RESOLVE and ECO: Finding Low-metallicity z ∼ 0 Dwarf AGN Candidates Using Optimized Emission-line Diagnostics

The RESOLVE survey (Kannappan & Wei 2008) is volume-limited in two equatorial footprints covering >50,000 Mpc³ within a redshift range of 0.015 < z < 0.023 (4500 < cz < 7000 km s⁻¹). RESOLVE-B covers a volume of ∼13,700 Mpc³ and is highly complete down to M_r = −17.0, as it has added redshift coverage from SDSS Stripe 82 and other sources (Eckert et al. 2015, henceforth E15).

The ECO catalog (Moffett et al. 2015, henceforth M15) is an archival, volume-limited data set designed to complement RESOLVE by performing a similar census in a volume that is an order of magnitude larger (>400, 000 Mpc³) in the northern spring sky, overlapping RESOLVE-A and spanning a redshift range of 0.01 < z < 0.023 (3000 < cz < 7000 kms⁻¹).

We define a baryonic mass-limited sample with M_bary > 10^9.2 M_⊙ for both surveys. There are 1202 RESOLVE galaxies and 7767 ECO galaxies that satisfy this criterion. RESOLVE-B is a highly complete sample due to its overlap with Stripe 82 with added redshift coverage. ECO, however, is less complete than RESOLVE-B since it does not have additional redshift coverage, and its redshift limits exclude some galaxies in massive groups/clusters with large peculiar velocities (Eckert et al. 2016, henceforth E16). We use the highly complete RESOLVE-B sample to calculate the baryonic mass incompleteness for ECO in the same manner as E16, but with a baryonic mass floor of 10^9.2 M_⊙. The incompleteness for ECO is calculated as the ratio of the number of RESOLVE-B galaxies above the ECO mass floor but below the ECO luminosity floor (M_r ∼ −17.33) to the total number of RESOLVE-B galaxies above the ECO mass floor. From this, we estimate the baryonic mass completeness for ECO to be ∼97%.

The data products of the RESOLVE and ECO surveys are mostly homogeneous with the biggest differences being the data quality and survey completeness. RESOLVE-B overlaps Stripe 82 and has deep optical data from SDSS and Medium Imaging Survey (MIS)-depth UV data from GALEX, whereas ECO (including RESOLVE-A) has only shallow optical SDSS data and mixed All Sky Imaging Survey (AIS)- and MIS-depth UV coverage, with MIS-depth for less than half of the survey footprint (including RESOLVE-A). Both RESOLVE and ECO have photometric magnitudes estimated by reprocessing existing photometry from the UV to the near-IR as described in E15 and M15. The improvements to the photometric reprocessing that are applied to both surveys are described in E15 and E16. Stellar masses and colors are estimated using a Bayesian spectral energy distribution (SED) fitting code (Kannappan et al. 2013; E15; E16).

For both surveys, baryonic mass, i.e., stellar mass + neutral atomic gas mass, is also estimated. For RESOLVE-B, the gas data from the 21 cm HI line come from the ALFALFA survey and from new observations with the Green Bank and Arecibo telescopes (Stark et al. 2016). Nearly all galaxies in the RESOLVE survey have high-quality HI data, i.e., either detections or strong upper limits of <5%–10% of the stellar mass. ECO galaxies within the region outside RESOLVE and overlapping the ALFALFA40 public catalog have flux-limited HI data (E16). For 21 cm observations that are missing, have weak upper limits, or cannot be deconfused, M_HI is estimated by the photometric gas fraction method (E15), which uses relationships between color, axial ratio, and gas-to-stellar-mass ratio. The total gas mass is estimated as 1.4M_HI to account for the Helium contribution.

To estimate the halo masses of galaxy groups, RESOLVE and ECO galaxies are associated to groups using the friends-of-friends group-finding algorithm described in M15 and E16. Once the galaxy groups are determined, halo masses are estimated by using halo abundance matching as detailed in Eckert et al. (2017).

To measure the star formation activity of our sample, we use a metric called fractional stellar mass growth rate (FSMGR; the ratio of newly formed stellar mass to preexisting stellar mass divided by the timescale separating new versus preexisting mass) instead of a specific star formation rate (sSFR), since the latter asymptotes at high SFRs (see Figure 19 in Kannappan et al. 2013). The long-term FSMGRs for both surveys are estimated from the same stellar population modeling code that we use to determine the stellar mass (see Kannappan et al. 2013 for more details), and is defined as:

$$\begin{eqnarray}&&{\mathrm{FSMGR}}_{\mathrm{LT}}=\displaystyle \frac{{\mathrm{mass}}_{\mathrm{formedinlastGyr}}}{1\ \mathrm{Gyr}\times {\mathrm{mass}}_{\mathrm{preexisting}}}.\end{eqnarray} \tag{ 1 }$$

To estimate star formation rates (SFRs) for RESOLVE and ECO galaxies, we use custom reprocessed WISE mid-IR and GALEX UV photometry (Paper II and E15, respectively). Extinction corrections for the UV photometry are estimated from SED fitting with optical (SDSS) + UV (GALEX) + near-IR (2MASS and UKIDSS) data (E15). As GALEX FUV imaging is incomplete for our surveys, our default UV-based SFRs are based on NUV calibrations from Wilkins et al. (2012). Our mid-IR SFRs are based on WISE calibrations from Jarrett et al. (2013). We use the prescription from Jarrett et al. (2013) to combine the nondusty (UV) and dusty (IR) SFRs to infer total SFR (adopting η = 0.17 and γ = 1; Buat et al. 2011). To compute SFR_IR we require galaxies to have S/N > 5 in the W1, W2, and W3 WISE bands, and we either omit W4 or use the data only if S/N > 5 in that band. For galaxies with inadequate WISE S/N, SFR_IR is not computed, and the total SFR is simply SFR_UV. As expected, we find that SFR_UV is typically dominant, except for very dusty (giant) galaxies where SFR_IR contributes significantly. For this work, we use these SFRs to compute a short-term FSMGR on a timescale of 100 Myr as:

$$\begin{eqnarray}&&{\mathrm{FSMGR}}_{\mathrm{ST}}=\displaystyle \frac{100\ \mathrm{Myr}\times \mathrm{SFR}}{[{{\rm{M}}}_{* }-(100\ \mathrm{Myr}\times \mathrm{SFR})]\times 0.1\,\mathrm{Gyr}}.\end{eqnarray} \tag{ 2 }$$

We have validated SFRs for RESOLVE and ECO by comparison to SFRs from Salim et al. (2016), who also performed SED fitting with GALEX and SDSS data (see Paper II for details). RESOLVE/ECO SFRs are on average ∼0.2–0.3 dex higher than those from Salim et al. (2016), consistent with expectations given the improved flux recovery by the reprocessed photometry for these surveys (Kannappan et al. 2013; E15). We tabulate the SFRs for all of RESOLVE and ECO in Paper II, which also examines mid-IR AGN detection.

Throughout this paper, we use the RESOLVE and ECO surveys in a nonduplicating way, i.e., without double-counting the galaxies common to both surveys.

