© ESO 2019

1. Introduction

Nearby dwarf star-forming galaxies (SFG) with extremely low metallicities are often considered as local counterparts of high-redshift galaxies. Their proximity allows us to study them in much greater detail than the high-redshift galaxies and to establish useful constraints on the physical conditions of their low-metallicity interstellar medium (ISM), the origin of chemical elements, and to develop models of stellar evolution. These studies can be used to analyse the physical properties of the primeval galaxies at redshifts z ∼ 5−10, which are thought to be dwarf systems responsible for the reionization of the Universe (Ouchi 2009; Wise & Cen 2009; Bouwens et al. 2015; Robertson 2013; Robertson et al. 2015; Khaire et al. 2016).

However, the number of known nearby lowest-metallicity SFGs with oxygen abundances 12 + log O/H ≲ 7.35 is very low. Among them are SFGs with 12 + log O/H ∼ 7.00. These are J0811+4730 with 12 + log O/H = 6.98 ± 0.02 (Izotov et al. 2018a), A198691 with 12 + log O/H = 7.02 ± 0.03 (Hirschauer et al. 2016) and SBS 0335−052W with 12 + log O/H ranging from 6.86 to 7.22 in different star-forming regions (Izotov et al. 2009). Recently, Annibali (2019) reported the range 6.96–7.14 in different H II regions of DDO 68. Additionally, Pustilnik et al. (2016) and Kniazev et al. (2018) derived 12 + log O/H ≲ 7.35 in nearly a dozen of galaxies in the Lynx-Cancer and Eridanus voids.

Large spectroscopic surveys, such as the Sloan Digital Sky Survey (SDSS) containing millions of galaxy spectra in its database, open an opportunity to considerably increase the sample of the extremely low-metallicity SFGs with 12 + log O/H ≲ 7.35 (e.g. Izotov et al. 2012; Guseva et al. 2017). In particular, the SFG with a record low luminosity-weighted metallicity, J0811+4730, was found in the SDSS.

The empirical diagnostic [O III]λ5007/Hβ–[N II]λ6584/Hα diagram (Baldwin et al. 1981) can successfully be used to preselect the candidates to extremely low-metallicity SFGs. On this diagram, they occupy the region with both very low [O III]λ5007/Hβ ≤ 3 and very low [N II]λ6584/Hα ≤ 0.1, far from the main sequence of low-redshift SFGs (Izotov et al. 2012). Similarly, the empirical R₂₃−O₃₂ diagram can be used, where R₂₃ = ([O II]λ3727+[O III]λ4959+[O III]λ5007/Hβ and O₃₂ = [O III]λ5007/[O II]λ3727 (Izotov et al. 2018a).

A variety of methods have been developed and can be used for the abundance determination. The most reliable method is the direct method, but it requires the detection, with good accuracy, of the [O III]λ4363 emission line for the electron temperature determination. For SFGs with weak or undetected [O III]λ4363 emission lines only strong-line methods can be used. These methods are based on combinations of strong emission-line intensities of various elements, such as O, N, S, and Ar. However, the problem with strong-line methods is that the intensities of emission lines depend not only on the metallicity but also on some other quantities, such as the ionization parameter. Many of these methods were constructed for application to galaxies in a wide range of metallicities, typically at oxygen abundances 12 + log O/H ≳ 7.4, and often on expence of the accuracy in the abundance determination at lowest metallicities (e.g. Pilyugin & Thuan 2005; Pilyugin & Grebel 2016; Nagao et al. 2006; Yin et al. 2007). This paper focuses on simple empirical strong-line methods for extremely low-metallicity objects based on oxygen lines. Emphasizing systems with abundances much lower than solar, we aim to increase the accuracy of abundance determinations in SFGs with undetected or weak [O III]λ4363 emission. The construction of this method is also motivated by the fact that recently a number of new extremely low-metallicity SFGs have been discovered for which the oxygen abundances were derived using high-quality spectra (Izotov et al. 2018a, and references therein) allowing for more reliable calibration with the direct method at lowest metallicities. The new method is applied in the present paper to search for extremely low-metallicity candidates in the SDSS Data Release 14.

The data base used for the selection of the lowest-metallicity SFGs is described in Sect. 2. Methods of the oxygen abundance determination are discussed in Sect. 3. The results of the abundance determination and the list of the SFGs with 12 + log O/H <  7.4 are presented in Sect. 4. The application of diagnostic diagrams for the selection of extremely low-metallicity SFGs is considered in Sect. 5. Our main conclusions are summarised in Sect. 6.

2. Data

We have searched for extremely low-metallicity galaxies from a sample of ∼30 000 SFGs selected from the spectroscopic database of the SDSS DR14 (Abolfathi et al. 2018). This is a continuation of studies which used earlier SDSS Data Releases and were published by Izotov et al. (2012) and Guseva et al. (2015, 2017). Details of SDSS sample selection can be found in Izotov et al. (2014a). This sample includes SFGs with equivalent widths EW(Hβ)≳10 Å of the Hβ emission line in their spectra indicating an active star formation and the presence of hot massive stars. The [O III] λ4363 Å emission line is present in spectra of ∼18 700 out of ∼30 000 SFGs. In ∼2000 SDSS spectra this line is measured with an accuracy better than 25% allowing for the reliable determination of element abundances, most often of oxygen, neon and nitrogen.

