Hidden AGNs in Dwarf Galaxies Revealed by MaNGA: Light Echoes, Off-nuclear Wanderers, and a New Broad-line AGN

The finding of hundreds of active galactic nuclei (AGNs) in low-mass dwarf galaxies (stellar mass M_* ≤ 3 × 10⁹ M_⊙) has revolutionized black hole and galaxy formation models. These AGNs are often found to be powered by low-mass or intermediate-mass black holes (IMBHs; 100 < M_BH ≲ 10⁶ M_⊙), which could be the relics of the early universe seed black holes from which supermassive black holes (M_BH > 10⁶ M_⊙) grow (see Mezcua 2017; Greene et al. 2019 for recent reviews). If these seeds have not significantly grown by mergers or AGN feedback through cosmic time (Mezcua 2019, 2020), characterizing the local population of IMBHs in dwarf galaxies, in particular the black hole occupation fraction and the low-mass end of the scaling relations, can tell us how seed black holes predominantly formed at z > 10 (e.g., van Wassenhove et al. 2010; Ricarte & Natarajan 2018).

Myriad studies have hence focused on identifying AGNs in dwarf galaxies. The first searches were based on optical spectroscopic surveys such as the Sloan Digital Sky Survey (SDSS³ ) and the use of narrow emission line diagnostic diagrams (BPT diagrams; Baldwin et al. 1981; Kewley et al. 2001, 2006; Kauffmann et al. 2003) to distinguish between gas ionization by AGNs or by stars, accompanied by the detection of broad Balmer emission lines from which a black hole mass measurement could be obtained (e.g., Reines et al. 2013; Moran et al. 2014). Such optical studies are, however, biased toward central luminous type 1 AGNs with high accretion rates and at low redshifts (z < 0.3). This can be circumvented by X-ray and radio searches, which are not only able to detect AGNs in dwarf galaxies at the pinnacle of cosmic star formation history (Mezcua et al. 2016, 2018a, 2019), but offer the most reliable measurements of the AGN fraction in dwarf galaxies (e.g., Mezcua et al. 2018a; Birchall et al. 2020).

In addition to the biases of optical searches, single-fiber BPT diagnostics can fail at identifying AGNs in dwarf galaxies with active star formation (e.g., Cann et al. 2019; Birchall et al. 2020) or with a strong host galaxy light that dilutes the AGN signatures (e.g., Moran et al. 2002). This can be circumvented by high-resolution optical spectroscopy (e.g., Dickey et al. 2019) or optical variability studies (e.g., Baldassare et al. 2018, 2020), which are able to reveal AGN signatures even in quiescent and star formation-dominated dwarf galaxies.

The advent of integral-field unit (IFU) spectroscopic surveys such as SDSS/MaNGA (Mapping Nearby Galaxies at APO; Bundy et al. 2015) has opened a new window in the identification of "hidden" AGNs (e.g., Wylezalek et al. 2018). Such sources are missed by single-fiber BPT diagnostics if other mechanisms rather than the AGN dominate the ionization at the galaxy center, if the AGN is displaced from the central region due to a galaxy merger (e.g., Comerford & Greene 2014; Barrows et al. 2018), or if the AGN has recently switched off and its ionization signatures are only observable as light echoes at large distances from the optical center (e.g., Keel et al. 2015). Yet, most AGN searches using IFU data are biased toward central AGNs or do not focus on dwarf galaxies (Rembold et al. 2017; Sánchez et al. 2018; Wylezalek et al. 2018). Penny et al. (2018) used the MaNGA survey to investigate AGNs in dwarf galaxies, but their sample included only 69 quenched systems. Hence, no systematic studies of AGNs in dwarf galaxies has been performed so far using IFU data.

In this Letter we present the largest dedicated study of AGNs in dwarf galaxies using MaNGA data of 4718 sources. This allows us to identify AGN candidates in 37 dwarf galaxies, in 23 of which no clear AGN signatures are revealed in the single-spectrum BPT, and includes the discovery of a new broad-line low-mass AGN. The sample and analysis are presented in Section 2, while the results obtained are discussed in Section 3. Conclusions are provided in Section 4.

The sample of dwarf galaxies studied in this paper is drawn from the SDSS/MaNGA survey Data Release⁴ 15, which includes 4718 galaxies. The MaNGA IFU is centered on the optical center of the galaxy and has a size that typically matches the angular size of the galaxy, so that most targets are covered spectroscopically to at least 1.5 effective radii (R_eff). The final data cubes provided by the MaNGA data-analysis pipeline⁵ have square spaxels 05 in size, and the median spatial resolution is of 25 full width at half maximum (FWHM). We use the emission line measurements (performed after subtraction of the stellar continuum and corrected for absorption) of the HYB10-GAU-MILESHC maps, in which the spaxels are binned to a signal-to-noise ratio (S/N) ∼ 10 but the kinematic properties are extracted from the individual spaxels.