We are currently in the process of extracting flux measurements from new RESOLVE 2D and 3D spectroscopy (in preparation). In the meantime, we have obtained results for this study using spatially unresolved flux measurements from the SDSS. We have crossmatched galaxies from the mass-limited RESOLVE and ECO surveys with three different SDSS-derived emission-line catalogs. The crossmatch statistics of the 1202 RESOLVE galaxies are: 1082 in the MPA-JHU catalog (Tremonti et al. 2004), 1082 in the Portsmouth catalog (Thomas et al. 2013), and 1075 in the NSA catalog (Blanton et al. 2011). The crossmatch statistics of the 7767 ECO galaxies are: 7221 in MPA-JHU catalog, 7221 in Portsmouth catalog, and 7316 in NSA catalog. The combined ECO and RESOLVE mass-limited data set (not double-counting overlap) has 8160 galaxies, and these galaxies have 7557/7557/7657 crossmatches in the MPA-JHU/Portsmouth/NSA catalogs.

We filter the SDSS data based on the presence of the following SELs: Hβ, [O iii] λ5007, [O i] λ6300, [N ii] λ6548, Hα, [N ii] λ6584, [S ii] λ6717, and [S ii] λ6731. To exclude spurious measurements, we require all SELs to have a "reliable" flag in the SDSS spectroscopic catalogs and to have positive finite fluxes and errors, both of which are less than 10⁵. The subsamples that pass these cuts in RESOLVE have 901/834/760 galaxies for the MPA-JHU/Portsmouth/NSA catalogs, and in ECO have 6005/5460/5205 galaxies for the MPA-JHU/Portsmouth/NSA catalogs. To reliably use the N ii, S ii, and O i plots, we also apply a requirement of S/N > 5 on all of the aforementioned SELs; the resulting sample is mainly limited by the S/N restriction for the weak [O i] line. These criteria reduce our mass-limited RESOLVE sample (of 1202 galaxies) to 382/202/209 SEL galaxies from the MPA-JHU/Portsmouth/NSA catalogs. Similarly, the mass-limited ECO sample (of 7767 galaxies) is reduced to 2507/1161/1363 SEL galaxies from the MPA-JHU/Portsmouth/NSA catalogs. The combined ECO and RESOLVE SEL samples, not double-counting the overlap, contain 2605/1207/1411 galaxies from the MPA-JHU/Portsmouth/NSA catalogs. Among these SEL galaxies, ∼60% are dwarfs in all catalogs. Figure 1 shows that our strict S/N cuts, especially on the weak [O i] line, greatly reduce our parent survey to ∼16%–35% of its original size. However, an S/N > 5 cut is critical to use the weak [O i] line effectively. Interestingly, the S/N cut on [O i] flux selects relatively more galaxies with a low [O i]/Hα ratio, which selects against AGN in the [O iii]/Hβ versus [O i]/Hα diagnostic plot, (contrary to naive intuition given the rising correlation between AGN luminosity and [O i] line flux taken alone). In Section 2.3, we will justify our choice to use the [O i] line, and in Section 6, we will show that despite the apparent selection bias against AGN implied by our S/N cuts, our new method still yields a higher AGN percentage for dwarfs than most other methods.

Zoom In Zoom Out Reset image size

Figure 2 shows the distributions of baryonic mass, halo mass, and (u-r) color for the mass-limited parent sample, the emission-line (EL) sample (only Hα S/N > 5), and the SEL sample in ECO. The EL and SEL samples have different baryonic masses, halo masses, and color distributions than each other and than the parent survey (at high confidence >4σ based on K-S tests for both samples). Nonetheless, these distributions span the full range of baryonic mass and halo mass, showing more bias in color. The differences we see are in line with expectations—requiring the presence of Hα emission biases the EL sample to have more dwarfs than giants and more blue galaxies than red galaxies, and further requiring the presence of all SELs biases the sample toward having more relatively high-mass dwarfs (M_bary ∼ 10^9.6−9.9 M_⊙) and bluer colors. Since dwarfs are generally gas-rich and actively star-forming, and giants are generally more gas-poor and not as actively star-forming, this bias is expected by design. The RESOLVE samples also follow the same trends. Nonetheless, since the EL and SEL samples span the entire range of baryonic and halo masses in the parent surveys, we expect our statistics to offer broad insight into the underlying z ∼ 0 galaxy population.

Zoom In Zoom Out Reset image size

Flux measurements in the MPA-JHU and NSA catalogs are corrected for Milky Way foreground extinction. Fluxes from the Portsmouth catalog need additional corrections. We follow the steps outlined in Thomas et al. (2013) and correct for (1) Milky Way foreground extinction, and (2) per-plate r-band flux rescaling, taking both correction factors from the MPA-JHU catalog.

None of the three SDSS catalogs account for galaxy internal extinction corrections, so we perform emission-line dereddening using Balmer decrements to estimate internal extinction. We determine each galaxy's color excess, E(B–V), from the ratio of the galaxy's foreground extinction-corrected Hα/Hβ flux ratio to the intrinsic Hα/Hβ ratio of 2.86 ⁹ (Domínguez et al. 2013). For galaxies with M_* < 10⁹ M_⊙, we use the SMC extinction curve given by Gordon et al. (2003) with a slight modification: since their polynomial does not fit the optical data well, we fit a line to their data for wavelengths greater than 3030 Å, and we redefine the extinction curve for optical wavelengths as:

$$\begin{eqnarray}&&\displaystyle \frac{{\rm{E}}({\rm{x}}-{\rm{V}})}{{\rm{E}}({\rm{B}}-{\rm{V}})}=1.91{{\rm{x}}}^{2}-0.80{\rm{x}}-2.75,\end{eqnarray} \tag{ 3 }$$

where x = 1/λ μ m. For galaxies with M_* > 10¹⁰ M_⊙, we use an unmodified O'Donnell (1994) Milky Way extinction curve. For galaxies with intermediate masses, we use a smoothly varying linear combination of the SMC and Milky Way extinction curves.

With these corrections, fluxes from the MPA-JHU and Portsmouth catalogs are comparable, even though the Portsmouth catalog uses only solar metallicity stellar population models to fit the continuum, while the MPA-JHU catalog includes low-metallicity models (see Figure 2 in Thomas et al. 2013). However, the flux ratios used in the N ii, S ii, and O i plots are on average ∼10% higher from the NSA catalog than from the MPA-JHU and Portsmouth catalogs. This difference mainly arises from an additional flux calibration in the NSA catalog to fix small-scale calibration residuals (Yan 2011). The NSA catalog seems to have superior flux calibration, but the MPA-JHU catalog has almost twice as many galaxies, and it uses low-metallicity continuum models better suited to dwarfs (see Section 4.3 for more details). Since all three catalogs have matches for roughly the same fraction of RESOLVE and ECO galaxies (∼90%), we believe that the differences in SEL subsample sizes are primarily driven by differences in spectral modeling methodologies that manifest as sample selection effects when S/N cuts are imposed (see Section 4.3). As we do not find a single catalog that is clearly better, we report statistics from all three catalogs. We use the MPA-JHU crossmatched sample for plots since it has the most RESOLVE/ECO galaxies of the three crossmatched samples.

The high degree of completeness and volume- and mass-limited survey definitions of RESOLVE and ECO allow us to normalize statistics for the SEL samples to the full parent survey, yielding "completeness-corrected statistics useful for comparisons to theory (e.g., Haidar et al. 2022). RESOLVE and ECO also allow us to study the properties of AGN hosts using extensive supporting data (especially gas content and environment), providing added value to SDSS emission-line fluxes.

