VLT/X-Shooter Spectroscopy of Lyman Break Analogs: Direct-method O/H Abundances and Nitrogen Enhancements

The targets studied as part of this paper were selected from a number of sources related to the LBA project and by no means represent a complete sample. A total of 17 targets (marked "LBA" in Table 1) were selected from the original sample of Heckman et al. (2005) and studied in detail by Overzier et al. (2009). Ten new LBAs (marked "LBA2" in Table 1) were taken from an extended sample of several hundred LBAs in the cross-match between SDSS data release 7 (DR7; Abazajian et al. 2009) and GALEX GR6. Of these 10 targets, nine (marked "BPT" in Table 1) were further selected based on having large BPT offsets and observability, while one (marked "COS" in Table 1) was selected because it has been observed with the HST/COS as part of a program to target LBAs having high FUV fluxes within the COS aperture. Finally, three more targets (marked "S2-Deficit" in Table 1) were selected on the basis of their relative weakness of the [S ii] λλ6717, 6731 optical emission lines. This was motivated by previous work on LBAs that showed that objects with abnormally low [S ii] λλ6717, 6731/Hα flux ratios compared to typical star-forming galaxies may have matter-bounded conditions in the ISM that can result in the escape of ionizing photons (e.g., Pellegrini et al. 2012). This has inspired a new category of local analogs that have potentially large escape fractions (Alexandroff et al. 2015; Wang et al. 2019). All three "S2-Deficit" objects satisfy the UV surface brightness criterion of LBAs, but one is of substantially lower FUV luminosity than LBAs, as defined by Heckman et al. (2005). This object is a known low-metallicity blue compact dwarf (e.g., it was part of Subsample 1 from Izotov et al. 2007).

Originally part of the LBA sample, two objects (001054 and 005439) were found to be Type 1 AGNs based on broad Mg ii lines detected in the UVB arm of Very Large Telescope (VLT)/X-Shooter. These two objects are considered to be contaminants of the LBA sample, as the emission from the unobscured nucleus likely dominates the FUV flux detected by GALEX. These two sources were removed from the sample. Object 210358 had problems with the observations, as the on-slit dither positions were chosen to be too close to each other, leading to problems with the background subtraction. This object is not studied further as part of this paper.

Several objects from the Overzier et al. (2009) sample have been observed with deep spectroscopic data before. Amorín et al. (2012) observed three of the sources with the GTC-OSIRIS spectrograph (004054, 113303, 232539), while Brown et al. (2014) observed four sources with the LBT-MODS1 spectrograph (004054, 005527, 020356, 092600), one of which (004054) is in common with Amorín et al. (2012). These six unique sources include two LBAs not covered by our spectroscopy (092600 and 113303) and four that overlap with our data. Although these authors have already derived many of the parameters similar to the ones that we will derive here (i.e., stellar populations, densities, temperatures, and abundances), we will use the stellar absorption and dust-corrected line fluxes as reported by these authors in order to compare their results with our own data. Figure 1 shows the rest-frame UV properties of the VLT/X-Shooter sample compared to the local galaxy population. The location of the objects in the BPT diagram as measured from the VLT/X-Shooter spectra or given by the literature data is shown in Figure 2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Spectra were taken using the X-Shooter spectrograph installed at UT2 (Kueyen) of the ESO VLT in Chile during two separate observing campaigns. The first sample of 17 objects was observed in service mode during 2010 (Program ID 085.B-0784(A)), and a second sample of 13 objects was observed in visitor mode during 2016 March (Program ID 096.B-0192(A)). The target list and log of observations are given in Table 1. We used X-Shooter in slit mode (11'' slit) to obtain simultaneous spectra from U to K band using the UVB (10 wide slit, R ≈ 5100), VIS (09 wide slit, R ≈ 8800), and near-IR (NIR; 09, R ≈ 5100) arms. The exposure times for each object were 4 × 690 s, 8 × 320 s, and 12 × 240 s in nodding-on-slit mode, except for very extended objects that were observed in offset mode and received half the exposure time. The slits were centered on the brightest starburst regions, but we note that the effective radii of most objects are comparable to the slit width. The seeing was estimated from the telluric standard-star spectra taken each night at the same air mass as the science observations.