The line fluxes and their errors in each spectrum were measured using the IRAF¹splot routine and then corrected for extinction. The internal extinction was derived from the Balmer hydrogen emission line fluxes after correction for the Milky Way extinction. The line fluxes were corrected for both reddening (Cardelli et al. 1989) and underlying hydrogen stellar absorption by the application of an iterative procedure (Izotov et al. 1994) and were used for the element abundance determination. The same SDSS spectra were used for the determination of some integrated characteristics such as Hβ luminosities, L(Hβ), star formation rates SFR, and stellar masses M_⋆, adopting a luminosity distance derived with a cosmological calculator (NED, Wright 2006), based on the cosmological parameters H₀ = 67.1 km s⁻¹ Mpc⁻¹, Ω_Λ = 0.682, and Ω_m = 0.318 (Planck Collaboration XV 2014).

3. Methods of oxygen abundance determinations

One of the commonly used methods of the oxygen abundance determination in low-metallicity SFGs is the direct method based on the determination of the electron temperature T_e(O III) from the [O III] λ4363/(λ4959 + λ5007) emission-line ratio and on the relation between T_e(O II) and T_e(O III) obtained by, for example, Izotov et al. (2006a) from the photoionized H II region models. The knowledge of the temperatures and the presence of [O II] and [O III] emission lines in the SDSS spectra allowed us to determine the abundances of these two most abundant oxygen ions in the H II regions and thus the total oxygen abundances for a large sample of SFGs. However, the [O III]λ4363 Å emission line is weak in most SDSS spectra of low-metallicity SFGs and is detected with low signal-to-noise ratio introducing large uncertainties in the determination of the electron temperature and oxygen abundance.

Therefore, for these galaxies, other methods based on strong emission lines are needed to more reliably derive their metallicities. In this paper we have developed a purely empirical strong-line method based on the SDSS DR14 sample aiming to apply it for selection of the extremely low-metallicity SFG candidates. This method uses the spectroscopic properties of well-studied lowest-metallicity galaxies and is calibrated with the direct method. For the strong-line method the most natural is to use the emission-line fluxes of oxygen, the most abundant heavy element. The common approach in the past was to use the relation between the metallicity and the sum of the fluxes of strong oxygen lines in the optical range, R₂₃ = ([O II]λ3727 + [O III]λ4959 + [O III]λ5007)/Hβ.

The problem with this method is that the relation between the metallicity and R₂₃ consists of low- (12 + log O/H ≲ 8.0) and high-metallicity (12 + log O/H ≳ 8.0) branches and thus the same value of R₂₃ corresponds to two metallicities (e.g. Pagel et al. 1980; Edmunds & Pagel 1984; McGaugh 1991). Therefore, other constraints separating low- and high-metallicity branches are needed to resolve this ambiguity. One of the possible solutions is to use the relation between the metallicities and [N II]λ6584/Hα emission line flux ratios which monotonically increase with metallicity and thus this relation is not degenerate (e.g. van Zee et al. 1998; Pettini & Pagel 2004). However, the use of this relation for the determination of the oxygen abundances is limited because the [N II]λ6584 emission line is very weak or undetected in low-metallicity galaxies with high-excitation H II regions. We have used this line (or the upper limit of its flux) to separate low- and high-metallicity branches. Adopting [N II]λ6584/Hβ ≲ 0.2 would then select objects located on the low-metallicity branch of the log R₂₃ − 12 + log O/H relation.

To construct the log R₂₃ − 12 + log O/H diagram we select ∼2000 SDSS DR14 SFGs in which the [O III] λ4363 emission line is detected with an accuracy better than 25% (see Sect. 2) allowing for a reliable oxygen abundance determination (grey dots in Fig. 1a). Additionally, all these galaxies show in their spectra [O II]λ3727 Å, [O III]λ4959 Å, λ5007 Å and have [O III]λ4959/Hβ ≥ 0.2. These data are supplemented by various samples of SFGs with reliably detected [O III]λ4363 Å emission lines in their spectra: a sample of extremely low-metallicity SFGs with 12 + log O/H ≤  7.35 (filled symbols in Fig. 1a; Hsyu et al. 2017; Annibali 2019; Izotov et al. 2018a, and references therein), a sample of SFGs with highest O₃₂ = [O III]λ5007/[O II]λ3727 ratios of ∼20–40 (asterisks in Fig. 1a; Izotov et al. 2017, and references therein), a sample of Lyman continuum leaking galaxies with O₃₂ ratios in the range ∼5–28 (crosses in Fig. 1a; Izotov 2018b, and references therein), and a sample of SFGs used for the primordial He abundance determination (open circles in Fig. 1a; Izotov et al. 2014b, and references therein). We note that two encircled filled circles indicate the location of star-forming regions 3 and 4 in SBS 0335−052W (Izotov et al. 2009) with undetected [O III] λ4363 Å line. They are shown because of their extremely low oxygen abundances, which are among the lowest known and considerably lower than the luminosity-weighted oxygen abundance 7.12 of SBS 0335−052W (Izotov & Thuan 2007). The electron temperatures and the oxygen abundances in these objects are derived by Izotov & Thuan (2007) using the semi-empirical method. However, we have not used them in our subsequent fitting of relations for the abundance determination. It is also worth to mention SFG J0811+4730 because of showing the lowest luminosity-weighted oxygen abundance 12 + log O/H = 6.98 ± 0.02 known so far (filled star in Fig. 1a; Izotov et al. 2018a).