We also take from the MaNGA data-analysis pipeline the star formation rate (SFR) within the IFU field of view based on the Hα flux and the flux-weighted mean stellar velocity dispersion (σ) of all spaxels within 1 R_eff.

R_eff, redshift (z), and M_* are taken from the NASA–Sloan Atlas (NSA catalog) version v1_0_1, as used by MaNGA for its targeting,⁶ which incorporates an expanded redshift range and an improved elliptical Petrosian aperture photometry compared to the previous NSA catalog v0_1_2. The SFR and M_* are provided in units of h⁻² M_⊙ yr⁻¹ and h⁻² M_⊙, respectively, and we assume h = 0.73.

To draw our sample we first select those sources for which the Petrosian and Sérsic NSA stellar masses are consistent within 0.5 dex (e.g., Secrest & Satyapal 2020). The target dwarf galaxies are then selected as having M_* < 3 × 10⁹ M_⊙, as commonly done in AGN studies of dwarf galaxies (e.g., Reines et al. 2013, 2020; Mezcua et al. 2018a, 2019). No magnitude cut is applied to avoid any biases toward galaxy type. This yields an initial sample of 1609 dwarf galaxies.

To identify the source of ionization of each spaxel we use the lines of Kewley et al. (2001, 2006) and Kauffmann et al. (2003) to distinguish between AGNs, star formation, or composite (combination of AGN and star formation ionization) in the [O iii]/H_β versus [N ii]/H_α ([N ii]-BPT) and [O iii]/H_β versus [S ii]/H_α ([S ii]-BPT) emission line diagnostics. We also use the line from Schawinski et al. (2007) for a LINER⁷ division in the [N ii]-BPT. We only consider those dwarf galaxies for which the S/N of the lines used in the [N ii]-BPT and [S ii]-BPT diagrams (H_α, H_β, [N ii]λ6583, [S ii]λ6718, [O iii]λ5007) is ≥3. For these we first plot the [N ii]-BPT and [S ii]-BPT and select those galaxies for which at least 5 spaxels are located in the AGN/LINER photoionization region of both these diagram (see Figure 1, top left panel). Then, for each galaxy, the individual spectra of all the spaxels identified as AGNs or LINERs according to the above criteria are stacked together in order to increase the signal. The stacking procedure is explained in detail in Appendix A. Those sources whose stacked spectrum is very noisy and/or for which the emission lines of interest are hardly identifiable or have inconsistent measurements are removed from the sample. We also exclude those galaxies for which the spaxels classified as AGN/LINER are spread at the edges of the S/N > 1 continuum map (gray area in Figure 1, top middle panel). This yields a sample of 102 AGN dwarf galaxy candidates.

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

View figure set (35 panels)

Tarball of figure set images

We derive a median BPT location of the 102 AGN candidates by taking the median of the [N ii]-BPT line fluxes of the same AGN/LINER spaxels used to obtain the stacked spectrum (black square in Figure 1, top left panel). Based on this median BPT location, 47 of the AGN candidates are classified as LINERs. The LINER emission can not only originate from AGNs but also be produced on galactic scales by hot old stars, which are expected to have an Hα equivalent width (EW) typically below 3 Å. A distinction between the two processes can be made using the WHAN diagram (Cid Fernandes et al. 2010), which suggests that sources with EW(Hα) > 3 Å and flux ratio log [N ii]/Hα  >  −0.4 can be classified as true AGNs. A further sub-classification can be performed among strong AGNs (EW(Hα) > 6 Å) and weak AGNs (3 Å < EW(Hα) < 6 Å; see Figure 2). Based on the WHAN diagram of the median value of the AGN/LINER spaxels and relaxing the EW(Hα) cut to >2.8 Å in order not to exclude bona fide AGNs (see Section 3), we find that the AGN/LINER spaxels of 37 out of the 102 AGN candidates are consistent with being dominated by AGN ionization. These 37 sources constitute our final sample of AGN dwarf galaxy candidates.