We used the ESO X-Shooter pipeline (Modigliani et al. 2010) and the EsoRex command-line recipes to reduce the data. The pipeline removes nonlinear pixels and subtracts the bias in the UVB and VIS arms and the dark frames in the NIR arm. It then predicts the positions of ar lines and order edges on a format-check frame, determines the order positions and the two-dimensional wavelength solution to resample the orders, divides the raw frames by a master flat field, measures the instrument resolving power, and computes the instrument flexures, the instrument response, and the total system efficiency. The spectra are then flux-calibrated using standard stars that were observed during each night. The pipeline combines the science frames and subtracts the sky background. One-dimensional spectra were extracted using the X-Shooter pipeline at the location of the continuum along the slit. The size of the apertures was set to the larger value of the seeing (FWHM) measured from the telluric standard in the UVB arm (averaged over all wavelengths) and the spatial extent of Hα in the VIS arm. The spectra extracted from each of the arms were joined, and small flux offsets were applied to the UVB and NIR spectra in order to correct for small flux normalization mismatches with overlapping wavelength ranges at either end of the VIS spectra. Although telluric standards were observed and reduced in the same way, ultimately the telluric absorption corrections were done using the ESO Molecfit tool (Kausch et al. 2015; Smette et al. 2015). Molecfit allows one to fit a model of the tropo- and stratospheric telluric features directly to the science spectra and subtract this model from the data in order to remove the telluric features. Molecfit also updates the input error spectra that are produced by the ESO pipeline.

The reduced, calibrated spectra are first corrected for foreground extinction assuming the Cardelli et al. (1989) extinction curve and reddening from Schlegel et al. (1998). Emission-line fluxes are determined by fitting single Gaussians in the case of isolated lines or series of Gaussians in the case of groups of lines. In order to correct the main lines for underlying stellar continuum absorption, we use the STARLIGHT code version 04 (Cid Fernandes et al. 2005, 2011; Mateus et al. 2006; Asari et al. 2007). We mask all emission lines, as well as the locations of possible W-R bumps (4600–4700 Å and 5700–5900 Å). We apply an additional mask around the locations where the UVB and VIS arm spectra were joined owing to increased noise, as well as possible residual offsets that exist in this wavelength range. The spectra are interpolated on a regular grid of 1 Å bins. We run the code with 115 single stellar populations (SSPs) with a Chabrier (2003) initial mass function for five different metallicities up to solar metallicity and ages up to 13 Gyr with the Padova 1994 tracks from Bruzual & Charlot (2003). For objects with strong nebular continuum, we include a nebular continuum template. We use the Calzetti (2001) extinction curve with a single reddening parameter that is applied equally to all SSPs with A(V) in the range of 0–4 mag. We fit the stellar kinematics allowing for a systematic shift of ±500 km s⁻¹ and a stellar velocity dispersion of up to 500 km s⁻¹. In order to obtain the best fits to the stellar continuum in the vicinity of the weakest, temperature-sensitive lines ([O iii] λ4363, [N ii] λ5755, and [O ii] λλ7319, 7330), we follow Andrews & Martini (2013) and limit the spectral range of the STARLIGHT fit to 400 Å around these lines (200 Å in either direction), which produced a good match to the observed continuum in all cases. For higher-metallicity sources, in which [O iii] λ4363 is weak, there can be substantial contamination of [O iii] λ4363 from [Fe ii] λ4352.78 and [Fe ii] λ4359.34 (e.g., Andrews & Martini 2013; Curti et al. 2017). We took this into account by including the possible presence of the [Fe ii] lines into our fits and fixing the widths of these weak lines to that of Hγ. The line fluxes were measured after subtracting the best STARLIGHT model and corrected for dust extinction based on the Balmer decrement and assuming a Cardelli et al. (1989) extinction curve. For [O ii] λλ7319, 7330 we were able to estimate the fraction of recombination excitation that contributes to the auroral line using Equation (2) from Liu et al. (2000) that uses the temperature, the Hβ line flux, and the ${{\rm{O}}}^{2+}/{{\rm{H}}}^{+}$ ratio. In all cases in which [O iii] λ4363 was detected (thus allowing a direct estimate of the temperature and ${{\rm{O}}}^{2+}/{{\rm{H}}}^{+}$), the recombination excitation fraction of the measured [O ii] λλ7319, 7330 line flux was at the 2%–4% level. We did not correct for this small contribution. The errors on the emission-line fluxes were obtained by performing a Monte Carlo simulation of the line-fitting process using the error spectrum. For the analysis involving the weak [O iii] λ4363, [N ii] λ5755, and [O ii] λλ7319, 7330 lines we required a detection of at least 5σ. The continuum-subtracted spectra and the best-fit results in the spectral region around [O iii] λ4363, [N ii] λ5755, and [O ii] λλ7319, 7330 are shown in Figure 3. The measurements used and derived in this paper are summarized in Tables 2–5.

