Blueberry Galaxies: The Lowest Mass Young Starbursts

One frontier in observational cosmology is understanding the reionization of neutral hydrogen in the universe. How did reionization occur and which sources caused it? Many simulations and observations suggest the very faint dwarf starburst galaxies (stellar mass $\sim {10}^{6\mbox{--}8}\,{M}_{\odot }$) contribute a significant fraction of ionizing photons (Lyman continuum, LyC) during reionization (e.g., Trenti et al. 2010; Salvaterra et al. 2011; Bouwens et al. 2015; Dressler et al. 2015). Thus, it is important to study these faint dwarf starbursts and understand their physical properties and formation mechanism.

High redshift Lyα emitters (LAEs) are an important population of low-mass star-forming galaxies at $z\gt 2$, increasing in fraction to constitute 60% of Lyman break galaxies at redshifts $z\gt 6$ (Stark et al. 2011). A large fraction of the dwarf starburst galaxies during reionization may be intrinsic LAEs, but their Lyα photons can be scattered by the HI in IGM, which makes Lyα line a powerful probe of reionization (e.g., Malhotra & Rhoads 2004; Hu et al. 2010; Ouchi et al. 2010; Jensen et al. 2013; Tilvi et al. 2014; Matthee et al. 2015; Zheng et al. 2017). These high-z LAEs have low metallicity, low stellar masses, low dust extinction, and compact sizes (Gawiser et al. 2007; Pirzkal et al. 2007; Finkelstein et al. 2008; Pentericci et al. 2009; Malhotra et al. 2012).

Studying LAEs and faint dwarf starbursts in the high-z universe is very challenging, because it usually requires long exposure times and difficult near-infrared observations (e.g., McLinden et al. 2011; Trainor et al. 2015). A complementary approach is to study the physical processes in low-z analogs of these high-z LAEs and faint dwarf starbursts.

The current best nearby analogs of high-z LAEs are green pea galaxies (e.g., Jaskot & Oey 2014; Henry et al. 2015; Yang et al. 2016, 2017; Verhamme et al. 2017). Green pea galaxies were discovered in the citizen science project Galaxy Zoo (Cardamone et al. 2009). They are compact galaxies that are unresolved in SDSS images. The green color is because the [O iii] doublet (EW(O[III]5007) ∼ 300–2500 Å) dominates the flux of the SDSS r-band, which is mapped to the green channel in the SDSS's false-color gri-band images. They are similar to high-z LAEs in many galactic properties—small sizes, low stellar masses (${10}^{8\mbox{--}10}\,{M}_{\odot }$), low metallicities for their stellar masses, high specific star formation rates (sSFR), and large [O iii]λ5007/[O ii]λ3727 ratios (Cardamone et al. 2009; Amorín et al. 2010; Izotov et al. 2011). Five green peas are observed and confirmed as LyC leakers (Izotov et al. 2016).

Green peas are relatively luminous and massive galaxies compared to the faint-end dwarf starbursts and LAEs. In the present study, we searched for closer and lower mass analogs of the faint-end LAEs in SDSS ugriz broadband images, and studied their properties with MMT and SDSS spectra.

The strong [O iii] emission line makes green peas show distinctive colors. Cardamone et al. (2009) selected green peas at $0.14\lt z\lt 0.36$ in SDSS. To select the lowest mass green peas in SDSS images, we focus on the lowest redshift green pea sample. While the SDSS images are magnitudes limited, these lower redshift green peas generally tend to have lower luminosities and lower stellar mass. In this paper, we select green peas at $z\lesssim 0.05$.

We use simple color criteria to select photometric candidates of $z\lesssim 0.05$ green peas. To obtain the color criteria, we made model spectra of green peas and generated corresponding redshift tracks in color–color space. The model spectra include both continuum and strong emission lines, with a range of line strengths. (1) The continuum component includes an age = 4 Myr young starburst and an age = 900 Myr old population model spectra from Starburst99 (Leitherer et al. 1999). The old to young mass ratio is fixed at 10. (2) The strong emission lines include [O ii]3727, Hβ, Hα, [O iii]4959, and [O iii]5007 at three different equivalent widths (EWs). The EW([O iii]5007) are 300, 800, and 1600 Å. At each EW([O iii]5007), the strength of the other lines ([O ii], Hβ, Hα, and [O iii]4959) are estimated using typical line ratios of green peas at the corresponding [O iii]5007 equivalent width (Yang et al. 2017). Then we multiply the model spectra with the SDSS throughput curves for ugriz filters and calculate the colors at different redshifts. When the [O iii]λ5007 and Hα lines move close to the wavelength edges of filters, the colors change dramatically with redshift.

