CANDIDATE ACTIVE NUCLEI IN LATE-TYPE SPIRAL GALAXIES

There is now overwhelming evidence that the vast majority of massive early-type galaxies harbor a supermassive black hole (BH) at their centers, and that BHs play an important role during the assembly and evolution of their host galaxies (see reviews in Ho 2004). The existence of the M_BH–σ_⋆ correlation between BH mass and central stellar velocity dispersion links the growth of the BH with the formation of the host bulge (Gebhardt et al. 2000; Ferrarese & Merritt 2000), perhaps through feedback mechanisms associated with an active galactic nucleus (AGN). The detection of BHs can be achieved by several methods, including direct kinematics of surrounding gas and stars (e.g., Kormendy 2004; Ghez et al. 2005), and optical spectroscopic identification of nuclear emission lines originating from an AGN (e.g., Ho et al. 1995). The first method is limited to nearby, inactive or mildly active galaxies, however, while optical spectroscopic identification is generally limited to moderately bright nuclei that are not strongly confused by circumnuclear star formation or heavily obscured by dust.

An extensive spectroscopic survey of nearby galaxies has revealed that ∼60% of galaxies of Hubble type E − Sb contain an AGN (Ho et al. 1997b), broadly consistent with the idea that BHs are ubiquitous in these systems (Ho 2008). For types later than Sc, where classical bulges disappear (Kormendy & Kennicutt 2004), only ∼15% of galaxies show optical AGN activity. This leads to the notion that classical bulges or ellipticals are necessary for BH formation. There are, however, a few striking counter examples. NGC 4395, an essentially bulgeless Sm galaxy, exhibits all the hallmarks of a Seyfert 1 nucleus (Filippenko & Sargent 1989; Filippenko et al. 1993), and it has a BH mass of ∼10⁵M_☉ (Filippenko & Ho 2003; Peterson et al. 2005). POX 52 shares an almost identical optical spectrum to NGC 4395, but is hosted in a spheroidal galaxy (Barth et al. 2004)³, a visually similar but distinct morphological type compared to elliptical galaxies or classical bulges. More recently, an AGN was discovered in NGC 3621, a late-type Sd galaxy, using infrared spectroscopic observations (Satyapal et  al. 2007). By contrast, the nearby Scd galaxy M 33 has a firm upper limit on the mass of any central dark object of <1500 M_☉ (Gebhardt et al. 2001), far below the typical 10⁵ M_☉ BHs found in galaxies such as NGC 4395. Thus BHs can indeed exist in bulgeless systems, but exactly how common such systems are remains unknown.

From a spectral analysis of SDSS data, Greene & Ho (2007a) have found a population of low-mass BHs (M_BH ≈ 10⁵–10⁶ M_☉) accreting at appreciable Eddington ratios. The host galaxies from the initial sample of 19 sources (Greene & Ho 2004), using HST imaging, are found to be mid- to late-type spirals or spheroidals similar to POX 52 (Greene et al. 2008). None are as late-type as NGC 4395, however, and so the incidence of BHs in very late-type systems remains uncertain.

For such systems, there is also the possibility that BHs are replaced by nuclear star clusters (NCs). Both Ferrarese et al. (2006) and Wehner & Harris (2006) suggest that BHs in the M_BH–σ_⋆ relation are replaced by compact massive objects at low mass, with a generalized correlation M_CMO–σ_⋆. Such compact objects could include NCs instead of BHs, commonly found in late-type spirals (Böker et al. 2002, 2004). It is therefore not immediately obvious that such late-type spirals should contain BHs. Establishing a BH census for late-type spirals is important to determine if BHs are indeed formed in earlier-type galaxies only (ellipticals or early-type spirals), or if they also exist in very late-type galaxies.

One complication in late-type spirals is the abundance of gas and dust, with strong levels of star formation. If these systems contain BHs, they are likely to be of relatively low mass, and hence the AGN, even when fully active, will be of quite low luminosity. While the mere presence of an AGN indicates a BH, energetically weak or heavily embedded AGNs are extraordinarily difficult to observe, particularly at optical wavelengths where most of these observations have historically occurred. Dust obscuration severely attenuates optical and ultraviolet flux, and central star formation can dominate the optical/infrared emission.

