MULTI-EPOCH SPECTROSCOPY OF DWARF GALAXIES WITH AGN SIGNATURES: IDENTIFYING SOURCES WITH PERSISTENT BROAD Hα EMISSION

While massive black holes (BHs; here defined as ${\text{}}{M}_{\mathrm{BH}}\gtrsim {10}^{4-5}$ M_⊙) are ubiquitous in the centers of galaxies of Milky Way mass and larger (Magorrian et al. 1998), little is known about the occurrence rate and properties of BHs in dwarf galaxies (i.e., galaxies with ${\text{}}{M}_{\star }\lesssim {10}^{9.5}$ M_⊙). Studying the population of BHs in dwarf galaxies is useful for understanding BH formation and growth in the early universe. Having had relatively quiet merger histories, their BHs may have masses similar to their "birth" mass (e.g., Volonteri 2010; Bellovary et al. 2011).

Clues to the formation of BH seeds may reside in the well-studied relation between BH mass and stellar velocity dispersion (see, e.g., Ferrarese & Merritt 2000; Gebhardt et al. 2000; Gültekin et al. 2009). Using semianalytic modeling, Volonteri & Natarajan (2009) find that while the relationship between BH mass and velocity dispersion (or the ${\text{}}{M}_{\mathrm{BH}}-\sigma$ relation) is in place by $z\sim 4$ for massive halos, systems with lower-mass BHs (${\text{}}{M}_{\mathrm{BH}}\lt {10}^{6}\,{\text{}}{M}_{\odot }$) evolve onto it at later times. If BH seeds form "light" ($\sim {10}^{2}\,{\text{}}{M}_{\odot };$ the remnants of Population III stars; see, e.g., Madau & Rees 2001; Alvarez et al. 2009; Madau et al. 2014), then BHs begin as undermassive with respect to the local ${\text{}}{M}_{\mathrm{BH}}-{\sigma }_{\star }$ relation and will evolve toward it from below. Alternatively, if BH seeds form "heavy" (${10}^{4-5}\,{\text{}}{M}_{\odot };$ through direct collapse of gas clouds; e.g., Begelman et al. 2006; Lodato & Natarajan 2006), then BHs are initially overmassive and systems evolve rightward onto the relation (i.e., toward higher velocity dispersions) as the host galaxies grow through mergers. The smallest $z=0$ halos (${\sigma }_{\star }\sim 20\mbox{--}50\,\,\mathrm{km}\,{{\rm{s}}}^{-1}$) have not yet undergone major mergers and thus may be present-day outliers on the local ${\text{}}{M}_{\mathrm{BH}}-{\sigma }_{\star }$ relation.

Finding and weighing relatively low-mass BHs is difficult due to the small gravitational sphere of influence of the BH, which is typically ≲1 pc for ${\text{}}{M}_{\mathrm{BH}}\sim {10}^{5}$ M_⊙. Thus, in order to identify BHs in dwarf galaxies outside the Local Group, it is necessary to look for signs of BH activity (i.e., active galactic nuclei, or AGNs). Large-scale surveys such as the Sloan Digital Sky Survey (SDSS) have made it possible to search for signatures of AGNs in samples composed of tens of thousands of galaxies. Early work by Greene & Ho (2004) (see also Greene & Ho 2007) identified galaxies in the SDSS with low-mass AGNs by searching for broad Hα emission, which can be produced by virialized gas near the central BH. These studies identified ∼200 low-mass AGN candidates with a median BH mass of $\sim {10}^{6}$ M_⊙.

More recent studies have pushed the search for low-mass AGNs into the dwarf galaxy regime. X-ray and radio observations led to the discovery of AGNs in the dwarf starburst galaxy Henize 2-10 (Reines et al. 2011; Reines & Deller 2012) and in the dwarf galaxy pair Mrk 709 (Reines et al. 2014). Optical spectroscopic observations revealed the well-studied AGNs in NGC 4395 (Filippenko & Sargent 1989; Filippenko & Ho 2003) and POX 52 (Barth et al. 2004) and more recently led to the discovery of a BH with a mass in the range of 27,000–62,000 M_⊙ in the dwarf galaxy RGG 118 (Baldassare et al. 2015). Additionally, multiwavelength studies using large samples of bona fide dwarf galaxies have produced a collective sample of hundreds of AGN candidates in dwarf galaxies (Reines et al. 2013; Moran et al. 2014; Lemons et al. 2015; Sartori et al. 2015; Pardo et al. 2016).

Reines et al. (2013, hereafter R13) searched for dwarf galaxies with spectroscopic signatures of AGN activity based on their narrow and/or broad emission lines. Out of ∼25,000 nearby ($z\leqslant 0.055;$ D ≲ 250 Mpc) dwarf galaxies in the SDSS, they identified 136 galaxies with narrow-line ratios indicating photoionization at least partly due to an AGN based on the BPT (Baldwin et al. 1981) diagram, i.e., they were classified as either AGN or AGN/star-forming composites (see also Kewley et al. 2001, 2006; Kauffmann et al. 2003). A fraction of these also had broad Hα emission. Additionally, R13 identified 15 galaxies with broad Hα emission in their SDSS spectroscopy (FWHM in the range of $\sim 600\mbox{--}3700\,\,\mathrm{km}\,{{\rm{s}}}^{-1}$), but with narrow-line ratios that placed them in the star-forming region of the BPT diagram. We are particularly concerned with the origin of the broad emission in these systems.