Zoom In Zoom Out Reset image size

The electron density (n_e) plays an important role in abundance measurements, since this parameter affects the fluxes of collisionally excited lines in star-forming regions. We determine n_e based on the sulfur [S ii] λλ6717, 6731 and [O ii] λλ3726, 3729 line ratios following Sanders et al. (2016). The results are shown in Figure 4 and show good agreement between the two values. However, given the wider separation of the [S ii] λλ6717, 6731 doublet, in this paper we adopt the electron density based on [S ii] λλ6717, 6731. For two sources (040208 and 232539), the [S ii] λλ6717, 6731 ratio produced an unphysical density, and in the analysis below we used their [O ii] λλ3726, 3729 based densities instead. For the LBAs with literature spectroscopic data, we were only able to calculate densities on the basis of [S ii] λλ6717, 6731, showing that they are in the same range as the other objects (green triangles drawn as lower limits on N_e(O ii) in Figure 4).

Zoom In Zoom Out Reset image size

We measure electron temperatures, T_e(O iii), through the $\mathrm{RO}3\equiv (I(4959)+I(5007))/I(4363)$ ratio that probes conditions in the innermost high-excitation zone in the H ii region, which is relatively insensitive to the density. The relative proximity of the LBAs allows us to detect the faint [O iii] λ4363 line and hence determine RO3 (see the spectra in Figure 3). However, the electron temperature decreases with increasing metallicity, making the auroral lines too weak to be measured, especially for the more massive systems. In addition, these higher-metallicity sources also often show an increased strength of [Fe ii] λ4359.34, which needs to be deblended from [O iii] λ4363. We detected [O iii] λ4363 in 12 of the cases (S/N of [O iii] λ4363 larger than 5). The remaining sources in which [O iii] λ4363 was not detected were excluded from the parts of the analysis that require RO3. To estimate T_e(O iii), we use the fitting function from Proxauf et al. (2014) that relates the temperature to RO3. The result is shown in Figure 5 and indicates that the temperatures are in the range of 9000–13,000 K, with the exception of S01_2, which has 19,000 K. In order to derive total oxygen and nitrogen abundances, we also require the temperatures of the lower ionization states of O and N, T_e(O ii) and T_e(N ii). These can be estimated from the density-dependent $\mathrm{RO}2\equiv (I(3726)+I(3729))/(I(7319)+I(7330))$ and $\mathrm{RN}2\equiv (I(6548)+I(6584))/(I(5755)$ line ratios, respectively (see Pérez-Montero 2017). Because these ratios depend on the usually weak auroral lines [O ii] $\lambda \lambda$7319, 7330 and [N ii] λ5755, many studies rely on standard conversions between T_e(O iii) and T_e(O ii) (e.g., Garnett 1992; Proxauf et al. 2014; Pérez-Montero 2017). Requiring at least a 5σ detection in each of the temperature-sensitive lines, we were able to measure T_e(O iii) in 12 sources, T_e(O ii) in 22 sources, and T_e(N ii) in 4 sources. The results are compared in Figure 6. The discrepancies between T_e(N ii) and T_e(O ii) are up to a factor of ∼2, i.e., much larger than the formal measurement errors, although there are only four sources with both determinations (left panel). The scatter between T_e(O iii) and T_e(O ii) is a few thousand kelvin, and the points lie around the T_e(O iii)–T_e(O ii) relations typically assumed for an H ii region with an electron density of ∼100 cm⁻³ (middle panel). Last, we compare the observed T_e(O ii) with that estimated from T_e(O iii) and the electron density using Equation (14) from Pérez-Montero (2017). The right panel of Figure 6 shows the difference between the two values as a function of the T_e(O ii) measured from RO2, with a full range of $\approx \pm 2000$ K. Because our abundance results below are quite sensitive to the correct value for T_e(O ii), in this paper we will evaluate our results using both the RO2-based values and the values inferred from T_e(O iii). For the literature LBAs, we always estimate T_e(O ii) from T_e(O iii) in the same way as done for our own spectra.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Having determined electron densities and temperatures, we now use the formalism presented in Izotov et al. (2006) to estimate the "direct-method" oxygen abundance, ${Z}_{\mathrm{direct}}\,=12+\mathrm{log}({\rm{O}}/{\rm{H}})$ under the assumption that the total oxygen abundance can be approximated by ${\rm{O}}/{\rm{H}}=({{\rm{O}}}^{+}+{{\rm{O}}}^{2+})/{{\rm{H}}}^{+}$ (see Equations (3) and (5) from Izotov et al. 2006). This calculation depends on the electron temperatures T_e(O iii) and T_e(O ii), as well as the electron density for which we will use that determined from the [S ii] λλ6717, 6731 doublet in Section 3.1. For the determination of T_e(O ii) we have two choices. In the right panel of Figure 6 we showed that the T_e(O ii) estimated from RO2 and that inferred from T_e(O iii) can show significant differences.