X-ray observations allow us to peer through the obscuring gas and dust column, since high column densities are required to suppress X-ray flux (particularly energetic X-rays). X-ray detections are therefore a useful indicator of AGN activity, and since the launch of the Chandra X-Ray Observatory, the sensitivity with which to carry out such studies has been greatly improved. In particular, the Advanced CCD Imaging Spectrometer (ACIS; Garmire et al. 2003) provides better than 05 resolution with very low background contamination (∼10⁻⁷ counts pix⁻¹ s⁻¹ between 0.5 and 8 keV). This allows low flux limits to be achieved with very short (few ks) exposures, enabling surveys of non-X-ray selected samples (Ho et al. 2001). Gallo et al. (2008) have used X-ray detections successfully to identify BHs in low-mass bulges, and have begun to establish statistics on AGN X-ray activity as a function of host bulge mass. X-ray observations have also been useful in searching for nuclear activity in quiescent (i.e., optically faint) BHs residing in massive elliptical galaxies (Di Matteo et al. 2000; Loewenstein et al. 2001; Pellegrini 2005; Soria et al. 2006; Wrobel et al. 2008), and in star-forming early-type spiral galaxies (Tzanavaris & Georgantopoulos 2007).

In this paper, we investigate the nuclear activity in a sample of late-type spiral galaxies using archival Chandra data, which provide a clearer method of detecting AGN activity in such star-formation-dominated, potentially heavily obscured environments. Systems with nuclear X-ray detections coincident with the optical nucleus are identified as AGN candidates, and after considering the possible contamination of our results from X-ray binaries, we estimate the prevalence of AGNs in late-type spiral galaxies. We discuss our sample selection in Section 2 and our results in Section 3. We conclude with a discussion of possible non-AGN sources and directions for future work in Section 4.

All data were obtained from the Chandra Data Archive, with the full list summarized in Table 1. We initially selected objects from the Palomar spectroscopic survey of nearby galaxies (Ho et al. 1997a) with morphological T types between 6.0 and 9.0 (representing Hubble types Scd to Sm, i.e., bulgeless or nearly bulgeless spiral galaxies), which were available in the archive. The Palomar survey contains the most sensitive and comprehensive statistics on local AGNs to date, but the reliability of optically derived statistics in late-type spirals is uncertain due to dust and star-formation contamination. We can therefore complement the survey with X-ray data. Out of 63 objects in the Palomar survey with T types between 6.0 and 9.0, 33 are present in the Chandra archives. The 33 Palomar objects in our sample have a median absolute magnitude of M_B = −19.0 (corrected for Galactic and internal extinction), and a median distance of 7.8 Mpc.

If we do not restrict ourselves to Palomar objects, we can expand our sample significantly. From the master list of all public Chandra observations (as of May 2008), we searched for any galaxies in the HyperLeda⁴ database within 4' of the aim point (i.e., half a chip size for ACIS-S). After restricting our search to objects with T type between 6.0 and 9.0 (de Vaucouleurs et al. 1991) and within 30 Mpc (Tully 1988), our sample grows to 64 objects. These Palomar and non-Palomar subsamples are denoted by "P" and "C," respectively, in Table 1. When multiple observations exist for a given object, we generally chose the longest ACIS-S exposure if possible, otherwise opting for the newest observation. In two cases ACIS-S images were not available. For M 33, we chose the HRC-S observation, since the detector does not suffer from pileup as a result of the high count rate (see below for more discussion on pileup).

Because our sample is selected only on the basis of availability within the Chandra Data Archive, there is a potential for selection biases (based on, for instance, the selection criteria of the original proposals)—although these objects were not necessarily targeted for their nuclear properties. In Figure 1, we compare the distributions of T type between our sample and the original sample of 63 Palomar objects from Ho et al. (1997a). The distributions agree reasonably well, although our sample is perhaps weighted slightly toward the most late-type spirals (i.e., T = 9.0).

Zoom In Zoom Out Reset image size

With two exceptions, all data were obtained using the Chandra ACIS detector, in either the I- or S-array. The I-array consists of four CCDs arranged for imaging, while the S-array consists of six CCDs used for either imaging or grating spectroscopy. The mirror assembly has a point-spread function (PSF) with a full width at half-maximum (FWHM) of 05, equal to the ACIS pixel size. At a distance of 10 Mpc, this corresponds to a physical scale of ∼25 pc. Two objects were observed with the high-resolution camera (HRC), which has 04 pixels.

