The Semiforbidden C iii] λ1909 Emission in the Rest-ultraviolet Spectra of Green Pea Galaxies

We selected all GPs that had existing archival HST/COS spectra covering the wavelengths that include the Lyα emission from two previous HST-GO programs that observed GPs. The sample consists of 10 GPs (Table 1) at redshifts z = 0.1–0.3 with low metallicities (<0.4 Z_⊙; Izotov et al. 2011), of which four are classified as "extreme" GPs based on their high optical emission line ratios, with [O iii] λλ4959, 5007/[O ii] λ3727 > 9 (Jaskot & Oey 2013). For eight of the objects, HST/COS G130M and G160M spectra are available in the archive covering the rest wavelengths λ ∼ 950–1450 Å from GO-12928 (PI: Henry), and two of the remaining galaxies have COS G160M spectra from GO-13293 (PI: Jaskot). The COS G160M spectra from both of these programs provide Lyα emission line detections with a signal-to-noise ratio (S/N) > 10. The sample of GPs represents a range of physical properties with EW (Lyα) ∼ 0–170 Å, nebular oxygen abundances with 7.8 < 12+log(O/H) < 8.3 (0.15–0.40 Z_⊙), and stellar masses (<3 × 10⁹ M_⊙). The GPs have high UV surface brightnesses (<20 mag arcsec⁻²) in the Galaxy Evolution Explorer (GALEX) near-UV (NUV) images and compact sizes (∼1'') that correspond to physical sizes of 2.5–3.5 kpc for their redshifts.

The HST/STIS observations were carried out using the G230L grating that covers the wavelength range 1700–3200 Å at a spectral resolution of R ∼300–600 across these wavelengths. The average dispersion with the G230L grating is 1.58 Å pixel^–1, and the 2 pixel resolution element yields a 3.16 Å wavelength resolution for the spectrum. The STIS is equipped with the NUV-MAMA detector, with a pixel scale of 0025 pixel⁻¹ (or 005 per 2 pixel resolution element). The GPs are barely resolved in the SDSS images, with sizes of ∼1''. We used the 52'' × 05 slit to optimize between slit loss and spectral resolution, resulting in R ∼ 500 over most of the wavelength range. The target acquisition images were taken using short exposures (∼100 s) on the STIS CCD detector. The exposure times for the STIS NUV spectra were estimated based on the GALEX NUV flux for the continuum and using the predicted C iii] λ1909 and O iii] λ1663 emission line fluxes from CLOUDY photoionization models (Ferland et al. 2013). The total exposure times varied from 2500 to 7500 s based on the target brightness and predicted emission line fluxes.

We retrieved STIS spectra from the MAST archive that are pipeline-processed and calibrated using CALSTIS. The pipeline produces flux-calibrated, rectified two-dimensional spectroscopic images and one-dimensional spectra of flux versus wavelength for each target. The default aperture used for spectral extraction in the pipeline is 11 pixels (0275) along the cross-dispersion axis. However, some of the GPs are clearly more extended in the spatial direction based on the two-dimensional spectroscopic image and the acquisition images. We reprocessed the raw data files using updated reference files and performed the spectral extraction using optimized extraction boxes. The standard extraction height of 11 pixels was used for the more compact GPs, and the spectra of the more extended sources were extracted using 21 pixels in the spatial direction. The extracted one-dimensional spectra were smoothed by 2 pixels in the wavelength direction to a final spectral resolution of 3.2 Å pixel^–1. In the cases where there were multiple exposures from two or more orbits, the one-dimensional spectra from multi-orbit visits were combined using scombine in pyRAF. The final combined spectra have S/N ≥ 4 pixel^–1 in the continuum at ∼1900–2000 Å  for all GPs.

The HST/STIS spectra covering rest-UV wavelengths from 1400 to 2200 Å for the sample are presented in Figures 1 and 2. The C iii] emission line is detected with greater than 3σ significance in seven of the 10 GPs. We measured the fluxes and EWs of the C iii] emission line using the splot task within the specred package in PyRAF. We used the HST/COS acquisition images to measure the extent of the UV continuum and correct for the slit losses due to the 05 × 0525 aperture used for HST/STIS spectral extraction for the GPs. In most cases, the UV continuum is compact and the extraction box includes 75%–80% of the flux, except for two galaxies that have multiple UV-bright knots, and only 62% of the flux is within the extraction box. There is a possibility that the nebular emission may be more extended than the continuum. We used the HST/ACS images taken in the narrowband ramp filter tuned to [O iii] λ5007 from HST-GO-13293 (PI: Jaskot) to measure the slit losses for four galaxies (J0303–0759, J0815+2156, J1219+1526, and J1457+2232) in our sample. The extent of the nebular emission and measured slit losses was found to be comparable to that measured from the UV continuum in all cases. The EW(C iii]) measurements were performed by using direct integration of the area under the line, as well as the Gaussian profile fitting method. The measurements from the two methods are in close agreement, and the average of the two is used for further analysis. The continuum flux (F_ν) is assumed to be flat in the vicinity of the C iii] line. The noise in the continuum spectrum on either side of the emission line contributes to the uncertainty in the measured EWs. We include this in the errors by using repeated measurements with different continuum levels to determine the rms uncertainty arising from the continuum noise.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We do not detect any measurable C iii] emission in three GPs, J0911+1831, J1053+5237, and J1137+3524, and we provide upper limits on their C iii] fluxes in Table 2. These GPs with no C iii] emission line in their rest-UV spectrum have dominant interstellar (and possibly stellar) absorption features, in particular, Si iv λλ1393, 1402 and C iv λλ1548, 1550. We detect strong C iv λλ1548, 1550 absorption lines with EW(C iv) = −10.53 ± 0.3, −4.55 ± 0.02, and −14.8 ± 1.91Å in J0911+1831, J1053+5237, and J1137+3524, respectively. In the case of J0911+1831 and J1053+5237, the spectra also show Si iv λλ1393, 1402 absorption. The GP 1137+3524 has a lower redshift (z = 0.194), and only a broad C iv absorption is detected. Because it falls in the low-S/N blue end of the spectrum, Si iv absorption cannot be identified for this galaxy.

Table 2.  Emission Line Fluxes of UV and Optical Nebular Lines

Name	Fλ(C iii] 1909)	(O iii] 1666)	Iλ(C iii] 1909)	(O iii] 1666)	Iλ([O iii] 5007)		(O iii] 1666)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
J030321−075923	0.849 ± 0.09	0.714	1.918 ± 0.21	1.428	52.88
J081552+215623	1.617 ± 0.01	0.460	3.210 ± 0.01	0.853	42.45
J091113+183108	0.446†	0.605	2.318†	2.991	27.84
J105330+523752	0.987†	2.505	2.262†	5.579	43.77
J113303+651341	0.368 ± 0.08	0.556	0.865 ± 0.18	1.278	16.32
J113722+352426	0.856†	1.479	1.666†	2.791	65.06
J121903+152608	1.992 ± 0.01	1.141	3.069 ± 0.02	1.708	61.37
J124423+021540	0.656 ± 0.06	0.742	1.794 ± 0.18	1.971	76.83
J124834+123402	0.998 ± 0.09	0.727	2.439 ± 0.25	1.748	33.13
J145735+223201	2.080 ± 0.02	0.622	5.846 ± 0.06	1.605	95.61

Note. Observed emission line fluxes (columns (2) and (3)) and fluxes corrected for internal reddening and Milky Way extinction (columns (4)–(6)) in units of 10⁻¹⁵ erg s⁻¹ cm⁻². The 3σ upper limits are indicated by the ^† symbol. The O iii] λ1666 fluxes are all upper limits (^†) from HST/STIS spectra, and the expected fluxes (^e; column (8)) are based on the O iii] λ1666/[O iii] λ4363 ratio (column (7)) from CLOUDY models. The [O iii] λ4363 and [O iii] λ5007 fluxes are from the SDSS spectra.