These different estimates for the same temperature can lead to significant differences in the derived oxygen abundances. In order to evaluate this effect, we show in Figure 7 the difference one would obtain when choosing one or the other method. The figure plots O/H calculated using the RO2-based T_e(O ii) and the RO3-based T_e(O iii) versus the difference between O/H estimated using this method and that using T_e(O ii) as inferred from T_e(O iii) (using the electron density and the TO3–TO2 relation from Pérez-Montero 2017). The discrepancy in O/H can be up to ∼0.4 dex. In the analysis below, we will adopt the T_e(O ii) values as inferred from T_e(O iii). However, in all relevant plots, we also include a second set of symbols (open circles) that give the results for the case where T_e(O ii) was estimated directly from RO2. While results on individual sources may vary, we show that our main conclusions are not affected by either choice. Further discussion of the possible sources of these discrepancies can be found in, e.g., Yates et al. (2020).

Zoom In Zoom Out Reset image size

We find values of 12 + log(O/H) in the range of ∼8.0–8.6 (with the exception of the extremely metal-poor object S01_2, which has 12 + log(O/H) ∼ 7.4). These direct-method abundances can be compared to the strong-line methods typically employed for objects at high redshift or objects with relatively shallow spectra at low redshift. We first compare two common strong-line methods based on the ${\rm{N}}2\equiv I(6584)/I({\rm{H}}\alpha )$ and ${\rm{O}}3{\rm{N}}2\equiv (I(5007)/I({\rm{H}}\beta ))/(I(6584)/I({\rm{H}}\alpha ))$ ratios. Using the Pettini & Pagel (2004) calibrations, which relate these line ratios to abundances determined using the direct method applied to a local calibration data set, the oxygen abundances for N2 and O3N2 are in general agreement (Figure 8). We find slightly higher abundances based on N2 than on O3N2. A similar result was found for z ∼ 2 galaxies from the MOSDEF survey by Sanders et al. (2015; red line in Figure 8), which appears to be a good fit to the LBAs as well. This offset between N2 and O3N2 for LBAs and Sanders et al. (2015) is about half the size of that found for z ∼ 2 KBSS galaxies by Steidel et al. (2014), especially at high metallicities. We also indicate the relation found for the local BPT offset analogs from Bian et al. (2018; blue solid line). This relation was obtained by taking the best fits between the direct-method abundance and N2 and O3N2 given by Bian et al. (2018) and converting these back to abundances in the Pettini & Pagel (2004) calibration.

Zoom In Zoom Out Reset image size

The comparison between the two strong-line method abundance estimates shown in Figure 8 does not necessarily tell us anything about the accuracy of these strong-line methods. Having determined the direct-method abundances for LBAs, we can now compare these to either strong-line method result. The results are shown in Figure 9, which compare Z(T_e) with $Z({\rm{O}}3{\rm{N}}2)$ (left panel) and $Z({\rm{N}}2)$ (right panel). The direct oxygen abundances do not appear to correlate very strongly with those based on O3N2 or N2, although the dynamic range in O/H of our sample is quite small, making it difficult to look for such trends. Overall, however, the direct-method abundances appear in general agreement with the strong-line methods within an intrinsic scatter of 0.1–0.2 dex and a small systematic offset of at most 0.1 dex. These findings are very similar to those found at z ∼ 2 by Patrício et al. (2018). The LBAs lie close to the relations for typical SDSS galaxies and BPT offset analogs from Bian et al. (2018), which have a very similar range of abundances (7.8 ≲ 12 + log(O/H)_direct ≲ 8.4) to our sources (blue and red lines in Figure 9).

Zoom In Zoom Out Reset image size

In Hoopes et al. (2007) and Overzier et al. (2010) it was shown, based on strong-line abundances, that LBAs lie below the local mass–metallicity relation of SDSS star-forming galaxies. We confirm this result using the direct-method abundances in Figure 10. At a given stellar mass, the LBAs for which direct-method abundances could be determined are offset toward lower oxygen abundances compared to the local mass–metallicity relation (MZR). We compare the LBAs with the direct-method abundances for stacked SDSS galaxies by Andrews & Martini (2013). The offset of the LBAs toward lower oxygen abundance is similar to the offsets observed for the SDSS stacks for high specific SFR objects (e.g., dashed black line in Figure 10).