Broad Balmer emission with FWHM up to several thousand kilometers per second can be produced by stellar processes such as Type II supernovae (SNe; Filippenko 1997; Pritchard et al. 2012), luminous blue variables (Smith et al. 2011b), or Wolf–Rayet stars. Follow-up spectroscopy, taken over a sufficiently long baseline, can help distinguish between an AGN and one of the aforementioned transient stellar phenomena.

Here we present follow-up optical spectroscopic observations of 14 of the 15 BPT star-forming galaxies with broad Hα presented in R13, as well as observations of 10 BPT AGNs (one with broad Hα) and three BPT composites (one with broad Hα). Follow-up observations for RGG 118—the current record holder for the lowest-mass BH in a galaxy nucleus—were presented separately in Baldassare et al. (2015). The goals of this work are twofold. First, for our targets that were identified by R13 to have broad Hα emission (broader than narrow lines such as [S ii] λλ6713 and 6731) with FWHM $\gt 500\,\mathrm{km}\,{{\rm{s}}}^{-1}$, we analyze their spectra and determine whether the broad Hα emission is still present and consistent with previous observations. Second, for galaxies with sufficiently high resolution follow-up spectroscopy, we measure stellar velocity dispersions.

In Section 2, we describe our observations and data reduction procedures. In Section 3, we discuss our emission-line fitting analysis and report stellar velocity dispersion measurements for objects with sufficiently high spectral resolution data. In Section 4, we present results from the follow-up spectroscopy for the broad-line objects.

The sample analyzed here is composed of 27 total targets. Of these, 16 compose our "SDSS broad Hα sample" or galaxies identified to have broad Hα emission in their SDSS spectroscopy.¹⁰ In the SDSS broad Hα sample, 14 objects are BPT star-forming, 1 is a BPT composite, and 1 is a BPT AGN. Additionally, we have follow-up observations for 11 objects (9 BPT AGNs and 2 BPT composites) that did not have broad Hα emission in their SDSS spectroscopy. In addition to the SDSS observations, each galaxy has one or two follow-up spectroscopic observations taken between 5 and 14 yr after the SDSS spectrum. Below, we outline our observations and data reduction procedures for each instrument used. See Table 1 for a summary of our sample and observations. All spectra were corrected for Galactic extinction.

2.1. Clay/MagE

2.2. APO/DIS

2.3. MDM/OSMOS

2.4. SDSS

3.1. Emission-line Modeling

For each galaxy, we modeled the Hα-[N ii] complex in order to ascertain whether broad Hα emission was present. We summarize the general procedure here, though more thorough discussions can be found in R13 and Baldassare et al. (2015). In this work, we are mainly interested in whether a broad Hα emission line is present. Thus, we simply fit the continuum as a line across the region surrounding the Hα-[N ii] complex. This also ensures that we do not artificially introduce a broad line by oversubtracting an absorption feature. We then create a model for the narrow-line emission using an intrinsically narrow line (e.g., [S ii] λλ6713 and 6731; forbidden lines are guaranteed not to be produced in the dense broad line region gas). We use the narrow-line model to fit the narrow Hα emission and the [N ii] lines simultaneously. The width of the narrow Hα line is allowed to increase by up to 25%. The relative amplitudes of the [N ii] lines are fixed to their laboratory values. We next add an additional Gaussian component to the model to represent broad Hα emission. If the preferred FWHM for the component is $\gt 500\,\mathrm{km}\,{{\rm{s}}}^{-1}$ and the ${\chi }_{\nu }^{2}$ is improved by at least 20%, then we classify the galaxy as having persistent broad Hα emission (Hao et al. 2005; R13). As noted in R13, this threshold is somewhat arbitrary, but found to be empirically suitable for this work. If the broad model is preferred, we attempt to add an additional broad component, effectively allowing the broad emission to be modeled with up to two Gaussians. Once we have our best-fit model (i.e., adding additional components no longer improves the ${\chi }_{\nu }^{2}$ by our threshold value), we measure the FWHM and luminosity of broad Hα, if present. We model and measure emission-line fluxes for Hβ and [O iii] λ5007. Below, we discuss the narrow-line models used for each instrument.

3.1.1. Narrow-line Models

MagE: For the MagE spectra, we follow R13 and model the narrow-line emission by fitting Gaussian profiles to the [S ii] λλ6713 and 6731 doublet. In our modeling, the two lines are required to have the same width.