Download table as:  ASCIITypeset image

We do not detect the O iii] λ1663 doublet in any of the GPs, although it is one of the most prominent nebular lines seen in the rest-UV spectra of low-metallicity galaxies. We provide 3σ upper limits for the O iii] λ1663 doublet in Table 2. The photoionization models from JR16 show a tight correlation between the [O iii] λ4363 optical line emission and the O iii] λ1663 emission, and this allows us to predict the expected O iii] fluxes. We estimate a value for the ionization parameter (U) using the O₃₂ versus C iii] EW diagram from JR16. Then, keeping U fixed, we interpolate the predicted O iii] 1663/[O iii] 4363 ratios as a function of metallicity (Z), given the GPs' calculated metallicities. The uncertainty in Z dominates over the uncertainty in our U estimate, and the errors provided in Table 2 show the uncertainty in the predictions given the GPs' metallicity uncertainties. The predicted O iii] λ1663 fluxes are below or close to the detection limits in most cases. In the LyC leaker GP, J1154+2443, Schaerer et al. (2018) detected O iii] λ1663 with EW = 5.8 ± 2.9 Å, which is ∼0.5 EW(C iii]). However, that galaxy has a much lower metallicity than our sample with 12+log(O/H) ∼ 7.65, closer to the metallicities of BCDs (Berg et al. 2016, 2019) that also show significant O iii] λ1663 detections. There are five galaxies in our sample with Z ≤ 0.2 Z_⊙, but even in such low-metallicity galaxies, the O iii] λ1663 can be weak with a median EW < 2.4 Å (Senchyna et al. 2017).

We measured the fluxes and EWs of the Lyα emission from their HST/COS far-UV (FUV) spectra and the [O iii] λ4363, [O iii] λ5007, and Hα emission lines from the SDSS spectra for all of the GPs. The line fluxes were corrected for Milky Way extinction using the Fitzpatrick (1999) extinction law and the Schlafly & Finkbeiner (2011) dust map. The fluxes were corrected for internal reddening using the Balmer decrement derived from the observed Hα/Hβ ratio and the Cardelli et al. (1989) reddening law, taking into account the variation of R_V in the presence of intense UV radiation as outlined in Izotov et al. (2017). All of the GPs exhibit very low internal reddening with  = 0.03–0.2. The extinction-corrected fluxes for rest-UV and rest-optical emission lines are provided in Table 2, and the EWs are presented in Table 3.

The stacked composite rest-UV spectrum of the GPs was created after applying a Doppler correction to bring each individual spectrum to the rest-frame wavelengths using the redshifts inferred from the [O iii] λ5007 emission line in the SDSS spectrum. Each GP spectrum was scaled to match the flux density between 2000 and 2100 Å, which is redward of the C iii] doublet, a continuum region that has high S/N, and free of absorption lines. The stacking was done using the scombine task in the PyRAF/Specred package by averaging at each dispersion point after applying a 3σ clipping factor. The composite spectrum created from all 10 GPs is shown in Figure 3. The composite spectrum reveals weak spectral features, particularly the broad emission feature at the location of the He ii λ1640 line. Only three GPs in the sample show a weak He ii emission feature in their individual spectra, with EW(He ii) = 0.77 ± 0.05 Å for J0815+2156, 2.32 ± 0.02 Å;for J1244+0215, and 1.02 ± 0.03 Å for J1457+2232. The EW(He ii) as measured in the composite spectrum is ∼20% of the C iii] emission flux. As noted in JR16, the photoionization models for GPs predict much weaker nebular He ii λ1640, <10% of C iii]. However, it is known that the models tend to underpredict the He ii emission strengths, as also evidenced by the observed higher optical He ii λ4686/Hβ line ratios for GPs. Additional sources, such as WR stars, shocks, or high-mass X-ray binaries, have been invoked to explain the higher-than-predicted He ii strengths (Shirazi & Brinchmann 2012; Jaskot & Oey 2013). The composite spectrum also reveals stellar photospheric and interstellar absorption features, including Si ii λ1526, C iv λ1548, 1550, and Al iii λ1854, 1862. The C iv absorption does not show the characteristic P Cygni profile shape in the composite spectrum because the stacking is performed using a small sample of 10 galaxies, of which only three have the strong C iv absorption feature and with widely different profile shapes.

Zoom In Zoom Out Reset image size

In Figure 3, the normalized composite spectrum is shown separately for the GPs with C iii] detections and nondetections. The interstellar absorption features are more prominent in the composite spectra of galaxies with C iii] nondetections compared to the stack of the entire sample. The C iv λλ1548, 1550 absorption profiles are stronger in the weak C iii] emitters and those with nondetections. The very broad wing of the C iv feature in the composite spectrum is mainly contributed by the absorption profile of GP 1137+3524 (Figure 1). The Si iv λ1393, 1402 interstellar absorption lines are also more easily identified in the composite for galaxies with C iii] nondetections. In the case of the C iii] emitters, there may be many factors that make the absorption lines less prominent, such as interstellar emission filling in the absorption features, the stronger nebular continuum from the ionized gas, or a difference in the age of the stellar population. Notably, the He ii λ1640 emission feature is prominent in the stack created using only the C iii] emitters and is not present in the composite from the C iii] nondetections. The O iii] λ1663 doublet remains undetected even in the stack created using only the C iii] detections.

By definition, GPs are a class of strong [O iii] λ5007 emitters with EW([O iii]) > 300 Å. The frequency of C iii] emitters (∼70%) we find among the GPs suggests that EELGs selected by the presence of strong high ionization lines (e.g., [O iii] λ5007) are very likely to also be strong C iii] emitters. It is interesting, therefore, to see how this fraction compares to SFGs selected using different criteria. In the sample of He ii emitters in the local universe, C iii] doublet emission is detected in 7/10 galaxies, with EW(C iii]) ∼ 3–14.8 Å (Senchyna et al. 2017). The He ii emitters span a wider range in metallicities (7.81 < 12+log(O/H) < 8.48), particularly toward lower metallicities than our GP sample. Local BCD galaxies with low metallicities (7.2 < 12+log(O/H) < 8.2) have a higher fraction of strong C iii] emitters (Berg et al. 2016, 2019), with EW(C iii]) reaching as high as >15 Å at the lowest metallicities.

At higher redshifts, C iii] emitters have been identified in surveys of SFGs that employ a broader range of selection criteria (e.g., UV luminosity or color selection). Using MUSE observations, Maseda et al. (2017) found 17 C iii] emitters, which represents only ∼3% of their photometric sample of SFGs in the range 1.5 ≤ z ≤ 4 in the Hubble Deep Field South and Ultra Deep Field. The C iii] EWs of these galaxies are in the range of 2–10 Å, similar to the GPs in most cases. However, a larger fraction (five out of 17) among these high-redshift SFGs are strong C iii] emitters with EW(C iii]) > 10 Å, and three of them have EW(C iii]) > 11.7 Å, which is the highest value observed for the GPs (Schaerer et al. 2018). Among the UV-selected star-forming population at 2 < z < 3.8 in the VUDS survey (Le Fèvre et al. 2019), 24% of the SFGs have EW(C iii]) > 3 Å, and of these, 4% have >10 Å. While the number of known high-redshift C iii] emitters at z > 6 is small, in almost all cases, they have high EWs for C iii], and their broadband colors suggest the presence of strong optical emission lines with EW([O iii]) > 500 Å (Stark et al. 2014, 2017). In summary, the frequency of C iii] emitters among GPs and other low-metallicity galaxies with strong optical emission lines is higher compared to populations of SFGs selected based on UV luminosity or colors, which likely span a wider range of metallicities.