Fig. 1. Panel a: (log R23)−(12 + log O/H), and panel b: log (R23 − 0.08O32)−(12 + log O/H) relations. In both these relations 12 + log O/H is derived by the direct method, R23 = ([O III]λ4959+λ5007+[O II]λ3727)/Hβ and O32 = [O III]λ5007/[O II]λ3727. The galaxy J0811+4730 with the lowest luminosity-weighted oxygen abundance (Izotov et al. 2018a) is shown by a filled star, while two knots in SBS 0335−052W with similar or lower oxygen abundances (Izotov et al. 2009) are represented by encircled filled circles. Other lowest-metallicity SFGs from Izotov et al. (2018a), the galaxy Little Cub (Hsyu et al. 2017) and the H II region DDO 68 #7 (Annibali 2019) are shown by filled circles. SFGs with the highest O32 ratios in the range ∼20–40 (Izotov et al. 2017) are shown by asterisks. Samples of SFGs used for the He abundance determination (Izotov et al. 2014a, and references therein) and Lyman continuum leakers (Izotov 2018b, and references therein) are shown by open circles and crosses, respectively. Grey dots represent SDSS SFGs at z >  0.02 from DR9 and earlier releases and at any redshift from DR10 and later releases, with the [O III]λ4363 Å emission line measured with an accuracy better than 25%. The dashed, dotted, solid, and dash-dotted lines in panel a are the empirical relations by Skillman (1989), Nagao et al. (2006), Yin et al. (2007) and Pilyugin (2000), respectively. The solid line in panel b is the linear most likelihood fit to the SDSS data with log (R23 − 0.08O32)≤0.8 (this paper).

Open with DEXTER

It is important that the data include the objects with the highest O₃₂ which are indicators of a very high ionization parameter U that is the measure of the number of ionizing photons per one ion in the H II region. The use of these galaxies allows us to eliminate the dependence of the calibration relation on U.

For the sake of comparison, we show in Fig. 1a some simple calibration relations produced in the past. Three of them, those by Skillman (1989), Pilyugin (2000) and Yin et al. (2007), are linear. We note that these relations reproduce the oxygen abundances reasonably well (within ∼0.2 dex) of the extremely low-metallicity SFGs with log R₂₃ <  0.5 shown by filled symbols. On the other hand, SFGs with highest O₃₂ at log R₂₃ >  0.5 (asterisks and three SFGs shown by filled circles) considerably deviate from both the SDSS SFGs and linear relations.

We note the relatively high scatter of filled symbols in the diagram which we attribute to the ionization parameter varying in a large range. To minimize the scatter caused by various ionization parameters we show in Fig. 1b the relation log (R₂₃ − 0.08O₃₂)−12 + log O/H. The distributions of SFGs in Fig. 1b show a much lower scatter than in Fig. 1a. The data in Fig. 1b can be reproduced by the linear relation

(1)

shown by a solid line.

4. Selection of candidates to the extremely low-metallicity SFGs in SDSS DR14

We applied Eq. (1) for selection of candidates to the extremely low-metallicity SFGs with 12 + log O/H <  7.4. It is seen in Fig. 1b that these low metallicities correspond to (R₂₃ − 0.08O₃₂)≲4. We also adopted an upper limit of [N II]λ6584/Hβ ≤ 0.2 to exclude the contamination of the sample from objects on the upper branch of the log R₂₃ − 12 + log O/H relation, and put a low limit [O III]λ4959/Hβ ≳ 0.2 to exclude the galaxies which were not tested in Sect. 3 with the direct method because of the very weak [O III]λ4363 Å emission line.

The [O II] λ3727 Å, [O III] λ4959 Å, λ5007 Å emission line intensities are needed to derive 12 + log O/H by the strong-line method discussed in Sect. 3. However, most of SDSS lowest-metallicity galaxies are at low redshifts. Therefore, the [O II]λ3727 Å emission line is outside the wavelength range of ∼3800–9200 Å in spectra of galaxies with z ≲ 0.02 selected in DR9 and earlier releases, precluding the determination of metallicity while spectra of DR10 and later releases are obtained in a larger wavelength range of ∼3600–10300 Å including the rest-frame wavelength of [O II]λ3727 Å emission line. To avoid this difficulty with spectra of DR9 and earlier releases some prescriptions are needed to estimate the intensity of the [O II]λ3727 Å emission line from intensities of other lines. In particular, Izotov et al. (2012) and Guseva et al. (2015, 2017) used for that [O II]λ7320 Å, λ7330 Å emission lines. However, these lines are relatively weak or undetected introducing large uncertainties. Additionally, for galaxies at redshifts z ∼0.02–0.03, their observed wavelengths are close to the wavelength ∼7500 Å of strong telluric absorption. Keeping in mind these caveats, we decide to use the [O II]λ7320 Å, λ7330 Å emission lines if detected for the determination of the [O II]λ3727 Å emission line flux from the relation (Aller 1984; Izotov et al. 2006a)

(2)

where I([O II] λ7325) = I([O II] λ7320) + I([O II] λ7330) and T_e(O II) is the electron temperature in the O⁺ zone of the H II region from where [O II] emission lines originate. Thus the electron temperature T_e(O II) is needed to derive I([O II] λ3727) using Eq. (2).