OSMOS: The wavelength coverage of the OSMOS spectra does not extend to the [S ii] doublet, so here we model the narrow emission with the [O iii] λ5007 line. The profile of [O iii] often has a core component and a wing component (Heckman et al. 1981; De Robertis & Osterbrock 1984; Whittle 1985; Greene & Ho 2005a; Mullaney et al. 2013; Zakamska & Greene 2014). We therefore fit the [O iii] line with two components if the addition of a second component improves the ${\chi }_{\nu }^{2}$ by at least 20%. If [O iii] is indeed best fit with both a core and wing component, then we use the width of the core component for our narrow-line model. If it is instead best fit with a single component, we retain the width of the single component for our narrow-line model. In order to determine whether the use of [O iii] as our narrow line would affect our fits to the Hα-[N ii] complex, we compared the profiles of [S ii] and [O iii] in the SDSS spectra for our OSMOS targets. We find that, for a given object, the FWHMs of the [O iii] and [S ii] lines differ by very little, with percent differences in FWHM ranging from 0.3 to 11%. Since we allow the width of the Hα narrow line to increase by up to 25%, we do not expect this choice of narrow line to affect our results.

DIS: For the DIS spectra, we also use the [S ii] lines to model the narrow-line emission. However, the emission lines in the DIS spectra display some additional instrumental broadening, which we observe in the red channel arc lamp spectra (see Figure 1). Since this instrumental broadening is not visible in low-S/N [S ii] lines but can still affect the fit to Hα, we created a basic model for the narrow-line emission to use in fitting the [S ii] lines. Using the galaxy spectrum for which the [S ii] lines had the most flux, we generated a narrow-line model consisting of two Gaussians with fixed relative amplitudes and widths. This basic shape was then used to model [S ii] lines in all other DIS galaxy spectra and, correspondingly, their narrow Hα and [N ii] lines. Figure 2 shows the [S ii] lines from a representative DIS spectrum.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Given the instrumental broad component seen in the emission lines, we tested the degree to which the narrow-line model affects the presence/properties of broad Hα emission to ensure that our results were not influenced by this broadening. To do this, we created two additional narrow-line models: one where the instrumental broadening is less prevalent (i.e., we reduced the relative width and relative amplitude of the "broad" narrow-line component by 10%), and one where it is more prevalent (i.e., the relative width and amplitude are increased by 10%). We then refit the spectra and assessed the security of our broad Hα detections. We found this to have little effect on the FWHM and luminosity of broad Hα for secure detections (i.e., FWHM varied by less than 10%), but to greatly influence these measured quantities for more ambiguous detections (with the FWHM varying by as much as 90% for a slightly different narrow-line model). We err on the side of caution in our classifications; objects that have ambiguous broad Hα detections in their DIS spectra are classified as ambiguous.

3.2. Stellar Velocity Dispersions

We measured stellar velocity dispersions for 15 galaxies with MagE observations using Penalized Pixel Fitting (pPXF; Cappellari & Emsellem 2004). As a reminder, of these, 9 are classified as BPT AGNs, 3 are BPT composites, and 3 are BPT star-forming galaxies. Four galaxies (one composite and three star-forming) overlap with our SDSS broad Hα sample.

PPXF uses a library of stellar templates to fit the stellar continuum and kinematics of a galaxy spectrum. We used a library of 51 stars from the ELODIE spectral database (Prugniel & Soubiran 2001). Our library consists of stars covering spectral classes from O through M and luminosity classes from bright giants to dwarfs (see Table 5 in Appendix A for a list of stars and spectral types). The ELODIE spectra span from 3900 to 6800 Å in wavelength and have a spectral resolution of R = 10,000.

We measured stellar velocity dispersions in a region of the spectrum surrounding the Mg b triplet, spanning from 5100 to 5250 Å (see Figure 3). We fit each region using combinations of low-order multiplicative and additive polynomials (orders range from 1 to 4; see, e.g., Woo et al. 2010, 2015) and report the mean value for all measurements. Uncertainties are reported by pPXF for each individual fit (i.e., each combination of multiplicative and additive polynomial). We take the mean for all velocity dispersion measurements of the Mg b band and add the errors in quadrature to obtain a velocity dispersion measurement and corresponding error estimate. We measure ${\sigma }_{\star }$ values in the range of $28\mbox{--}71\,\mathrm{km}\,{{\rm{s}}}^{-1}$ with a median ${\sigma }_{\star }$ of $41\,\mathrm{km}\,{{\rm{s}}}^{-1}$. These values correspond to galaxies with NASA-Sloan Atlas stellar masses ranging from $\sim 5\times {10}^{8}$ to $3\times {10}^{9}$ M_⊙. All stellar velocity dispersion measurements are presented in Table 2

Zoom In Zoom Out Reset image size

.

Table 2.  Stellar Velocity Dispersion Measurements

R13 ID	${\sigma }_{\star }$
	($\,\mathrm{km}\,{{\rm{s}}}^{-1}$)
6	62 ± 12
16	33 ± 5
22	39 ± 6
27	28 ± 6
28	32 ± 4
29	49 ± 7
31	31 ± 4
33	46 ± 5
34	58 ± 3
119	28 ± 6
120	71 ± 8
128	43 ± 5
E	37 ± 12
G	41 ± 13
H	60 ± 11

Note. Stellar velocity dispersion measurements for objects with MagE data. Stellar velocity dispersions were measured in the Mg b region of the spectrum.

