PROBING THE BALANCE OF AGN AND STAR-FORMING ACTIVITY IN THE LOCAL UNIVERSE WITH ChaMP

Chandra imaging with ACIS provides energy resolution sufficient to constrain the X-ray spectral properties as well. To characterize the X-ray activity of the ChaMP-SDSS galaxies included in our sample, we perform direct spectral fits to the counts distribution using the full instrument calibration, known redshift, and Galactic 21cm column⁶N^Gal_H. Source spectra are extracted from circular regions with radii corresponding to energy encircled fractions of ∼90%, while the background region encompasses a 20''  annulus, centered on the source, with separation 4'', from the source region. Any nearby sources are excised, from both the source and the background regions.

The spectral fitting is done via yaxx⁷ (Aldcroft 2006), an automated script that employs the CIAO Sherpa⁸ tool. Each spectrum is fitted in the range 0.5–8 keV by two different models: (1) a single power law plus absorption fixed at the Galactic 21cm value (model "PL"), and (2) a fixed power law of photon index Γ = 1.9 plus intrinsic absorption of column N_H (model "PLfix"). These fits use the Powell optimization method and provide a robust and reliable one-parameter characterization of the spectral shape for any number of counts. Spectra with less than 100 net counts were fit using the ungrouped data with Cash statistics (Cash 1979), while those with more than 100 counts were grouped to a minimum of 16 counts per bin and fit using the χ² statistic with variance computed from the data. For the nine objects with more than 200 counts, we employ a third model in which both the slope of the power law and the intrinsic absorption are free to vary (model "PL_abs").

Many of the X-ray-detected galaxies in our sample have relatively few net counts (mean 76, median 19). In such cases, instrumental hardness ratios are often used, in the belief that genuine spectra fitting is not warranted by the data quality. We stress, however, that spectral fitting provides the most consistent and robust estimates of the physical parameters of interest, the power-law slope, and intrinsic absorption. Because the ChaMP X-ray exposures span a variety of intervening Galactic columns, include data from both back- and front-side ACIS CCDs, and span six years of observations, the spectral response between sources varies significantly. While constraints from spectral fitting may not be tight for low count sources, the use of unbinned event data and the appropriate response gives an optimal and unbiased estimate of the fit parameters and their uncertainties, especially important when absorption may be present at different redshifts. A classical hardness ratio analysis, on the other hand, amounts to grouping the data into two rather arbitrary bins, introducing potential biases and statistical complexity. Interpreting the hardness ratio value for ChaMP sources in disparate fields requires incorporating the instrument response in any case, so we strongly prefer spectral fitting. Nevertheless, when there is only one free parameter, only the overall spectral shape is constrained.

This simple parametrization proves generally sufficient to model the 0.5–8 keV spectra of these objects. Comparisons of 0.5–8 keV fluxes f_x (and luminosities L_X) obtained from the PL and PLfix models show good agreement, for the whole sample of galaxies, the average (median) of the difference in these values being 0.07 (0.01) dex. We caution that the simple power-law fits we use here could be misleading for objects in which the absorption is complex (i.e., a partial covering, with one or more absorber potentially being ionized). However, the data quality is insufficient to show that the situation is more complex than a simple power law.

We compile a set of "best" measurements for Γ or N_H, by using the values obtained from the PL (intrinsic N_H fixed at zero) and PLfix (Γ fixed at 1.9) models, respectively. For objects with more than 200 counts, we use the Γ and N_H values obtained from the PL_abs. The mean Γ for the whole sample of 107 galaxies is 2.03 ± 1.38, with a median of 2.04. The level of intrinsic absorption is generally low. More than 85% of the sample exhibits N_H < 1 × 10²² cm⁻², while for 60% of the objects the spectral fits are consistent with zero intrinsic absorption. Note that, given the simplified model used in fitting the X-ray spectra, these values might not necessarily represent the true distribution of absorption in these objects. Individual measurements of all of these X-ray properties, together with their observational parameters, like the total number of X-ray counts, the exposure time, the off-axis angle, together with their corresponding X-ray source ID, are listed in Table 1 for all 107 objects.