There is no need to use Eq. (2) if [O II]λ3727 Å is observed. If it is not observed but [O III]λ4363 Å is detected, then both T_e(O III) and T_e(O II) can be derived. The most complicated case for SDSS spectra of low-metallicity galaxies is when both [O II]λ3727 Å and [O III]λ4363 Å emission lines are not detected. Then some estimates of the electron temperature T_e(O II) are needed to use the Eq. (2). To do that we consider the relation between T_e(O II) and log (R₂₃ − 0.08O₃₂) for the galaxies with spectra where the [O II]λ3727 Å line is present and the [O III]λ4363 Å line is detected with a relative error better than 25%. This allowed us to derive both T_e(O II) with good accuracy and R₂₃ − 0.08O₃₂. The relation is shown in Fig. 2a. It is seen that T_e(O II) at high R₂₃ − 0.08O₃₂ varies in a wide range, but it is nearly constant with the value of ∼15 500 K at lower R₂₃ − 0.08O₃₂. This constancy is due to the fact that T_e(O III) at low R₂₃ − 0.08O₃₂ asymptotically approaches the value ∼22 000 K. It is also worth to note that CLOUDY photoionization models of hottest H II regions predict T_e(O III) and T_e(O II) of ∼22 000 K–24 000 K and ∼15 500 K, respectively (Ferland et al. 1998; Ferland 2013). We adopt the value T_e(O II) = 15 500 K for the determination of I([O II] λ3727) in the lowest-metallicity galaxy candidates for which the condition R₂₃ − 0.08O₃₂ <  5 is satisfied corresponding to 12 + log O/H ≲ 7.5.

	Fig. 2. Panel a: log (R23 − 0.08O32)−Te(O II) relation for SFGs. Panel b: relation between the extinction-corrected [O II]λ3727/Hβ and [N II]λ6584/Hβ flux ratios. The solid line corresponds to [O II]λ3727 = 20 × [N II]λ6584. Symbols in both panels are as in Fig. 1.
Open with DEXTER

We used the extinction-corrected [N II]λ6584 emission line to estimate the [O II]λ3727 emission-line flux for objects with undetected [O II]λ7325 emission in their spectra. To find the relation between [N II]λ6584 and [O II]λ3727 fluxes we show in Fig. 2b the diagram [N II]λ6584/Hβ–[O II]λ3727/Hβ for SDSS DR14 SFG spectra with detected [N II]λ6584 and [O II]λ3727 lines. We see that

(3)

at low [N II]λ6584/Hβ ≤ 0.1. The determination of the [O II]λ3727 flux for higher [N II]λ6584/Hβ is subject to much larger uncertainties and the use of Eq. (3) would in general result in overestimation of the [O II]λ3727 flux and correspondingly in overestimation of the oxygen abundance. We applied this relation for selection of extremely low-metallicity candidates keeping in mind that the oxygen abundance in selected SFGs with [N II]λ6584/Hβ = 0.1 − 0.2 might be overestimated. Nevertheless, additional spectroscopic observations of selected galaxies including the [O II]λ3727 Å emission line are needed to verify and improve our indirect determination of its flux.

Three methods are used to derive the oxygen abundances. First is the direct method using prescriptions by Izotov et al. (2006a). A two-zone photoionised H II region model was adopted: a high-ionisation zone with temperature T_e(O III), where [O III] lines originate, and a low-ionisation zone with temperature T_e(O II), where the [O II] lines originate. For T_e(O II), the relation between the electron temperatures T_e(O III) and T_e(O II) from Izotov et al. (2006a) is used. Ionic and total oxygen abundances are derived using expressions for oxygen ionic abundances by Izotov et al. (2006a).

The second, semi-empirical method proposed by Izotov & Thuan (2007) is based on the determination of the electron temperature from the strong oxygen emission lines in the galaxies with undetected [O III]λ4363 Å emission line, while ionic and total oxygen abundances are derived in the way used by the direct method.

Finally, the third method is the strong-line method (Eq. (1)) developed in this paper. This simplest method is purely empirical and does not require determination of physical conditions in the H II region, at variance with the two other methods.

We selected 66 extremely low-metallicity SFGs in the entire SDSS spectroscopic data base for which at least one of the three values of 12 + log O/H derived by each of three methods is less than 7.4. The list of these SFGs is shown in Table A.1 and oxygen abundances derived by all three methods to check their mutual consistency are presented in Table A.2. The tables also include coordinates, redshifts, line intensities, equivalent widths of the Hβ emission line, and integrated characteristics, such as SDSS absolute magnitudes M_g in the g-band and stellar masses M_⋆ derived from the fitting of SDSS spectra. The selected galaxies are located mainly at very low redshifts, have faint absolute magnitudes and low stellar masses, characterising them as dwarf star-forming galaxies. The [O III]λ4363 Å emission line is detected in most selected galaxies allowing the 12 + log O/H determination by the direct method.