According to the tracks in g − r versus r − i color–color diagram (Figure 1), green peas at $z\lt 0.08$ and $0.14\lt z\,\lt 0.36$ have very different colors from stars, normal galaxies, and quasars. To select a clean sample of green peas at $z\lesssim 0.05$, we use the following color selection criteria:

$$\begin{eqnarray*}&&g-r\lt -0.5\,\mathrm{and}\,r-i\lt 1.0\,\mathrm{and}\,g-i\lt -0.5\end{eqnarray*}$$

and

$$\begin{eqnarray*}&&(g-r\lt -0.7\,\mathrm{or}\,g-i\lt -1.0)\,\mathrm{and}\,g-u\lt -0.3.\end{eqnarray*}$$

The g − u selection is added to exclude some very blue young stars or white dwarfs that fall in the selection region. Those stars have very blue u − g colors in comparison to the red u − g colors of green peas with strong [O iii]λ5007 in g-band. As shown in Figure 1, these color criteria select green peas with EW([O iii]λ5007) ≳ 800 Å at $z\lesssim 0.05$. The selection was conducted with the SDSS DR12 photometric catalog. To exclude imaging artifacts from the selected catalog, we add some photometric flags to the selection. The exact selection criteria can be found in the Appendix. Especially because green peas are compact sources, we use the flag "NOT CHILD" to select small sources that are not deblended as a part of an underlying diffuse galaxy.

Zoom In Zoom Out Reset image size

Among all sources in the SDSS DR12 photometric catalog, only 51 sources satisfy the selection criteria. Figure 2 shows their SDSS gri images (Blue-Green-Red jpg files generated by SDSS, Lupton et al. 2004). Most of them are unresolved or marginally resolved in SDSS images (seeing ∼13), although a few sources have some weak diffuse emission around the central bright knots. Because their blue/purple colors, we call them "blueberry" galaxies.

Zoom In Zoom Out Reset image size

The Pan-STARRS1 (PS1) survey (Chambers et al. 2016) has deep z- and y-band images, so we also get the PS1 grizy photometric data for this sample to aid in spectral energy distribution (SED) fitting. The photometric data are presented in Table 1 as well as the attached machine-readable table.

We obtained spectra of this sample from SDSS, MMT, and LBT. Out of the total sample of 51 sources, 10 sources already have SDSS spectra. These 10 sources (the first 10 objects in Figure 2, with objID $\lt 15$) all show strong [O iii] emission lines and are at the bright end of this sample. One object was observed using LBT/MODS (5-minute exposure) in the morning of 2016 May 07, after twilight started. The spectra shows strong emission lines. Another 36 objects were observed with MMT/BlueChannel on 2017 Jan 07. The remaining four objects do not have optical spectra. In all spectra, the wavelength range of [O ii]3727 to Hα was covered. Out of the 47 objects with spectra, 40 objects show extreme emission lines. The other seven objects are stars, quasars, or unknown sources (with low S/N spectra).

The MMT observations were made using the medium-resolution grating (500GPM) and 125 × 180'' slit, giving a spatial scale along the slit of 06 per pixel, a spectral range of 3700–6900 Å, and a spectral resolution of ∼3.8 Å (FWHM). The seeing during the MMT observations was ∼07–10. The slit orientations were along the parallactic angles for all objects. The exposure time is about 10–15 minutes for each object.

The MMT/BlueChannel and LBT/MODS long-slit spectra were reduced using IRAF following the standard steps. The raw data were bias subtracted and flat-fielded. The sky background at the upper and lower regions of the object were subtracted. The continuum was detected in most objects and was used for determining the spectral trace. The traces of two faint objects were determined using other spectra as references. The 1D spectra were extracted using optimal extraction. The spectra of HeNeAr lamps were used for wavelength calibration. The standard star LB227 was observed with 5'' slit width and was used for the flux calibration. To get better absolute spectral fluxes, we calculated a g-band magnitude by multiplying the spectra with the g-band throughput curve and then scaled each spectrum to make the spectroscopic g-magnitude equal the SDSS g-band magnitude.

The optical spectra (Figure 3) of blueberry galaxies show very strong [O iii]5007 emission lines and large [O iii]5007/Hβ and [O iii]/[O ii] ratios. Figure 4 shows the spectra in 3600–4800 Å. The [O iii]4363 line is very strong and well detected. The [O ii]3727 line is relatively weak.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

From the optical spectra, we measured the properties of emission lines. First, we estimate the continuum by fitting a cubic polynomial function to wavelength regions without emission lines and subtract the continuum from the spectra. Then we fit the emission lines with Gaussian functions and get the line fluxes. We correct the measured line fluxes for Milky Way extinction using the attenuation of Schlafly & Finkbeiner (2011; obtained from the NASA/IPAC Galactic Dust Reddening and Extinction tool) and the Fitzpatrick (1999) extinction law. Notice that the Hα/Hβ ratios for most of the MMT spectra are smaller than the case-B value 2.86, because the flat field calibration in the red end of spectra (6000–6900 Å) is very poor and the Hα fluxes are probably underestimated. Therefore, we did not correct the line fluxes for extinction by dust within the blueberry galaxies. The properties of emission lines are shown in Table 2 and additional columns are included in the attached machine-readable table.