Zoom In Zoom Out Reset image size

Of the 37 AGN candidates identified above using MaNGA data, 36 have also single-fiber (3'' aperture) SDSS spectra available. For one of the sources (9881-1901) the Hα region is missing, so we consider only 35 SDSS spectra. Two of them (8446-1901 and 9000-1901) are known broad-line AGNs (e.g., Reines & Volonteri 2015, from which we take the narrow emission line fluxes to obtain the SDSS BPT classification), while for the remaining 33 sources we obtain an SDSS BPT classification based on the emission line measurements of the SDSS DR12 spectra provided by the Portsmouth Group⁸ catalog. We classify the sources as quiescent if the S/N of any of the SDSS H_α, H_β, [N ii]λ6584, or [O iii]λ5007 emission lines used in the BPT is <3 (e.g., Secrest & Satyapal 2020). If S/N  ≥ 3, the sources are classified as either star-forming, AGNs, or composite following the same classification as for the MaNGA BPT (see Section 2). We find that 12 out of the 33 sources are classified as AGNs according to the SDSS (single-fiber) BPT, in agreement with the MaNGA BPT classification. However, two of these AGNs have a median MaNGA EW(Hα ∼ 2.8 Å) and would be missed by the WHAN diagram classification. To ensure we do not exclude any bona fide AGNs we thus relax the EW(Hα) cut to >2.8 Å. The remaining 21 sources out of the 33 are classified as either quiescent, star-forming, or composite. Using IFU MaNGA data we are thus able to identify 23 new AGNs in dwarf galaxies whose AGN signatures are hidden (or partly hidden in the case of composite objects) or not available in the BPT of single-fiber SDSS spectra. The use of integrated spectra to search for AGNs can therefore fail in low-mass galaxies, biasing the demographics of dwarf galaxies hosting an AGN (see Section 3.6; e.g., Reines et al. 2013; Cann et al. 2019; Birchall et al. 2020).

3.1. Host Galaxy Properties

3.2. Other Wavelengths

3.3. Off-nuclear Wanderers, Hidden, or Switched-off AGNs?

3.4. AGN Luminosity

The bolometric luminosity (L_bol) of the 37 AGN candidates can be estimated from the [O iii] luminosity (${L}_{[{\rm{O}}{\rm{\small{III}}}]}$). The latter is derived from the median [O iii] flux of the AGN/LINER spaxels and converted to L_bol assuming a bolometric correction of 1000 (Moran et al. 2014). We find a range of log L_bol = 38.9–41.4 erg s⁻¹, which is below the typical AGN bolometric luminosity threshold of ∼10⁴² erg s⁻¹ and indicates that these are low-luminosity AGNs (e.g., Mezcua & Prieto 2014). We note that excluding those sources classified as AGNs by the SDSS BPT, the median bolometric luminosity would be log L_bol = 39.6 erg s⁻¹ (see Figure 3), which is nearly one order of magnitude lower than the median log L_bol = 40.5 erg s⁻¹ of the 14 SDSS AGNs. The L_bol is also more than two orders of magnitude lower than those of X-ray- and radio-selected AGNs (Figure 3; e.g., Mezcua et al. 2018a, 2019), which highlights the power of IFU spectroscopy in revealing very faint AGNs in dwarf galaxies.

Zoom In Zoom Out Reset image size

3.5. Black Hole Mass: Revealing a Hidden Type 1 Low-mass AGN

3.6. Demographics

Using MaNGA IFU data we have identified a sample of 37 AGN candidates in dwarf galaxies. In 23 of them the AGN ionization signatures are largely missed by single-fiber (SDSS) spectroscopy, which classifies them as either quiescent, star-forming, or composite. While this suggests that star formation is the dominant source of ionization in the central region, most of the galaxies are not star-forming. The AGNs could thus be either switched off or wandering in the host dwarf galaxy, as predicted by models of seed black hole formation. A black hole mass ∼10⁵ M_⊙ is derived for one of the 23 new AGN candidates based on the detection a broad Hα emission line component, which constitutes the first hidden type 1 low-mass AGN to be revealed by IFU data. The AGN bolometric luminosity of the 23 new AGN candidates is ≲10⁴⁰ erg s⁻¹, nearly one order of magnitude fainter than that of the dwarf galaxies with single-fiber AGN emission and several orders of magnitude lower than that of X-ray and radio AGN dwarf galaxies. IFU surveys offer thus a potential tool for identifying hidden and faint AGNs, which is crucial for population studies of AGNs in dwarf galaxies and for understanding whether these host the seed black holes of the early universe. The detailed kinematic an stellar population properties of the sample will be presented in a forthcoming paper, while radio and X-ray observations are planned to confirm the AGN nature.

