A Window on the Earliest Star Formation: Extreme Photoionization Conditions of a High-ionization, Low-metallicity Lensed Galaxy at z ∼ 2*

Neither the components of the double Lyα emission feature nor the individual lines of the C iii] λλ1907, 1909 or C iv λλ1548, 1550 doublets are resolved, but they are still well fit by two blended Gaussians. C iv and He ii are further complicated by the potential for their profiles to be modified by multiple components. In the case of He ii, which can have both nebular and stellar emission, we used a two-component fit with narrow and wide profiles, respectively. Both single- and multicomponent fits to the He ii profile are demonstrated in Figure 5. While the two-component model may offer a more visually appeasing fit to the He ii profile, it does not significantly reduce the fit residuals, and so the significance of the wide (∼700 km s⁻¹) stellar component is difficult to assess with the current S/N and resolution. We therefore suggest that the reported fits are upper/lower limits for the stellar/nebular components. Regardless of the correct fit, the He ii emission in SL2S0217 is clearly dominated by a narrow component whose width is consistent with nebular emission (∼280 km s⁻¹).

In the case of C iv, which is typically an ISM absorption feature or a P Cygni profile from the stellar winds of massive stars, it is clearly seen in SL2S0217 as an emission doublet. Because the C iv doublet, like Lyα, can also be resonantly scattered, its profile widths were not constrained to the pure nebular line widths. The resulting best fit had an observed-frame Gaussian FWHM 6.0 Å or about 410 km s⁻¹. Compared to the nearby profiles of the O iii] nebular lines (285 km s⁻¹), this 45% increase in profile width indicates that C iv is likely dominated by nebular emission but is broadened by resonant scattering. Additionally, strong emission may be masking an absorption component, and therefore we report the C iv emission-line strengths as lower limits.

Electron temperature is typically determined by observing a temperature-sensitive auroral-to-strong-line ratio. Historically, the [O iii] λ4363/[O iii] λ5007 emission-line ratio has been the ideal measure of electron temperature in the high-ionization zone of nebulae. However, the temperature-sensitive [O iii] λ4363 auroral line is not resolved in the HST grism spectrum.

Fortunately, the electron temperature can also be determined from the O iii] λ1666/[O iii] λ5007 ratio, as is commonly done in high-redshift targets where the intrinsically faint optical auroral line is often undetected. In the case of SL2S0217, this diagnostic combines space- and ground-based observations, potentially introducing flux and aperture mismatching issues, and so this calculation must be done carefully. Given the excellent flux calibration of the HST grism spectrum and the match of the flux-corrected LRIS continuum to the HST ACS F606W AB magnitude, we used the O iii] λ1666/[O iii] λ5007 ratio to measure T_e,λ1666 = 15,400 ± 200 K. Adopting this T_e,λ1666 measurement for the high-ionization zone electron temperature, the intermediate- and low-ionization zone temperatures (T_e [S iii] and T_e [O ii], respectively) were then determined from the theoretical temperature relationships of Garnett (1992). The temperatures used for each ionization zone are listed in Table 8.

Table 8.  Ionic and Relative Nebular Abundances

ne used	100 cm−3	4500 cm−3
Te [O iii]opta	15,700 ± 3200 K
Te [O iii]combb used	15,400 ± 200 K
Te [S iii] used	14,500 ± 200 K
Te [O ii] used	13,800 ± 200 K
ne C iii] measured	${300}_{-300}^{+1300}$ cm−3
ne Si iii] measured	${4500}_{-1,400}^{+1,500}$ cm−3
O+/H+a,c	5.21 × 10−7	8.77 × 10−7
O+2/H+a	3.14 ± 0.12 × 10−5	3.13 ± 0.12 × 10−5
ICFd	1.055
12 + log(O/H)a	≥7.50e	≥7.51e
C+3/C+2	0.86	0.86
C+2/O+2	0.122 ± 0.005	0.124 ± 0.005
log U	−1.50
ICF	1.27 ± 0.27
log(C/O)	−0.81 ± 0.09	−0.80 ± 0.09
N+2/O+2	0.033 ± 0.062	0.029 ± 0.054
log(N/O)	−1.48 ± 0.46	−1.53 ± 0.45
Si+2/C+2	0.072 ± 0.002	0.073 ± 0.002
ICF	2.31 ± 0.32
log(Si/C)	−0.76 ± 0.09	−0.77 ± 0.09
log(Si/O)	−1.59 ± 0.17	−1.58 ± 0.17

Notes. Physical conditions and abundances derived for the nebular gas in SL2S0217 assuming either n_e = 100 cm⁻³ (consistent with the n_e from C iii]) or n_e = 4500 cm⁻³ (from Si iii]). Ionic abundances are reported for both the optical and UV species when available.

^aDetermined from the HST optical grism spectrum values. ^bDetermined from the combined UV and optical λ1666/λ5007 line ratio. ^cThe [O ii] λ3727 emission-line intensity was inferred from the observed [O iii] λ5007 flux and the [O iii] λ5007/[O ii] λ3727 ratio from cloudy photoionization modeling. ^dThe oxygen ICF accounts for unseen O⁺³, estimated from the 3σ flux upper limit of O iv] λλ1401, 1404. ^eBecause we do not detect [O ii], the estimated oxygen abundance is reported as a lower limit. However, the O⁺ contribution is expected to be small at such high ionization.