We used type 2 event files output by standard pipeline processing. We used 0.2–8 keV counts in our analysis, limiting background contamination at high energies. All data were corrected for known astrometry offsets, following standard CIAO threads⁵ (using CIAO version 3.3). We then used celldetect, with default parameters, to automatically find all point sources within each image and determine the centroids. These detections have absolute astrometry accurate to 1'', although it can be marginally worse for off-axis objects with larger PSFs and few counts. From these centroids, we measured background-subtracted counts in a 2'' radius aperture using the CIAO task dmextract, with a background annulus ranging from 5'' to 10''. In some cases, the background region was slightly altered to avoid nearby point sources. For NGC 3079 and 4945, the aperture is 1'' to avoid nearby sources and strong background, and for NGC 3184 the aperture is 3'' since this object is far off-axis (and thus has a large PSF). A 2'' aperture ensures that we measure virtually all of the point-source counts, since the on-axis PSF of Chandra contains 95% of the encircled energy with 1''.

Finally, we used Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) positions (obtained from the NASA/IPAC Extragalactic Database (NED)), with typical 1 σ uncertainties of ∼05 for the photometric center of extended sources (Jarrett et al. 2000), to determine whether X-ray point-source detections were coincident with the nucleus. While the nominal accuracy for 2MASS is often quoted as 01, this only applies to point sources. The distribution of offsets between the nearest X-ray detection and photometric center from 2MASS is highly bimodal, and we chose to consider objects belonging to the smaller-offset group as AGN candidates. Our formal criterion requires that the 2MASS positions and X-ray detections lie within 5'' of each other. In practice almost all candidates have better than 25 agreement, while nearly all nondetections have no X-ray point sources within 10'' of the 2MASS position. (NGC 4625 has an X-ray source within 6'' of the optical nucleus, while NGC 3395, NGC 3556, NGC 4038, and NGC 4631 have an X-ray source within 10'' of the optical nucleus.) The minimum X-ray signal-to-noise (S/N) threshold accepted was 1.5, although in practice only three sources were detected with such low S/N. We detected 18 X-ray point sources in our sample of 64 objects, 12 sources with S/N > 3, three sources with 3 > S/N > 2, and three sources with 2 > S/N > 1.5.

For all AGN candidates except IC 342, M 33, NGC 55, and NGC 4395, we estimate the total X-ray luminosity similarly to Ho et al. (2001), assuming a photon index of Γ = 1.8, typical of most low-luminosity AGNs (see Ho 2008), and an average Galactic interstellar medium column density of N_H = 2 × 10²⁰ cm⁻². This gives L_X(2–10 keV) = 3.6 × 10³⁷ erg s⁻¹ (ACIS counts ks⁻¹) (D/10 Mpc)². This is only valid, however, for ACIS-S data. For IC 342 and NGC 55 (i.e., HRC-I and ACIS-I data), we use the PIMMS tool⁶ (version 3.9d) with the same assumptions to obtain the 2–10 keV luminosity. The luminosities would increase by a factor of 1.8 and 6.7 for intrinsic column densities of obscuring gas and dust of 2 × 10²¹ and 2 × 10²² cm⁻², respectively. For M 33 (HRC-S data), we again use PIMMS but assume N_H = 2 × 10²¹ cm⁻² (Takano et al. 1994; Parmar et al. 2001; Dubus & Rutledge 2002; La Parola et al. 2003). NGC 4395 is a known AGN with strong absorption (N_H ≈ 2 × 10²² cm⁻²; Iwasawa et al. 2000; Moran et al. 2005), and so we use this value instead to estimate L_X. None of our ACIS objects suffer from significant pileup (multiple detections of lower energy photons within a single frame time that is registered as a single event at higher energy, possibly even rejected as a cosmic ray hit). NGC 4395 has the highest count rate of 7.5 × 10⁻² counts s⁻¹, which corresponds to a pileup of only 2% (calculated using PIMMS), roughly equal to the measured uncertainty in total counts.

We obtained an upper limit for those sources without an X-ray point-source detection. We used a circular aperture with a 2'' radius, centered at the optical position of the nucleus. We measured the background counts with its uncertainty, and determined⁷ the counts necessary for a theoretical source to have a S/N equal to 1.5 and thus be detected by celldetect. These counts were then converted to a luminosity as described above. In some cases, because of the very low exposure times, no counts were detected in this aperture size. In these cases, we used larger apertures and scaled down to a 2'' aperture, up to a 12'' aperture in the most extreme cases. Fields were sparse enough that no contamination from nearby X-ray point sources was present. For objects with a detected X-ray point source, the limiting luminosity was similarly found using the background measurements, scaled to a 2'' aperture.