A notable feature is that all three methods give consistent results with a dispersion less than 0.1 dex in most cases (Table A.2). In Fig. 3 we compare oxygen abundances derived by our strong-line method with those derived by the direct, semi-empirical, and some other strong-line methods in the literature. The agreement between 12 + log O/H derived with the direct method and that derived with our newly developed strong-line method is somewhat better than that derived with the semi-empirical method. This is more clearly seen in Figs. 3a and b, in which we show only galaxies with detected [O III]λ4363 Å emission line. On average, the oxygen abundances derived by the strong-line and semi-empirical methods are respectively by ∼0.04 dex and ∼0.08 dex higher than those derived by the direct method. In Fig. 3c we compare oxygen abundances derived by the strong-line and the semi-empirical methods which on average are consistent within ∼0.05 dex. The strong-line method developed in this paper and semi-empirical method are likely more preferable compared to the direct method for galaxies with weak [O III]λ4363 Å emission, detected with poor signal-to-noise ratio.

Fig. 3. Panel a: distribution of differences between oxygen abundances 12 + log O/H derived by the semi-empirical and the direct methods for the sample from Table A.2 excluding galaxies with non-detected [O III]λ4363 Å emission lines. Panel b: as in panel a but for differences between 12 + log O/H derived by our new strong-line and the direct methods. Panel c: as in panel a but for differences between 12 + log O/H derived by the strong-line and the semi-empirical methods for all SFGs from Table A.2. Panel d: as in panel c but for differences between 12 + log O/H derived by the strong-line method and the strong-line method by Yin et al. (2007). Panel e: as in panel c but for differences between 12 + log O/H derived by the strong-line method and the P-method (Eq. (23) in Pilyugin & Thuan 2005). Panel f: as in panel c but for differences between 12 + log O/H derived by the strong-line method and the P-method (Eq. (24) in Pilyugin & Thuan 2005). In all panels, galaxies with observed and non-observed [O II]λ3727 emission line are shown by filled and open circles, respectively. Solid horizontal lines indicate average abundance differences and dotted horizontal lines indicate ±1σ dispersions of the sample around the average values.

Open with DEXTER

In Figs. 3d–3f we compare our strong-line method with some previously proposed strong-line methods from the literature. Yin et al. (2007) presented a simple relation between 12 + log O/H and R₂₃ which on average gives oxygen abundances consistent with our strong-line method (Fig. 3d). However, at lowest 12 + log O/H ≲ 7.1, the relation by Yin et al. (2007) predicts oxygen abundances by as much as ∼0.2 higher than those obtained with our strong-line method. Pilyugin & Thuan (2005) develop their strong-line method by introducing the parameter P = [O III]λ5007/([O II]λ3727 + [O III]λ5007) which takes into account the dependence on the ionization parameter. They present two relations for their method which we compare in Figs. 3e and 3f with our strong-line method. Two features of this comparison are worth to note: 1) a larger dispersion of objects compared to that in Fig. 3c and 2) a systematic offset of average values of Δ(12 + log O/H) from the zero value indicating that the two modifications of the P-method give systematically higher 12 + log O/H, by ∼0.02 (Fig. 3e) and by ∼0.10 (Fig. 3f). Finally, we note that there is no difference between the galaxies with observed and non-observed [O II] λ3727 emission line in all panels of Fig. 3 (filled and open circles, respectively) implying reliability of calculated intensities of this line. From the above comparison we conclude that our simple strong-line method is likely the most reliable for the oxygen abundance determination of extremely low-metallicity galaxies.

The lowest-metallicity segment of the log (R₂₃ − 0.08 × O₃₂)−12 + log O/H relation with 66 selected extremely low-metallicity SDSS SFGs (open circles and open triangles) and some well-studied galaxies from the literature (filled symbols) is shown in Fig. 4. Some well-known, extremely low-metallicity SFGs are labelled in the figure. The oxygen abundances for all galaxies shown in the Figure are derived by 1) the direct method if the [O III]λ4363 Å emission line is detected (open circles) or by the 2) semi-empirical method otherwise (open triangles) while the relation for the strong-line method determined by Eq. (1) is shown by the solid line. Thirty six SDSS galaxies shown in Table A.2 were previously recovered by our team (e.g. Guseva et al. 2015, 2017; Izotov & Thuan 2007; Izotov et al. 2012, 2018a). Three objects, I Zw 18SE, I Zw 18NW, and J0811+4730, are presented in both the SDSS and comparison samples and connected by solid red lines.

Fig. 4. Log(R23 − 0.08O32)−(12 + log O/H) relation for SFGs with 12 + log O/H ≤ 7.4 from Table A.2. The SFGs from the SDSS with 12 + log O/H derived by our newly introduced direct method and by the semi-empirical method are shown by open circles and open triangles, respectively. Other symbols and the solid line are the same as in Fig. 1b. Solid lines connect different observations of the same object. Some extremely low-metallicity SFGs from the literature are labelled.