Download table as:  ASCIITypeset image

Interestingly, Brammer et al. (2012) reported excess emission from the Hγ + [O iii] λ4363 blend in the grism spectrum, which can be attributed to [O iii] λ4363 and used to constrain the optical electron temperature diagnostic. We determined the Hγ flux contribution to the blend using the measured Hβ grism flux ((74 ± 2) × 10⁻¹⁷ erg s⁻¹ cm⁻²) and the Case B Hγ/Hβ value from Hummer & Storey (1987) assuming T_e = 15,000 K and n_e = 100 cm⁻³ (conditions representative of SL2S0217; see Section 7.1.2 for n_e). Subtracting this Hγ value from the blend ((40 ± 3) × 10⁻¹⁷ erg s⁻¹ cm⁻²; B12), the flux corresponding to [O iii] λ4363 is F_λ4363 = 4.9 × 10⁻¹⁷ erg s⁻¹ cm⁻². We use the standard [O iii] λ4363/[O iii] λ5007 temperature diagnostic to estimate the high-ionization zone electron temperature to be T_e,λ4363 = 15,700 ± 3200 K, in agreement with the combined UV/optical temperature diagnostic. A higher-S/N and higher-resolution rest-frame optical spectrum would be useful to more securely measure [O iii] λ4363 and compare the two temperature diagnostics.

In nearby objects, the gas-phase electron density is most commonly determined from the [S ii] and [O ii] optical collisionally excited doublets. For SL2S0217, the [O ii] lines were not detected and the [S ii] lines were outside the wavelength range covered by the HST grism spectrum. Instead, we observe two sets of density-sensitive UV emission-line doublets in our LRIS spectrum: Si iii] λλ1883, 1892 and C iii] λλ1907, 1909. We calculate densities of ${4500}_{-1,400}^{+1,500}$ cm⁻³ and ${300}_{-300}^{+1300}$ cm⁻³ for the Si iii] and C iii] ions, respectively.¹⁰ Since the relevant lines are all high S/N and should result purely from nebular emission, we find no obvious explanation for this discrepancy.

The critical densities of C iii] and Si iii] are both of the order of 10⁴ cm⁻³, and so they are generally not useful density diagnostics for n_e values below ∼10³ cm⁻³. This is evident in Figure 9, where we plot the emissivity ratios of common rest-frame UV (Si iii] and C iii]) and optical ([O ii] and [S ii]) electron density diagnostic ratios, assuming a general electron temperature of 15,000 K. When measured, the optical line ratios provide constraints below n_e ∼10³ cm⁻³; however, they probe a different (low-)ionization zone and indicate significantly lower densities relative to the highly ionized gas in high-redshift star-forming galaxies (e.g., Christensen et al. 2012; Bayliss et al. 2014; James et al. 2014). Traditionally, the high-ionization zone (∼35–55 eV) is represented by O⁺⁺, whereas the low-ionization zone (∼15–35 eV) is represented by the O⁺ or N⁺ ions, with the intermediate zone, based on S⁺⁺, partially overlapping (∼23–35 eV). By these definitions, the UV Si iii] and C iii] density diagnostics span the low- to intermediate-ionization and intermediate- to high-ionization zones, respectively. Perhaps not surprisingly, the mixed-zone electron densities measured for high-redshift targets, including SL2S0217, are significantly larger than the pure low-ionization zone densities typical of H ii regions in nearby galaxies. Further, Sanders et al. (2016) find that z ∼ 2 galaxies have mean electron densities that are an order of magnitude higher relative to local galaxies at fixed stellar mass.

Zoom In Zoom Out Reset image size

While the insensitivity of the UV density diagnostics to low densities inhibits the density determination for SL2S0217, the C iii] ratio is at least in agreement with the low-density regime, and so we assumed a standard value of n_e = 100 cm⁻³. Fortunately, the gas temperature and line emissivities are practically independent of the gas density in the low-density regime, and so this choice of density makes little difference. In fact, subsequently derived abundance ratios for both n_e = 100 cm⁻³ (consistent with the n_e from C iii]) and n_e = 4500 cm⁻³ (from Si iii]), given in Table 8, show differences that are significantly smaller than the respective uncertainties. However, detailed studies of these diagnostics must be prioritized to understand whether targets such as SL2S0217 have abnormally high electron densities/density inhomogeneities, the UV diagnostics probe a higher-density regime than their optical counterparts, or some other effect is at play.

Abundance determinations for other elements require ICFs to account for their unobserved ionic species. The prominence of the high-ionization-potential He ii and C iv emission lines observed in SL2S0217 indicate that ICFs are particularly important for C. Although the relatively narrow emission profiles of the C iv doublet indicate the dominance of nebular emission, it is complicated by potential contributions from a range of stellar features, from pure absorption to P Cygni profiles. To avoid this complication, we used the ICF presented in Berg et al. (2016) to estimate the C⁺³ contribution at log U = −1.50 (our best model value; see Section 8.1). Despite the strong C iv emission present in the UV spectrum of SL2S0217, we determine a modest C ICF = 1.27 ± 0.27. The C/O abundance is then determined using

$$\begin{eqnarray}&&\displaystyle \frac{{\rm{C}}}{{\rm{O}}}=\displaystyle \frac{{{\rm{C}}}^{+2}}{{{\rm{O}}}^{+2}\ }\times \ {\rm{ICF}}\end{eqnarray} \tag{ 3 }$$

to be log(C/O) = −0.81 ± 0.03, which is typical of star-forming dwarf galaxies with similar oxygen abundance (Garnett et al. 1995b; Berg et al. 2016).

Carbon and oxygen are produced on different timescales, by different mass populations of stars, and so the C/O trend with metallicity has been studied by many authors with observations for a variety of astrophysical objects, and using a variety of chemical evolution models and stellar yields. Oxygen is synthesized by SNe II (massive stars) and returned to the ISM quickly, whereas carbon is primarily produced by He burning through the triple-α process, a reaction that can occur in both massive (M > 8 M_⊙) and low- to intermediate-mass (1 M_⊙ < M < 8 M_⊙) stars.