Contribution from thermal emission is expected for some of the objects included in this sample of nearby galaxies, whether or not they show line emission activity. Such a component may provide a reasonable contribution to the total (X-ray) emission even in objects in which the dominant ionization mechanism, as identified optically, is a compact nuclear source, i.e., an AGN. LINERs, for instance, have frequently been associated with photoionization by hot, young stars (Filippenko & Terlevich 1992; Shields 1992; Barth & Shields 2000), clusters of planetary nebula nuclei (Taniguchi et al. 2000), or more recently (and perhaps more consistent with their older stellar populations), by hot post-AGB stars and white dwarfs (Stasińka et al. 2008).

We attempted fitting the 0.5–8 keV spectra with a Raymond–Smith (R-S) thermal plasma model, with the abundance fixed at 0.5 solar. The choice of abundance level is inspired by previous investigations of LLAGNs (e.g., Ptak et al. 1999; Terashima et al. 2002), even though in many cases the abundance remains poorly constrained. This R-S model fit results seem physically feasible for only about the third of the sample. Reasonable values, in agreement with previous findings, i.e., kT ≲ 2 keV, are only found for 35 sources. For another third the best fit kT>10 keV. We discuss in more detail the results of fitting this model as a function of the optical spectroscopic properties of these sources in Section 2.5.

As probably expected, it is the passive galaxies that show the softest power-law slopes we measure (Γ>3.5), suggesting that, for these cases in particular, a power-law representation may be incorrect. If we instead fit an R-S model, we derive reasonable typical temperatures near ∼0.7 keV. Nonetheless, since even these objects are likely to have some power-law contribution from X-ray binaries, the R-S model is perhaps no better a characterization of the true spectrum than a power law. The X-ray flux distribution we derive from the R-S model fits to passive galaxies is not significantly different (±25%), so we prefer to retain the power-law fits everywhere to facilitate a more direct comparison of the different spectral classes.

We identify and classify accretion sources and other types of active systems in both the parent galaxy sample and the X-ray-detected subsample, based on their optical emission line properties. It has been argued (Ho et al. 1997a; Constantin & Vogeley 2006; Kewley et al. 2006) that the best way to separate accretion sources from starbursts or other types of active systems is via a set of three diagnostic diagrams, which employ four line flux ratios: [O iii]λ5007/Hβ, [N ii]λ6583/Hα, [S ii]λλ6716, 6731/Hα, and [O i]λ6300/Hα. Thus, for both samples, we first select a subset of strong emission-line sources that show significant emission in all six lines used in the type classification (Hα, Hβ, [O iii], [N ii], [S ii], and [O i]), and a set of passive objects that show insignificant line emission activity. An emission feature is considered to be significant if its line flux is positive and is measured with at least 2σ confidence. Following Kewley et al. (2006) classification criteria, the emission-line objects are separated into Seyferts (Ss), LINERs (Ls), transition objects (Ts), and star-forming, or H ii galaxy nuclei.

This method of classifying low luminosity actively line-emitting galaxy nuclei has the disadvantage that it leaves unclassified a high fraction (∼40%) of galaxies, which show strong emission features, but not in all six lines considered here. The condition for strong emission in [O i] in particular is significantly restrictive. Moreover, another quite large (∼25%) fraction of the emission-line objects remains unclassified, as their line ratios, although accurately measured, do not correspond to a clear spectral type in all three diagrams. In the majority of such cases, while the [N ii]/Hα ratio shows relatively high, S-like values, the corresponding [S ii]/Hα and/or [O i]/Hα ratios place them in the T- or H ii-like object regime; thus, because the [S ii] and [O i] emission lines are better AGN-diagnostics than [N ii], these systems are likely to be excluded from the AGN samples selected via these classifications. As a consequence, our samples based on the six-line classification are small.