Download table as:  ASCIITypeset image

We show the locations of all 16 SDSS broad Hα targets on the BPT diagram in Figure 4. Based on our follow-up spectroscopy, we classify our targets as having SDSS broad Hα emission that is transient (broad Hα is not present in follow-up spectra), persistent (broad Hα is present in follow-up spectroscopy and consistent with prior observations), or ambiguous (state of broad emission is unclear). In the following subsections, we describe our classifications in more detail. Table 3 summarizes the results of this analysis. Figures showing the fits to Hβ, [O iii], and Hα are given for each observation in Appendix B. See Section 6 for a discussion on aperture effects and intrinsic variability as they relate to this work.

Zoom In Zoom Out Reset image size

Table 3.  Broad Hα Candidate Results

R13 ID	NSAID	R13 $\mathrm{log}{L}_{{\rm{H}}\alpha }$ (erg s−1)	R13 FWHM ${}_{{\rm{H}}\alpha }$ (km s−1)	R13 BPT Class	Follow-up	Δt (yr)	Persistent Broad Hα?
9	10779	40.15	703	AGN	DIS	13	yes
119	79874	40.16	1043	Comp	DIS, MagE	5	yes
C	109990	40.10	3690	SF	DIS, OSMOS	8	no
D	76788	39.56	4124	SF	OSMOS	8	no
E	109016	39.26	935	SF	MagE, OSMOS	9	no
F	12793	40.09	598	SF	OSMOS	14	no
G	13496	39.56	2027	SF	MagE	11	no
H	74914	39.97	3014	SF	MagE	8	no
J	41331	39.99	1521	SF	DIS	10	no
L	33207	39.51	3126	SF	DIS	11	no
M	119311	39.75	3563	SF	DIS, OSMOS	6	no
N	88972	40.42	645	SF	OSMOS	11	no
O	104565	38.88	1653	SF	OSMOS	9	no
B	15952	40.67	1245	SF	DIS, OSMOS	14	ambiguous
I	112250	39.39	994	SF	DIS, OSMOS	8	ambiguous
K	91579	39.58	774	SF	DIS, OSMOS	10	ambiguous

Note. Summary of our analysis for the SDSS broad Hα sample. All objects here were observed to have broad Hα in their SDSS spectroscopy in R13. Columns (1) and (2) list the R13 and NASA-Sloan Atlas ID for each object. Columns (3) and (4) list the luminosity and FWHM of the broad emission measured in R13. Column (5) lists the instrument(s) used for the follow-up spectroscopy. Column (6) lists the time between the SDSS observation and most recent follow-up observation. Column (7) presents the classifications determined in this work.

Download table as:  ASCIITypeset image

4.1. Transient Broad Hα

Out of the 16 objects with SDSS broad Hα, we find 11 to have transient broad Hα emission, i.e., they lacked broad emission in their follow-up spectroscopic observations (see Figure 5 for an example). All 11 fall in the "Hii" region of the BPT diagram, which indicates the presence of recent star formation in the galaxy. We note that several objects (e.g., RGG G, RGG F; see Appendix B) had additional components in their Hα line models with FWHMs of a few hundred kilometers per second; since these are all narrower than the broad emission in the SDSS spectra and fall below our FWHM cut of $500\,\mathrm{km}\,{{\rm{s}}}^{-1}$, we do not classify those objects as having persistent broad emission. Given the narrow emission line ratios and host galaxy properties, we consider Type II SNe to be the most likely origin for the transient broad Hα (see Section 5 for more details). We note that, as discussed in detail in R13, luminous blue variables (Smith et al. 2011b) and Wolf–Rayet stars can also produce transient broad Hα emission. However, Wolf–Rayet stars typically produce other notable spectral features, such as the Wolf–Rayet bump from 4650 to 5690 Å. Moreover, their spectra do not typically have strong hydrogen lines (Crowther 2007; Crowther & Walborn 2011). Finally, recent work has revealed the existence of "changing-look" quasars, i.e., quasars with broad Balmer emission lines that fade significantly over the course of 5–10 yr (see, e.g., Ruan et al. 2016; Runnoe et al. 2016). In the case of changing-look quasar SDSS J101152.98+544206.4, the broad Hα luminosity dropped by a factor of 55 on this timescale. This behavior is suspected to be caused by a sudden drop in accretion rate.

Zoom In Zoom Out Reset image size

4.2. Persistent Broad Hα

We identified 2/16 objects with secure, persistent broad Hα emission. RGG 9¹⁵ (NSA 10779) is classified as an AGN on the BPT diagram based on its SDSS spectrum. We observe broad Hα emission in both the SDSS spectrum and the DIS spectrum, taken 13 yr apart (see Figure 6 for the best fit to the DIS spectrum). The second secure broad-line object, RGG 119 (NSA 79874), is classified as a composite object on the BPT diagram. This object has broad Hα emission in the SDSS, MagE, and DIS spectra, spanning 5 yr (see Figure 7).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We compute the mass of the central BH using standard virial techniques that employ only the FWHM and luminosity of the broad Hα emission (Greene & Ho 2005b; Bentz et al. 2009, 2013). The calculation also includes a scale factor intended to account for the unknown geometry of the broad-line region; we adopt = 1 (see Equation (5) in R13). This method makes use of multiple scaling relations between parameters (e.g., the relation between FWHM_Hα and FWHM_Hβ), each of which has its own intrinsic scatter, giving a systematic uncertainty on the measured BH mass of 0.42 dex (Baldassare et al. 2015). To ensure that our choice of continuum subtraction method would not affect our BH mass estimates, we measured BH masses for the SDSS spectra using both R13 and our continuum subtraction and found the masses to be consistent with one another.