Historically, C/O has been difficult to measure, but recently C/O has been measured in dozens of nearby, metal-poor, dwarf galaxies using the UV lines (e.g., Berg et al. 2016; Senchyna et al. 2017). Similar to the early work of Garnett et al. (1995b), these studies find a general trend of increasing C/O with oxygen abundance, but with a significant unexplained scatter. These data can also be interpreted as a constant trend in C/O at low metallicities. We plot SL2S0217 (open star) on this C/O versus O/H trend in the left panel of Figure 10, where nearby dwarf galaxies are plotted as filled circles for comparison. The trend is extended to higher oxygen abundances by incorporating C/O determinations from optical recombination lines (filled squares: Esteban et al. 2002, 2009, 2014; Pilyugin & Thuan 2005; García-Rojas & Esteban 2007; López-Sánchez et al. 2007).

Zoom In Zoom Out Reset image size

Significant UV C and O have also been measured for a handful of z ∼ 2−3 galaxies (gold triangles in Figure 10: Pettini et al. 2000; Fosbury et al. 2003; Erb et al. 2010; Christensen et al. 2012; Bayliss et al. 2014; James et al. 2014; Stark et al. 2014; Steidel et al. 2016; Vanzella et al. 2016; Amorín et al. 2017; Rigby et al. 2018), among which the C/O ratio of SL2S0217 appears to be average. Interestingly, C/O values for the z ∼ 2−3 galaxies are consistent with the nearby dwarf galaxies of similar metallicity, but potentially with a lower average value. The distinction between these populations of different ages may support the notion put forth by Berg et al. (2016) that higher C/O values may be due to a delayed pseudo-secondary carbon contribution from low- and intermediate-mass stars.

We estimate the N/O ratio by combining the UV O iii] λλ1661, 1666 and C iii] λλ1907, 1909 emission lines and the N iii] λ1750 multiplet. Since the N iii] λ1750 feature is only weakly detected in the SL2S0217 LRIS spectrum, it is appropriate to think of the following N/O determination as an upper limit. Utilizing the fact that the ionization potentials of N⁺⁺ and C⁺⁺ are nearly the same (47.448 and 47.887 eV, respectively), we used the N⁺⁺/C⁺⁺ ion ratio to measure a relative N/C enrichment of log(N/C) = −0.68 ± 0.02. Then, combining this ratio with the ionization-corrected C/O ratio (see Table 8), we inferred a total N/O of log(N/O) = −1.48 ± 0.46, which is consistent with the range of log(N/O) seen for local dwarf galaxies with similar oxygen abundance (e.g., van Zee & Haynes 2006; Berg et al. 2012).

Relative N/O abundances are well studied in nearby dwarf and spiral galaxies, providing a robust sample to compare measurements from more distant targets. In the top right panel of Figure 10 we compare the N/O of SL2S0217 to nearby dwarf (red points: van Zee & Haynes 2006; Berg et al. 2012, 2016; D. A. Berg et al. 2018b, in preparation) and spiral galaxies (blue points: the CHAOS project; Berg et al. 2015; Croxall et al. 2015, 2016; D. A. Berg et al. 2018a, in preparation), as well as nearby galaxies with oxygen abundance measurements from recombination lines (green squares). SL2S0217 lies among the N/O range of both nearby dwarf galaxies and the handful of z ∼ 2−3 galaxies with N/O and O/H measurements (gold triangles). Note that the N/O value for SL2S0217 falls on the plateau expected for primary (metallicity-independent) N production, as expected for a low oxygen abundance of 12+log(O/H) ∼ 7.5 (see, e.g., the discussion in Pettini et al. 2002a; Pettini & Cooke 2012).

In the bottom right panel of Figure 10 we examine the C/N ratio, noting the remarkably flat trend for both nearby and distant galaxies. Berg et al. (2016) found that the constant ratio of C/N seen over a large range in O/H for nearby galaxies may indicate that carbon is predominantly produced by similar nucleosynthetic mechanisms to nitrogen (see also Steidel et al. 2016). However, a larger sample of C/N measurements in distant galaxies is needed to determine whether the hypothesis of Berg et al. (2016) also applies at z ∼ 2−3, or whether different nucleosynthetic processes dominate C and N.

Recently, observations of the standard rest-frame optical emission-line [O iii]/Hβ versus [N ii]/Hα diagnostic plot (the BPT diagram; Baldwin et al. 1981) have revealed a pronounced offset of ∼2–3 galaxies from the main sequence of local star-forming galaxies (e.g., Masters et al. 2014; Steidel et al. 2014; Shapley et al. 2015). Several studies have investigated the physical origin of this offset by analyzing the BPT parameter space in terms of other physical properties for large surveys of local galaxies. Using ∼100,000 galaxies from the Sloan Digital Sky Survey Data Release 12, Masters et al. (2016) found that the region of the BPT occupied by ∼2–3 galaxies corresponds to galaxies with elevated SFR surface densities, stellar masses, and N/O ratios, concluding that the higher N/O ratios at fixed [O iii]/Hβ are the proximate cause of the offset. In contrast, the sample of z ∼ 2−3 galaxies plotted in Figure 10 occupies the same range in N/O as the local comparison sample. SL2S0217, in particular, has an average N/O abundance relative to the local dwarf galaxies; however, this value is an upper limit estimated from the weak detection of the N iii] λ1750 line. Further, despite the unusual strength of its UV and optical emission lines, the [O iii]/Hβ and [N ii]/Hα (inferred from photoionization modeling; Section 8.1.1) ratios of SL2S0217 indicate that it may not coincide with the offset region of the BPT. Clearly there is still much to understand about the z ∼ 2−3 BPT diagram.