To enlarge our samples of galaxy nuclei of all spectral types, we also explored an emission-line classification based on only the [O iii]/Hβ versus [N ii]/Hα diagram, i.e., a four-line classification method, for the X-ray-detected sources. Thus, the emission line galaxy samples comprise all objects showing at least 2σ confidence in the line flux measurements of these four lines only. The delimitation criteria of H ii's and T's remain unchanged, while Ss and Ls are defined to be all objects situated above the Kewley et al. (2006) separation line, and with [O iii]/Hβ >3 and <3, respectively. The four-line and six-line classifications result in significantly different classes when applied to optically selected galaxies in general (Constantin & Vogeley 2006). In particular, true properties become heavily blended into the dominant population of LINERs (or Ts, depending on the separation lines used in the diagnostic diagrams). Interestingly, however, when applied to the X-ray sample, the four-line classes fall well within the six-line loci; although Ss and Ls are separated only by their [O iii]/Hβ line flux ratio, they remain clearly separated into the [S ii]/Hα and [O i]/Hα diagrams as well. Figure 1 shows how the six-line (top) and the four-line (bottom) classifications compare for the ChaMP X-ray-detected galaxies.

Zoom In Zoom Out Reset image size

Although the sample of X-ray-detected galaxies is small, this comparison indicates that adding X-ray detection makes the four-line classification more secure, and that the need for the (usually unavailable) [O iii]/Hβ versus [S ii]/Hα and [O iii]/Hβ versus [O i]/Hα diagrams is not as stringent as in the cases in which only optical information is available. We will thus consider for the analysis presented in this paper only the four-line classification.

The "cleaning" role of the X-ray detection in finding and defining LLAGNs is even more obvious when we compare the fractions of X-ray-detected objects by spectral type, relative to both the parent sample of nearby optically selected objects and the subsample of X-ray-detected galaxies. Table 2 lists these percentages, where for the parent optically selected sample we consider only the SDSS galaxies on ACIS chips (excluding chip S4, ccd \#8), and with off-axis angle θ < 0.2 deg, consistent with the conditions used in compiling the X-ray samples. The first two columns show the number (and fraction) that each spectral type represents, among (1) the optical parent sample and (2) the X-ray-detected subsample. The third column lists the raw fraction of X-ray detections per spectral type.

That the X-ray detection is very efficient in finding LLAGNs, particularly Seyferts, is quite apparent. While narrow-line Ss usually make up only <2%–4% (Ho et al. 1997b; Constantin & Vogeley 2006) of the optically selected nearby galaxies, the X-ray detection increases the chances of finding them tenfold. Ts and Ls, where an AGN contribution to the total ionization power is expected, are also much better represented in the X-ray-detected sample of galaxies, their fractions being ∼3× larger than when only optical selection is employed. Some 65% of the optically defined Ss are detected in X-rays, while the other spectral types hardly reach an X-ray detection fraction of 20%. Ls are the second most X-ray active sources within nearby galaxies, while Ts and the galaxies that show some/weak emission line activity account for less than 1/5th of the sample. As discussed in Section 2.5, only the luminous H iis are detected in X-rays. While the X-ray detection rate of the H iis is basically consistent with zero, when detected, their X-ray emission is moderately strong, L_X = 10³⁹–2.5 × 10⁴¹ erg s⁻¹; half of the H ii detections show X-ray luminosities higher than the level that can be reached without contribution from AGNs (10⁴⁰ erg s⁻¹), suggesting once more that the nuclear emission in these sources might not be completely driven by stellar processes.

The only previous studies that encompass the whole spectral variety of LLAGNs are Roberts & Warwick (2000) and Parejko et al. (2008), which employ ROSAT (HRI and RASS, respectively). Their search for X-ray emitting nearby galaxy nuclei, optically characterized via the Palomar and SDSS surveys respectively, concluded in soft X-ray detection rates of ∼70% of both Ss and Ls (HRI), ∼70% of Ss and ∼60% of Ls (RASS), ∼40% (HRI) and <10% (RASS) of H ii's, and ∼30% of passive galaxies (both cases). A comparison of these detection rates with ours is problematic because of their softer instrument bandpass, lower sensitivity, but wider sky coverage.

Hard X-ray studies of homogeneously selected samples, including all spectral types of nearby active galaxies are practically nonexistent. Ls, and particularly those found in the Palomar survey, have been clearly privileged in terms of X-ray targeting (e.g., Ho et al. 2001). Their detection rates are found to be significantly higher than what we report here based on the serendipitous ChaMP survey. Ho et al. (2001) reports a ∼70% detection rate, while, when chosen for having a flat-spectrum radio core, Ls are found to be 100% X-ray active (Terashima & Wilson 2003). Later studies claiming better accounting for selection and classification as LINER (Satyapal et al. 2004; Dudik et al. 2005; Pellegrini 2005; Satyapal et al. 2005; Flohic et al. 2006; Gonzalez-Martin et al. 2006) conclude with lower fractions, ∼50%.