As a proxy for spectral shape, we compute X-ray colors for those objects with detected X-ray sources, with counts measured in three bands: soft (S, 0.3–1 keV), medium (M, 1–2 keV), and hard (H, 2–8 keV). The soft and hard colors are then defined to be H₁ ≡ (M − S)/T and H₂ ≡ (H − M)/T, respectively, where T is the total counts in all three bands. This was only achievable with ACIS data, which have accurate registration of incoming photon energy, whereas the HRC instrument has a spectral resolution of only E/ΔE ≈ 1.

Four objects in our sample have sufficient counts (≳200) to do a preliminary spectral analysis: NGC 4039, NGC 4395, NGC 4490, and NGC 6946. NGC 4945 has most of its counts in a strong Fe Kα feature at 6.4 keV, resulting in few continuum photons. To accomplish this analysis, we used the CIAO task psextract to extract the X-ray spectra along with the associated response files. The aperture and background used are the same as those for the X-ray photometry. Bins are selected to have a minimum of 20 counts. Using XSPEC (Arnaud 1996), we then fit simple models to the data, ignoring bins with energies below 0.3 keV and above 5 keV (where there are few counts). In the case of NGC 4395, a strong source, we ignored bins with energies below 1 kev (there is a soft excess at low energies; see also Moran et al. 2005) and above 10 keV. We tested various models, including an absorbed power law, a blackbody, and an absorbed power law + blackbody.

Out of our sample of 64 objects, we detect 18 sources coincident with the optical position of the nucleus. We consider these as AGN candidates (except for M 33, which we discuss in Section 4). Our full sample is summarized in Table 1, while the AGN candidates are summarized in Table 2, along with their measured X-ray colors. Their images are presented in Figure 2. Of the 17 candidates, three objects are known AGNs from optical diagnostics (NGC 3079, NGC 4395, and NGC 4945, although they are weak sources), and one more is a known low-ionization nuclear emission-line region (LINER; NGC 6503). NGC 4654, NGC 4701, and NGC 5457 were detected above a S/N of 1.5. NGC 925, NGC 4713, NGC 6503 were detected above a S/N of 2. The rest of the AGN candidates were detected with a S/N above 3. NGC 45 has a possible nuclear X-ray source coincident with the optical position, but it was not detected above a S/N of 1.5, and thus is not included in our candidate list.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The 17 AGN candidates have 2–10 keV luminosities ranging from L_X = 10^37.1 to 10^39.6 erg s⁻¹. The distribution of L_X is shown in Figure 3. The three known AGNs (NGC 3079, NGC 4395, and NGC 4945) have L_X = 10^38.2, 10^39.6, and 10^37.7 erg s⁻¹, respectively. The median offset between the X-ray detection and the optical/near-infrared position is δ = 16. For those objects with enough counts, our X-ray spectral fits are summarized in Table 3 and Figure 4, and are discussed below.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For the AGN candidates with available measurements of σ_⋆ in the literature, we estimate a possible BH mass using the M_BH–σ_⋆ relation of Tremaine et al. (2002). We also estimate the corresponding X-ray luminosity to Eddington luminosity ratio. These are presented in Table 2.

The evidence for the majority of these candidates being true AGNs is circumstantial. For most of these systems, the physical scale of an X-ray point source in Chandra data is of order tens of parsecs, making it extremely unlikely that the vast majority of detections are serendipitous field X-ray binaries along the line of sight to or near the nucleus. Furthermore, the probability of detecting a background/foreground X-ray source within the Chandra PSF is extremely low, of order 10⁻⁷ (Gallo et al. 2008). With a few exceptions, the positional agreement between the X-ray source and the 2MASS position of the nucleus is excellent (less than 2'', compared to the Chandra and 2MASS positional uncertainties of ∼1''). NGC 4945 has a larger offset (δ≈ 32), but the presence of bright diffuse X-ray emission makes point-source detection more challenging, and once again this object is a known Seyfert 2.