Zoom In Zoom Out Reset image size

The ionization state of the gas can be quantified by the ionization parameter $q\equiv Q({{\rm{H}}}^{0})/{n}_{{\rm{H}}}$, the ratio of the hydrogen-ionizing photon rate at the surface of a plane-parallel slab, and the density of hydrogen, also expressed as the dimensionless ionization parameter $U\equiv {n}_{\gamma }/{n}_{{\rm{H}}}=q/c$, the ratio of the ionizing photon and hydrogen densities. We follow Kojima et al. (2017) in determining the ionization parameter using the ionization-sensitive ${{\rm{O}}}_{32}\equiv (I(4959)+I(5007))/I(3727)$ ratio. Although O₃₂ depends on both q and the oxygen abundance (Kewley & Dopita 2002), we can use our direct-method estimate of the oxygen abundance to break this degeneracy. We calculate q (and hence log(U)) using Equation (13) from Kobulnicky & Kewley (2004) based on the Kewley & Dopita (2002) photoionization model grids. In Figure 11 we show the relation between O₃₂ and log(U), finding that LBAs lie a small distance (0–0.1 dex) below the relation determined by Strom et al. (2018) based on photoionization modeling of KBSS galaxies at z ∼ 2 (red line). Strom et al. (2018) showed that the log(U) or q in KBSS galaxies correlate well with O₃₂, Ne3O2, and O3, but most strongly with O₃₂.

Zoom In Zoom Out Reset image size

In typical local star-forming galaxies, the ionization parameter is strongly anticorrelated with oxygen abundance. Kojima et al. (2017) provide a best-fit relation based on the direct-method oxygen abundances measured in the SDSS stacks from Andrews & Martini (2013). This relation is indicated in Figure 12 along with the LBA data. Although the dynamic range of our sample of LBAs is too small to see whether they follow the same general trend, the majority of the LBAs lie above the local relation at a given oxygen abundance. This result is very similar to that obtained by Kojima et al. (2017) for a sample of star-forming galaxies at z = 1.4–3.6. They also noticed that this behavior is similar to that found for a sample of GP galaxies (Amorín et al. 2012; Jaskot & Oey 2013). As pointed out above, these include the three objects from Amorín et al. (2012) that are also in our LBA sample. The implications of this average offset in the ionization parameter q at a fixed O/H will be discussed in Sections 3.6 and 3.7 below.

Zoom In Zoom Out Reset image size

It is expected that the pressure in the ISM of starburst galaxies is set by the feedback from massive stars (Chevalier & Clegg 1985; Heckman et al. 1990). The electron density is then the result of the compression of the shocked medium from the rate of momentum injection from stellar winds and SNe. The latter can be approximated by the SFR surface density. In the left panel of Figure 13 we plot the electron density estimated from the [S ii] λλ6717, 6731 doublet against the SFR surface density. The SFR was estimated from Hα corrected for dust using the Balmer decrement, while the surface area was computed using the seeing-deconvolved half-light radius in the r' band from SDSS. The two quantities are indeed correlated, confirming our hypothesis that the star formation feedback sets the pressure in the ISM in these compact starburst galaxies. With knowledge of both the electron density and the ionization parameter we can furthermore estimate the total hydrogen-ionizing photon rate surface density, N_ecU. This is plotted in the right panel of Figure 13, again versus the SFR surface density. This ionizing photon rate surface density can be converted directly into an SFR surface density as well, assuming the standard relation between SFR and the ionizing photon rate of $\mathrm{SFR}(Q({{\rm{H}}}^{0}))=7.4\times {10}^{-54}Q({{\rm{H}}}^{0})$. These estimates for the SFR surface densities can be read from the top axis of Figure 13, and they are of very similar order of magnitude to the more direct measure of SFR surface density shown on the vertical axis. The dotted line shows the one-to-one relation, indicating that most objects lie within a factor of a few or better from the expected values. This analysis confirms that the conditions in the ISM are directly related to the star formation activity in these compact starburst galaxies.

Zoom In Zoom Out Reset image size

Nitrogen and oxygen have similar ionization potentials, making ${{\rm{N}}}^{+}/{{\rm{O}}}^{+}$ a good proxy of the nitrogen-to-oxygen abundance ratio, N/O. We use the electron temperatures derived in Section 3.1 above, together with the relations for N⁺/H⁺ and O⁺/H⁺ from Izotov et al. (2006) (Equations (3) and (6)), to estimate N/O. Besides this direct method for N/O, there also exist convenient strong-line methods for measuring N/O, such as N2O2 ≡ [N ii] λ6584/[O ii] λ3727, which can be easily measured in optical (low redshift) or near-infrared (high-redshift) spectra of moderate signal-to-noise ratio. Besides N2O2, N/O also correlates well with other strong-line ratios involving nitrogen, such as N2S2 ≡ [N ii] λ6584/[S ii] λλ6717, 6731 and N2 ≡ [N ii] λ6584/Hα (e.g., Pérez-Montero & Contini 2009; Kojima et al. 2017; Strom et al. 2017, 2018).