To aid in interpretation of the optical emission-line diagnostic plots and estimation of gas-phase metallicities, we have computed new photoionization model grids with Cloudy (Ferland et al. 2017) as presented in Paper Ib (accepted). These models use the BPASS stellar population synthesis code (Stanway & Eldridge 2018) to account for binary stellar populations and to take advantage of BPASS's flexibility with respect to metallicity.

Our grids span metallicities (Z) of [0.05, 0.1, 015, 0.2, 0.3, 0.4, 0.5, 0.7, 1.0, 1.5, 2.0] Z_⊙, and ionization parameter log(U) values in the range [−4.00, −1.75] in increments of 0.25 dex. We selected the Z and U values based on the availability of models in BPASS and the range of values we can potentially find in local surveys like RESOLVE and ECO. Our models include AGN mixing fraction values of [0, 4, 8, 16, 32, 50, 64, 100] percent, assuming coincident mixing with an open geometry (see Paper Ib, accepted, for more details). An AGN fraction of 8% means that 8% of the incident radiation field is due to the AGN, and 92% is due to SF. To simulate IMBHs in z ∼ 0 dwarfs, we use the M_BH =10⁵ M_⊙ AGN SED given by the QSOSED model (Kubota & Done 2018). We choose this AGN model as it is more physically realistic than other popular AGN SEDs, especially for modeling IMBHs in dwarfs (see Paper 1b, accepted, for a detailed discussion; also see Panda et al. 2019).

We create two grids with different choices of star formation histories (SFHs)—one with a continuous star formation history (CSF) viewed at 250 Myr, and one with an instantaneous SFH or simple stellar population (SSP) viewed at 20 Myr. In Figure 3, we show the two most important emission-line diagnostic plots, the N ii and O i plots, with the full photoionization grids for CSF and SSP SFHs. Grid points above the Kewley et al. (2001, henceforth Ke01) demarcation line (solid black line) are traditionally classified as AGN. As will be discussed in Section 3.1, the green star (a fiducial model dwarf representative of the low metallicities of RESOLVE and ECO SEL galaxies) falls below this Ke01 line even for a 100% AGN model. In panel (a), we can see that our grids are offset ∼0.2–0.3 dex lower than the Ke01 demarcation line, which was derived from a PEGASE-based CSF model (details in Ke01). We have tested our grids with different settings and concluded that the BPASS and PEGASE SEDs are similar, with PEGASE producing a slightly harder continuum at energies >10.4 keV, which may result in a portion of the offset (Levesque et al. 2010). We attribute the rest of the offset to differences in abundances and depletion factors between our model and that of Ke01 (see Paper Ib, accepted).

Zoom In Zoom Out Reset image size

A point of concern is the fact that the maximum starburst lines of our grids do not follow the loci of data points. This difference can be attributed to the fact that real dwarfs have complicated SFHs. The choice of SFH for the model affects the predicted line ratios. For the CSF grid with 0% AGN (Figure 3(a)), all grid points fall below the Ke01 maximum theoretical starburst line in both plots, as expected, but the grid falls below the real data in the O i plot. In contrast, the SSP model with 0% AGN (Figure 3(c)) overshoots the Ke01 line due to a harder continuum from younger stars, and much of the grid falls above the real data in the O i plot. Both SFH choices are unrealistic, and ideally we would use a combined continuous + bursty SFH to model a realistic dwarf. However, bracketing reality using the simplified cases of the CSF and the 20 Myr SSP, we see that the O i plot is always better than the N ii plot at identifying AGN in metal-poor star-forming dwarfs.

Despite the SFH caveat, we note that our models have better overlap with the data than some widely used standard models in the literature (e.g., Groves et al. 2004; Levesque et al. 2010). The classification scheme we will propose in Section 3, however, is unaffected by how well our new models match real data because in what follows, we continue to use the Ke01 maximum theoretical starburst lines for classification. Nonetheless, in the following sections, we will use our new models to demonstrate the need for a more optimized classification scheme using those same lines, especially for dwarf AGN, and to compute gas-phase metallicities for our sample.

To estimate the metallicities and ionization parameters of our galaxies, we run the Bayesian inference code, NebulaBayes (Thomas et al. 2018), with models from a slice of our new photoionization grid with pure star formation (0% AGN contribution) and a CSF SFH. We defer exploration of the effect of varying the AGN fraction and other parameters given in Paper 1b to a future paper. Currently, we only consider the six SELs present in the three diagnostic plots we are using: Hβ, [O iii], [O i], [N ii], Hα, and [S ii] doublet. For each galaxy, the code compares observed emission lines to predicted emission lines from the photoionization grid, then calculates the posterior probability of each combination of Z and U given flat priors. After evaluating all combinations of Z and U, the code marginalizes over nuisance parameters and obtains the best estimates for Z and U from the 2D joint probability distribution functions. We use the Bayesian-inferred metallicity estimates to explore the physical properties of the newly identified dwarf AGN in Section 5. Based on the Bayesian estimates, we define a fiducial dwarf as one having Z = 0.4 Z_⊙ and log(U) = −3.25, the median values of Z and U for the combined RESOLVE and ECO SEL sample.

The N ii plot alone is an inadequate classifier because its abscissa, [N ii]/Hα, is a crude proxy for metallicity. Typical metal-poor dwarfs have relatively high [O iii]/Hβ emission and relatively low [N ii]/Hα (Moustakas et al. 2006), usually placing them on the left side of the plot below the demarcation line from Kauffmann et al. (2003a). Thus, AGN in low-metallicity dwarfs are likely placed in the star-forming region of the N ii plot (Groves et al. 2006; Ludwig et al. 2012).

In Section 3.2, we will present a new classification scheme that optimally uses the N ii, S ii, and O i plots together to classify every galaxy uniquely. The O i plot is not commonly used since [O i] is a weaker emission line than [N ii], but in this section, we show the value of using this less metallicity-sensitive plot as well as the S ii plot. In Figure 3, our models show that the metallicity-sensitive N ii plot does not identify even 100% AGN contributions to the spectra of metal-poor galaxies, regardless of their SFHs. For a fiducial dwarf representing the RESOLVE and ECO SEL sample with Z = 0.4 Z_⊙ (lime-green star; see Section 2.4), even with a 100% AGN contribution, it falls below the traditional AGN demarcation line in the N ii plot, instead landing in the "composite" SF+AGN region (between the dashed and solid curves in Figure 3). A study by Groves et al. (2006) also found that the model grids for low-metallicity AGN and low-metallicity starbursts overlap on the N ii plot. In contrast, the O i plot easily identifies a 100% AGN contribution for all metallicities due to its metallicity insensitivity, regardless of the SFH.

Figure 4 shows that a simulated fiducial z ∼ 0 dwarf AGN with low metallicity (Z = 0.4 Z_⊙) is not classified as an AGN by the N ii plot but is easily identifiable as an AGN by the S ii and O i plots, which are much less sensitive to metallicity. Figure 4 also shows that the S ii and O i plots outperform the N ii plot at identifying AGN with low (8%–16%) spectral contributions in a fiducial metal-poor dwarf. This fiducial dwarf is modeled with a CSF SFH, but the same result stands for a dwarf with an SSP SFH. Dwarfs are expected to have central BH masses ∼10³–10⁵ M_⊙ (Greene et al. 2020). At such low masses, even if these BHs were accreting at the Eddington limit, they would be quite faint in optical wavelengths (Greene & Ho 2007). Even without intense star formation, these faint IMBH signatures may be diluted by the host galaxy (Moran et al. 2002), so in a highly star-forming galaxy like a z ∼ 0 dwarf, BH signatures are easily masked by much stronger star formation signatures (Reines et al. 2013). Nonetheless, the O i plot can easily identify 16% AGN spectral contributions. It can even identify AGN contributions down to almost 8%, making it well suited to finding AGN in highly star-forming galaxies like typical z ∼ 0 dwarfs (Figure 4). The S ii plot is also better than the N ii plot at identifying AGN in metal-poor and/or star-forming galaxies, albeit it is more sensitive to star formation dilution than the O i plot.