Open with DEXTER

It is seen in Fig. 4 that selected SDSS galaxies (open circles and open triangles) are evenly distributed around the solid line implying that our newly proposed strong-line method (Eq. (1)) reasonably well reproduces galaxy metallicities. A considerable number of SFGs have 12 + log O/H lower than in the prototypical galaxy I Zw 18. Two of these galaxies, J0811+4730 (see also Sect. 3) and J1234+3901 are very close to or below the luminosity-weighted value 12 + log O/H = 7.0, representing the lowest-metallicity galaxies known so far. In fact, using Large Binocular Telescope observations that have high signal-to-noise ratio Izotov et al. (2018a) have confirmed that J0811+4730 has the lowest luminosity-weighted metallicity among all low-redshift SFGs with z ≲ 0.1 (Izotov et al. 2018a). We note that metallicities in regions 3 and 4 of the galaxy SBS 0335−052W are even lower (Izotov et al. 2009). However, the [O III]λ4363 Å emission line is not detected in spectra of these regions and their oxygen abundances are derived by the semi-empirical method. Furthermore, two other brighter regions 2 and 1 of SBS 0335−052W have higher oxygen abundances (Fig. 4) indicating an abundance gradient in this galaxy. This gradient results in a higher luminosity-weighted oxygen abundance of 7.12 averaged over the entire galaxy SBS 0335−052W. Annibali (2019) reported the very low oxygen abundance 12 + log O/H = 6.96 ± 0.09 in the H II region #7 of DDO 68 derived by the direct method. However, the oxygen abundance in this region is derived with low accuracy and needs to be confirmed with higher precision. Furthermore, oxygen abundances in other H II regions of DDO 68 are considerably higher than in the region #7 (Pustilnik et al. 2005; Izotov & Thuan 2007, 2009). We note that spectra of all these regions are not present in the SDSS data base.

One of the lowest-metallicity SFGs among selected SDSS galaxies is J1234+3901. A single high-excitation H II region is observed in this galaxy with an oxygen abundance of 6.98 and 7.03 derived by the strong and direct methods, respectively. The SDSS spectra of this galaxy and another lowest-metallicity galaxy J0811+4730 from Table A.2 are shown in Fig. 5. The [O III]λ4363 Å emission line is clearly detected in both spectra allowing the oxygen abundance determination by the direct method. The galaxy J1234+3901 at the redshift of 0.1330 is one of the most distant and luminous galaxies in Table A.2. Additional observations with higher signal-to-noise ratio are needed to confirm the extremely low metallicity of this galaxy.

	Fig. 5. SDSS spectra of the two lowest-metallicity galaxies represented in Table A.2. Insets in both panels show expanded segments of the spectra for a better view of the [O III] λ4363 Å emission lines.
Open with DEXTER

5. Diagnostic diagrams

Izotov et al. (2012, 2018a) and Guseva et al. (2017) proposed to use the Baldwin, Phillips and Terlevich (BPT; Baldwin et al. 1981) diagnostic diagram [O III]λ5007/Hβ vs. [N II]λ6584/Hα for the selection of the extremely low-metallicity SFGs. They have shown that these galaxies are located in the lower-left corner of the diagram far from the main sequence of SFGs in the region defined by relations [O III]λ5007/Hβ <  4 and [N II]λ6584/Hα <  0.03 corresponding to [N II]λ6584/Hβ <  0.1. However, the use of this diagram is limited because of the weakness of the [N II]λ6584 Å emission line. In many galaxies this line is not detected. Alternatively, the diagram [O III]λ5007/Hβ vs. [O II]λ3727/Hβ can be used as the [O II]λ3727 Å emission line in galaxies with low metallicities is approximately 20 times brighter than the [N II]λ6584 Å emission line.

We note that the [O II]λ3727/Hβ ratio is sensitive to dust extinction. However, in low-metallicity galaxies the internal extinction is low. It is derived from the Balmer hydrogen decrement after correction of emission line intensities for the Milky Way extinction. Both extinctions are applied to obtain extinction-corrected line intensities in the [O III]λ5007/Hβ–[O II]λ3727/Hβ diagram. We show this diagram in Fig. 6a. The shaded region indicates the location of lowest-metallicity galaxies with 12 + log O/H ≤ 7.35. Dashed lines indicate the lines of equal R₂₃ − 0.08O₃₂ corresponding to equal metallicities according to Eq. (1). The SFGs selected in this paper and shown in Table A.2 are represented in the Figure by open circles and open triangles and are located in the shaded region. The range of [O II] λ3727/Hβ <  2 is somewhat larger than ≲1 tested by galaxies with 12 + log O/H derived by the direct method (filled symbols in Fig. 6a). Therefore, metallicities in galaxies with [O II]λ3727/Hβ = 1 − 2 need to be confirmed with the direct method.