In Figure 14 we first compare N2O2 with N2S2 with various relations from the literature. Strom et al. (2018) showed that KBSS galaxies at z ∼ 2 lie along a relation that is more similar to that found for local H ii regions (Pérez-Montero & Contini 2009; Strom et al. 2017) than typical star-forming galaxies from SDSS. LBAs occupy the region in between the relations for z ∼ 2 and H ii regions on one hand and local SDSS galaxies on the other, with most LBAs being closer to the high-redshift/local H ii lines. At a given N2O2, LBAs have higher N2S2 compared to typical SDSS galaxies. This behavior is very similar to that observed for local H ii regions and z ∼ 2 KBSS galaxies and can be understood by the fact that LBAs and KBSS galaxies have star formation histories that are much simpler and more similar to H ii regions compared to typical SDSS star-forming galaxies (Sanders et al. 2016; Kashino et al. 2017; Strom et al. 2018). Part of the [S ii] emission may arise from diffuse emission regions in between the H ii regions, lowering N2S2 in the case of more complex galaxies (Sanders et al. 2017).

Zoom In Zoom Out Reset image size

In Figure 15 we plot the strong-line indicators for N/O (N2O2, N2S2, and N2) versus the direct N/O ratio. The majority of LBAs are concentrated in the region between ${({\rm{N}}/{\rm{O}})}_{\mathrm{direct}}\gtrsim -1.5$ and the solar value (horizontal dotted line). The smallest scatter is found for N2O2, which is not surprising because it is the most direct measure of N⁺/O⁺ among these three indicators. The LBAs lie along the relations found for KBSS galaxies and local H ii regions. The N2O2–N/O relation for H ii regions and galaxies from Pérez-Montero & Contini (2009) is known to be too steep, overestimating N/O for high N2O2 for both local H ii regions and high-redshift galaxies (Strom et al. 2018). The relation between N/O and the strong-line indicators found for the LBAs is more similar to the calibrations found for H ii regions and high-redshift galaxies (Strom et al. 2017, 2018). Using only those objects in our sample for which the direct-method N/O ratio is available, we fit the relation between N/O and N2O2, finding $\mathrm{log}{({\rm{N}}/{\rm{O}})}_{\mathrm{direct}}=0.73\times {\rm{N}}2{\rm{O}}2-0.58$.

Zoom In Zoom Out Reset image size

We will now determine whether the nitrogen-to-oxygen ratios of LBAs are in any way markedly different compared to either typical local or high-redshift sources. In Figure 16 we show the direct-method nitrogen-to-oxygen ratio log(N/O)_direct versus the stellar mass (left panel) and the direct-method oxygen abundances. In both panels, we again give two sets of points, one calculated using T(O ii) as inferred from T(O iii) (filled circles) and the other using T(O ii) estimated from RO2 (open circles). In the left panel of Figure 16, we are furthermore able to show a larger set of points by taking advantage of the best-fit relation between N/O and N2O2 determined for LBAs above (see left panel of Figure 15). This allows us to include LBAs for which no direct-method N/O abundances could be measured. These estimates are indicated by the open squares. In the left panel, it can be seen that the LBAs scatter around the typical N/O found for stacks of SDSS galaxies of a similar stellar mass (solid black line), which was also determined through the direct method by Andrews & Martini (2013). Kojima et al. (2017) find that their small sample of z ∼ 2 galaxies have N/O values comparable to or smaller than this local relation, while Strom et al. (2017) find a somewhat lower N/O on average by  ∼0.1–0.2 dex for KBSS galaxies at z ∼ 2 compared to local SDSS galaxies with stellar masses in the range of ≃10¹⁰–10¹¹ M_⊙ (red solid line in the left panel). However, in any case, these similar or somewhat lower values of N/O at fixed stellar mass are smaller than the typical downward offsets observed for LBAs in the O/H–M_* relation (see Figure 10). This implies that LBAs and the z ∼ 2 samples from Kojima et al. (2017) and Strom et al. (2017) likely lie along a different N/O–O/H relation compared to typical local star-forming galaxies. We illustrate this in the right panel of Figure 16, where we have indicated again the local sequence based on the SDSS stacks from Andrews & Martini (2013) as determined by Kojima et al. (2017) (solid black lines). The LBAs are, on average, offset toward higher N/O at fixed O/H.

Zoom In Zoom Out Reset image size

The N/O excesses observed are of similar order of magnitude to those found for galaxies at z ∼ 2 by Kojima et al. (2017). The red dashed line in the right panel of Figure 16 is the upper limit on the average N/O ($\mathrm{log}{({\rm{N}}/{\rm{O}})}_{\mathrm{direct}}=-1.26$) for their z ∼ 2 sample, while the blue dashed line is the average N/O ($\mathrm{log}{({\rm{N}}/{\rm{O}})}_{\mathrm{direct}}=-1.08$) they found for a sample of local galaxies selected to have similar (low) stellar masses and (high) SFRs to their z ∼ 2 sample. Kojima et al. (2017) also noted that a stacked spectrum of the Steidel et al. (2016) sample of KBSS galaxies at z ∼ 2 falls close to their sample average (red dashed line), while GPs (which include several of the LBAs from our sample) span the range from no N/O excess to the average N/O for local, low-mass, high-SFR objects (blue dashed line). Our results for LBAs are in general agreement with those trends.