Many of these galaxies have H ii nuclei as identified in the Palomar survey, but most of the optical observations have slit apertures of 2'' × 4'' (Ho et al. 1995), much larger than the Chandra PSF. The Palomar data are therefore probing a much larger region than the X-ray point sources, and we must be careful when making comparisons. As these are gas- and dust-rich systems, there is clearly some level of star formation in the circumnuclear region. If we consider, for example, typical Hα luminosities emitted by such low-luminosity AGNs, they are significantly lower than the measured Hα luminosities in the Palomar data. Assuming that the bolometric, X-ray, and Hα luminosities of AGNs are related by L_bol = 16L_X = 220L_Hα (Ho 2008), a range L_X = 10³⁷–10⁴⁰ erg s⁻¹ corresponds to L_Hα = 7 × 10³⁵–7 × 10³⁸ erg s⁻¹. Palomar objects in our sample with H ii nuclei have L_Hα = 10^37.5–10^38.8 erg s⁻¹, which can explain why such low-luminosity AGNs, if present, were missed in the Palomar data.

The galaxies in our sample are low-mass, late-type spirals, and thus likely harbor an NC at the center of the gravitational potential well. Böker et al. (2002, 2004) found that ∼75% of late-type spirals (T = 6–9) host a compact, photometrically distinct source located near the photometric center of the galaxy, with the surface brightness profile deviating from a pure disk profile. These NCs have mean effective radii of r_eff = 3.5 pc and I-band luminosities of L_I = 10^6.2 L_☉. Rossa et al. (2006) found that for NCs in galaxies within a similar range in T type, the average cluster mass is M = 10^6.25 M_☉. The increased stellar density at the center therefore implies an increased chance of detecting, in particular, a low-mass X-ray binary (LMXB) as a nuclear X-ray point source, confusing our results. M 33, for example, likely has no central massive BH (Gebhardt et al. 2001), but does possess an X-ray source coincident with the optical nucleus, with properties suggestive of an X-ray binary (Dubus et al. 2004). This is why we do not included M 33 as part of our AGN candidates. NGC 2403 also hosts an X-ray binary in its known NC (Yukita et al. 2007), although the X-ray luminosity is not high enough to be confused with an AGN (in this work, no detection was found above the limiting luminosity of 10^36.1 erg s⁻¹). X-ray binaries must therefore be carefully considered when attempting to detect low-luminosity AGNs. Although NCs also occur frequently in more massive early-type spiral galaxies (Carollo et al. 1997, 1998) and intermediate-luminosity elliptical galaxies (Côté et al. 2006), LMXB contamination is not important in these cases because at these larger BH masses L_X > 10⁴⁰ erg s⁻¹ due to nuclear activity, much greater than typical LMXB X-ray luminosities.

An X-ray detection in an NC is much more likely to be an LMXB rather than a high-mass X-ray binary (HMXB) because the mean age of NCs in late-type hosts is ∼10^8-9 yr (Rossa et al. 2006), older than typical high-mass stars in HMXBs. Although many late-type galaxies have H ii nuclei (Ho et al. 1997a), the slit aperture used, in most cases, is significantly larger than the Chandra PSF. Given that nearly all galaxies in the survey exhibit some emission lines in their nuclear spectra, and that late-type galaxies in particular are gas and dust rich with strong levels of star formation, the prevalence of H ii nuclei is not surprising. This does not necessarily imply, however, that the NCs have ongoing star formation (and therefore HMXBs are more likely), as evidenced by the inferred NC ages from Rossa et al. (2006).

For our purposes, we will approximate the NCs as globular clusters (GCs) in order to estimate the incidence of LMXBs. Although the environments of both types of clusters are widely different, NCs and GCs share remarkably similar structural and dynamical properties. Viewed in the fundamental plane (relating effective radius, surface density, and velocity dispersion), NCs occupy a region very distinct from classical bulges and elliptical galaxies, and fall on the high-mass extension of the GC sequence (Walcher et al. 2005). With some important exceptions (see below), approximating NCs as GCs is therefore well justified.