For RGG 9, we measure a BH mass from the SDSS observation of $3.5(\pm 0.7)\times {10}^{5}\,{{M}}_{\odot }$. From the DIS observation, we measure a BH mass of $3.7(\pm 0.3)\times {10}^{5}\,{\text{}}{{M}}_{\odot }$. Using these two measurements, we compute a mean mass of ${\text{}}{M}_{\mathrm{BH}}=3.6(\pm 0.8)\times {10}^{5}\,{M}_{\odot }$. Taking into account the systematic uncertainty of 0.42 dex, this gives us a final estimate of ${\text{}}{M}_{\mathrm{BH}}={3.6}_{-2.3}^{+5.9}\times {10}^{5}\,{\text{}}{M}_{\odot }$. Using the stellar mass of RGG 9 from R13 (${\text{}}{M}_{\star }=2.3\times {10}^{9}\,{\text{}}{M}_{\odot }$), we compute a ratio of BH mass to stellar mass of ${\text{}}{M}_{\mathrm{BH}}/{\text{}}{M}_{\star }=1.6\times {10}^{-4}$.

For RGG 119, we measure a BH mass of $2.1(\pm 0.5)\times {10}^{5}\,{\text{}}{{\rm{M}}}_{\odot }$ from the SDSS spectroscopy. The MagE observation yields a BH mass of $2.9(\pm 0.2)\times {10}^{5}\,{{M}}_{\odot }$, and the DIS observation gives a BH mass of $3.6(\pm 0.3)\times {10}^{5}\,{{M}}_{\odot }$. We compute a mean BH mass of ${\text{}}{M}_{\mathrm{BH}}=2.9(\pm 0.6)\times {10}^{5}\,{M}_{\odot }$ using the three spectroscopic observations. With the additional systematic uncertainty, our final BH mass estimate is ${\text{}}{M}_{\mathrm{BH}}={2.9}_{-1.8}^{+4.9}\times {10}^{5}\,{\text{}}{M}_{\odot }$. R13 reports a stellar mass of ${\text{}}{M}_{\star }=2.1\times {10}^{9}\,{\text{}}{M}_{\odot }$ for RGG 119, giving a ratio of BH mass to stellar mass of ${\text{}}{M}_{\mathrm{BH}}/{\text{}}{M}_{\star }=1.3\times {10}^{-4}$. See Table 4 for the measured broad Hα FWHM and luminosities and corresponding BH masses for each observation. We note that our measured BH masses are consistent with those measured in R13 ($2.5\times {10}^{5}\,{\text{}}{M}_{\odot }$ and $5.0\times {10}^{5}\,{\text{}}{M}_{\odot }$ for RGG 9 and RGG 119, respectively). We also refer the reader to Section 6 for a discussion of the potential effect of instrumentation/aperture on the measured properties of the broad Hα emission.

Table 4.  Broad Hα Parameters for Galaxies with Persistent Broad Hα

R13 ID	Obs.	$\mathrm{log}{L}_{{H}\alpha }$ (erg s−1)	FWHM${}_{{H}\alpha }$ (km s−1)	${\text{}}{M}_{\mathrm{BH}}({10}^{5}\,{\text{}}{M}_{\odot })$
9	SDSS	40.36 ± 0.01	744 ± 69	3.5 ± 0.7
9	DIS	${40.40}_{-0.02}^{+0.01}$	758 ± 27	3.7 ± 0.3
119	SDSS	40.36 ± 0.01	589 ± 69	2.1 ± 0.5
119	MagE	40.55 ± 0.01	616 ± 23	2.9 ± 0.2
119	DIS	40.59 ± 0.01	670 ± 27	3.6 ± 0.3

Note. Best-fit broad Hα FWHM and luminosity from each observation for objects found to have persistent broad Hα emission. We also present the corresponding BH mass calculated using the FWHM and luminosity. These values are all measured using the fitting code described in this paper, but the SDSS spectra are continuum subtracted using continuum fits obtained by R13, while the continua in the DIS and MagE spectra were modeled with a straight line.

Download table as:  ASCIITypeset image

Using the MagE spectrum of RGG 119, we measure a stellar velocity dispersion of $28\pm 6\,\,\mathrm{km}\,{{s}}^{-1}$ from the Mg b triplet and place this galaxy on the M_BH-${\sigma }_{\star }$ relation (Figure 8). RGG 119 sits close to the extrapolation of M_BH-${\sigma }_{\star }$ to low BH masses, similar to well-studied low-mass AGNs NGC 4395 (Filippenko & Ho 2003) and Pox 52 (Barth et al. 2004). Our BH-to-galaxy stellar mass ratios are also consistent with other low-mass AGNs (Reines & Volonteri 2015).