Previous studies have highlighted the empirical relation between the C iii] and Lyα emission line EWs (Shapley et al. 2003; Stark et al. 2014, 2015a; Rigby et al. 2015; Nakajima et al. 2018a). Both C iii] and Lyα emission lines are produced in the ionized gas and are powered by the ionizing radiation from young, massive stars. The Lyα is a resonant line and sensitive to neutral gas, while C iii] emission escapes freely. The relation between C iii] and Lyα EWs is important in the context of the potential use of C iii] emission as a redshift indicator and a tracer of the ionizing populations during the reionization epoch. At z > 6, the Lyα photons are resonantly scattered by the neutral hydrogen in the IGM, making the detection of Lyα emission difficult (Konno et al. 2014; Tilvi et al. 2014; Mason et al. 2018). The C iii] emission from actively SFGs at these redshifts, however, is often strong and easily observable (Stark et al. 2015a, 2017; Ding et al. 2017; Hutchison et al. 2019), making it a useful diagnostic emission line at high redshifts.

In Figure 4 (left), we present the observed relation between EW(C iii]) and EW(Lyα) for the GPs, along with other samples at low and high redshift from the literature. The GPs mostly lie with the high-redshift z = 2–6 galaxies, with an EW(C iii]) that spans the range ∼2–10 Å and EW(Lyα) ∼ 0–175 Å. The Pearson correlation coefficient suggests a strong linear relationship between EW(C iii]) and EW(Lyα) for SFGs in Figure 4, with r_P = 0.61, and a low probability (p = 7.80 × 10⁻⁸) that the two quantities are uncorrelated. However, there is no strong correlation that is evident within the GP sample alone. By definition, the GP selection isolates strong emission line galaxies, and the scatter in Lyα EW among the GPs is likely driven by the variations in Lyα optical depth due to the neutral gas. The SFGs that are non-GPs are not necessarily extreme emission line objects, and their weak Lyα emission may indicate that the flux in the emission lines is intrinsically low. One of the two most deviant points in Figure 4 is J1457+2232, which has the highest EW(C iii]) among the GP sample, but the measured EW(Lyα) is weak or absent. Jaskot & Oey (2014) proposed that the dominant broad absorption feature of the Lyα profile in this galaxy implies the presence of a high column density of neutral gas along the line of sight, and the weak Lyα emission probably escapes via scattering, perhaps in a bipolar outflow. On the other hand, J1219+1526 has the highest EW(Lyα) = 174 ± 9 Å among the GP sample and rest-frame EW(C iii]) = 5.66 ± 0.48 Å, which is lower than expected from the empirical EW(C iii])–EW(Lyα) relation. There are various factors that can impact the C iii] line strengths, and they are discussed in detail in Section 3.5.

Zoom In Zoom Out Reset image size

Among the non-GPs, the noticeably deviant points from the empirical correlation between EW(C iii]) and EW(Lyα) are the highest-redshift galaxies: EGS-zs8-1 at z = 7.73 with high EW(C iii]) = 22 ± 2 Å and weak EW(Lyα) = 21 ± 4 Å (Stark et al. 2017) and z7-GND-42912 at z = 7.50 with EW(C iii]) = 16.23 ± 2.32 Å and EW(Lyα) = 33.2 ± 3.2 Å (Hutchison et al. 2019). Both galaxies exhibit very red [3.6]–[4.5] color in the Spitzer IR bands, which results from the large rest-frame EW of the [O iii]+Hβ in the 4.5 μm band, a selection criterion that is comparable to the GP selection at low redshift. For EGS-zs8-1, Stark et al. (2017) found that the velocity offset of the Lyα emission line, Δv_Lyα = 340–520 km s⁻¹, implies that the Lyα profile is modulated by the presence of dense neutral gas close to the systemic redshift of the galaxy. The high column density of neutral gas is responsible for the observed low EW(Lyα). The case of the GP J1457+2232 is similar, as the Lyα profile shows a large velocity separation of ∼750 km s⁻¹ between the emission peaks, which suggests a high column density along the line of sight. The galaxies EGS-zs8-1, z7-GND-42912, and J1457+2232 have higher EW(C iii]) than expected for their EW(Lyα) from the EW(C iii])–EW(Lyα) empirical relation. Such galaxies illustrate the utility of the C iii] emission line as an alternate diagnostic for the spectroscopic redshifts and ISM conditions when Lyα is absorbed by the neutral gas in the ISM and IGM.

The SDSS spectra of GPs provide a suite of rest-frame optical emission lines, which are useful to derive electron temperatures (T_e) and estimate the gas-phase metallicities or nebular oxygen abundances (12+log(O/H)). The nebular abundances used in our analysis are based on the direct method using T_e ([O iii]) calculated from the [O iii] line fluxes from SDSS spectra. In Figure 4 (right), we present EW(C iii]) versus metallicity for the GPs, along with various samples of C iii] emitters at low and high redshifts from the literature (Stark et al. 2014; Rigby et al. 2015; Berg et al. 2016, 2019; Vanzella et al. 2016; Maseda et al. 2017; Senchyna et al. 2017, 2019; Nakajima et al. 2018a). Previous studies have highlighted the significant trend of increasing EW(C iii]) with decreasing metallicity, reaching values ≥5 Å only at 12+log(O/H) < 8.4 or 0.5 Z_⊙ (Rigby et al. 2015; Maseda et al. 2017; Senchyna et al. 2017; Nakajima et al. 2018b). Using photoionization models, it has been shown that low metallicities create ISM conditions that are necessary to produce significant C iii] emission (JR16; Feltre et al. 2016; Gutkin et al.2016; Byler et al. 2018; Nakajima et al. 2018b). At low metallicities, stars have higher effective temperatures with weaker stellar winds, and their harder ionizing SED boosts the supply of C⁺ ionizing photons. Also, since the ionized nebulae predominantly cool via the forbidden emission of metal lines, the low-metallicity nebulae tend to have higher T_e favoring high collisional excitation rates and stronger C iii] emission. At a fixed C/O ratio or carbon abundance, the photoionization models require a metallicity threshold of Z ∼ 0.006 for the C iii] to be strong with EWs in excess of 3 Å. The GPs have metallicities Z/Z_⊙ ≤ 0.4 (or Z ≤ 0.006), and their EW(C iii]) shows a large spread at a given metallicity, indicating that the C iii] emission is also influenced by other factors, such as ionization parameter, age of the stellar population, and C/O ratio (JR16, Nakajima et al. 2018b).