As in Gallo et al. (2008), we can estimate the expected number of LMXBs brighter than 3.2 × 10³⁸ erg s⁻¹ for a GC (and thus NC) of given mass, size, and g − z color, using an expression derived by Sivakoff et al. (2007):

$$\begin{equation} n_{\rm X} = 8 \times 10^{-2} \left(\frac{M}{10^6 \;M_\odot } \right)^{1.237} 10^{0.9(g-z)} \left(\frac{r_{\rm h,corr}}{1 {\rm \ pc}} \right)^{-2.22}. \end{equation} \tag{ 1 }$$

The quantity r_h,corr specifies the corrected half-mass radius, where r_h,corr = r_h × 10^{0.17[(g − z)−1.2]}. Since this depends rather weakly on color, we can approximate r_h,corr ≈ r_e = 3.5 pc as a mean value for NCs in late-type spirals. Rossa et al. (2006) find a mean B − V color for such NCs of order ∼0.5–0.6 mag. In order to get g − z colors, we use approximate color transformations from Fukugita et al. (1995) by approximating the stellar population in an NC as that between an Scd and an Sbc galaxy, which have B − V colors similar to those of NCs. This gives g − z colors ranging from 1.05 to 1.17 mag; we adopt g − z = 1.1 mag. Therefore, the mean number of LMXBs we expect is n_X ≈ 0.1. This estimate is specifically for sources brighter than ∼10³⁸ erg s⁻¹, which corresponds to the completeness limit of Sivakoff et al. (2007). The limiting luminosity for our X-ray detections varies from L_X = 10^36.2 to 10^38.1 erg s⁻¹. For objects in the Sivakoff et al. sample with a lower limiting luminosity (∼10³⁷ erg s⁻¹ or lower), however, they detected only a factor of a few more sources.

An important caveat to this estimate is the question of stellar age. GCs generally have old stellar populations, formed via a single burst of star formation, whereas NCs have repeated star-formation episodes since the reservoir of gas can be replenished. This leads to, for a given cluster mass, a younger mean stellar age for NCs compared with GCs, and thus bluer colors (Rossa et al. 2006; Walcher et al. 2006). As a result, the expected number of LMXBs estimated above is an upper limit, since it was derived for GCs, and the true number of LMXBs in NCs is likely lower. This is due to the ∼1 Gyr timescale necessary for the stellar companion to come into Roche lobe contact with its BH companion (and thus turn on the LMXB), either by stellar evolution or via orbital decay (White & Ghosh 1998).

Using n_X ≈ 0.1 as our estimate, this would imply only ∼5 X-ray detections from LMXBs in a sample of 64 late-type spirals (assuming 75% have NCs with mean properties), of which M 33 is a typical example. This is a factor of 3–4 lower than our observed detection rate. The observed X-ray luminosities in our sample are similar to the known AGNs, making our other detections reasonable AGN candidates. This implies that NCs and BHs need not be necessarily mutually exclusive, if the majority of our AGN candidates are true AGNs. Indeed, Seth et al. (2008) find that over all Hubble types roughly 10% of all galaxies hosting an NC also host an AGN, based on optical spectroscopy, with a fraction that increases strongly with galaxy and NC mass.

It is important to note that these detections are AGN candidates, and are worthy of further detailed observations to properly identify these sources. Ghosh et al. (2008) provide some evidence from a multiwavelength analysis that NGC 3184, NGC 4713, and NGC 5457 are true AGNs. Since most of our candidates have very few counts, long-integration X-ray spectroscopy will be useful in differentiating true AGNs from LMXBs, particularly at very soft and hard energies. As described above, AGNs have X-ray spectra composed of an absorbed power law (with Γ ≈ 1.8), whereas X-ray binaries have spectra that can be separated into three states: a high/soft state with a ∼1 keV thermal component and a weak nonthermal tail at high (> 10 keV) energies; a very high state with a steep power law (Γ ≈ 2.5) combined with a ∼1 keV thermal blackbody, which can constitute a significant amount of the total flux; and a low/hard state with a single power law (1.4 < Γ < 2.1) (Remillard & McClintock 2006). Of these three states, only the low/hard state could be confused with an AGN, but X-ray binaries in these states typically have extraordinarily low luminosities (L_X = 10^30.5–10^33.5 erg s⁻¹), which would distinguish them from true AGNs.

Zoom In Zoom Out Reset image size

Our spectral fits (Table 3 and Figure 4) are insufficient to definitively rule out one model over another, though they remain very suggestive. For NGC 4039, a pure thermal blackbody is likely ruled out, whereas an absorbed power law is plausible, and an absorbed power law + blackbody model (indicative of a very high state LMXB) is consistent with the data. Given that the reduced χ² of the latter model is <1, however, we may be overfitting the data, and so we must be cautious. Although it may appear as though this is more likely an X-ray binary (of extremely high luminosity >10³⁹ erg s⁻¹), longer-integration data are needed to distinguish between the two models.