Zoom In Zoom Out Reset image size

Previous studies have also used the N ii, S ii, and O i plots together for spectral classification. (Kewley et al. 2006, hereafter K06) designed a method of using the N ii, S ii, and O i plots to classify galaxies as star-forming H ii regions (abbreviated as SF), composite, Seyfert, and LINER galaxies. However, their method was not optimized for dwarfs, and it does not robustly classify all of the galaxies in a sample. The K06 scheme marks galaxies as "ambiguous" either if they are classified differently as Seyfert versus LINER by the S ii and O i plots, or if they are classified as "composite" by the N ii plot and as Seyfert/LINER in the S ii and/or O i plots. It is unclear whether these "ambiguous" galaxies are always treated as AGN candidates in the rest of the K06 analysis. Additionally, there is no classification that explicitly includes galaxies classified differently as SF by the N ii plot and as AGN by the S ii and/or O i plots or vice versa. These galaxies, as Figure 4 shows, may include many dwarf AGN.

Our optimized classification scheme assigns a unique category to every galaxy to create a more systematic classification scheme that does not exclude any galaxies, as shown in Figure 5 using the RESOLVE sample. Following the convention of K06, the S ii plot uses the sum of the doublet [S ii] λ6717 + [S ii] λ6731 (i.e., [S ii] λ6720), and the N ii plot uses only the [N ii] λ6584 line flux. However, to enhance S/N, we combine the [N ii] λ6548 and λ6584 doublet fluxes and then scale them back to get the [N ii] λ6584 flux given that the ratio of [N ii] λ6584 to [N ii] λ6548 is approximately 3:1 (Acker et al. 1989).

Zoom In Zoom Out Reset image size

The theoretical maximum starburst lines for the N ii, S ii, and O i plots are defined by Equations (5), (6), and (7) from Kewley et al. (2001), and we refer to them as Ke01 lines in the respective plots. The composite line is given by Equation (1) in Kauffmann et al. (2003a), which we refer to as the Ka03 line hereafter. The dividing lines between Seyfert and LINER galaxies are given by Equations (9) and (10) for the S ii and O i plots, respectively, in K06, so we hereafter call these the K06 Seyfert/LINER dividing lines. Below we detail the mutually exclusive categories in our optimized classification scheme. Figure 6 shows representative images of galaxies from the RESOLVE survey in the categories of our optimized classification scheme. We note that we do not explicitly account for the errors on the flux ratios while classifying galaxies in this scheme, similar to general practice in the literature. Our strict S/N cuts minimize error bars, and we have verified that any changes in statistics due to shifting classification of galaxies close to the dividing lines are within the reported errors on the classification percentages.

Zoom In Zoom Out Reset image size

• 1.  
  SF or Definite Star-forming Galaxies: These are galaxies that lie below the Ka03 composite line in the N ii plot and to the left of the Ke01 theoretical maximum starburst line in both the S ii and O i plots.
• 2.  
  Composite Galaxies: These are galaxies that lie between the Ke01 and the Ka03 lines in the N ii plot, regardless of location in the O i or S ii plots, unlike K06.
• 3.  
  Seyfert Galaxies: These are galaxies that lie above the Ke01 theoretical maximum starburst lines in the N ii, S ii, and O i plots, and above the K06 Seyfert/LINER dividing lines in the S ii and O i plots.
• 4.  
  LINERs: These are galaxies that lie above the Ke01 theoretical maximum starburst lines in the N ii, S ii, and O i plots and below the K06 Seyfert/LINER lines in the S ii and O i plots.
• 5.  
  Ambiguous-type AGN Galaxies: These are galaxies that lie above the Ke01 line in the N ii plot, but have different AGN types in the S ii and O i plots—Seyfert in (either) one and LINER in the other. Our "Ambiguous-type" galaxies are a subset of the "Ambiguous" galaxies defined by K06 (see Section 3.1 for discussion of the K06 definition).
• 6.  
  SF-AGN Galaxies: These are galaxies that are classified as SF (below the Ka03 line) in the N ii plot, but as Seyfert or LINER in the S ii and/or O i plots (above the Ke01 lines). We will show that these galaxies are mostly metal-poor, gas-rich, and highly star-forming dwarfs that likely host AGN (see Sections 4 and 5).
• 7.  
  Low-S ii , Low-O i , or Low-S ii +O i AGN Galaxies: These are galaxies that are classified as AGN in the N ii plot (above Ke01 line), but as SF in either the S ii or O i plots or both (below the Ke01 lines). In our sample, all galaxies in this category are AGN in the N ii and O i plots, but SF only in the S ii plot, i.e., they are low-S ii AGN. Our sample does not contain any low-O i AGN or low-S ii+O i AGN, although examples could exist in a larger sample especially given observational errors. In the rest of this paper, we will refer to this category as low-S ii AGN. We suspect that the SF classification from the S ii plot is spurious and that these galaxies are truly AGN hosts (see below).

The above scheme has introduced two new categories of galaxies: SF-AGN and low-S ii AGN. Most of the new AGN candidates we identify are in the SF-AGN category (see inset table of Figure 5 and Section 4.2). Regardless of the SDSS catalog used, <1% of RESOLVE and ECO galaxies fall in the low-S ii AGN category. As mentioned above, all low-S ii AGN galaxies in our sample have AGN-like (Seyfert or LINER) line ratios in the N ii and O i plots, but not in the S ii plot. All low-S ii AGN galaxies have high mass and high metallicity, except for one galaxy, rs0672. The high metallicity can boost [N ii]/Hα but not [S ii]/Hα. We do not have evidence of high SF for the low-S ii AGN hosts, but we also do not have detailed enough information to determine whether SF dilution can cause the differential classifications for these galaxies. Factors like dust and metallicity may drive these differential classifications, but it is beyond the scope of the data we have to disentangle these factors. Images from DECaLS DR8 ¹⁰ (Figure 5) also confirm that all of these galaxies have bright nuclei—rs0181 even has a broad Hα feature—suggestive of real AGN activity. Since the O i plot is a more sensitive indicator of AGN than the S ii plot, and it corroborates the AGN classification from the N ii plot, we choose to consider low-S ii AGN galaxies as AGN candidates.

Based on the classifications of RESOLVE and ECO SEL galaxies, we find that the S ii plot identifies far fewer dwarf AGN compared to the O i plot. This result is likely because the S ii plot cannot identify AGN spectral contributions lower than ∼16%, while the O i plot identifies contributions down to almost 8%, as seen in Figure 4. Paper Ib (accepted) also shows that [S ii]/Hα is less sensitive to low AGN contributions than [O i]/Hα. However, if the O i plot cannot be used due to practical/observational constraints, the S ii plot can still be valuable and recover some AGN in metal-poor and/or SF galaxies that would be missed by the N ii plot.