Fig. 6. Panel a: diagnostic diagram [O III]λ5007/Hβ–[O II]λ3727/Hβ. Open circles and open triangles are the extremely low-metallicity SFG candidates selected in this paper with oxygen abundances derived by the direct and semi-empirical methods, respectively. Other symbols are the same as in Fig. 1. The shaded region indicates the location of the extremely low-metallicity SFGs with 12 + log O/H in the range 6.62–7.39. Dotted and dashed lines are the lines of constant O32 and of constant R23 − 0.08O32 (or constant 12 + log O/H), respectively, labelled with their values. The values R23 − 0.08O32 = 0.75, 1.5, 2.5, and 3.75 correspond to oxygen abundances 12 + log O/H = 6.69, 6.97, 7.18, and 7.35, respectively. Panel b: diagram O32–R23. Symbols, lines, labels and shaded region follow the definitions provided for panel a.

Open with DEXTER

Additionally, for selection of extremely low-metallicity galaxies, the O₃₂–R₂₃ diagram can be used, as proposed by Izotov et al. (2018a). This diagram for selected SDSS galaxies (open circles and open triangles) and galaxies from the comparison sample (filled symbols) is shown in Fig. 6b. Similar to the diagram in Fig. 6a these galaxies are located far from the main-sequence SFGs (grey dots). We note a wide range of O₃₂ with the highest values reaching values up to ∼15. However, most of the extremely low-metallicity SFGs have O₃₂ <  3. The shaded region corresponds to that in Fig. 6a. It indicates the location of extremely low-metallicity SFGs with 12 + log O/H ≤ 7.35.

6. Conclusions

We present results of a search for the extremely low-metallicity star-forming galaxies (SFGs) from the Data Release 14 of the Sloan Digital Sky Survey (SDSS DR14). Our main results are as follows.

1. A new, simple, purely empirical strong-line method based on the [O II]λ3727 Å and [O III]λ4959 Å, λ5007 Å emission lines is developed for the oxygen abundance determination at low metallicities and to search for such extremely low-metallicity galaxies. This method is a modification of the well-known R₂₃ method and takes into account the dependence of oxygen emission-line fluxes not only on the metallicity but also on the ionization parameter in the H II region. We adopt that the O₃₂ = [O III]λ5007/[O II]λ3727 emission-line ratio serves as a measure of the ionization parameter. This method was calibrated using observations with high signal-to-noise ratios (S/N) of a large sample of SDSS galaxies with detected [O III]λ4363 Å emission applying the direct method in combination with other samples of SFGs with reliably derived oxygen abundances via the direct method.

2. Several selection criteria were applied to build a sample of the extremely low-metallicity SFGs. First, the SDSS galaxies were preselected using a criterion [N II]λ6584/Hβ ≤ 0.2. This allows us to exclude high-metallicity objects located on the upper branch of the R₂₃ − 12 + log O/H diagram. Then, all SFGs with [O III]λ4959/Hβ ≤ 0.2 were excluded because the [O III]λ4363 Å emission line was not detected in these SFGs and thus their abundances derived by the strong method were not tested by the direct method. Finally, the remaining SFGs satisfying condition R₂₃ − 0.08O₃₂ ≲ 4, corresponding to 12 + log O/H <  7.4, were included in the final list of candidates to the lowest-metallicity SFGs.

3. We selected 66 emission-line galaxies with 12 + log O/H ≤ 7.35 in the entire SDSS DR14 using a new strong-line method. Some selected galaxies have 12 + log O/H as low as 7.0 and therefore belong to the lowest-metallicity SFGs known so far. Results of the metallicity determination by the new strong-line method are compared with those derived by a semi-empirical method (Izotov & Thuan 2007), by a direct method, and by some strong-line methods from the literature. We find a good agreement between 12 + log O/H derived by all three first methods with deviations less than 0.1 dex, on average. The agreement with some methods selected from the literature is not as good. However, the SDSS spectra of many selected SFGs have relatively low S/N. Furthermore, in 61 percent of SFGs the [O II]λ3727 Å emission line is outside the wavelength range of SDSS spectra because of their low redshifts z ≲ 0.02. Therefore, its fluxes were estimated from the fluxes of weak [O II]λ7320 Å, 7330 Å or [N II]λ6584 Å emission lines. We conclude that new observations with higher S/N and including [O II]λ3727 Å emission line are needed to confirm the low metallicity of these selected galaxies.

4. We propose the use of two diagnostic diagrams, [O III]λ5007/Hβ–[O II]λ3727/Hβ and O₃₂–(R₂₃ − 0.08O₃₂) for the selection of the extremely low-metallicity SFGs.

Acknowledgments

Y.I.I. and N.G.G. thank the hospitality of the Max-Planck Institute for Radioastronomy, Bonn, Germany. They acknowledge support from the Program of Fundamental Research of the Department of Physics and Astronomy (Project No. 0117U000240) of the National Academy of Sciences of Ukraine. Funding for the Sloan Digital Sky Survey (SDSS) has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Aeronautics and Space Administration, the National Science Foundation, the U.S. Department of Energy, the Japanese Monbukagakusho, and the Max Planck Society. The SDSS Web site is http://www.sdss.org/. The SDSS is managed by the Astrophysical Research Consortium (ARC) for the Participating Institutions. The Participating Institutions are The University of Chicago, Fermilab, the Institute for Advanced Study, the Japan Participation Group, The Johns Hopkins University, Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, University of Pittsburgh, Princeton University, the United States Naval Observatory, and the University of Washington.