Silicon abundances relative to carbon were determined following the logic presented in Garnett et al. (1995a): the Si iii] and C iii] doublets arise from comparable energy levels (∼6.57 and 6.50 eV, respectively), and therefore the Si/C abundance ratio is relatively insensitive to uncertainties in the electron temperature. Additionally, Si/C depends little on the assumed density, as both Si iii] and C iii] have very high critical densities, meaning that their volume emissivities have a similar dependence on density. Finally, the proximity of the Si iii] and C iii] emission lines in wavelength eliminates strong effects of differential extinction.

The Si iii] and C iii] emission-line features are extremely prominent in SL2S0217; we measured their strengths with S/N > 20 (see Figure 5). As described in Section 7.1.2, we assume that the ionized gas of SL2S0217 is in the low-density limit (<10³ cm⁻³), where collisional de-excitation does not significantly impact the Si⁺²/C⁺² ratio derived from the UV lines. Because C⁺² has a larger ionization potential than Si⁺² (47.9 and 33.5 eV, respectively), we expect a larger fraction of Si to be in a higher, unobserved ionization state than C, and so an ICF is needed to determine the Si/C abundance. We determine the Si/C element abundance using the equation

$$\begin{eqnarray}&&\displaystyle \frac{{\rm{Si}}}{{\rm{C}}}=\displaystyle \frac{{{\rm{Si}}}^{+2}}{{{\rm{C}}}^{+2}}\ {\left[\displaystyle \frac{X({{\rm{Si}}}^{+2})}{X({{\rm{C}}}^{+2})}\right]}^{-1}=\displaystyle \frac{{{\rm{Si}}}^{+2}}{{{\rm{C}}}^{+2}}\ \times {\rm{ICF}},\end{eqnarray} \tag{ 4 }$$

where X(Si⁺²) and X(C⁺²) are the Si⁺² and C⁺² volume fractions, respectively. The ICF is modeled as a function of the ionization parameter using the photoionization code cloudy (Ferland et al. 2013), with BPASS models as the input ionizing radiation field (see Section 8.1). We sought simple models that reproduced the conditions in SL2S0217 as closely as possible; see Section 8.1 for a detailed discussion.

The ionization fractions of Si and C species as a function of ionization parameter are shown in Figure 11. As we will show in Section 8.1.1, we can reproduce many aspects of the observed spectrum using cloudy models of a stellar population with binaries and an ionization parameter of log U = −1.5; this value is reasonable given the very hard ionizing radiation field predicted by the SED modeling of B12. We estimate the uncertainty in the ICF as the scatter among the different models considered at a given C⁺² volume fraction. The resulting Si ICF = 2.31 ± 0.32 for 12+log(O/H) = 7.5 and t = 10^7.0 yr is significant, correcting the relative Si abundance from log(Si/C) = −1.14 to log(Si/C) = −0.76 ± 0.09. When combined with the C/O abundance derived for SL2S0217, we estimate log(Si/O) = −1.59 ± 0.17. The relevant Si abundance values are reported in Table 8.

Zoom In Zoom Out Reset image size

Due to the need of far-UV spectroscopic observations of nearby galaxies, there exists a general paucity of nebular measurements of Si/O to compare to. Perhaps the best-known study of Si/O comes from Garnett et al. (1995a), who measured the UV Si iii] emission lines for seven extragalactic H ii regions using the Faint Object Spectrograph on HST. These authors found Si/O to be constant over the observed range in oxygen abundance, with a weighted mean value of log(Si/O) = −1.59 ± 0.07. Our log(Si/O) = −1.59 ± 0.17 measurement for SL2S0217 is remarkably consistent with this average from the Garnett et al. (1995a) study but indicates significant Si/O enrichment or a reduced Si depletion onto dust relative to the z ∼ 2.4 galaxies of Steidel et al. (2016; log(Si/O) = −1.81 ± 0.10).

Note that our Si/O value does not account for the relative depletion of Si and O onto dust grains. While one might expect this effect to be small given the low predicted dust reddening in SL2S0217 and the probability of grain destruction or erosion in the presence of a hard ionizing radiation field, our Si/O value is significantly lower than the solar ratio of log(Si/O)_⊙ = −1.15 (Jenkins 2009). Many studies argue that both Si and O are predominantly products of massive star nucleosynthesis, and so their ratio is expected to vary little with time (e.g., Timmes et al. 1995). Note, however, that some studies find that Si production can have a substantial contribution from SN Ia outbursts (see Matteucci & Greggio 1986; Kobayashi et al. 2006). Since oxygen does not deplete as strongly, the observed underabundance can likely be attributed to the depletion of Si onto dust (Steidel et al. 2016).

We estimated the logarithmic depletion of Si in the ISM following the prescription laid out in Jenkins (2009):¹¹

$$\begin{eqnarray}&&{[{\rm{Si}}/{\rm{H}}]}_{{\rm{gas}}}={[{\rm{Si}}/{\rm{H}}]}_{{\rm{obs}}}-{[{\rm{Si}}/{\rm{H}}]}_{{\rm{ref}}},\end{eqnarray} \tag{ 5 }$$

where, assuming that the observed subsolar Si/O ratio is solely due to Si depletion, we use Si/O in place of Si/H and use the solar values from Asplund et al. (2009) for our reference. This results in a dust depletion of [Si/H]_gas = −0.43 for SL2S0217, a value typical with respect to Garnett et al. (1995a), but smaller than the value derived by Steidel et al. (2016). From observations of nearby dwarf galaxies we have come to expect [Si/Fe] enrichment at low [Fe/H] due to early contamination of Si by SNe II. However, this exercise indicates that a significant fraction of the Si produced in SL2S0217 is missing, perhaps resulting in the subsolar Si/Fe ratio measured from the absorption lines (Section 6).