Zoom In Zoom Out Reset image size

4.3. Ambiguous Galaxies

We consider Type II SNe the most likely explanation for the fading broad Hα emission observed in 11/16 objects in our SDSS broad Hα sample. In this section, we further examine our SN II candidates and discuss this hypothesis in more detail. All 11 objects with transient broad Hα have narrow-line ratios consistent with recent star formation. Moreover, their broad emission properties and galaxy properties are offset from the remainder of the SDSS broad-line objects in two important ways.

As seen in Figure 9, SDSS broad emission that we found to be transient tended to have larger FWHM and broad lines that were more offset from the Hα line center in velocity space. All SDSS broad Hα emission lines with FWHM $\gtrsim \,2000\,\mathrm{km}\,{{s}}^{-1}$ in our sample were found to be transient, as well as all those with absolute velocity offsets of more than $\sim 200\,\mathrm{km}\,{{s}}^{-1}$.

Zoom In Zoom Out Reset image size

Additionally, galaxies with transient broad Hα also tend to be bluer with respect to the sample of broad- and narrow-line candidate AGN host galaxies from R13 (Figure 10). The narrow-line AGN candidates were found to have a median galaxy color of $g-r=0.51$ (R13), while the transient broad Hα galaxies have a median host galaxy color of $g-r=0.22$ (with all transient broad Hα galaxies having $g-r\lt 0.4$). This suggests that the transient broad Hα galaxies have, in general, younger stellar populations more likely to produce SNe II.

Zoom In Zoom Out Reset image size

Using time-domain spectroscopy, we have empirically identified targets most likely to be SNe based on the disappearance of broad emission lines over long ($\sim 5\mbox{--}10$ yr) timescales. A complementary analysis was performed by Graur et al. (2015) and Graur & Maoz (2013). They searched through most SDSS galaxy spectra for SN-like spectra and used SN template matching, rather than broad emission line fitting, to identify potential SNe. Here we compare the results of the two methods.

Similar to Graur et al. (2015) and Graur & Maoz (2013), R13 did identify nine SN candidates (in a sample of ∼25,000 objects) directly from the single-epoch SDSS spectra (Table 5 in R13; none of those have been followed up in this work). Those objects were identified because they exhibited P Cygni profiles in Hα. Of the 15 BPT star-forming galaxies with broad Hα identified in R13 as AGN candidates (14 of which are considered in this paper), one was detected by Graur et al. (2015) and classified as an SN II (RGG M, identified in Table 1 of Graur et al. 2015 as SDSS J131503.77+223522.7 or 2651-54507-488). The remainder did not meet all of the detection criteria set by Graur et al. (2015) and Graur & Maoz (2013), and thus were not detected by Graur et al. (2015). The Graur et al. (2015) and Graur & Maoz (2013), method is superior at detecting bona fide SNe II near maximum light, when these objects display prominent P Cygni profiles. The nature of the objects considered in this paper is generally more ambiguous, as no P Cygni profiles are seen and the emission-line profiles are relatively symmetric (although generally broader and more asymmetric than the bona fide AGNs; Figure 9).

During their survey, Graur et al. (2015) also identified five ambiguous broad-line objects in dwarf galaxies, which their pipeline classified as either SNe or AGNs. Of these, three were not in the R13 parent sample due to the stellar mass cut in two cases and a glitch in the NSA catalog in the third. Two are in R13; one was identified as a transient (Koss et al. 2014), and the other (RGG J) is included in this work.

To understand why the majority of the transient sources presented here were not identified by Graur et al. (2015) and Graur & Maoz (2013), and to try to learn something about their nature, we pass our continuum-subtracted SDSS spectra through the same SN identification scheme used by Graur et al. First, the stellar continuum and narrow emission line fits are removed. We employ a different procedure from that used by Graur et al. to fit the continuum, but the overall result should be similar. Second, we de-redshift the spectra, remove their continua, and manually ran them through the Supernova Identification code (SNID;¹⁶ Blondin & Tonry 2007). Four objects were classified as old SNe II, i.e., when their spectra are dominated by a broad Hα feature, but with relatively low rlap values. The SNID "rlap" parameter is used to assess the quality of the fit and is equal to the height-to-noise ratio of the normalized correlation function times the overlap between the input and template spectra in ${ln}\,\lambda$ space (see Blondin & Tonry 2007 for details). A value of rlap = 5 is the default minimum, and fits with rlap values close to 5 are regarded as suspect (see Blondin & Tonry 2007 for details). Of the four objects identified as SNe II, two had rlap values <5 and two had rlap values of ∼7. One object was classified as an old SN Ia at z ∼ 0.1. As the spectra were de-redshifted, this classification is not trustworthy. The other three objects failed to be classified by SNID. See Figure 11 for an example of the fit obtained by SNID to RGG C. We will discuss the special case of RGG J in some detail in Section 5.1.