References

Appendix A: Selected extremely low-metallicity galaxies

All Tables

All Figures

	Fig. 1. Panel a: (log R23)−(12 + log O/H), and panel b: log (R23 − 0.08O32)−(12 + log O/H) relations. In both these relations 12 + log O/H is derived by the direct method, R23 = ([O III]λ4959+λ5007+[O II]λ3727)/Hβ and O32 = [O III]λ5007/[O II]λ3727. The galaxy J0811+4730 with the lowest luminosity-weighted oxygen abundance (Izotov et al. 2018a) is shown by a filled star, while two knots in SBS 0335−052W with similar or lower oxygen abundances (Izotov et al. 2009) are represented by encircled filled circles. Other lowest-metallicity SFGs from Izotov et al. (2018a), the galaxy Little Cub (Hsyu et al. 2017) and the H II region DDO 68 #7 (Annibali 2019) are shown by filled circles. SFGs with the highest O32 ratios in the range ∼20–40 (Izotov et al. 2017) are shown by asterisks. Samples of SFGs used for the He abundance determination (Izotov et al. 2014a, and references therein) and Lyman continuum leakers (Izotov 2018b, and references therein) are shown by open circles and crosses, respectively. Grey dots represent SDSS SFGs at z >  0.02 from DR9 and earlier releases and at any redshift from DR10 and later releases, with the [O III]λ4363 Å emission line measured with an accuracy better than 25%. The dashed, dotted, solid, and dash-dotted lines in panel a are the empirical relations by Skillman (1989), Nagao et al. (2006), Yin et al. (2007) and Pilyugin (2000), respectively. The solid line in panel b is the linear most likelihood fit to the SDSS data with log (R23 − 0.08O32)≤0.8 (this paper).
Open with DEXTER
In the text

	Fig. 2. Panel a: log (R23 − 0.08O32)−Te(O II) relation for SFGs. Panel b: relation between the extinction-corrected [O II]λ3727/Hβ and [N II]λ6584/Hβ flux ratios. The solid line corresponds to [O II]λ3727 = 20 × [N II]λ6584. Symbols in both panels are as in Fig. 1.
Open with DEXTER
In the text

	Fig. 3. Panel a: distribution of differences between oxygen abundances 12 + log O/H derived by the semi-empirical and the direct methods for the sample from Table A.2 excluding galaxies with non-detected [O III]λ4363 Å emission lines. Panel b: as in panel a but for differences between 12 + log O/H derived by our new strong-line and the direct methods. Panel c: as in panel a but for differences between 12 + log O/H derived by the strong-line and the semi-empirical methods for all SFGs from Table A.2. Panel d: as in panel c but for differences between 12 + log O/H derived by the strong-line method and the strong-line method by Yin et al. (2007). Panel e: as in panel c but for differences between 12 + log O/H derived by the strong-line method and the P-method (Eq. (23) in Pilyugin & Thuan 2005). Panel f: as in panel c but for differences between 12 + log O/H derived by the strong-line method and the P-method (Eq. (24) in Pilyugin & Thuan 2005). In all panels, galaxies with observed and non-observed [O II]λ3727 emission line are shown by filled and open circles, respectively. Solid horizontal lines indicate average abundance differences and dotted horizontal lines indicate ±1σ dispersions of the sample around the average values.
Open with DEXTER
In the text

	Fig. 4. Log(R23 − 0.08O32)−(12 + log O/H) relation for SFGs with 12 + log O/H ≤ 7.4 from Table A.2. The SFGs from the SDSS with 12 + log O/H derived by our newly introduced direct method and by the semi-empirical method are shown by open circles and open triangles, respectively. Other symbols and the solid line are the same as in Fig. 1b. Solid lines connect different observations of the same object. Some extremely low-metallicity SFGs from the literature are labelled.
Open with DEXTER
In the text

	Fig. 5. SDSS spectra of the two lowest-metallicity galaxies represented in Table A.2. Insets in both panels show expanded segments of the spectra for a better view of the [O III] λ4363 Å emission lines.
Open with DEXTER
In the text

	Fig. 6. Panel a: diagnostic diagram [O III]λ5007/Hβ–[O II]λ3727/Hβ. Open circles and open triangles are the extremely low-metallicity SFG candidates selected in this paper with oxygen abundances derived by the direct and semi-empirical methods, respectively. Other symbols are the same as in Fig. 1. The shaded region indicates the location of the extremely low-metallicity SFGs with 12 + log O/H in the range 6.62–7.39. Dotted and dashed lines are the lines of constant O32 and of constant R23 − 0.08O32 (or constant 12 + log O/H), respectively, labelled with their values. The values R23 − 0.08O32 = 0.75, 1.5, 2.5, and 3.75 correspond to oxygen abundances 12 + log O/H = 6.69, 6.97, 7.18, and 7.35, respectively. Panel b: diagram O32–R23. Symbols, lines, labels and shaded region follow the definitions provided for panel a.
Open with DEXTER
In the text