Jenkins (2009) interpreted the abundances of 17 different elements from more than 100 studies of sight lines to 243 different stars to construct a unified representation of depletions that is linearly dependent on the severity of depletion along a given sight line, F_*. Using this model and the inferred depletion value of Si, we estimated the depletion of the other significant elements in the SL2S0217 UV spectrum—C, N, and O. Using the fit from Jenkins (2009) for Si, [X/H]_gas = B_X + A_X(F_* − z_X) (coefficients reported in their Table 4), we found a depletion severity of F_* = 0.182. Then, we determine the following dust depletion factors: δ(Si) = 0.628, δ(C) = 0.259, δ(N) = 0.222, and δ(O) = 0.111¹² ([C/H]_gas = −0.130, [N/H]_gas = −0.109, and [O/H]_gas = −0.051). While these depletion values imply a systemically higher relative Si/O abundance (Δ ∼ 0.4 dex), the O/H, C/O, and N/O ratios would remain relatively unchanged (Δ ∼ 0.05 dex).

To determine the best model from our grid, we first eliminated sets of models that were obviously poor matches. The single-star model sets do not reproduce the UV nebular emission-line feature strengths (see also Jaskot & Ravindranath 2016) and so were not investigated further. For the young ages considered here (t < 10^7.2 yr), the continuous star formation models are consistent with the burst models, and so only the burst models were examined as representative of both types of star formation. Note, however, that continuous star formation models do produce very different results from bursts for older systems. Finally, the models with an IMF cutoff of 300 M_⊙ do produce more He ii ionizing photons than the 100 M_⊙ models, yet they still fall short of producing the He ii flux observed for SL2S0217 and significantly overpredict the Si iii]/C iii] ratio. Therefore, we employed an error-weighted χ²-minimization routine of the grid of binary burst models with an IMF cutoff of 100 M_⊙. We find we can reproduce most of the features in the UV emission-line spectrum of SL2S0217 with a model similar to the SED modeling of Brammer et al. (2012): the best model is a high-ionization (log U = −1.5), metal-poor (12+log(O/H) = 7.75; CNSi = 0.3) gas cloud, ionized by Z = 0.001 star formation with binaries at an age of t = 10^7.0 yr.

In Figure 12 we compare the normalized LRIS spectrum of SL2S0217 to the output model spectrum at the spectral resolution of the observed spectrum. The C iv, O iii], N iii], Si iii], and C iii] emission lines are all relatively well matched, while the He ii emission feature is clearly discrepant. Notably, Fosbury et al. (2003) also observed strong, narrow nebular He ii and C iv emission from the z ∼ 3.4 Lynx Arc, reportedly due to a young, low-metallicity stellar cluster with an extreme ionization parameter of log U = −1.0. However, to date, no pure stellar photoionization studies have reported the levels of nebular He ii emission observed in SL2S0217 or the Lynx Arc. Interestingly, we do not see the N iv] emission line in our SL2S0217 spectrum despite the fact that we see strong C iv emission and C⁺⁺ and N⁺⁺ have similar ionization potentials (47.888 and 47.445 eV, respectively). The N iv] emission line is also absent from the best model, indicating that N emission is mostly sensitive to metallicity and, therefore, not expected in metal-poor galaxies like SL2S0217, whereas C iv is more sensitive to temperature/cooling. This fact has motivated the use of the strong-line [N ii]/Hα metallicity indicator in nearby galaxies; however, the line ratio is not available from the ground at high redshifts (z ≳ 2.5) and is highly sensitive to ionization parameter at larger metallicities (e.g., Kewley & Dopita 2002). Since exceptionally low nitrogen abundances seem to be common among distant galaxies with similar extreme emission-line features (e.g., Bayliss et al. 2014; James et al. 2014; Smit et al. 2017), the lack of UV N iii] and N iv] could help identify additional metal-poor galaxies at high redshifts.

Zoom In Zoom Out Reset image size

The predicted nebular emission-line ratios from our photoionization models are plotted relative to the ratios measured from the LRIS spectrum of SL2S0217 in Figure 13. Different gas-phase metallicity models are indicated by color, where relative C/O, N/O, and Si/O abundances increase with point size. The statistically best model is indicated by the dashed yellow line (12+log(O/H) = 7.75, CNSi = 0.3, t = 10^7.0 yr), where the corresponding yellow-shaded region extends from t = 10^6.4 to 10^7.0 yr. Note that nebular He ii emission peaks before t = 10^7.0 yr (∼10^6.8 yr) in the models, and so the dashed yellow line in the He ii/C iii] plot lies within the yellow shaded region. Interestingly, although C iv also has a very high ionization potential, its nebular line strength continues to increase past t = 10^7.0. Measured line ratios and uncertainties for SL2S0217 are shown as solid black horizontal lines and gray shaded boxes. Good agreement between model and measurement is found near an ionization parameter of log U = −1.50 for all of the UV emission-line ratios except He ii/C iii].

Zoom In Zoom Out Reset image size

Just as Stark et al. (2014), Berg et al. (2016), and Senchyna et al. (2017) observed strong C iii] emission resulting from subsolar C/O in metal-poor systems (Z ≤ 0.5 × Z_⊙), reproducing the observed emission lines in SL2S0217 also requires subsolar C, N, and Si abundance ratios relative to O. The low gas metallicity prevents efficient metal cooling by oxygen, resulting in a high electron temperature that secures a stable production of the collisionally excited O, Si, and C nebular emission lines. If a portion of the metals are also locked up in dust, C and Si may be underabundant as a result. However, no evidence of dust reddening exists, and the extreme ionizing radiation field likely destroyed or eroded the dust in the H ii regions (but also see discussion of dust depletion in Section 7.2.3).