ChaMP has the distinct advantage of presenting a large, homogeneous serendipitous sample of LLAGNs. The detection fraction is not an intrinsic property of galaxies, but rather a convolution of galaxy properties with optical survey depth, and the X-ray sensitivity versus sky area curve. As described by Green et al. (2009), the ChaMP is characterizing the X-ray sensitivity at the position of every SDSS galaxy, which will enable us to compile the unbiased fraction of galaxies by optical spectral type (e.g., Ls) that fall in X-ray luminosity bins, from sky volumes complete to those limits. We will present the results of such an investigation in a subsequent paper.

X-ray and optical emission are correlated. Thus, while the X-ray selection is one of the most powerful tools to exploit in detecting accretion sources, it is also expected that this selection picks up, selectively, the brightest (optical) sources. Due to its serendipitous character, ChaMP should however reduce such effects. Note that, out of 107 X-ray detections that this ChaMP sample of galaxies provides, only 13 are targets.

In Figure 2, we explore the biases that X-ray detection potentially adds to our sample. Comparisons of the distributions of redshift z, apparent and absolute r-band (SDSS) magnitudes, r and M_r respectively, and of the concentration index C⁹ as a proxy for the morphological type of these galaxies, both for the whole optically and X-ray-selected samples (histograms on the left panel) and separately per spectral type (the right panel) show, pleasingly, that biases are not strong.

Zoom In Zoom Out Reset image size

However, H ii galaxies are of lower z and brighter M_r when detected in X-ray. As shown in the next section, the H iis are generally weak X-ray sources. The tendency for X-ray-detected galaxies to appear brighter in apparent magnitude (r, by ∼0.5 mag) seems to be caused by the large difference in r between the H ii galaxies alone, as all the other types of sources show very similar ranges, averages, or medians, when analyzed separately.

The concentration index C appears to be somewhat larger for the X-ray objects. Several factors are likely to account for this. The H iis are the least concentrated optically and have the lowest X-ray detection fraction. X-ray detection is sensitive to the AGN activity, which is more prevalent in the early-type (massive bulge-dominated) galaxies (e.g., Ho et al. 2003; Kauffmann et al. 2004). Nuclear activity is also expected to increase C simply by adding light to the core. Note that there is no significant difference in this parameter in regard to Ss' X-ray selection. However, Ss make up for a tiny fraction of z ≈ 0 galaxies, and the morphology of their hosts spans quite a range.

X-ray information about the nearby galaxy centers constrains the contribution of accretion-related processes to their optical spectral characteristics. In this section, we present the distributions of a variety of X-ray properties for the whole sample as well as for subsamples by optical spectral class. Figure 3 shows the distributions of the X-ray counts, the 0.5–8 keV unabsorbed X-ray fluxes, the best-fit X-ray photon indices Γ, and the intrinsic neutral hydrogen column densities N_H. These measurements are shown for the whole sample of X-ray detected nearby galaxy nuclei (left column) and separately per spectral type (right column).

Zoom In Zoom Out Reset image size

For X-ray fluxes, we show the values derived using the primary power-law fitting models discussed in Section 2.1: (1) a single power law with no intrinsic absorption and (2) a fixed power law (Γ = 1.9) with absorption. The f_x values are generally consistent with each other for the whole range of brightness and optical spectral type. As expected, the X-ray brightest objects are among Ss; however, even for this spectral type the range of values remains pretty broad, spanning 3 orders of magnitude. Note also that the few H ii galaxies that are X-ray detected are, in general, brighter than the passive systems.

In terms of the X-ray spectral shape, the ChaMP nearby galaxies are quite a diverse population. The mean photon index per optical spectral type shows however a rather clear dependence on the spectral type: Ss show the hardest 0.5–8 keV spectral shape, becoming softer and softer from Ts to Ls to the passive galaxies, which are clearly the softest. The H ii galaxies are unexpectedly hard in average Γ = 1.46; however, two particularly hard detections clearly weight the subsample in this direction. The Ts and Ls average at Γ ≈ 2.