NGC 4490 and NGC 6946 are both well fitted by simple absorbed power law models (χ² ≈ 1), consistent with the AGN interpretation. In the case of NGC 4490, the unabsorbed blackbody model results in a poorer fit, particularly at low energies, while for NGC 6946, both models are consistent with the data. Unfortunately, the data are not of high enough quality to differentiate between models with more parameters, such as an absorbed power law + blackbody, since χ² ≲ 1 in all cases. Once again, these results underscore the need for higher S/N data if we are to use a χ² statistic to differentiate between various models. As a check, we also analyzed NGC 4395, a known Seyfert 1, which is unsurprisingly well fitted by an absorbed power law, whose parameters are consistent with those given by Moran et al. (2005).

In Figure 5, we show the X-ray colors of those AGN candidates with sufficient counts. We also include various absorbed power-law models and a 1 keV blackbody (representative of the high/soft state from Remillard & McClintock 2006), computed for observations with ACIS-S using XSPEC. All AGN candidates are consistent with absorbed power-law models, and inconsistent with the 1 keV blackbody. Although the uncertainties are large, it appears as though most of these candidates exhibit softer X-ray spectra than is typically assumed for low-luminosity AGNs. Low-mass AGNs can exhibit a range of X-ray spectra, however, with Γ≈ 1.0–3.0 (Greene & Ho 2007c). Clearly, detailed X-ray spectroscopy is needed to further characterize these systems.

Once AGNs are positively identified, determining a BH mass estimate will be challenging. Dynamical measurements are currently not feasible in these objects. Reverberation mapping (Peterson 2007) or virial estimates (Greene & Ho 2005) based on the strength of optical emission lines (if detected) will likely require excellent seeing conditions with a narrow slit or adaptive optics. As these are all nearly bulgeless systems, using the bulge velocity dispersion σ_⋆ or the bulge luminosity L_B (Marconi & Hunt 2003) to determine M_BH is challenging and highly uncertain. NCs with BHs, however, fall on the M_BH–σ_⋆ relation—for example NGC 4395 (Filippenko & Ho 2003; Peterson et al. 2005) and G1 (Gebhardt et al. 2002, 2005)—consistent with the extrapolation of M_BH–σ_⋆ to lower masses (Barth et al. 2005). The velocity dispersions of NCs (Walcher et al. 2005) could therefore be used to estimate the BH mass. BH masses can also be estimated from X-ray variability timescales derived from complete power density spectra (McHardy et al. 2006), obtained over long monitoring campaigns. Alternatively, the BH mass can be estimated from the excess variance measured in shorter X-ray data sets (Nikolajuk et al. 2004; Gierliński et al. 2008).

Our detection rate suggests an AGN fraction in late-type spiral galaxies of ∼20–25% (depending on the LMXB contamination), higher than that observed by Ho et al. (1997b). (We should caution, however, that several detections are at low S/N. Using only S/N > 3 detections, our AGN fraction becomes ∼17%.) This would imply that we are indeed missing many low-mass AGNs in optical surveys, and that BHs in such systems are more prevalent than previously suggested. Optical surveys of low-mass AGNs can only find systems with high accretion rates, such that their bolometric luminosity L_bol approaches a substantial fraction of the Eddington luminosity L_Edd ≡ 1.26 × 10³⁸(M_BH/M_☉) erg s⁻¹ (Greene & Ho 2007b). Greene & Ho (2007a) found a median L_bol/L_Edd = 0.4 and a minimum L_bol/L_Edd ≈ 10⁻² in a sample of 174 optically selected low-mass AGNs. In a pilot study drawn from this sample, Greene & Ho (2007c) measured L_X(0.5–2 keV) ≈ 10⁴¹–10⁴³ erg s⁻¹, much higher than any of our measurements. Most of our candidates with available M_BH estimates exhibit L_bol/L_Edd ratios orders of magnitude lower than those of Greene & Ho, assuming a correction factor of 16 to convert from L_X to L_bol (Ho 2008). The one exception is NGC 4395, a known Seyfert nucleus, with L_bol/L_Edd ≈ 10^−2.1, consistent with the value of 2 × 10⁻³ to 2 × 10⁻² found by Filippenko & Ho (2003). In comparison, POX 52 has L_bol/L_Edd ≈ 1 (Barth et al. 2004.) Our estimated Eddington ratios are similar, however, to those seen in massive BHs in earlier-type hosts with low-luminosity AGNs (Ho 2008), suggesting that we are detecting the low-mass analogs of weakly accreting BHs. X-ray surveys therefore allow us to probe much fainter down the luminosity function of low-mass AGNs, detecting weakly accreting systems that would be completely swamped by galactic emission at optical wavelengths, and obtain a much higher detection rate. The Greene & Ho sample, for instance, identified only 174 low-mass AGNs out of 8435 optically identified AGNs (from a parent sample of ∼ 600,000 galaxies).