Zoom In Zoom Out Reset image size

Since we consider SNe II the best explanation for the transient broad emission, in the following we compute the expected number of SNe II in the R13 parent sample and check for consistency with our results. R13 identified and removed nine galaxies with SNe II from their sample based on P Cygni profiles in the broad Hα emission. Considering the 11 objects with transient broad Hα and three ambiguous objects identified in this work, we have 20–23 SN II candidates. The Lick Observatory Supernova Search (Li et al. 2000) calculated the rate of SNe II in the local universe as a function of color and Hubble type by observing more than 10,000 galaxies over the course of 12 yr (Li et al. 2011a). Using representative SDSS gri magnitudes and $(g-r)$ and $(r-i)$ colors for the star-forming galaxies from R13, we use filter transformations (Jester et al. 2005; Bilir et al. 2008) to determine a typical (B–K) color in order to select the correct SN rate from Li et al. (2011a). Choosing a galaxy stellar mass of $1.2\times {10}^{9}$ M_⊙and using the equation for $(B-K)\lt 2.3$, we find that the expected rate of SNe II in a given galaxy over the course of 6 months is ${0.002}_{-0.001}^{+0.003}$. We then convert this to a probability using Poisson statistics. We choose to compute the rate over 6 months since SNe II typically begin to fade at 100–150 days post explosion (see, e.g., Kasen & Woosley 2009).

Similarly, Graur et al. (2015) measured a mass-normalized SN II rate of ${5.5}_{-2.4}^{+3.7}\,{statistical}{\,}_{-0.7}^{+1.2}\,\mathrm{systematic}\,\times {10}^{-12}\,{\mathrm{yr}}^{-1}\,{M}_{\odot }^{-1}$ in galaxies with stellar masses of ${0.08}_{-0.05}^{+0.06}\,\times {10}^{10}\,{\text{}}{M}_{\odot }$ (see their Table 2). Thus, for a galaxy of ${10}^{9}\,{\text{}}{{M}}_{{\odot }}$, we would expect ${0.0026}_{-0.0016}^{+0.0025}$ SNe II in a period of 6 months, consistent with the rate derived from Li et al. (2011a).

R13 analyzed spectroscopy of ∼25,000 dwarf galaxies. For a sample of this size, the rate derived from Li et al. (2011a) suggests that we can expect ${50}_{-25}^{+75}$ SNe II. Using the rate from Graur et al. (2015), we expect ${65}_{-40}^{+63}$ SNe II in a sample of 25,000 galaxies. Both rates are roughly consistent with the number we detect in our sample. Using either rate, we compute that in a sample of 25,000 galaxies, we expect less than 1 to have an observed SN signature in the original SDSS observation and a signature from a new SN in the follow-up observation.

5.1. A Possible SN IIn in RGG J

In the following, we discuss the interesting case of RGG J. This particular source, classified in this work as having transient broad Hα emission, has fooled many of us. It was identified as a broad-line AGN in Greene & Ho (2007), Stern & Laor (2012), and Dong et al. (2012). Thus, it serves as an excellent cautionary tale that there may well be transient sources lurking in our broad-line AGN samples even at higher mass.

We also have more epochs for this target than others. It was observed twice by the SDSS, once by Xiao et al. (2011) (spectrum taken in 2008), and once by us in 2013. The SDSS spectra were taken on 2003 March 11 and 2003 April 05 (see Figure 12). Thus, our observations span a full decade. The SDSS flagged the second epoch as the "primary" spectrum, and so the fits in the literature are most likely made to that later epoch. During their survey, Graur et al. (2015) classified this source as either an SN IIn or an AGN based on a fit to the April 2003 spectrum. By 2008, when Xiao et al. (2011) observed this source with the Echellette Spectrograph and Imager on Keck II, there was still a (rather marginal) detection of broad Ha. The broad line is undetectable in our DIS spectrum, so we conclude that the broad line may be produced by an SN IIn (Figure 12).

Zoom In Zoom Out Reset image size

Unlike SNe II-P and II-L, which exhibit broad H emission lines with P Cygni profiles, Type IIn SNe (IIn) are characterized by narrow lines (FWHM ≲ 200 km s⁻¹; hence the "n" in "IIn") with intermediate-width bases (FWHM ∼ 1000–2000 km s⁻¹) and bluer continua (see Filippenko 1997 for a review). SNe IIn compose ≈5% ± 2% of all SNe and show a wide spread of peak luminosity (Li et al. 2011b). The narrow lines are thought to be the result of strong interaction between the SN ejecta and a dense circumstellar medium. Though some SNe IIn have been associated with luminous blue variables (e.g., SN 2005gl, Gal-Yam & Leonard 2009; SN 2010jl, Smith et al. 2011a; SN 2009ip, Foley et al. 2011), the exact nature of their progenitors is still debated (see Smith 2014 for a review).

We analyzed follow-up spectroscopy of 27 dwarf galaxies with AGN signatures identified in a sample of ∼25,000 dwarf galaxies in the SDSS (see R13). Of our follow-up targets, 16 were found by R13 to have broad Hα emission in their SDSS spectra, and 14 had narrow emission line ratios indicative of AGN activity (there is some overlap between the broad- and narrow-line objects).