In the rest of this paper, we will not distinguish between the Seyfert, LINER, and ambiguous-type AGN categories. We consider all of these to be traditionally identified AGN as they lie above the Ke01 line in all three plots, and we call them "traditional AGN" hosts. Even though there is some contention about the true nature of LINERs, there is still substantial evidence that they are some flavor of AGN (Ho et al. 2003; Kewley et al. 2006; Ho 2008; Goulding & Alexander 2009). We count all galaxies in the SF-AGN, composite, traditional AGN, and low-S ii AGN categories as AGN candidates. Throughout this paper, we use the terms AGN and AGN candidates interchangeably since no AGN detection method is completely foolproof by itself.

Several other mechanisms produce AGN-like line ratios in galaxies and could be mistaken for AGN activity. Here we explore such mechanisms and describe why they likely do not explain the behavior of SF-AGN galaxies. We also present ancillary data to argue that SF-AGN are indeed AGN candidates.

• 1.  
  Could SF-AGN be due to hard radiation from stars? Theoretically, extreme starbursts can lead to emission-line signatures similar to those from AGN. In Figure 3, our photoionization model with an SSP viewed at 20 Myr shows that AGN-like [O i]/Hα ratios can be observed even with 0% AGN spectral contribution, especially for galaxies with a high metallicity and high ionization parameter. The hardening of spectra from young starbursts can potentially be explained by the presence of Wolf–Rayet (W-R) stars within the lower-metallicity regime of our galaxies (e.g., Brinchmann et al. 2008). Extremely starbursting galaxies like blue compact dwarfs or blue nuggets have high sSFRs generally with log(sSFR) [Gyr⁻¹] > −0.5 (e.g., Hopkins et al. 2002) and high equivalent widths for the Hα line (>80 Å; Lee et al. 2009). From Table 2 of Lee et al. (2009), we estimate their median log(sSFR) [Gyr⁻¹] to be ∼−0.18, meaning their starburst definition (Hα EW > 80 Å) yields more extreme starbursts than that of Hopkins et al. (2002). The SF-AGN in our samples have sSFRs (computed from global photometry reflecting 100 Myr timescales; see Sections 2.1 and 5.2 for details on SFRs) comparable to starbursts with a median of log(sSFR) [Gyr⁻¹] ∼−0.27, but only ∼7% of SF-AGN have EW(Hα) >80 Å (computed from SDSS fiber spectroscopy). Given that our global SFRs are on average slightly higher than those from Salim et al. (2016; see Section 2.1), these results do not suggest we are underestimating the role of SF in producing [O i]/Hα in SF-AGN.For a more direct comparison on the nuclear scale of the SDSS fibers, we have checked the properties of SF-AGN against the largest local W-R galaxy database compiled by Brinchmann et al. (2008, henceforth B08). Figure 7 shows the emission-line ratio trends of SF-AGN and B08 W-R galaxies in a different diagnostic plot whose x-axis is a ratio of neutral-to-doubly ionized oxygen. B08 found that W-R galaxies were offset from the rest of the sample toward the bottom left of the diagnostic plot (see Figure 11 in B08). We see that only a small number of SF-AGN (<10%) with low [O i]/[O iii] are found in the region populated by W-R galaxies. In fact, in this diagnostic, most SF-AGN have emission-line trends that are consistent with traditional AGN and composites. Of course, we cannot rule out the coexistence of starbursting nuclear star clusters and AGN, both of which would be much smaller than the SDSS fiber. To assess possible mixtures, we have performed a preliminary analysis of AGN spectral contributions using NebulaBayes with our most starbursting photoionization model (20 Myr old SSP). We find that ∼92% of SF-AGN require a nonzero AGN contribution to their spectra to explain their observed emission lines. High mass XRBs (HMXBs) found in the presence of extreme star formation can also harden ionizing radiation causing elevated [O i] fluxes. Lehmer et al. (2021) found that HMXB spectral contributions are expected to increase with decreasing metallicity for galaxies with SFRs spanning a wide range from 0.01 to 100 M_⊙ yr⁻¹. Senchyna et al. (2020) studied 11 metal-poor SF dwarfs (0.05–0.35 Z_⊙) with nebular He II indicating the presence of high-energy photons. They compared the dwarfs with photoionization grids with simple blackbody SEDs to model HMXBs and found little significant contribution of high-energy photons from HMXBs. In contrast, Simmonds et al. (2021) used photoionization grids with more realistic SEDs and found that HMXBs in low-metallicity (<0.2 Z_⊙) dwarfs can potentially elevate [O i]/Hα if a specific SED is assumed. The new SF-AGN in this work have metallicities between 0.3 and 0.4 Z_⊙, higher than those of the Senchyna et al. (2020) sample (with a similar range of SFRs) and also higher than the Simmonds et al. (2021) models. We have an X-ray luminosity for only one SF-AGN, which has a lower L_X−ray/ SFR ratio than the model analyzed in Simmonds et al. (2021); all other SF-AGN are either not targeted or not detected by Chandra or X-ray Multi-Mirror Mission (XMM) (see #3). Based on their metallicity range and lack of exceptionally high X-ray luminosities, we conclude that SF-AGN are unlikely to have elevated [O i] fluxes solely due to HMXB-hardened spectra.In summary, the emission-line ratios and trends for a majority of SF-AGN are difficult to explain without the presence of AGN activity, even assuming extreme star formation.
• 2.  
  Do we have evidence of "nuclear" activity in SF-AGN? A crossmatch of all unique RESOLVE and ECO SEL galaxies with the SAMI DR2 catalog (Bryant et al. 2015) yielded integral field unit (IFU) data for two SF-AGN. For each IFU data cube, we filter spaxels to have both continuum S/N > 5 and emission line S/N > 5 for all SELs. This filtering is necessary because a high continuum S/N ensures reliable continuum fitting and signal decomposition, and a high emission-line S/N ensures reliable use of the three diagnostic plots. Figure 8 shows spatially resolved high-S/N data for rs0010, revealing that high AGN-like [O i]/Hα ratios (red spaxels) are centrally located and clearly separated from SF-like [O i]/Hα ratios (blue spaxels).Figure 9 shows spatially resolved diagnostic plots for rs0010, confirming AGN-like line ratios in the O i plot in several spaxels within the central 2 kpc. Only the O i plot is able to identify the AGN-like line ratios; the N ii and S ii plots fail, likely due to bias against low metallicity and SF dilution. Another SF-AGN, rs0775, shows similar line ratio behavior, but with a worse S/N for all lines, and the S/N of the continuum is much too weak to form any reliable conclusions. In summary, reliable IFU data for the SF-AGN rs0010 confirm that AGN-like line ratios are centrally located.
• 3.  
  Do SF-AGN have nonoptical or other known counterparts? Table 1 reports the statistics of RESOLVE and ECO SF-AGN dwarfs with nonoptical or other known counterparts. We do not find any crossmatches (within a 5'' radius) of RESOLVE and ECO SF-AGN in two optical AGN catalogs (Véron-Cetty & Véron 2006; Flesch 2015) and a comprehensive broadline AGN catalog (Liu et al. 2019).We have also crossmatched the RESOLVE and ECO SEL samples with the 3XMM-DR8 catalog (Rosen et al. 2016) and the Chandra Source Catalog (CSC) Release 2.0 (Evans et al. 2020). We find an X-ray match from 3XMM for only one RESOLVE SF-AGN, rf0477, and none from the CSC. To assess whether this detection represents an X-ray AGN or simply X-ray binaries, we put the catalog X-ray fluxes on a common basis with those used in the L_X−ray − SFR relation of Ranalli et al. (2003) as shown in Figure 10. We multiply the 0.5–12 keV 3XMM flux by a factor of 0.9 (based on a photon index, Γ = 1.7) to obtain 0.5–10 keV flux (Agostino & Salim 2019). In Figure 10, we also show X-ray crossmatches for RESOLVE and ECO SEL galaxies that are not SF-AGN for reference, including the composite that is an X-ray candidate shown in Figure 6. For galaxies with CSC matches, we multiply the 0.5–7.0 keV CSC flux by a factor of 1.21 to obtain 0.5–10 keV fluxes (LaMassa et al. 2013). The SF-AGN rf0477 borderline qualifies as an X-ray AGN candidate.We have also looked for radio counterparts of RESOLVE and ECO SEL galaxies in the HEASARC Master Radio Catalog, ¹¹ finding crossmatches for one ECO SF-AGN and two RESOLVE SF-AGN. Based on visual examination of radio continuum cutouts of these three SF-AGN, one galaxy, rs1038, shows signs of extended emission, and two galaxies, rs0124 and ECO05128, have unresolved emission. With the available data, we cannot test whether the emission is due to AGN or SF.Finally, we use our own recomputed WISE photometry (Paper II) available for 1324 of the 2605 RESOLVE and ECO SEL galaxies from the MPA-JHU catalog to assess mid-IR color selection criteria for AGN. We consider galaxies as mid-IR AGN candidates if they cross the AGN color threshold using any one of the three widely used criteria from Jarrett et al. (2011), Stern et al. (2012), and Satyapal et al. (2018). In practice, the Jarrett et al. (2011) criterion is the most restrictive while the Satyapal et al. (2018) criterion is the least restrictive. None of the SF-AGN is classified as a mid-IR AGN by any of the criteria.In summary, among all SF-AGN in the combined RESOLVE and ECO samples, one galaxy has an AGN counterpart from X-rays. There are also three SF-AGN with radio crossmatches, albeit the radio emission cannot be classified as having an SF or AGN origin with the available data. Generally, we conclude that most SF-AGN do not have counterparts (see Table 1), but this result is not unexpected, as most AGN identification techniques are sensitive to finding AGN representing higher-metallicity hosts, more massive BHs, and/or AGN with high spectral contributions. Also, the reliability of mid-IR selection for dwarf AGN detection has been debated (see Section 1); we investigate this issue further in Paper II.
• 4.  
  Could SF-AGN be galaxies with shocks? Shocks can cause enhanced optical emission-line ratios and are found in galaxies with high star formation (Heckman et al. 1987; Rupke et al. 2005), AGN (Cecil et al. 2002; Rupke & Veilleux 2011; D'Agostino et al. 2019), or galaxy mergers (Rich et al. 2011; Rupke & Veilleux 2013). Shocks and AGN line ratios are expected to behave similarly in the N ii plot, but shock emission is localized in the LINER region in the S ii and O i plots (Allen et al. 2008; Davies et al. 2017). However, SF-AGN mostly avoid the LINER regions of these plots (Figure 11). Additionally, as noted under #3, broad Hα features have not been cataloged for any of our SF-AGN, although this result does not rule out low-velocity shocks with v < 500 kms⁻¹ (Reines et al. 2013). Recent work (e.g., D'Agostino et al. 2019; Molina et al. 2021) has shown that in galaxies with AGN-like narrow emission-line ratios, the AGN can be the origin of observed shocks.In summary, if SF-AGN hosts do have low-velocity shocks, they could potentially originate from the AGN, but such shocks cannot easily explain the Seyfert-like line ratios in the S ii and O i plots.
• 5.  
  Could SF-AGN be diffused ionized gas (DIG) galaxies? DIG is low surface density Hα gas, typically found on the outskirts of face-on galaxies or in the extraplanar regions of disk galaxies. DIG can comprise up to ∼60% of the total gas mass (Zhang et al. 2017; Vogt et al. 2017; Lacerda et al. 2018). DIG can cause elevated ratios of [N ii], [S ii], and [O i] with respect to Hα (Kaplan et al. 2016) and can push galaxies to the composite/AGN side of the N ii plot (Zhang et al. 2017). The SF-AGN in our sample, by definition, do not have elevated [N ii]/Hα, but their low-metallicity nature could mask any potential [N ii]/Hα enhancement from DIG. Regardless, the SDSS spectra we use sample only the central 2'' or 3'' where the galaxy light should not be DIG dominated. Additionally, DIG does not easily explain the spatial trends observed in Figure 9 where we see AGN-like [O i]/Hα line ratios in only the central 2 kpc of an SF-AGN galaxy.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 2 gives the statistics of RESOLVE and ECO galaxies in our new mutually exclusive categories using fluxes from the three SDSS catalogs. Including the new categories (SF-AGN and low-S ii AGN), the AGN percentage in all z ∼ 0 SEL galaxies is ∼16%–30% depending on the SDSS catalog used. We note that it is standard in most AGN studies to express AGN percentages relative to the search sample (i.e., not including galaxies excluded during sample selection). We will usually follow that practice and state statistics as a fraction of the number of galaxies in the search sample, i.e., SEL galaxies. We also examine the AGN percentage normalized to the full galaxy population, and in both cases, we explicitly specify the population under consideration to avoid confusion.