High nebular temperatures, high ionization parameters, and hard ionizing radiation from metal-poor, young stellar populations with ages <5 Myr enhance the C iii] λ1909 emission. The same physical conditions that produce strong C iii] emission also favor the emission of strong forbidden lines from collisionally excited species in the optical wavelengths. The relation between EW(C iii]) and the EW of [O iii] λ5007 (hereafter EW([O iii])) provides insight into the frequency of C iii] emission among the [O iii] emitters and how the relative emission line fluxes inform us about the ISM properties. In Figure 5, we show the empirical relation between EW(C iii]) and EW([O iii]) for the GPs, along with other SFGs for which measurements are available in the literature. The sample presented in Figure 5 is inhomogeneous, in the sense that some of the measurements correspond to individual compact star-forming regions, while the others are integrated measures over the entire galaxies. Also, the apertures used for measuring the rest-UV and rest-optical fluxes are not matched in all cases. However, it is clear that there is a strong trend between the C iii] and [O iii] EWs. The EW(C iii])–EW([O iii]) relation is almost linear, since both C iii] and [O iii] are collisionally excited lines that are favored by higher nebular temperatures and ionization parameters. The Spearman rank correlation statistic between EW(C iii]) and EW([O iii]) indicates a strong correlation with r_Sp = 0.65 (p = 3.02 × 10⁻⁶), and the Pearson correlation statistic suggests a strong linear correlation with r_P = 0.62 (p = 1.41 × 10⁻⁵). As seen from the photoionization models (JR16; their Figure 8), the relation between the two EWs exhibits a large spread that reflects the range in ionization parameter, nebular temperature, and dependence on metallicity. Extreme values of EW([O iii]) > 2000 Å and EW(C iii]) > 10 Å require both high ionization parameters (log U > −2) and young stellar populations with ages less than 3 Myr. The energy requirements for the excitation of these two emission lines are different, although similar physical conditions tend to favor them. Both emission lines require an ionizing spectrum that has high-energy photons, since the ionization energy for C⁺² is 24.4 eV and that for O⁺² is 35.1 eV. However, because of the temperature dependence of the collisional excitation rates, the C iii] emission exhibits a greater sensitivity to metallicity. This may also explain the offset of the GPs in our sample from the other low-redshift samples. The BCDs (Berg et al. 2016) and He ii emitters (Senchyna et al. 2017) mostly have lower metallicities with 12+log(O/H) < 8.0 for the strong C iii] emitters. The location of the low-metallicity GP J1154+2443 (Schaerer et al. 2018), which lies along with the BCDs and He ii emitters in Figure 5, also seems to suggest that the offset for the GPs in this work may be due to their relatively higher metallicities.

Zoom In Zoom Out Reset image size

The EW(C iii]) also shows a trend with EW(Hα) as expected, because the nebular emission lines both depend on the amount of ionizing radiation. The Spearman rank correlation statistic between EW(C iii]) and EW(Hα) indicates a strong correlation with r_Sp = 0.72 (p = 2.85 × 10⁻⁵), and the Pearson correlation statistic suggests a strong linear correlation with r_P = 0.71 (p = 4.75 × 10⁻⁵). The EW(C iii]) versus EW(Hα) relation is tighter than the EW(C iii]) versus EW(Lyα) relation (Figure 4), which is significantly modified by radiative transfer effects, as the Lyα is resonantly scattered. Even the GPs that show large deviations in the EW(C iii]) vs. EW(Lyα) relation follow a tight correlation with EW(Hα). Since Hα is a well-calibrated SFR indicator (Kennicutt 1998), it is encouraging to consider the possibility of using the C iii] emission as a diagnostic for the ionizing flux and SFR at high-z. The C iii] emission depends on the ionization parameter and ionizing flux, as does the Hα emission, but the dependence of C iii] on metallicity and density makes it a weaker diagnostic for SFR. However, when Hα emission is redshifted beyond the NIR wavelengths at z > 7, the C iii] emission may serve as a useful diagnostic for estimating the production rate of ionizing photons (Chevallard et al. 2018; Schaerer et al. 2018).

The elemental carbon abundance is one of the parameters that influences the C iii] emission. Previous studies have used the C iii] λ1909 and O iii] λ1663 emission line doublets to derive the C/O ratios in SFGs (Garnett et al. 1995; Erb et al. 2010; Berg et al. 2016, 2019). Since we do not detect the O iii] λ1663 doublet, we use the dust-corrected C iii] and optical [O iii] doublets to derive the C/O ratios. Following the equation in Izotov & Thuan (1999),

where t = T_e/10⁴ and

The correction factor ICF(C/O) was derived from the CLOUDY photoionization models for log U = −2 and −3. The measured C/O ratios are presented in Table 4, and the range of observed values is consistent with the C/O ratio = 0.20 assumed in the JR16 models. We also used the C iii] fluxes along with the O iii] λ1663 upper limits from the HST/STIS spectra to compute the lower limits on the observed C/O using the equations from Erb et al. (2010). Taking advantage of the tight correlation between O iii] λ1663 and [O iii] λ4363 emission (JR16), we also used the observed [O iii] λ4363 fluxes to get the predicted O iii] λ1663 fluxes from the photoionization models. The C/O ratios derived based on the predicted O iii] fluxes and observed C iii] are also presented in Table 4. The C/O ratios measured using the observed [O iii] and predicted O iii] fluxes are consistent and range from log(C/O) ∼ −0.6 to −1.1, similar to the range of C/O ratios found in high-redshift galaxies (Erb et al. 2010; Stark et al. 2014; Amorín et al. 2017) and local BCDs (Berg et al. 2016).

In this section, we examine how the ionization parameter and optical depth of the ISM affect the detectability of C iii] emission. In Figure 6, we show the relation between EW(C iii]) and the O₃₂ ratio. The [O iii] and [O ii] lines originate from different ionization levels of oxygen; hence, the O₃₂ ratio serves as a proxy for the ionization parameter. A high value for O₃₂ indicates a high ionization parameter, and for SFGs, this could imply that the nebular gas is powered by the ionizing radiation from massive stars of very young ages. Recent studies have suggested that O₃₂ is a useful diagnostic of LyC leakage (Jaskot & Oey 2013; Nakajima & Ouchi 2014; Izotov et al. 2016b, 2018b). The SFGs at low redshifts (z < 0.3) selected from the SDSS based on their high O₃₂ ≥ 5 are mostly found to be LAEs, with few exceptions (Jaskot et al. 2017; McKinney et al. 2019). In a sample of 11 GPs with O₃₂ ≥ 5 targeted with HST/COS FUV spectroscopy to look for escaping ionizing radiation, all of them showed evidence for LyC leakage with escape fractions (f_esc) in the range 2%–72% (Izotov et al. 2016a, 2016b, 2018a, 2018b). In general, the f_esc(LyC) is found to be higher for GPs with higher O₃₂, although the relation between the two quantities shows considerable scatter (Izotov et al. 2016b, 2018b). High O₃₂ has been proposed to be a diagnostic for density-bounded H ii regions, which are potential LyC leaker candidates (Guseva et al. 2004; Jaskot & Oey 2013; Nakajima & Ouchi 2014; Izotov et al. 2018b). It is, therefore, interesting that EW(C iii]) also shows a positive trend with O₃₂; hence, LyC leakers should be strong C iii] emitters because the high ionization parameter would favor C iii] emission. However, the observed strength of the C iii] emission can be reduced due to other factors, such as the decrease in absorbed ionizing flux at low optical depth, the transition of C⁺² to C⁺³ for log U > −2, and a low elemental carbon abundance. The EW(C iii]) versus O₃₂ diagram shows a linear relation for the GPs when C iii] emission is detected, with C iii] strength increasing as the O₃₂ ratio increases, and EW(C iii]) ≥ 3 Å for O₃₂ ≥ 5. For similar O₃₂ values, even the He ii emitters that overlap with GPs in their metallicities have significantly higher EW(C iii]), which may be due to contributions from additional ionizing sources (such as shocks or X-ray binaries).

Zoom In Zoom Out Reset image size

The velocity separation between the peaks of the double-peaked Lyα emission profile shows a tight relation with the escape fraction of Lyα and LyC emission (Henry et al. 2015; Verhamme et al. 2015, 2017; Izotov et al. 2018b). According to the radiative transfer models from Verhamme et al. (2015), the velocity separation is strongly correlated with the neutral hydrogen column density. At low column densities, the Lyα photons experience less scattering in the surrounding neutral medium and are observed closer to the systemic velocity. The galaxies that have small separations (≤300 km s⁻¹) have low column densities (N_{H i} ≤ 10¹⁸ cm⁻²), while those with larger separations (≥600 km s⁻¹) have higher column densities (N_{H i} ≥ 10²⁰ cm⁻²). We used the velocity separations of the Lyα profile peaks measured from the HST/COS spectra for the GP galaxies in our sample to examine the ISM conditions that allow the detectability of the LyC escape. As shown in Figure 6, GPs with small velocity separations have higher escape fractions for Lyα emission and are also potential LyC leakers. Since the EW(C iii]) correlates with EW(Lyα) for SFGs, we examine how the EW(C iii]) relates to velocity separation. There is no obvious correlation overall between EW(C iii]) and the peak velocity separation of Lyα profiles for the GP sample.