Given the low-estimated accretion rates of our AGN candidates, radio observations can help to clarify the nature of these X-ray sources. If these are indeed low-mass and low-accretion-rate systems, we should expect them to have a radio signature, much like the more massive low-luminosity BHs (Ho 2008). Extrapolating from higher masses, we would expect 2 cm radio fluxes of order 10–100 μJy (Nagar et al. 2002). If these X-ray sources are X-ray binaries instead, however, they would be accreting at Eddington or super-Eddington rates to achieve such high luminosities. X-ray binaries in such states do not have a radio counterpart (Remillard & McClintock 2006).

Satyapal et al. (2008) have conducted a similar survey of AGNs in the infrared, searching for emission-line signatures of AGN activity (in particular, Ne v emission; see Abel & Satyapal 2008) in late-type galaxies that are optically quiescent. This should identify weakly accreting systems similar to our candidates. In a sample of 32 objects, nine of which overlap with our Chandra sample, they find seven AGN candidates, a similar detection fraction. Interestingly, of the nine overlap objects, six (IC 342, NGC 925, NGC 3184, NGC 4490, NGC 4559, and NGC 6946) are identified in this paper as AGN candidates, but not by Satyapal et al. It is possible that the weakly active nuclei picked out by our X-ray search method are too faint to emit much Ne v emission. We can estimate the total Ne v luminosity L_NeV for our sample assuming the bolometric luminosity L_bol = 16L_X and L_bol = 9L₅₁₀₀ (Ho 2008), where L₅₁₀₀ is the continuum luminosity at 5100 Å, coupled with the correlation between L₅₁₀₀ and L_NeV found by Dasyra et al. (2008). For our range in X-ray luminosities (L_X = 10³⁷–10⁴⁰ erg s⁻¹), this corresponds to Ne v luminosities of L_NeV = 10³⁵–7 × 10³⁷ erg s⁻¹. These luminosities are orders of magnitude fainter than those measured by Satyapal et al. (2008). Even NGC 3184, which appears to be a bona fide AGN (Ghosh et al. 2008), was not detected. The Satyapal et al. detections of Ne v are in excess of the Dasyra et al. relation, however, thus it is possible these relations do not hold for very late-type spiral galaxies.

Another factor that might be relevant is the ionization parameter; since Ne v is a high-ionization line, its strength decreases dramatically when the ionization parameter decreases (Abel & Satyapal 2008). By analogy with the local population of LINERs and Seyferts (Ho 2008), the very low Eddington ratios of our low-mass systems should correspond to low ionization parameters, and thus plausibly weak Ne v emission. Finally, we note that NGC 3556 is identified by Satyapal et al. as an AGN candidate but has no X-ray detection in this work. Wang et al. (2003) identify an X-ray source as a candidate for the galactic nucleus in NGC 3556 (which they number source 35); we chose not to include this source in our candidate list as the offset from the optical nucleus is 10''. More detailed study is clearly warranted, and it will likely take the combined efforts of large multiwavelength surveys to find such low-mass and low-luminosity AGNs, and to properly assess the global AGN fraction in these late-type systems.

L.-B.D. would like to thank Eliot Quataert for the many helpful and insightful discussions, and Jenny Greene for advice and guidance. We also thank an anonymous referee for careful, helpful comments that greatly improved this manuscript. Support for this work was provided by NASA through Chandra Award Number 06700184 issued by the Chandra X-Ray Observatory Center, which is operated by the Smithsonian Astrophysical Observatory for and on behalf of NASA under contract NAS8-03060, and through grant HST-GO-11130.01 from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS5-26555. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. We also acknowledge the usage of the HyperLeda database (http://leda.univ-lyon1.fr).