It has been known for some time that intensely star-forming galaxies at high redshift such as LBGs, star-forming BzK galaxies, and distant red galaxies are often offset toward higher line ratios with respect to the mean SDSS star-forming population at low redshift in a standard BPT diagram (e.g., Teplitz et al. 2000; Shapley et al. 2005; Erb et al. 2006; Liu et al. 2008; Hayashi et al. 2009; Lehnert et al. 2009). Such offsets were also found in some populations of low-redshift galaxies, such as the "warm" infrared-luminous galaxies (Kewley et al. 2001) and W-R galaxies (Brinchmann et al. 2008a). These BPT offsets have been observed for LBAs as well (Hoopes et al. 2007; Overzier et al. 2008, 2009; see also Figure 2 in this paper), further strengthening the conclusion that they are good local analogs of the high-redshift star-forming population. Overzier et al. (2009) showed that the BPT displacement is in the sense of enhancements in one or both of [N ii] λ6584/Hα and [O iii] λ5007/Hβ, and that the size of the perpendicular offset from the local star-forming ridge increases with both the luminosity of the brightest starburst clump and the electron density, indicating that higher SFR densities, interstellar densities and pressures, and ionization parameters may be related to the BPT offsets (e.g., Brinchmann et al. 2008b; Liu et al. 2008; Shirazi et al. 2014).

In recent years, it has become feasible to address in more detail the physical origins of these BPT offsets, driven mainly by the wealth of near-infrared spectroscopic data from new surveys of z ∼ 2 galaxies (e.g., Masters et al. 2014; Steidel et al. 2014; Shapley et al. 2015; Kashino et al. 2017). Understanding how objects move through the BPT diagram, including the nature of the offsets, is extremely important, because the measurement of gas-phase abundances critically depends on the interpretation of ratios of the strong emission lines. This is particularly important at high redshift, where the temperature-sensitive methods are often not available. There is therefore no guarantee that locally established calibrations are directly applicable. Various authors have found that high-redshift objects display BPT offsets that are often larger than that which can be explained by their increased ionization parameters and densities, and they have shown that the BPT offset objects often have an increased N/O abundance and/or ionization parameter at high redshift compared to local galaxies for the same oxygen abundance (e.g., Masters et al. 2014; Shapley et al. 2015; Sanders et al. 2016; Kojima et al. 2017). Steidel et al. (2016) and Strom et al. (2017, 2018) showed that z ∼ 2 galaxies indeed have a higher N/O (at fixed excitation) and that N/O varies with O/H in a similar manner to that observed for local H ii regions. Besides a higher N/O, they furthermore show the need for a harder ionizing spectrum at high redshift to fully explain the BPT offsets. These harder spectra could be a result of the differences in the star formation (and thus chemical enrichment) histories between typical local and high-redshift galaxies.

In the previous sections, we have shown that at fixed O/H, LBAs have higher ionization parameters (Figure 12) and higher N/O (Figure 16, right panel) compared to typical local star-forming galaxies. Kojima et al. (2017) showed that galaxies at z ∼ 2, as well as a small sample of local analogs consisting of LBAs, GPs, and low-mass/high-SFR objects, display BPT offsets when either one of N/O or q is increased, or both are increased. Positions along the BPT diagram are further modulated by the different O/H of the galaxies. The change in N/O (at fixed O/H) only affects the [N ii] λ6584/Hα ratio by an amount that directly corresponds to that change (a 0.3 dex increase in N/O gives a 0.3 dex change in [N ii] λ6584/Hα). The increase in q is in the opposite direction to N/O, with some component along [O iii] λ5007/Hβ that is roughly aligned along the star-forming ridge. At 12 + log(O/H) = 8.10, a positive change of 0.3 dex in q gives an increase in [O iii] λ5007/Hβ of ∼0.1 dex but a decrease in [N ii] λ6584/Hα of ∼0.2 dex. Combined with the 0.3 dex increase in N/O, the total offset in the BPT diagram is toward higher values of both [O iii] λ5007/Hβ and [N ii] λ6584/Hα, and in the direction observed for the z ∼ 2 galaxies studied.