Among the sample of GPs in this work, J0815+2156 and J1219+1526 have small velocity separation and high EW(C iii]). The narrow Lyα profiles of J0815+2156 and J1219+1526 indicate that the Lyα radiation escapes more easily with less resonant scattering. Both of these GPs are suggested to be possible LyC leakage candidates by Jaskot & Oey (2014). Henry et al. (2015) also noted that J1133+6513 and J1219+1526 are good candidates for LyC leakers based on the narrow velocity separations (see Table 3). The galaxy J1219+1526 has a high Lyα escape fraction (f_esc = 58%), large EW(Lyα) = 174 ± 9 Å, and EW(C iii]) = 5.7 ± 0.48 Å, while J0815+2156 also has strong Lyα emission (f_esc = 28%) with EW(Lyα) = 68 ± 4 Å and EW(C iii]) = 8.27 ± 0.53 Å. The galaxy J1133+6513 has low EWs, EW(Lyα) = 35 ± 2 Å and EW(C iii]) = 1.67 ± 0.44 Å, even though the velocity separation of the Lyα emission peaks is small (330 km s⁻¹) and f_esc = 31%. Henry et al. (2015) argued that the low EWs for Lyα and Hα emission of this GP are consistent with higher LyC leakage, because the lack of any prominent interstellar (IS) absorption lines supports that the system is optically thin along the line of sight. The small observed EW(C iii]) is also consistent with the high escape fraction of ionizing radiation in optically thin, density-bounded systems. However, the low C iii] values can also be accommodated by models with high optical depth but older ages for the stellar population (JR16). The galaxy J1133+6513 has a relatively high escape fraction, and the low EW(Hα) suggests a low intrinsic EW(Lyα) before scattering, which may reflect an older age for the current starburst or a continuous star formation history.

The galaxies J0303–0759 and J1457+2232 have low EW(Lyα) and the lowest Lyα escape fractions (f_esc ≤ 5% for Lyα) among the GP sample but high observed EW(C iii]). Their neutral column densities are likely high based on the large velocity separations, 460 and 750 km s⁻¹ for J0303–0759 and J1457+2232, respectively. While the velocity separation for GP 0303–0759 is very similar to the range of velocities seen for the LyC leakers (Izotov et al. 2016b, 2018b), the escape fraction for Lyα is low, ∼5%, and it is likely that f_esc(LyC) is also low. The high EWs of the optical emission lines ([O iii] λ4363 and Hα) along with the high O₃₂ ratio suggest that the intrinsic EW(Lyα) should be larger than observed for both of these GPs. Jaskot & Oey (2014) used the C ii λ1334 and Si ii λ1260 IS absorption lines to infer the optical depth and geometry of these systems. Both galaxies show strong IS absorption lines that confirm their high neutral column densities consistent with the observed weak Lyα emission. Both J0303–0759 and J1457+2232 are good local analogs that emphasize the utility of the strong C iii] λ1909 emission in optically thick systems. The galaxies J0815+2156, J1219+1526, and J1457+2232 have comparable O₃₂ (indicating high log U), low 12+log(O/H), and large optical emission line EWs and exhibit the highest EW(C iii]) among the GPs in our sample, while their Lyα emission suggests low optical depths in J0815+2156 and J1219+1526 and a high optical depth in J1457+2232.

In Figure 7, we show the predicted EW(C iii]) as a function of metallicity for the instantaneous burst models with different burst ages (Figure 15 from JR16), along with the measured EWs for the GPs. The EW(C iii]) measurements from the literature for high-z (Erb et al. 2010; Stark et al. 2014; de Barros et al. 2016; Vanzella et al. 2016; Maseda et al. 2017; Le Fèvre et al. 2019) and low-z (Rigby et al. 2015; Berg et al. 2016; Senchyna et al. 2017) galaxies are also shown. The BPASS models are able to fully reproduce the range of C iii] EWs observed for GPs, and for high C iii] EWs >5 Å, the models require very young burst ages (<3 Myr) and high ionization parameters (log U ≳ −2) even at low metallicities. The older burst ages are not able to reproduce the high C iii] EWs > 5 Å even for high ionization parameter values (log U ∼ −1) at the observed metallicities of the GPs. The BPASS models with the youngest ages, ∼1–3 Myr, are also required to accommodate the high EW(C iii]) values observed for the He ii emitters and BCDs with lower metallicities (<1/3 Z_⊙) and most of the C iii] emitters at z > 2. The lower EW(C iii]) values can be accommodated by the models with log U ∼ −2. Although the GPs in our sample only span a narrow range of metallicities, it is evident that their observed EWs follow the expected trend of EW(C iii]) with metallicity from the CLOUDY models. In addition to metallicity and ionization parameter, the C/O ratio also affects the observed EW(C iii]). The models presented in Figure 7 assume a fixed C/O = 0.2, which is consistent with the C/O values inferred for our GP sample (Section 3.4).

Zoom In Zoom Out Reset image size

In Figure 8, the observed EWs of the collisionally excited emission lines C iii] λ1909 and [O iii] λ5007 presented in Figure 5 are compared to the predictions from the photoionization models (Figure 8 of JR16). Among the z > 1 galaxies for which the EWs have been measured, the 1.8 < z < 2.5 galaxies from Maseda et al. (2017) have metallicities that overlap with the GPs and comparable EW(C iii]) and EW([O iii]). The one exception is the galaxy UDF 10-164 from Maseda et al. (2017), with high C iii] EW = 14.80 ± 3.10 Å but lower [O iii] EW = 1170 ± 292 Å compared to the predicted value of ≥2000 Å expected from the models for its metallicity (Z ≥ 0.004). In Figure 8, this galaxy lies with the galaxies that have much lower metallicities (Z ≤ 0.003).

Zoom In Zoom Out Reset image size

The observed EWs for low- and high-redshift non-GPs are clearly offset from the photoionization model grids in the EW(C iii]) versus EW([O iii]) diagram. As noted in JR16, the EW([O iii]) is the likely source of the discrepancy because the models do not include the redder continuum emission from an older population. So, the predicted [O iii] EWs are higher than what is observed, while the EW(C iii]) remains almost unaffected. However, this effect alone may not account for all of the offsets seen for the measured EWs, in particular, for the galaxies that have higher EW(C iii]) than predicted by the models and are only accommodated by models with the youngest ages (∼1–2 Myr) and highest ionization parameters. The He ii emitters from Senchyna et al. (2017) have metallicities similar to the GPs but possibly have additional sources of nebular heating, such as shocks or WR stars, that can explain their higher EW(C iii]). The dwarf galaxies from Berg et al. (2016) have very low nebular oxygen abundances, with 12+log(O/H) < 8.0, ranging from ∼1/5 to ∼1/20 Z_⊙, which are lower than the average metallicities of the GP sample in this study. In Figures 5 and 8, these dwarf galaxies have consistently higher EW(C iii]) and low EW([O iii]) than the models. On the SDSS images, the dwarf galaxies show bright star-forming regions and diffuse continuum from the underlying galaxy, which may partly explain the observed EW([O iii]) being lower than the model predictions compared to the GPs. In addition, the dwarf galaxies show a range in C/O ratios ranging from 0.15 to 0.50, while the photoionization models use C/O = 0.20. The low-metallicity GP, J1154+2443, shows a similar offset as the BCD galaxies, which suggests that the physical conditions in the ionized gas may be different from the input model parameters. Berg et al. (2016) inferred electron temperatures that are higher (∼15,200–19,600 K) for the BCDs compared to the GPs (∼14,100–15,500 K; Jaskot & Oey 2013). For three of the dwarf galaxies, the ionization parameters derived from the UV spectra are −2.15 < log U < −1.5, consistent with the high ionization parameters required by the models to reproduce the higher EW(C iii]). The comparisons with model predictions in Figures 7 and 8 show that the observed range and trends of EW(C iii]) for GPs can be explained by a combination of stellar ages, metallicities, and ionization parameters. While older ages (>3 Myr) fail to reproduce the highest observed EWs, younger ages can produce the observed range of EW(C iii]), with the highest EWs occurring for low Z and high U.