Even in star-forming galaxies, a substantial portion of the nebular emission may be generated by shocks (e.g., Krabbe et al. 2014; Rich et al. 2014). This fact motivated Jaskot & Ravindranath (2016) to incorporate contributions from shocks into their pure photoionization models. In their study, shock+precursor models were taken from Mappings III (Allen et al. 2008), assuming a metallicity of Z ∼ 0.003 (the Small Magellanic Cloud models), a velocity range of 125–1000 km s⁻¹, and magnetic field strengths of 0.5–10 μG. The authors find that the addition of the shock models results in an increase of both C iii] and He ii emission relative to Hβ; therefore, they suggest that nebular He ii may serve as a useful diagnostic of shocks for UV spectra.

In light of these findings, we investigated emission from shock ionization as a source for the significant He ii emission seen in SL2S0217. Following the methodology of Jaskot & Ravindranath (2016), we scaled the shock emission and added it to the emission from our photoionization models such that the shocks contribute 10%, 30%, 50%, 70%, or 90% of the total predicted emission in Hα (see their Section 2.3 for a more detailed description).

The emission-line ratios for a varying shock contribution to our best photoionization model are depicted in Figure 14(a) relative to those measured for SL2S0217. These results demonstrate that the addition of shocks to the BPASS photoionization does not improve the general match of models to the SL2S0217 features. While the extreme He ii/C iii] ratio can be reproduced when the ionizing radiation is dominated by shocks, the other observed line ratios are generally overpredicted and cannot be matched with the same conditions. Further, it is physically unlikely that the gas in SL2S0217 is 90% shock ionized, and so we dismiss shocks as a main source of ionization in SL2S0217.

Zoom In Zoom Out Reset image size

The complex suite of spectral features observed in extreme galaxies such as SL2S0217 may be due to a complex combination of ionizing sources, such as AGN emission diluted by stellar radiation. For supermassive black holes (SMBHs) in massive galaxies, numerous studies have found that the SMBH mass is well correlated with the velocity dispersion of the host-galaxy bulge (e.g., Gebhardt et al. 2000; Gültekin et al. 2009). However, it is unclear whether this well-known scaling relation extends to the galaxies as small as SL2S0217 (see discussion in Kormendy & Ho 2013). Black hole masses in dwarf AGNs have been measured in several low-mass star-forming galaxies in the local universe (e.g., Reines et al. 2011, 2014), with the lowest reported mass being just ∼50,000 M_⊙ (Baldassare et al. 2015). Recently, Baldassare et al. (2017) found evidence for dwarf AGNs in 11 nearby (z < 0.055) composite dwarf galaxies. These authors measured black hole masses ranging from 10^4.9 to 10^6.1 M_BH/M_⊙ for galaxies with stellar masses of 10^8.1–10^9.4 M_⋆/M_⊙, on the order of SL2S0217. While these recent studies of local dwarf galaxies indicate that a non-negligible fraction of low-mass galaxies might harbor AGNs, we caution that we do not know how these results extrapolate to galaxies at high redshifts.

Given the uncertainties surrounding AGNs in low-mass galaxies, it is worthwhile to explore the possibility of producing the extreme emission-line ratios observed in SL2S0217 with the addition of emission from AGNs. For this purpose, we used the dust-free, low-metallicity (Z_⋆ = 0.25 Z_⊙), narrow-line AGN photoionization models from Groves et al. (2004), assuming simple power-law radiation fields (f_ν ∝ ν^β) with a range of powers. As in Section 8.3.1, we scale the AGN emission such that the AGNs contribute 10%, 30%, 50%, 70%, or 90% of the total emission in Hα.

Emission-line ratios for the best BPASS + AGN models are plotted in Figure 14(b) relative to those measured for SL2S0217. The resulting shifts in line ratios due to the contribution of emission from AGNs look very similar to those seen for the BPASS + Shock models. In either case, the additional hard radiation increases the relative emission from the highest-ionization lines: He ii, O iii], and C iv. This allows the C iv/C iii] ratio to be reproduced with small AGN contributions. Typical narrow-line Type II AGN observations have stronger C iv than C iii] emission (e.g., Humphreys et al. 2008; Matsuoka et al. 2009; Hainline et al. 2011; Cassata et al. 2013), but our models suggest that the C iii] dominant emission in SL2S0217 (C iii]/C iv = 1.46) could be produced by a gas-enshrouded AGN.

On the contrary, Figure 14 shows that the Si iii]/C iii] ratio is significantly less affected by contributions from AGNs than from shocks. This is due, in part, to the fact that the full range of shock velocities is considered at each value of log U, such that the maximum contribution from shocks for any ion is always included (near v ∼ 175 km s⁻¹ for Si iii]). On the other hand, the AGN contribution to Si iii] emission peaks at low-ionization parameter; however, it has little effect on the total Si iii]/Hα emission, which is dominated by photoionization.

An additional difference is seen in the He ii/C iii] ratio, where the value observed in SL2S0217 is only reached by AGN-dominated models (>90% contribution; purple points), yet such models significantly overestimate the C iv/C iii] and O iii]/C iii] ratios. Further, if the strong nebular emission lines in SL2S0217 were powered by a Type II AGN, we would also expect to detect several high-ionization emission lines that are missing from the rest-frame UV spectrum: N v λ1240, Si iv λ1403, N iv λ1486 (e.g., Vanden Berk et al. 2001; Hainline et al. 2011). Again, the suite of lines observed for SL2S0217 cannot be simultaneously matched by BPASS photoionization models with an additional source of hard radiation, in this case, from AGNs.