The authors would like to thank they anonymous referee for his/her helpful comments. M.M. would like to thank A. Olmo-García for sharing her emission line fitting code, S. Sánchez for helpful discussions, and D. Wylezalek for sharing the stellar masses of her AGN sample. H.D.S. thanks M. Bernardi for her help with the stacking spectra procedure. M.M. acknowledges support from the Beatriu de Pinos fellowship (2017-BP-00114). H.D.S. acknowledges support from the Centro Superior de Investigaciones Científicas PIE2018-50E099.

For each individual galaxy, the spectra of all the spaxels that fall in the AGN/LINER region are stacked together. We use the HYB10-GAU-MILESHC maps from the MaNGA DR15, which provide the observed flux, the emission line component, the stellar continuum, and the best-fitting model. We stack both the observed flux (in blue in Figures 1, B1 and B2) and the emission line components (shown in red) to better visualize the emission line fluxes. The stacking procedure is performed as follows: the spectra are first rest-frame corrected (including the redshift and the correction due to the rotational velocity) and then the wavelengths are converted to air units following the standard convention. The flux is normalized in two regions free of emission lines in the blue and red part of the spectrum (4600–4700; 6780–6867 Å) to remove the intrinsic slope of the spectra. The stacked spectrum is computed as the median value of the 3σ clipped flux at each wavelength. The emission line component is shifted by 1 to match the observed flux. Skylines or masked wavelength regions are not used in the stacking (see Domínguez Sánchez et al. 2019 for additional details).

We perform a visual inspection of all the AGN/LINER candidates in order to check the consistency of the emission line measurements provided by the MaNGA data-analysis pipeline. The visual inspection of the stacked spectrum of 8442-1901 shows that the MaNGA emission line model seems not to correctly fit the emission lines used in the BPT diagnostic diagram, which could be biasing its BPT classification as an AGN (see Figure B1). This is likely due to the possible presence of a broad Hα emission line component in the stacked AGN spectrum of 8442-1901. To probe this, we perform a multi-Gaussian fit to the Hα+[N ii] complex (and also of the Hβ and [O iii] emission lines used in the BPT diagram) using the Python package LMFIT as in Olmo-García et al. (2017). We follow the common procedure (e.g., Reines et al. 2013) of assuming the same width for the [N ii]λ6583, [N ii]λ6548, [O iii]λ5007, [O iii]λ4958, narrow Hα, and narrow Hβ emission line Gaussian components and allowing for a secondary broad Hα and Hβ component. The relative separations between the centers of the narrow components are fixed using their theoretical wavelengths and the flux ratios of the [N ii] and [O iii] doublets are set to the theoretical values of 3.06 and 2.99, respectively. The best model (that with the lowest χ²) is obtained when including a broad Hα component but no Hβ broad component (see Figure B1, bottom). The resulting FWHM is corrected for instrumental resolution, thermal and natural broadening, which yields a FWHM (broad Hα) = 1131 ± 378 km s⁻¹. This, together with the luminosity (log L(Hα) = 38.6 erg s⁻¹) derived from the flux of the broad Hα component, allows us to derive a black hole mass under the assumption that the broad-line region from which the broad Hα originates is virialized. Using a geometrical factor  = 1 (Equation (5) in Reines et al. 2013), we find log M_BH = 5.07 M_⊙ (with a typical uncertainty of 0.5 dex). The resulting BPT classification, based on the derived narrow emission lines fluxes of the Hα, Hβ, [N ii]λ6583, and [O iii]λ5007, is shown as a yellow star in Figure B1 (top left panel) and confirms the AGN classification of 8442-1901.

Zoom In Zoom Out Reset image size

For 8446-1901 the MaNGA emission line model seems, as for 8442-1901, not to correctly fit the Hα+[N ii] emission line complex due to the presence of a broad Hα emission line component already present in the SDSS spectrum (see Figure B2). To confirm the AGN BPT classification of MaNGA we thus also perform a multi-Gaussian fit of the Hα+[N ii] complex that includes a broad Hα component (Figure B2, bottom). We find FWHM (broad Hα) = 3116 ± 205 km s⁻¹ and log M_BH = 5.73 M_⊙. Using the narrow emission line fluxes of Hα and [N ii]λ6583, 8442-1901 is classified as an AGN/LINER in the BPT diagram (yellow star in the top left panel of Figure B2).

Zoom In Zoom Out Reset image size

The BPT diagram and stacked spectrum of the 35 AGN dwarf galaxy candidates not shown here in the appendix are available in the figure set of Figure 1.