The high LyC escape fractions observed for GPs may indicate that these galaxies are density-bounded systems (Guseva et al. 2004; Jaskot & Oey 2013; Nakajima & Ouchi 2014; Izotov et al. 2018b). Galaxies with highly concentrated star formation, as in the compact GPs, have a high surface density of star formation, and the feedback from such systems can be very effective in clearing out pathways that allow the escape of LyC and Lyα (Heckman et al. 2011; Verhamme et al. 2017). The high LyC escape fractions also have implications for the observed emission line ratios. The density-bounded nebulae are optically thin, and compared to the radiation-bounded nebulae, they have lower column densities of surrounding gas in the outer layers where the low-ionization lines originate (Pellegrini et al. 2012; Jaskot & Oey 2013; Zackrisson et al. 2013). In JR16, we found that predicted C iii] EWs from the photoionization models are lower for density-bounded nebulae and suggested that the C iii] could be a possible diagnostic for optical depth. The transition from radiation-bounded to density-bounded nebulae is equivalent to truncating the nebular gas at different outer radii within the Stromgren sphere. To characterize the effect of varying optical depth in the models, we followed Stasinska et al. (2015) and used the f_Hβ parameter, which is the ratio of the total Hβ produced inside the nebular radius to the total integrated Hβ in a radiation-bounded Stromgren sphere. In this parameterization, f_Hβ = 1 for optically thick radiation-bounded nebulae, and f_Hβ < 1 for optically thin density-bounded nebulae. The photoionization models show that the C iii] flux declines with decreasing f_Hβ, and the effect is strongest for the models with the highest ionization parameters. For high values of U, the C iii] originates at larger radii because the higher-ionization C iv emission dominates in the inner regions of the nebulae closer to the ionizing source.

In Figure 9, the model grids for C iii] EWs versus O₃₂ are shown as a function of age, ionization parameter, and optical depth for four different metallicities. At a given metallicity, the O₃₂ ratio depends on the ionization parameters, but there is a weak dependence on optical depth. The C iii] EW also depends on both the ionization parameter U and the optical depth f_Hβ at a given metallicity. Since the O₃₂ ratio serves as a proxy for U, the JR16 models predict that galaxies with low EW(C iii]) for a given O₃₂ tend to be optically thin, with high escape fractions for LyC. Both age and metallicity also affect the scaling of C iii] EWs with optical depth. In each panel, we show the location of the GPs on the diagnostic grids corresponding to the metallicities derived from the optical spectra. The model grids in Figure 9 are based on an instantaneous burst scenario and assume a C/O ratio = 0.2.

Zoom In Zoom Out Reset image size

The GPs J0815+2156 and J1219+1526 have similar O₃₂ ratios of ∼10 and are likely optically thin based on the velocity separation measured for the Lyα profiles. Both galaxies also have high C iii] EWs, as expected given their high ionization parameters and low metallicities. These GPs, similar to J1154+2443 from Schaerer et al. (2018), are high EW(C iii]) galaxies that have high f_esc(Lyα) > 25% and are expected to be optically thin to the LyC. The strong C iii] emission in all of these GPs is primarily driven by the very young ages, low metallicity, and high ionization parameter that may be a common property among the LyC leakers, and the effect of optical depth (f_Hβ) may be relatively small. Some of the LyC leaker candidates have such high C iii] EWs that they can only be reproduced with starburst ages of 1–2 Myr, suggesting that the LyC leakers are very young or have harder ionizing spectra than the nonleakers.

The galaxies J0303–0759 and J1457+2232 also have low metallicities and comparably high O₃₂ values of ∼7.2, but their Lyα profiles suggest that they are optically thick with f_esc(Lyα) ∼ 0. Both J0303–0759 and J1457+2232 have high EW(C iii]) > 3 Å. Interestingly, the C iii] EWs of J0815+2156 and J1219+1526 are slightly lower than that of the optically thick galaxy J1457+2232, in spite of their high O₃₂ values. It is plausible that there is a slight suppression of the C iii] EW due to the lower optical depth to LyC in J0815+2156 and J1219+1526, although the dominant factors that influence the C iii] strength appear to be the age and ionization parameter. On the other hand, J0303–0759 has lower EW(C iii]) than the other three GPs, although it has low metallicity (Z ≤ 0.003) and high O₃₂ = 7.1. The EWs of the optical nebular lines [O iii] λ5007 and Hα are also relatively lower for this galaxy, which suggests a lower intrinsic ionizing flux and possibly an older burst age for the stellar population. For the remaining GPs in our sample, with low C iii] EW and low O₃₂ < 5, the combined effects of an older average starburst age and higher metallicities influence the EW(C iii]) and do not offer sufficient constraints on the optical depth. Since EW(C iii]) depends on various factors, the model grids presented in Figure 9 only serve as a diagnostic for the effect of optical depth for C iii] emitters with comparable ages, ionization parameters, and metallicities. The analysis of a larger sample is required to isolate the influence of LyC escape on the strength of the C iii] emission.

The semiforbidden C iii] nebular line is one of the most prominent emission features in the rest-UV spectra of low-metallicity SFGs. The C iii] emission appears to be ubiquitous in z > 2–6 low-metallicity galaxies selected by various criteria, including the z > 2 Lyman-break galaxies (Shapley et al. 2003; Steidel et al. 2016), LAEs at z ∼ 3 (Erb et al. 2010; de Barros et al. 2016; Vanzella et al. 2016), gravitationally lensed low-mass galaxies at z > 6 (Stark et al. 2015a, 2017), and UV luminosity-selected SFGs at 1.5 ≤ z ≤ 4 (Maseda et al. 2017; Le Fèvre et al. 2019). As discussed in the previous sections, the interpretation of the C iii] emission depends on various parameters, including metallicity, shape of the ionizing spectrum, starburst age, LyC optical depth, and dust extinction. The high fraction of C iii] emitters among the local GPs (8/11, including J1154+2443) and the IRAC color-selected galaxies at high redshifts show that low-metallicity galaxies that are selected by their strong [O iii] emission line are likely to also show C iii] emission. The higher effective temperatures of low-metallicity ionized nebulae favor the collisionally excited [O iii] and C iii] emission lines. In the presence of hard ionizing radiation with high ionization parameters (log U ≥ −2) powered by very young (∼1 Myr) metal-poor massive stars (Figure 7) or a weak AGN (Le Fèvre et al. 2019), the C iii] emission can be very strong, with EW(C iii]) > 20 Å. In the inner regions of such highly ionized nebulae, C iii] emission can be lowered by the transition to triply ionized carbon resulting in C iv emission. At metallicities Z < 1/5 Z_⊙, the C iv emission line has been observed in local star-forming dwarf galaxies (Berg et al. 2016; Senchyna et al. 2017). Strong C iv emission frequently detected in z > 6 galaxies suggests that such hard ionizing SEDs may be common for SFGs in the reionization epoch (Stark et al. 2015b; Mainali et al. 2017; Schmidt et al. 2017).