From the optical spectra, we measured some physical properties of blueberry galaxies. These properties are shown in Figures 5–8.

Zoom In Zoom Out Reset image size

Stellar mass: the stellar mass is estimated by fitting Starburst99 models (Leitherer et al. 1999) to the SDSS ugri and PS1 zy-bands photometric data. The strong emission lines are subtracted from the SDSS gri-band photometry. We model the star formation history with two starburst components—one young component with age <20 Myr and one old component with age <14 Gyr. From the Starburst99 single stellar population model (generated with Kroupa initial mass function and Geneva 2012/2013 tracks with zero rotation at metallicity Z = 0.002), we make 12600 composite spectra at a grid of three different parameters—the age of the young component, the age of the old component, and the mass ratio of old to young components. Then we fit each composite spectrum to the SED and compute the likelihood of each spectrum as $\exp (-{\chi }^{2}/2)$. From the likelihood distribution, we estimate the median value (50% probability on each side) as the final stellar masses and ages. The majority of the blueberry galaxies have stellar mass about ${10}^{6.5}\,{M}_{\odot }\mbox{--}{10}^{7.5}\,{M}_{\odot }$.

Note that a faint diffuse stellar disk may exist around the blueberry galaxies. The SDSS r-band images have a surface brightness limit of ∼26 mag arcsec⁻², which corresponds to $\sim {10}^{6}\,{L}_{\odot }\,{\mathrm{kpc}}^{-2}$. Assuming a mass to light ratio ${M}_{r}/{L}_{r}=1$ for low-mass galaxies (Kauffmann et al. 2003), the diffuse stellar disk has a surface limit of $\sim {10}^{6}\,{M}_{\odot }\,{\mathrm{kpc}}^{-2}$.

Star formation rate: assuming the Case-B condition, we convert the Hβ luminosity to SFR using the formula $\mathrm{SFR}({M}_{\odot }\,{\mathrm{yr}}^{-1})={L}_{{\rm{H}}\beta }(\mathrm{erg}\,{{\rm{s}}}^{-1})\times 2.86\times {10}^{-41.27}$ (Kennicutt & Evans 2012). (We estimate star formation using Hβ rather than Hα because our spectra are better calibrated at Hβ.) The resulting star formation rates are about $\mathrm{SFR}\,\approx 0.05\mbox{--}2\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ and sSFR are about ${10}^{-8}\,{\mathrm{yr}}^{-1}$. Figure 5 shows the mass−SFR relation of blueberry galaxies.

Gas metallicity: we calculate the metallicity using [O iii]λ4363, [O iii]λ5007, and [O ii]λ3727 line fluxes following the T_e method described in Izotov et al. (2006) and Ly et al. (2014). Blueberry galaxies show very low metallicities ($7.1\lt 12\,+\mathrm{log}({\rm{O}}/{\rm{H}})\lt 8.0$). A few can be classified as extremely metal-poor galaxies (XMP). Additionally, two blueberry galaxies have $12+\mathrm{log}({\rm{O}}/{\rm{H}})\lt 7.2$ and are among the most metal-poor galaxies ever found (e.g., Guseva et al. 2017). Thus, searching for blueberry galaxies may be a good method to find XMP with the lowest gas metallicity. Figure 6 shows the mass–metallicity relation of blueberry galaxies.

Zoom In Zoom Out Reset image size

Ionization: we also calculate the [O iii]/[O ii] ratios (([O iii]4959 + [OIII5007)/[O ii]3727) (Figure 7). For two galaxies where the [O ii] line is detected at ≲2σ significance, we use the 3σ upper-limit of [O ii] line to calculate the [O iii]/[O ii] ratio. The [O iii]/[O ii] ratios for the full sample range from ∼8 to ≳60.

Zoom In Zoom Out Reset image size

Environment: to characterize the environments of blueberry galaxies, we find the nearest galaxy neighbor in the SDSS DR12 spectroscopic catalog for each one. The distances to the nearest neighbor are between 300 kpc and >10 Mpc (Figure 8). We also calculate the nearest neighbors for a comparison sample of 5000 galaxies randomly selected at similar redshift ($0.03\lt z\lt 0.05$) in the SDSS DR12 spectroscopic catalog. Most galaxies in the comparison sample have neighbors within 2Mpc. Blueberry galaxies and the comparison sample have different distributions of distances to nearest neighbor (with K-S test p-value = 10⁻³⁹). So blueberry galaxies are generally in low density environments and some are in the outskirts of galaxy groups.