In Figure 17 we show again the location of LBAs in the BPT diagram, but with a color-coding determined by the offsets measured in N/O and q relative to the local relations expectation given the O/H determined for each object through the direct method. The objects with the largest offsets in either parameter (or both) are expected to have the largest BPT offsets, which is indeed the case. The objects in our sample with the largest offsets lie around the average relation found by Steidel et al. (2014) for galaxies at z ∼ 2. In the lower right corner of the diagram, we find a number of massive and (likely) more metal-rich LBAs for which the [O iii] λ4363 flux could not be determined and thus were not included in the analysis. We show the offsets measured perpendicularly from the local star-forming ridge as a function of the offsets in ionization parameter and N/O in Figure 18. Objects with the largest perpendicular offsets (ΔBPT_⊥) indeed tend to have the largest offsets in q (left panel) or log(N/O) (right panel) or both. From the analysis presented here, we thus conclude that the physical origin of the BPT offsets in LBAs is likely to be the same as those that explain the galaxies at z ∼ 2 and other types of local analogs thereof.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As shown in Figure 16, LBAs span a wide range of almost 1 dex in N/O values at a given O/H ratio. This wide scatter is similar to that observed for local H ii regions and the sample of KBSS galaxies at z ∼ 2 by Strom et al. (2018). These N/O values furthermore lie, on average, several tenths of a dex above the typical N/O expected for local galaxies. This is similar to the trends observed for z ∼ 2 galaxies and other local analogs (e.g., Kojima et al. 2017; Strom et al. 2018). What could be responsible for this N/O excess? The two most straightforward explanations are that N/O is increased either as a result of the presence of W-R stars that enhance nitrogen on a timescale that is shorter than the SN enrichment (e.g., Hawley 2012) or as a result of a decrease in O/H from the infall of a large quantity of low-metallicity gas, with the consequence that N/O appears spuriously high for the resulting O/H (e.g., Amorín et al. 2010; Masters et al. 2014).

3.7.1. Evidence for Wolf-Rayet Stars

Galaxies with W-R features often have enhanced N/O. Brinchmann et al. (2008a) compared the average N/O between W-R-enhanced galaxies and galaxies without W-R features as a function of the equivalent width of Hβ in order to control for differences that could arise owing to secondary nitrogen production. In Figure 19 we show the excess N/O as a function of the EW(Hβ) for our sample. Brinchmann et al. (2008a) found that the W-R phase can be responsible for about 0.15 ± 0.1 dex excess N/O for objects with EW(Hβ) below ∼100 Å, while above that the starbursts are typically younger than the phase where enrichment by W-R winds occurs (thick solid line in Figure 19). Several LBAs are known to have W-R features. We have measured the strengths of the W-R blue and red bumps in our sample, with two examples of objects with strong bumps shown in Figure 20. In total we find six objects with clearly identifiable W-R features (005527, 015028, 020356, 040208, BPT10, and HST03). These are marked in Figure 19 with the large pentagons. Brown et al. (2014) also showed evidence for W-R features in two additional LBAs (004054 and 092600). While some of the objects with the largest N/O excesses include objects with strong W-R features in their spectra, the N/O excess is much larger than expected for W-R galaxies. Furthermore, there appears to be no trend in Figure 19 that relates the N/O excess to W-R features or EW(Hβ).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.7.2. Evidence for Infall of Metal-poor Gas

Besides nitrogen enhancement from W-R stars, the rapid inflow of relatively metal-poor gas is an attractive alternative (or additional) scenario (e.g., Hwang et al. 2019), especially given that most LBAs appear to be starbursts triggered by a recent interaction event (Overzier et al. 2008, 2009). We showed that at a given stellar mass, LBAs have relatively low O/H with respect to the local MZR (Figure 10), while they have normal N/O for their stellar mass (left panel of Figure 16). They also have a lower O/H than expected based on their N/O (from the right panel of Figure 16). This is exactly what would be expected for the accretion scenario. We can show this more clearly in Figure 21, where we plot the offset in O/H with respect to the local MZR versus the offset in O/H with respect to the local average at a fixed N/O (from the right panel of Figure 16). The good one-to-one correspondence between them implies that these offsets are one and the same and related to the infall of a large quantity of metal-poor gas. In order to decrease O/H by 0.3 dex, the galaxy would need to accrete a quantity of metal-poor gas equal to the mass of the ISM prior to the infall. On the secondary axes of Figure 21 we have translated the offsets in O/H observed to the quantities of accreted metal-poor gas in units of the pre-infall gas mass. We have shown that the effect of the presence of W-R stars on N/O is expected to be relatively limited and furthermore is at odds with the normal N/O versus stellar mass relation (left panel of Figure 16). We conclude that the recent gas accretion scenario can largely, or completely, explain the apparent N/O excesses observed.

Zoom In Zoom Out Reset image size