The burst ages and star formation history can affect the observed emission line EWs. As shown in JR16, for an instantaneous burst, the high C iii] EWs ∼ 5 Å at young ages <3 Myr are primarily driven by the ionization parameter and metallicity. The same is true for EW(C iii]) > 10 Å and a continuous star formation history. In both star formation scenarios, for a given metallicity and ionization parameter, the EW(C iii]) initially declines quickly with age due to the decrease in the production rate of high-energy photons. However, in the case of continuous star formation, an equilibrium is reached between the birth and death of massive stars beyond ∼20 Myr, such that the C iii] nebular emission remains approximately constant with age and only the increasing nonionizing UV continuum at 1909 Å from the stellar population lowers the EW(C iii]). The EWs of the optical lines are lowered by the continuum from the current star formation, in addition to the contribution to the optical continuum from the underlying older stellar population. Therefore, when comparing the C iii] EWs against EW([O iii]) or EW(Hα), the contribution to the continuum from the older stellar component has to be considered, although the photoionization models used here do not include multiple stellar populations (Figures 5 and 8). For the C iii] emitters at z = 2, Stark et al. (2014) proposed photoionization models with two-component stellar populations, one young (≤3 Myr) and one older, to provide a better fit to their SEDs. Multicomponent star formation histories with a recent burst (<10 Myr) that powers the nebular emission and an older few hundred Myr stellar population that contributes to the UV–optical continuum have been proposed to consistently explain the observed UV spectra and IRAC (rest-optical) colors of z = 6–7 galaxies (Stark et al. 2015a). The GPs and EELGs at lower redshifts also show evidence for multiple stellar populations. Amorín et al. (2012) found that the star formation history of GPs indicates the presence of an evolved stellar component with an age between 10⁸ yr and several Gyr. For example, in the case of J1133+6513 with EW(C iii]) = 1.67 Å and EW([O iii]) = 394 Å, which is in common with our sample, they noted that the presence of the broad Mg i λλ5167, 5173 absorption feature confirms the presence of old stars. The high EWs of the nebular lines, however, are powered by the ionizing continuum from the young stellar component, which may only be <20% of the total mass. The high SFR (>4–60 M_⊙ yr⁻¹) and short mass doubling time of ≤1 Gyr imply that the GPs are currently experiencing a powerful starburst phase that dominates their UV–optical continuum. Izotov et al. (2011) found that the GPs and similar luminous compact SFGs have M_young/M_total ∼ 0.03–0.05, on average, for a galaxy with a total stellar mass of M_total = 10⁹ M_⊙. Although there is a higher fraction of young stars in the lower-mass galaxies, the range in the fraction of young stars for a given galaxy mass can be large based on their star formation histories. Among the GPs in this work, four (J0303–0759, J0815+2156, J1219+1526, and J1457+2232) have O₃₂ > 5 and low-metallicity Z ≤ 0.003, and their large optical emission line EWs are consistent with a very young starburst (<3 Myr). The lower C iii] and optical line EWs of the other GPs may result from older burst ages, lower ionization parameters, or high metallicities, and the effect of each individual parameter cannot be entirely disentangled using the present data.

The same physical conditions that produce strong intrinsic LyC and Lyα emission also favor strong C iii] emission at low metallicities. However, while LyC and Lyα are absorbed or scattered by neutral gas, C iii] can escape unimpeded. Therefore, C iii] is a potentially useful probe for systems where LyC and Lyα are suppressed due to a surrounding neutral ISM or IGM. For instance, J1457+2232 and J0303–0759 in our GP sample are optically thick, with low escape fractions for Lyα and possibly to LyC, and have high EW(C iii]). Many GPs have lower optical depths, allowing the escape of the ionizing continuum, and the LyC leakage can lower the observed EW(C iii]). At fixed age, metallicity, and ionization parameter, a lower EW(C iii]) can be interpreted as arising from a star-forming region that is optically thin to the LyC. As discussed in Section 4.2, the trends seen for the EW(C iii]) with optical depth for a subset of the current GP sample appear to be consistent with this interpretation and should be revisited using a larger sample of C iii] emitters.

The overall carbon abundances and dust content in the ionized nebular regions are also expected to influence the C iii] EWs. Since carbon acts as a coolant in ionized regions, the lower carbon abundance in the low-metallicity galaxies results in higher nebular temperatures and increases the C iii] collisional excitation rates. As shown in JR16, the C iii] emission does not scale linearly with the C/O ratio, but the higher nebular temperatures at low C/O ratios can partially compensate for the lower carbon abundance and result in strong C iii] emission. The LyC-emitting GP J1154+2443 with a high EW(C iii]) = 11.7 ± 2.9 Å has low C/O ∼ 0.13 (Schaerer et al. 2018) compared to most GPs in our sample. The enhanced C iii] emission in this low-metallicity galaxy with 12+log(O/H) ∼ 7.6 is consistent with the JR16 models that predict the strongest C iii] emission at Z < 0.002 because of the higher electron temperature, even if there are fewer carbon atoms. The dust content in ionized nebulae also affects the emergent C iii] EWs through its dependence on the dust extinction and the role of photoelectric heating versus cooling via the forbidden lines. The photoelectric heating can enhance the C iii] emission, but as the dust content increases, the extinction begins to be dominant and lowers the EW(C iii]). However, low-metallicity SFGs are relatively dust-poor systems. The GPs in our sample have very low nebular extinction, with  = 0.03–0.2, as inferred from the Balmer decrement.

Although a large number of C iii] observations have become available in recent years, the nature of the ionizing sources that provide the high ionization parameters to account for the C iii] emission strengths is not fully understood. The C iii] emission requires high-energy photons (>24.4 eV) that can be provided by young, massive stars in low-metallicity SFGs or by AGNs (Feltre et al. 2016; Gutkin et al. 2016; Nakajima et al. 2018b). In the case of SFGs, the effects of binary stellar evolution have to be incorporated to successfully reproduce the high C iii] EWs measured in SFGs consistent with their ages (JR16; Nakajima et al. 2018b). The X-ray observations of GPs reported by Svoboda et al. (2019) show that some GPs have X-ray luminosities that are a factor of 6 higher than expected for SFGs, which may be attributed to a hidden AGN, ultraluminous X-ray sources, or a higher fraction of high-mass X-ray binaries. However, the optical emission line ratios of GPs are compatible with SFGs in the BPT diagram (Baldwin et al. 1981) and inconsistent with an AGN contribution, although some extreme GPs can lie close to the maximum line for starburst (Jaskot & Oey 2013). While the emission lines from GPs in our sample can be accommodated by models with ages ≳3 Myr and the ionization parameter log U ≳ −2, some of the strong C iii] emitters in the literature require extremely young ages (∼1 Myr) and very high ionization parameters (log U ∼ −1). The role of other exotic ionizing sources, such as very massive stars with masses >100 M_⊙ (Smith et al. 2016), and the contribution from low-luminosity AGNs (Nakajima et al. 2018b; Le Fèvre et al. 2019) cannot be ruled out for the most extreme C iii] emitters.

