SE and ∼ 4.8 × 10 14 M for NW, and these two substructures are likely to merge into a Coma-like massive cluster with ∼ 2.3×10 15 M at the present day.