Of the 16 SDSS broad Hα objects for which we have follow-up spectroscopy, we found that the broad Hα emission faded for 11, all of which fall in the star-forming region of the BPT diagram. As stated above, we consider SNe II to be the most likely cause of the original broad emission for these objects, though luminous blue variables and changing-look quasars also produce transient broad Hα (see R13 for a more complete discussion of potential contaminants). We also find two of the SDSS broad Hα objects, both of which had narrow-line AGN signatures, to have persistent broad Hα emission, suggesting that it is emission from virialized gas around a central BH. This also suggests that the broad emission for the remainder of the R13 narrow-line AGN/composite objects is due to an AGN. R13 calculates the type 1 fraction (or the fraction with detectable broad Hα emission) for BPT AGNs to be ∼17% (6/35 objects). The type 1 fraction for BPT composites is lower; including RGG 118 (Baldassare et al. 2015), ∼5% of the composites (5/101) are type 1. Finally, three SDSS broad Hα objects (all BPT star-forming) are ambiguous and will require further observations to determine the source of broad emission.

In the case of RGG J (the expected Type II SN; see Section 5.1), the broad emission was persistent over at least 5 yr, but the FWHM and velocity shift varied considerably. Thus, in addition to determining whether broad emission is persistent, we also check for consistency between epochs. We note that differences in aperture could slightly affect broad emission measurements since the measured luminosity and FWHM of broad Hα depend on the fit to the narrow emission lines. Additionally, the instruments used all have different spectral resolutions, which can affect the measured FWHM. Finally, broad-line regions can vary intrinsically on these timescales (Tremou et al. 2015), making it difficult to determine the degree to which aperture effects play a role. Nevertheless, for RGG 9, we find the two measured FWHM to be consistent with one another within the errors. For RGG 119, the measured FWHM of Hα increases by ∼13% over the course of our three observations (though the first and last measurements are consistent within the errors).

In summary, we were able to securely detect broad Hα arising from an AGN in the follow-up observations of two SDSS broad Hα galaxies; both of these objects also have narrow-line ratios that support the presence of an AGN. Both objects (RGG 9 and RGG 119) have BH masses of a few times ${10}^{5}\,{\text{}}{{M}}_{\odot }$. Chandra X-ray Observatory observations (Cycle 16, PI: Reines) will provide further valuable information about the accretion properties of these AGNs. Three SDSS broad-line objects (all BPT star-forming) are classified as ambiguous and will require additional observations to determine whether an AGN is present. For the remainder of the star-forming SDSS broad Hα dwarf galaxies, we find that transient stellar phenomena are likely to be the source of detected broad Hα emission. Given the relatively low FWHMs, faint flux levels, and the prevalence of young stars, we confirm that, in dwarf galaxies (and perhaps more massive galaxies), broad Hα alone should not be taken as evidence for an AGN—another piece of evidence (such as narrow-line ratios) is required (see also discussions in R13 and Moran et al. 2014).

We also measure stellar velocity dispersions for 12 galaxies with narrow-line AGN signatures. One of these—RGG 119—also has persistent broad Hα emission. In this case, we use the broad emission to measure the BH mass and find that RGG 119 lies near the extrapolation of the ${{M}}_{{BH}}$–${\sigma }_{\star }$ relation to low BH masses. The measured stellar velocity dispersions for these targets are in the range of $28\mbox{--}71\,\mathrm{km}\,{{s}}^{-1}$ with a median value of $41\,\mathrm{km}\,{{s}}^{-1}$. With galaxy stellar masses of just a few times 10⁹, these stellar velocity dispersions correspond to some of the lowest-mass galaxies with accreting BHs. If the remaining narrow-line AGNs/composites similarly follow the ${{M}}_{{BH}}$–${\sigma }_{\star }$ relation, their BH masses would be in the range of $\sim 6\times {10}^{4}\mbox{--}2\times {10}^{6}\,{\text{}}{M}_{\odot }$ (using the relation reported in Gültekin et al. 2009).

V.F.B. is supported by the National Science Foundation Graduate Research Fellowship Program grant DGE 1256260. Support for A.E.R. was provided by NASA through Hubble Fellowship grant HST-HF2-51347.001-A awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS 5-26555. R.C.H. acknowledges support from an Alfred P. Sloan Research Fellowship and a Dartmouth Class of 1962 Faculty Fellowship.

This work makes use of observations obtained at the MDM Observatory, operated by Dartmouth College, Columbia University, Ohio State University, Ohio University, and the University of Michigan; observations obtained with the Apache Point Observatory 3.5 m telescope, which is owned and operated by the Astrophysical Research Consortium; and data gathered with the 6.5 m Magellan Telescopes located at Las Campanas Observatory, Chile. We thank Aaron Barth for providing us with the Keck ESI spectrum of RGG J.

In Table 5, we present the library of standard stars, which were used in measuring stellar velocity dispersions for objects with MagE spectroscopy.

Here we present our emission line fits for all available spectroscopy for each galaxy in our broad Hα sample. See Figure 13 for an example; the Figure Set contains fits for the entire broad Hα sample.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (16 panels)

Tarball of figure set images