Zoom In Zoom Out Reset image size

Number density: among all sources in the SDSS DR12 photometric catalog, only 51 sources satisfy the selection criteria (40 blueberries, 7 contaminants, 4 no spectra). Assuming 3 objects without spectra are blueberries, we get 43 objects in a sky area of 14555 deg² at redshift <0.05. Thus, the number density of blueberry galaxies is 43/(14355/41259 × 0.04 Gpc³) = 3.0 × 10⁻⁶ Mpc⁻³.

In summary, these blueberry galaxies are currently undergoing a young starburst. They have very small sizes ($\lt 1\,\mathrm{kpc}$), very low stellar masses and metallicities, and very high ionization.

Compared to the many different classes of star-forming "dwarf" galaxies, are blueberry galaxies a new class? The star-forming dwarf galaxies include the local volume dwarf galaxies (e.g., Lee et al. 2009), the blue compact dwarf galaxies (BCDs, e.g., Zwicky & Zwicky 1971; Thuan & Martin 1981; Gil de Paz et al. 2003), ultracompact blue dwarf galaxies (e.g., Corbin et al. 2006), H ii galaxies (e.g., Terlevich & Melnick 1981; Melnick et al. 2017), extreme metal-poor galaxies (e.g., Guseva et al. 2017; Sánchez Almeida et al. 2016), HI selected dwarfs (e.g., Huang et al. 2012), emission line dots (e.g., Werk et al. 2010; Kellar et al. 2012), and blue diffuse galaxies (James et al. 2017). Compared to those star-forming dwarf galaxies, blueberry galaxies have similar stellar mass and luminosity but much stronger [O iii] line strength and gas ionization. Because blueberry galaxies are selected by the strong [O iii] emission lines, they represent the star-forming dwarf galaxies with the highest emission line strength and gas ionization.

On the other hand, compared to green pea galaxies at $z\sim 0.2\mbox{--}0.3$ and typical high-z LAEs found in the current narrow-band surveys, blueberry galaxies have similarly strong emission lines but about 10–100 times smaller stellar mass, SFR, and luminosity. So blueberry galaxies represent the faint-end of green peas and LAEs.

We searched for blueberry galaxies in SDSS broadband images and studied their properties with MMT and SDSS spectra. Our main results are as follows.

(1) Using our color selection criteria, we find 51 photometric candidates at $z\lesssim 0.05$ from the whole SDSS DR12 photometric catalog. Optical spectra confirm that 40 of these sources show strong emission lines and can be classified as blueberry galaxies. The remaining 11 candidates consist of 7 contaminants and 4 without spectra. The number density of blueberry galaxies is about 3.0 × 10⁻⁶ Mpc⁻³.

(2) These blueberries are dwarf starburst galaxies with very small sizes ($\lt 1\,\mathrm{kpc}$), low stellar masses ($\mathrm{log}(M/{M}_{\odot })\,\sim 6.5\mbox{--}7.5$), small SFR ($0.05\mbox{--}2\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$), high sSFR ($3\,\mbox{--}100\,{\mathrm{Gyr}}^{-1}$), very low gas metallicities ($7.1\lt 12+\mathrm{log}({\rm{O}}/{\rm{H}})\lt 8.0$) estimated with the [O iii]4363 lines, and very high gas ionization ([O iii]/[O ii]] ∼ 10–60). Two blueberry galaxies have $12+\mathrm{log}({\rm{O}}/{\rm{H}})\lt 7.2$ and are among the most metal-poor galaxies known. Their nearest-neighbor distances in the SDSS spectroscopic catalog are between 300 kpc and 10 Mpc, larger than the galaxy population in general. Thus, they are in low density environments. These blueberry galaxies represent the faint-end sample of green peas and LAEs.

We thank Chun Ly, Lin Lin, Chenwei Yang, and Tianxing Jiang for help with the data reduction and observations. Observations reported here were obtained at the MMT Observatory, a joint facility of the University of Arizona and the Smithsonian Institution. H.Y. and J.X.W. are thankful for support from NSFC 11233002, 11421303, and CAS Frontier Science Key Research Program (QYZDJ-SSW-SLH006). This work has also been supported in part by NSF grant AST-1518057 and by support for HST program #14201, which was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555.

We use the following criteria to select the photometric candidates in the SDSS DR12 CasJobs tool. To make the selection robust, we used both the cModelMag and psfMag. We also require the Galactic latitude $\gt 20$ or $\lt -20$ degree. The "calibStatus" flag is the status of photometric calibration.

Algorithm 1.