The new SF-AGN category makes up ∼3%–9% of the full RESOLVE and ECO SEL samples (i.e., including dwarfs and giants). Most SF-AGN hosts (75%–95% depending on the catalog) are dwarfs. Figure 11 shows the dwarf AGN candidates in the overall RESOLVE and ECO SEL sample. Most dwarf AGN are in the new SF-AGN category (blue squares). SF-AGN are mainly identified by the O i plot, which is relatively insensitive to both metallicity and SF dilution. On comparison with Figure 4, most SF-AGN seem to have AGN spectral contributions in the 8%–16% range. Similar dwarf SF-AGN would have been missed in past studies that use only the N ii plot due to their low metallicity as well as their high SFRs, typical for z ∼ 0 dwarfs (see Section 5). The previous work of K06 using all three of the N ii, S ii, and O i plots likely missed these AGN because none of their categories explicitly includes them.

Zoom In Zoom Out Reset image size

Figure 12 shows that depending on the SDSS catalog used, the new overall dwarf AGN percentage in SEL galaxies is now ∼3%–16%, much higher that the <1% in previous studies (Reines et al. 2013; Sartori et al. 2015; Reines & Volonteri 2015). AGN in dwarf SEL hosts make up ∼0.6%–3.0% of the full baryonic mass-limited RESOLVE and ECO samples. The percentage of AGN in SEL giants varies between 36% and 47% depending on the sample and catalog. AGN in giant SEL hosts make up around 3%–4% of the full baryonic mass-limited sample.

Zoom In Zoom Out Reset image size

Table 2 shows noticeable variation in the AGN percentages derived from the three different SDSS catalogs. The catalogs have different data sources—the MPA-JHU and NSA catalogs are based on SDSS DR8, while the Portsmouth catalog is based on SDSS DR12—but we do not find evidence that the data source affects the sample selection or the AGN statistics. Crossmatching all RESOLVE and ECO galaxies (including non-EL galaxies), the Portsmouth and MPA-JHU catalogs include exactly the same 7557 galaxies, while the slightly larger NSA catalog has 95% overlap in galaxies. As shown visually in Figure 12, less than half as many Portsmouth measurements pass our SEL S/N cuts as do MPA-JHU measurements (16.0% versus 34.5%, see Section 2.2), apparently due to higher error estimates (for our six emission lines of interest, the median Portsmouth errors are 2.5–3.5× higher than the median MPA-JHU errors). However, the Portsmouth-identified dwarf AGN have substantial overlap with the MPA-JHU-identified dwarf AGN, and the two catalogs yield consistent dwarf AGN percentages within their uncertainties, despite being based on different SDSS data releases. In contrast, Figure 12 shows that the NSA and MPA-JHU catalogs have much lower dwarf AGN overlap and yield discrepant dwarf AGN frequencies, despite being based on the same SDSS data release.