The C iii] emission line in combination with other UV and optical emission lines forms an important diagnostic that helps to reveal the ionizing sources that are responsible for the nebular emission (Feltre et al. 2016; Byler et al. 2018). In addition to photoionization, the contribution of shocks can enhance the flux of C iii] emission and low-ionization optical emission lines. Among the six GPs in Jaskot & Oey (2013), five of them showed He ii λ4686 emission that may be produced by WR stars or shocks. The radiative shocks originating from supernova explosions can generate He ii emission even without the WR stars and are most efficient in the dense, low-metallicity systems that are characteristic of GPs (Guseva et al. 2000; Thuan & Izotov 2005). In order to understand the ionizing spectrum and optical depth effects, the contribution from shocks should be subtracted. Although GPs are young enough to host large numbers of ionizing O stars or WR stars, supernovae and stellar winds from an ongoing or previous burst of star formation can reshape their ISM. In JR16, we showed that the presence of shock always contributes to an increase in the observed C iii] flux and proposed a diagnostic involving C iii]/He ii versus C iv/ He ii to distinguish nebular emission from purely photoionized and shock-ionized gas. However, we only detect weak He ii λ1640 in the composite STIS spectrum from this study. Future spectroscopic observations of GPs with higher spectral resolution and high S/N to measure various shock diagnostics in the UV–optical wavelengths will be required to calibrate the shock diagnostics.

The UV spectra of GPs can constrain the input ISM parameters used in the photoionization models, such as C/O ratios and electron densities. The sample of ∼eight GPs with detectable C iii] emission (including J1154+2443 from Schaerer et al. 2018) shows that the C/O ratios range from ∼0.08 to 0.35 and are similar to the other C iii] emitters locally (Garnett et al. 1995; Berg et al. 2016, 2019; Senchyna et al. 2017) and at z ≥ 2 (Shapley et al. 2003; Erb et al. 2010; Stark et al. 2014; Amorín et al. 2017). The variation of C iii] emission over the same range of C/O ratios has been explored in JR16 and found to have a significant effect on the EW(C iii]). The electron densities, n_e ∼ 100 cm⁻³, used in the photoionization models (JR16) are similar to that seen in typical H ii regions. However, various measurements of electron densities from the [C iii] λ1906 + C iii] λ1908 doublet give n_e values that are about 2 orders of magnitude higher than derived using optical diagnostics (e.g., [O ii] or [S ii] doublets) for galaxies at low and high redshifts (Bayliss et al. 2014, Berg et al. 2018; James et al. 2014, 2018). A possible reason for the different n_e values is that the density-sensitive doublets in the optical are tracing the low-density regions compared to the C iii] doublet, which originates in higher-density regions closer to the ionizing source (Berg et al. 2018; James et al. 2018 ). The existing HST/STIS observations of GPs do not resolve the C iii] doublet lines, and high-resolution UV spectra of GPs would be required to constrain the electron densities used for the model predictions of C iii] emission.

One of the key goals of the JWST and upcoming large 20 m class ground-based telescopes is to reveal the physical properties of the galaxies responsible for the reionization of the universe and quantify their contribution of ionizing photons. The relative contributions of SFGs and AGNs to the total ionizing budget for reionizing the universe are still debated (Madau & Haardt 2015; Finkelstein 2016; Hassan et al. 2018; Matsuoka et al. 2018). The UV luminosity function from deep surveys suggests that SFGs are the primary agents for reionization (Finkelstein 2016), but recent faint AGN surveys suggest that AGNs may have also contributed to the transition (Madau & Haardt 2015). The spectroscopic capabilities of the JWST instruments will play a critical role in identifying AGNs and starbursts in the reionization epoch. While the commonly employed optical nebular diagnostics of the BPT diagram (Baldwin et al. 1981) will be redshifted to the mid-IR wavelengths at z > 8, the bulk of the JWST spectroscopic data at NIR wavelengths obtained using NIRSpec and slitless spectroscopy with NIRISS and NIRCam will provide access to the rest-UV emission lines (e.g., Lyα, C iv, O iii], He ii, and C iii]). For the gravitationally lensed galaxies at z > 8, the NIRSpec IFU, NIRISS, and NIRCam grism modes will provide spatially resolved emission line maps over physical scales of a few tens to hundreds of pc.

The SFGs in the reionization epoch are compact, have high SSFRs and low metallicities similar to the GPs, and are likely to show strong C iii] and nebular UV emission lines. The young ages of the starbursts and low metallicities of the ISM would favor high C iii] EWs (Figures 5 and 8), making C iii] one of the most easily detected emission features in the NIR spectra of z > 6 galaxies. Since the observability of Lyα emission drops at z > 6 due to the increased IGM absorption, the C iii] emission could be used as an alternate probe of the ionizing flux and SFR based on the empirical relation between C iii] and Hα emission lines (Figure 5). The C iii] emission line in combination with other nebular lines, such as N v λ1240, C iv λ1548, 1550, He ii λ1640, O iii] λ1661, 1666, and Si iii] 1883, 1892 Å, can be used to reveal the nature of the ionizing sources and derive the nebular abundances (Feltre et al. 2016; Gutkin et al. 2016; JR16; Nakajima et al. 2018b). The C iv/C iii] ratio that uses UV emission lines that originate from different ionization states of carbon can constrain the ionization parameter, similar to the O₃₂ optical emission line ratio. The C iii] and Si iii] doublet lines can both be used for deriving the electron densities (JR16; Gutkin et al. 2016; Byler et al. 2018), while the commonly used C iii]/O iii] can be used to determine the elemental carbon abundances (Garnett et al. 1995; Berg et al. 2016, 2018). Feltre et al. (2016) used photoionization models to show that the AGNs and SFGs separate out well in the C iv/C iii] versus C iv/He ii and C iii]/He ii versus C iv/He ii diagnostic diagrams. The He ii line serves as a key diagnostic for hard ionizing radiation with high-energy photons (>54.4 eV), since the enhanced He ii emission would lower the C iii]/He ii ratio. The models that include shock contribution can produce emission line ratios that overlap with the AGNs in the C iii]/He ii versus C iv/He ii diagnostic diagram, but the pure photoionization models occupy an entirely different part of the diagram (JR16). As seen from the composite spectrum of C iii] emitters among GPs, the He ii emission line can be strong enough (Section 2.3) to be observable in the low-metallicity, high-redshift galaxies with JWST.

The semiforbidden C iii] nebular emission doublet serves as an important diagnostic spectral line to infer the physical properties of reionizers. However, extensive calibration by applying the UV diagnostics to low-redshift galaxies is necessary for them to be used effectively to interpret sources in the reionization epoch. Currently, the number of galaxies in the low-metallicity regime with the required wavelength coverage to test the diagnostics based on rest-UV emission line ratios at z ≥ 2 is very sparse. Only in recent years have the rest-UV spectroscopic data become available for local analogs of the high-z galaxies, which include GPs, BCDs, and EELGs at lower redshifts. In this work, we have explored the conditions that favor the C iii] emission and examined diagnostics involving the C iii] and optical emission lines for the GPs. We have shown that the GPs with high O₃₂ also have high EW(C iii]) because the high ionization parameters favor C iii] emission, but there is also a weak dependence on the optical depth to LyC (Figure 9). The C iii] emission tends to be weaker for optically thin density-bounded nebulae compared to radiation-bounded nebulae for similar values of the ionization parameter. If these trends are calibrated for a sample of confirmed LyC-leaking GPs, the dependence of the C iii] EW on the optical depths can be used to quantify the LyC escape from SFGs in the reionization epoch. Future work will require deep UV spectroscopy to detect the weaker UV spectral lines for a larger sample of GPs and other local analogs that extend these analyses to lower metallicities and a wide range of Lyα emission profiles. Detailed calibrations of the rest-UV nebular diagnostics in combination with the commonly used optical emission line diagnostics is a crucial step to be able to characterize and interpret the spectroscopic observations of the sources of reionization obtained with JWST.