Typically galaxies show C iv in (i) absorption from the surrounding ISM or circumgalactic medium (CGM) gas or (ii) a P Cygni profile from the stellar winds of massive stars (e.g., Leitherer et al. 2001). To the contrary, AGNs can produce C iv emission, but line widths are typically broader than nebular emission lines (hundreds of kilometers per second). To date, narrow C iv has only been observed in a handful of strongly lensed high-redshift galaxies (e.g., Christensen et al. 2012; Stark et al. 2014; Smit et al. 2017); therefore, the significant C iv emission detected in SL2S0217 is another interesting piece of this peculiar puzzle.

Since the C iv λλ1548, 1550 doublet is expected to be produced as a combination of nebular, stellar, and ISM/CGM components, we can use a simple combination of absorption and emission profiles to offer a possible model for the C iv profile observed in SL2S0217. Despite the absence of apparent C iv absorption, if we assume that the absorption in both C iv and Si iv is due to high-ionization clouds in the surrounding ISM of the galaxy, we can use the average absorption profile (which is nearly identical to the Si iv λ1403 profile; see Figure 7) to model the C iv absorption. Although the oscillator strength of Si iv λ1394 is roughly two times larger than that of Si iv λ1403, their relative absorption profile depths are closer to a factor of 1.5, likely due to saturation. Similarly, C iv λ1548 absorption is expected to be twice as strong as that of C iv λ1550. Following the observed model of Si iv, we assume relative absorption profile strengths of 3:2 for both Si iv λλ1394, 1403 and C iv λλ1548, 1550. The total profile is then produced by co-adding the absorption profile with the emission predicted by the best photoionization model, n × F_λ, where n is an arbitrary scaling of the emission to best match the observed spectrum of SL2S0217. The resulting model profiles and their components are plotted against the SL2S0217 spectrum in Figure 15. These model fits offer a solution in which the Si iv and C iv observed profiles could be produced, but we caution that this interpretation cannot be confirmed.

Zoom In Zoom Out Reset image size

Arguably the most aberrant feature of the SL2S0217 UV spectrum is the impressive nebular He ii emission strength. Nebular He ii emission is produced by recombination from the He⁺² ionization state and so requires very hard ionizing radiation with photons at energies ≥54 eV in order to ionize He⁺⁺. The interpretation of He ii emission is further complicated by the fact that it can have both nebular and stellar origins. Stellar He ii is produced by the stellar winds of massive stars, such as Wolf-Rayet (W-R) stars, and so is broadened in relation to the wind velocity (e.g., Brinchmann et al. 2008; Erb et al. 2010; Cassata et al. 2013).

Leitherer et al. (1999) showed that the number of W-R stars relative to O stars peaks around 10^6.6 yr, and so stellar winds may no longer be prominently contributing to the He ii emission at a starburst age of 10⁷ yr, in line with the fact that we do not observe a significant broad component to the He ii emission in SL2S0217. Further, Eldridge & Stanway (2009) demonstrated that the W-R He ii λ1640 feature is only prominent in binary models with stellar metallicities of Z_⋆ = 0.004 or 0.008, which do not pair well with the gas-phase oxygen abundance of SL2S0217 (Z_neb ∼ 0.001). Instead, the He ii emission in SL2S0217 appears to be nebular in origin. In the photoionization models of Jaskot & Ravindranath (2016), the narrow, nebular He ii emission seen in SL2S0217 only becomes strong for more metal-rich systems during the W-R phase or when ionized by shocks. Other sources of high-energy photons may be required to produce strong nebular He ii, such as AGNs, exotic Pop III stars, or high-mass X-ray binaries (e.g., Shirazi & Brinchmann 2012; Kehrig et al. 2015).

As discussed in Section 8.3.1, we dismiss a significant contribution to the ionizing budget from shocks, as the models significantly overpredict the O iii]/C iii] budget. In addition to overshooting O iii]/C iii], the AGN models (Section 8.3.2) also produce an excess of C iv relative to C iii] emission, but this argument is not definitive, as some of the emission may be obscured by overlapping C iv absorption. However, AGNs typically have large He ii EWs (≳8 Å; Hainline et al. 2011; Cassata et al. 2013), which are generally greater than those measured for z ∼ 2 galaxies (Scarlata et al. 2009; Erb et al. 2010) and for SL2S0217 (EW(He ii_n) = 2.4 Å or EW(He ii_tot) = 2.8 Å, where the difference is due to the difficulty in deblending the narrow and wide components).

Photoionization by AGNs has also been investigated using diagnostic plots of the C iv/He ii and C iii]/He ii line ratios, both of which are sensitive to metallicity and ionization parameter (Villar-Martin et al. 1997). In Figure 16 we plot our stellar photoionization models described in Section 8.1.1 in comparison to photoionization from pure shocks and AGNs for the C iv/He ii versus C iii]/He ii ratios. The observations for SL2S0217 (black star) are extended to larger values (shaded region) in case C iv is underestimated (due to overlapping C iv absorption) and/or He ii is overestimated (due to a stellar contribution), clearly suggesting a stellar origin of the photoionization budget. Note, however, that the stellar photoionization models still fail to reproduce the absolute strength/EW of the He ii emission. Motivated by the uncertainties of these line ratios, we will present a preferable analysis in a forthcoming paper that allows us to distinguish between sources of ionizing radiation using line ratios that are not complicated by stellar and CGM components.