Table 2. Statistics of SEL Galaxy Categories in Our Optimized Emission-line Classification Scheme

Category	Classification of All SEL Galaxies (Dwarfs + Giants)
	MPA-JHU			Portsmouth			NSA
	RESOLVE	ECO	Overall a	RESOLVE	ECO	Overall	RESOLVE	ECO	Overall
(No. of galaxies)	(382)	(2507)	(2605)	(202)	(1161)	(1207)	(209)	(1363)	(1411)
Definite SF	79.3%	81.8%	81.6%	82.7%	83.8%	83.5%	70.3%	70.4%	70.2%
SF-AGN	3.9%	4.0%	4.0%	2.0%	3.4%	3.4%	7.7%	8.6%	8.5%
Composite	7.3%	8.0%	7.9%	7.9%	6.8%	6.7%	8.6%	11.4%	11.2%
Low-S ii AGN	0.8%	0.4%	0.4%	0.5%	0.8%	0.7%	0.5%	0.4%	0.4%
Seyfert	2.9%	1.8%	1.9%	3.0%	2.4%	2.7%	8.1%	4.3%	4.5%
LINER	4.4%	3.5%	3.7%	3.0%	2.4%	2.6%	4.3%	4.3%	4.4%
Ambiguous-type AGN	1.3%	0.5%	0.5%	1.0%	0.2%	0.2%	0.5%	0.7%	0.8%
Traditional AGN b	8.6%	5.8%	6.1%	7.0%	5.0%	4.7%	12.9%	9.3%	9.7%
All AGN c	${20.6}_{-2.0}^{+2.1} \%$	${18.2}_{-2.5}^{+2.8} \%$	${18.4}_{-1.0}^{+1.1} \%$	${17.4}_{-2.5}^{+2.8} \%$	${16.0}_{-1.0}^{+1.1} \%$	${16.3}_{-1.0}^{+1.1} \%$	${29.7}_{-3.0}^{+3.2} \%$	${29.7}_{-1.2}^{+1.2} \%$	${29.8}_{-1.2}^{+1.2} \%$
	Classification of SEL Dwarf Galaxies (M* < 109.5 M⊙)
(No. of galaxies)	(226)	(1525)	(1577)	(129)	(749)	(776)	(114)	(738)	(761)
Definite SF	93.4%	93.1%	93.2%	96.9%	94.0%	94.1%	85.1%	84.3%	84.4%
SF-AGN	5.8%	6.2%	6.1%	2.3%	5.1%	4.9%	13.2%	14.6%	14.5%
Composite	0.4%	0.4%	0.4%	0.8%	0.4%	0.5%	0.0%	0.5%	0.5%
Low-S ii AGN	0.4%	0.1%	0.1%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%
Seyfert	0.0%	0.3%	0.3%	0.0%	0.5%	0.5%	1.8%	0.5%	0.7%
LINER	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%
Ambiguous-type AGN	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%
Traditional AGN	0.0%	0.3%	0.3%	0.0%	0.5%	0.5%	1.8%	0.5%	0.7%
All AGN	${6.6}_{-1.5}^{+1.8} \%$	${6.9}_{-0.6}^{+0.7}{\rm{ \% }}$	${6.9}_{-0.4}^{+0.4} \%$	${3.1}_{-1.2}^{+1.9} \%$	${6.0}_{-0.8}^{+0.9} \%$	${5.9}_{-0.8}^{+0.9} \%$	${15.0}_{-3.0}^{+3.6} \%$	${15.6}_{-1.3}^{+1.4} \%$	${15.6}_{-1.3}^{+1.3}{\rm{ \% }}$
	Full-Population Dwarf SEL AGN Statistics
(No. of dwarfs)	(648)	(3931)	(4161)	(648)	(3931)	(4161)	(648)	(3931)	(4161)
Dwarf AGN	${2.3}_{-0.5}^{+0.7} \%$	${2.7}_{-0.2}^{+0.3} \%$	${2.6}_{-0.2}^{+0.2} \%$	${0.6}_{-0.2}^{+0.4} \%$	${1.1}_{-0.2}^{+0.2} \%$	${1.1}_{-0.2}^{+0.2} \%$	${2.6}_{-0.6}^{+0.7} \%$	${3.0}_{-0.2}^{+0.3} \%$	${2.9}_{-0.2}^{+0.3} \%$

Notes. All error bars are computed using binomial confidence intervals.

^a Overall RESOLVE+ECO sample does not double-count the overlap between the two surveys. ^b Sum total of Seyfert, LINER, and Ambiguous-type AGN categories. ^c Sum total of SF-AGN, composite, low-S ii AGN, and traditional AGN categories.

Download table as:  ASCIITypeset image

The choice of stellar population models for spectral decomposition can certainly affect emission-line ratios and consequently the sample selection and AGN statistics. The MPA-JHU catalog uses Bruzual & Charlot (2003, hereafter BC03) models with varying metallicities to fit the stellar continuum, including a low-metallicity model. Both the Portsmouth and NSA catalogs use only solar metallicity models from Maraston & Strömbäck (2011, hereafter MS11_solar) and try to exploit the age–metallicity degeneracy to model metallicity dependence by using different ages. The authors claim that this method should not greatly affect flux estimates in galaxies with very strong emission lines, like the sample in our study. However, Reichardt et al. (2001) found that the age–metallicity degeneracy disappears while simultaneously fitting the absorption lines and the continuum. This may lead to fitting absorption features that are too shallow or too deep, yielding emission fluxes that are too low or too high, as seen in Chen et al. (2018).

Apart from the continuum modeling, the MPA-JHU and Portsmouth catalogs rely on similar procedures for extracting fluxes and applying corrections (given our homogenization of extinction corrections; see Section 2.2). For RESOLVE and ECO SEL galaxies that are common to both catalogs, we find a tight correlation between the SEL fluxes with a low spread in values. This trend has also been observed by previous studies that have compared the MPA-JHU and Portsmouth fluxes (e.g., Thomas et al. 2013; Chen et al. 2018; Zaw et al. 2019). However, despite the tight correlation between the fluxes, small variations in the flux ratios are enough to change the AGN versus SF classification of galaxies close to the demarcation lines. Such small variations, whether due to modeling differences or different SDSS data releases, may explain the imperfect overlap between dwarf SEL AGN in the Portsmouth and MPA-JHU catalogs (Figure 12).

AGN statistics are also dependent on the calibrations applied to the data sets. The NSA catalog has an additional flux calibration to fix small-scale calibration residuals that arise from the standard SDSS pipeline (Yan 2011). With this reprocessed spectroscopy, the NSA fluxes for RESOLVE and ECO SEL galaxies are on average ∼30% higher than corresponding MPA-JHU or Portsmouth fluxes, and the NSA flux ratios are ∼10% higher. With NSA fluxes, the dwarf AGN percentage among SEL galaxies comes out ∼3× higher than with MPA-JHU or Portsmouth fluxes (see Figure 12). This difference is somewhat expected considering that Yan (2011) applied the NSA flux calibrations to the MPA-JHU catalog, and they found that ∼7% of LINERs have different classifications in the K06 system using MPA-JHU fluxes with the additional flux calibrations versus without them, indicating that the flux differences are nonnegligible.

Another likely contributor to the higher dwarf AGN frequencies derived from the NSA versus MPA-JHU or Portsmouth catalogs may be selection effects. The emission-line measurements from the three SDSS catalogs are affected by differences in continuum fitting methodology, and our SEL samples are selected based on S/N cuts on these measurements, consequently inheriting selection effects related to the differences in fitting methodology. Comparing the properties of SEL galaxies in the three catalogs (baryonic mass, halo mass, and (u-r) color), we see a consistent trend where the NSA SEL catalog has relatively fewer blue-sequence dwarfs in lower-mass halos and more red-sequence giants in higher-mass halos compared to the other two catalogs (Figure 2 left panel). We suspect that this trend may be because the ratio of NSA catalog errors to MPA-JHU catalog errors is negatively correlated with mass for [N ii] (the NSA errors are also, on average, ∼2–3× larger than the MPA-JHU errors for all emission lines, but this fact is also true for Portsmouth errors, which show no mass-dependent trends relative to MPA-JHU errors). The NSA errors for [N ii] are ∼3× the MPA-JHU errors on the low-mass end but ∼0.5× the MPA-JHU errors on the high-mass end. This mass-dependency of errors and therefore S/N likely results in fewer dwarfs passing our SEL S/N cuts. We cannot speculate as to why this trend in errors arises between the catalogs, but it may contribute to the higher dwarf SEL AGN percentage estimated from the NSA catalog (∼16% compared to ∼3%–6% from the JHU and Portsmouth catalogs) if it reflects modeling differences that preferentially lead to rejecting non-AGN dwarfs from the NSA SEL sample.

In summary, the differences we see in AGN percentages for SEL dwarfs seem to be primarily due to irreducible (as per our current knowledge) systematics due to different choices in spectral processing methodologies. Chen et al. (2018) and Zaw et al. (2019) explore these differences and find that modeling choices dramatically affect AGN categorization. Using only the N ii plot, they find that BC03-based fluxes from the MPA-JHU catalog identify more AGN than MS11_solar-based fluxes from the Portsmouth catalog. However, we find that the dwarf SEL percentage from the NSA catalog (using our new scheme) is the highest despite having MS11_solar-based fluxes, possibly due to the additional flux calibrations in the NSA catalog. We would like to stress that all three catalogs are state of the art and that we have no evidence that the differences in statistics are due to mistakes in any of the catalogs. Rather, the large discrepancies in emission-line-based AGN statistics are an important result of this paper, associated with the many methodological choices made during the spectral fitting process. Such discrepancies are unavoidable without consensus on methodological choices and should be represented in any statistical conclusions.

In this paper, we display the MPA-JHU SEL sample in all plots since this sample has the most galaxies and its SEL properties lie between the properties from the other two catalogs. However, we cannot determine whether any catalog is clearly better, so we report the statistics of all three catalogs.

Figure 13 shows the gas-to-stellar-mass ratios (G/S) of all RESOLVE and ECO SEL galaxies. Most of the traditional AGN and composites are in gas-poor giant galaxies, and almost all SF-AGN are in gas-dominated dwarfs. The high G/S of SF-AGN is typical of dwarfs in the local universe (Kannappan 2004; Kannappan et al. 2013; Stark et al. 2016).

Zoom In Zoom Out Reset image size

Figure 14 shows the mass–metallicity relation of RESOLVE and ECO SEL galaxies, using the gas-phase metallicities obtained from the Bayesian inference code, NebulaBayes, as detailed in Section 2.4. The median metallicity of composites and traditional AGN, 0.8 Z_⊙, is higher than the median metallicity of the dwarf-dominated RESOLVE SEL sample as a whole, 0.7 Z_⊙. On the other hand, most SF-AGN are hosted by dwarfs and have a median metallicity of ∼0.45 Z_⊙, slightly higher than the fiducial dwarf metallicity used in our models in Section 2.3. We find that SF and AGN galaxies follow different mass–metallicity trend lines, as has been observed in other work (e.g., Thomas et al. 2019). However, the metallicities of AGN in this work are not optimally modeled because we use a pure SF photoionization grid. We recognize that the metallicities would change if our modeling included AGN contributions. However, adaptive modeling of AGN metallicities is beyond the scope of this paper, and Figure 14 is purely demonstrative of trends.

Zoom In Zoom Out Reset image size

Importantly, despite SF-AGN being selected only based on differential classifications between the diagnostic diagrams, between 75% and 95% of SF-AGN (depending on catalog) belong to the category of low-metallicity gas-rich dwarfs.

Figure 15 shows that in our SEL sample, composites and traditional AGN have low long- and short-term SFRs. We trace SFHs using long- and short-term FSMGRs, defined as the ratio of newly formed stellar mass to preexisting stellar mass per timescale, where the timescale dictating the division of new and preexisting is at 1 Gyr and 100 Myr, respectively (see Section 2.1).

Zoom In Zoom Out Reset image size

Most SF-AGN are gas-rich dwarfs that have more than doubled their stellar mass in the past 1 Gyr lying above the line in Figure 15, so they are much more actively star-forming than composites and traditional AGN. This high star formation activity implies a dilution of AGN signatures and consequently makes them hard to identify with the N ii and S ii plots, as discussed by Moran et al. (2002) and Reines et al. (2013), and in Section 3.1 and Figure 4.

We note that the handful of outliers in the SEL sample at low FSMGR_LT and moderate FSMGR_ST are part of a larger population of very dusty galaxies whose FSMGR_ST is well measured from UV and IR photometry, but whose FSMGR_LT is likely underestimated by SED fitting that does not include mid-IR photometry representing dusty SF (see Section 2.1). These outliers are not associated with mid-IR detected AGN (see Paper II).

Figure 16 investigates the relationship between the group halo mass and galaxy baryonic mass of RESOLVE and ECO SEL galaxies. We see shifts at two important mass scales—the gas-richness threshold mass, defined as M_halo ∼ 10^11.5 M_⊙, which corresponds to M_bary ∼ 10^9.9 M_⊙ and M_* ∼ 10^9.5 M_⊙ (Dekel & Silk 1986; Kannappan et al. 2013), and the bimodality mass, defined as M_halo ∼ 10¹² M_⊙ (Dekel & Birnboim 2006), which corresponds to M_bary ∼ 10^10.6 M_⊙ and M_* ∼ 10^10.5 M_⊙ (Kauffmann et al. 2003b; Kannappan et al. 2013). The gas-richness threshold scale marks the onset of gas heating within 0.1 R_virial of the dark matter halo, while the bimodality scale corresponds to the mass where the entire halo is shock heated (Dekel & Birnboim 2006). Figure 16 shows a sharp transition from AGN candidates being mostly SF-AGN below the gas-richness threshold mass to their being mostly traditional AGN above the bimodality mass. Composites span the transition range and overlap the other AGN categories. We also find that SF-AGN mainly occupy single-galaxy low-mass halos, whereas traditional AGN are more commonly found in multigalaxy massive halos.

Zoom In Zoom Out Reset image size