Zoom In Zoom Out Reset image size

In observations of star-forming galaxies, He ii emission is commonly attributed to W-R stars, which typically show strong and broad (Δv > 1000 km s^–1; Crowther 2007) emission lines produced by their fast stellar winds. For example, Chandar et al. (2004) measured an extreme He ii λ1640 EW in the local universe from W-R stars in the massive cluster NGC 3125-1, with a broad width of ∼1000 km s⁻¹, and further supported by strong detections of the N iv λ1488 and N v λ1720 WN W-R features. We therefore rule out W-R stars as the source of the He ii emission on the account that we do not observe significant broad He ii emission in SL2S0217, or any other indicative feature, and that the W-R phase occurs in a very short and early window that is unlikely the period of our observations. In binary star models, where the effects of mixing and mass transfer extend the life of these massive stars, He ii emission peaks ∼50 Myr after a burst of Z_⋆ = 0.004 star formation. However, the He ii emission from W-R stars depends on the metallicity, and the lower-metallicity stellar spectra representative of SL2S0217 are not capable of reproducing the observed He ii (see Figure 13).

Theoretically, gravitational cooling radiation from gas accretion is another mechanism that can produce He ii (as well as Lyα) emission profiles like those in SL2S0217. In these models, infalling gas onto dark matter can be heated up to temperatures of T ∼ 10⁵ K and then is cooled by H and He line emission (Kereš et al. 2005; Yang et al. 2006). It has been suggested that Lyα halos (Lyα "blobs" that show extended Lyα emission on scales of tens of kiloparsecs or more; e.g., Caminha et al. 2016; Patrício et al. 2016) are the observational signature of this cooling radiation in forming galaxies (Haiman et al. 2000; Fardal et al. 2001). While inflowing gas would also help explain the strong blue peak in the Lyα profile, it is unlikely that SL2S0217 is a Lyα halo. Our observations and lensing model show a small galaxy with Lyα emission coming from three dense star-forming regions, where the Lyα emission extends beyond the continuum on much smaller spatial scales (as detected in narrowband HST imaging; D. K. Erb et al. 2018, in preparation).

The current arsenal of stellar population models in the literature are only able to reproduce strong He ii emission lines with pure photoionization for extreme stellar populations. Due to its high ionization potential, He ii emission has been used as a probe to find Pop III stars (e.g., Visbal et al. 2015). A top-heavy IMF in extremely low metallicity regions can produce large He ii line luminosities (Tumlinson & Shull 2000; Schaerer 2003; Yoon et al. 2012). However, the He ii EW is very sensitive to metallicity and so requires Z < 10⁻⁵ for this model. Steidel et al. (2016) predicted that H ii regions with normal oxygen abundances can exhibit extreme UV features if they are powered by stars that are O/Fe enhanced relative to solar. Such a scenario would be expected if the ISM was enriched by core-collapse SNe, and then the best-matching stellar model would actually have a metallicity several times lower than the gas-phase oxygen abundance. Since stellar spectra are primarily dependent on the total opacity, which is largely modulated by the Fe abundance (e.g., Rix et al. 2004), much harder ionizing radiation would be produced than is expected from the nebular metallicity. Alternatively, Tornatore et al. (2007) predicted that zero-metallicity regions can survive in low-density regions around large overdensities, allowing Pop III stars to form, until as recently as z ∼ 2.

Other studies do not require large modifications to relative abundances or metal-free stars to produce narrow He ii emission. Kudritzki (2002) showed that certain O stars are indeed hot enough to ionize He ii, and Brinchmann et al. (2008) suggested that at low metallicities nebular He ii is predominantly produced by O stars where optically thin winds can be penetrated by ionizing photons. The energy production of low-metallicity massive stars may also be increased if they are fast rotators (Meynet & Maeder 2007). Further support is provided by the study of Kehrig et al. (2015), who found that the He ii ionization budget of I Zw 18 can only be produced by models of either low-metallicity supermassive O stars or rotating metal-free stars. However, the details of very low metallicity O stars are still widely debated and further complicated by the fact that some He ii nebulae do not appear to have surviving O or W-R stars (Kehrig et al. 2011). Gräfener & Vink (2015) posed an alternative scenario for the origin of narrow He ii emission in which metal-poor, very massive stars form strong but slow W-R-type stellar winds owing to their proximity to the Eddington limit. Post-AGB stars from an older stellar population may also produce He ii emission. The combined radiation field of post-AGB stars will dominate the ionizing spectrum about 100 Myr after a burst of star formation and would be strong enough to ionize He ii (Binette et al. 1994). However, the He ii luminosities produced by post-AGB stars are insufficient to explain the emission-line strength seen for SL2S0217 (Cassata et al. 2013).

It seems likely that very massive stars play a significant role in the He ii line strength. Unfortunately, the nature of metal-poor massive stars remains poorly understood, and therefore very massive stars are often withheld from current population synthesis models. Theoretical work on rotation and binary evolution of massive stars (e.g., Eldridge & Stanway 2009; de Mink et al. 2014) has produced a vast range in the predicted lifetimes and energy output of low-metallicity massive stars. Further, these models make simplifying assumptions about the rotation velocities and distribution of binary separations that are poorly motivated. In the future, more sophisticated models of metal-poor massive stars may be able to explain the strength of high-ionization nebular lines observed for SL2S0217. However, since narrow He ii emission is more common at z ∼ 3 (Cassata et al. 2013) than at z ∼ 0 (Kehrig et al. 2011), it is possible that the stellar populations that produce He ii ionizing radiation are more common at higher redshifts and are fundamentally different from typical populations in the local universe.