The N_H values are poorly constrained for this sample, and there is no obvious correlation with the optical spectral type. It is, however, obvious that the Ss are the sources with the highest fraction of nonzero absorption. Since all our Ss are of type 2, i.e., lack broad emission lines in their spectra, the unification paradigm predicts that many will show signature of absorption in X-rays. A typical unabsorbed power law Γ = 1.9 requires a column density of N_H = 8 × 10²¹ cm⁻² to yield an apparent Γ = 1 similar to the mean value found for our S subsample, which is consistent with the observed mean N_H for these particular systems.

The relation between the X-ray photon index Γ and the Eddington ratio for the ChaMP X-ray-detected galaxies is illustrated in Figure 6. The Eddington luminosity L_edd is calculated as indicated in Section 3, using M_BH values estimated based on stellar velocity dispersion σ_* via Tremaine et al. (2002), while the bolometric luminosity is calculated based on L_X, using the average bolometric correction of L_bol/L_X = 16 (Ho 2008). There is a rather clear trend of spectral hardening with increasing accretion rate. Γ and L/L_edd are found to be negatively correlated.

Zoom In Zoom Out Reset image size

Defining an accretion rate, i.e., calculating L/L_edd, for galaxies with L_X < 10⁴² erg s⁻¹ may be misleading if the X-ray emission in these objects is dominated by X-ray sources other than an accreting super massive BH (e.g., individual compact binaries or hot diffuse gas). Note, however, that the x-axis of Figure 6 is simply the measure of L_X/M_BH (or better, L_X/σ_*⁴), where both L_X and the BH mass M_BH (or σ_*) are measured in exactly the same manner for all of the objects involved, and the trend remains even if this parameter is not interpreted as an accretion rate. The strongest likely dilution to accretion emission comes from contributions of the hot interstellar medium (ISM) in passive galaxies. Hot gas emission is soft and relatively stronger in more massive hosts, both of which would push the passive galaxy points toward softer X-ray spectra (larger Γ) and lower L_X/M_bh, which might spuriously accentuate the observed trend even in the absence of significant accretion power. Indeed, the observed trend is weakened once the passive galaxies are removed (Table 4). So, while we rule out significant extended emission contributions in this sample, a more detailed examination of such objects is warranted to determine the relative fractions of nuclear versus extended hot gas contributions.

We measure the significance of the Γ–L/L_edd anticorrelation using the Spearman-rank test. The Spearman-rank coefficient, the chance probability, and the number of sources for each correlation are listed in Table 3. We fit the anticorrelation points with a linear least-squares method for the whole sample and for the subsamples of galaxies corresponding to different spectral types, using the mpfit¹⁰ routine, being able to account for the errors in Γ. We show, for comparison, both the error weighted (continuous line) and the unweighted (dotted line) best fits in Figure 6. The scatter around the best linear fit is large, and as expected the points with the largest error bars show the largest deviation. However, within the errors, the results of the fit remain unchanged when we use only objects with small measurements errors (i.e., ΔΓ/Γ < 50%). The results for the linear regression coefficients and the corresponding χ² and degreesoffreedom(dof) values are listed in Table 4. For the sake of considering "cleaner" AGN-like activity only, we also list here the results of such a fitting technique to samples that exclude the passive galaxies and the H iis, and for samples of luminous X-ray systems (L_X ≳ 10⁴² erg s⁻¹) only, and indications of an anticorrelation, albeit weaker, remain.

The linear regression fits might appear at odds with the conclusion of the Spearman test, which indicates that Γ and L/L_edd are possibly anticorrelated for all the subsamples presented here. Note, however, that such a discrepancy appears for the subsamples in which the Spearman test remains rather inconclusive as the probability that an anticorrelation appears by chance is large. Moreover, the Spearman test ignores the errors, and thus tests the unweighted data, for which linear regression fits are always consistent with negative slopes. It is clear however that for Seyferts in particular, there is no evidence for either positive or negative correlation between Γ and L/L_edd. This has been seen in other samples as well (Winter et al. 2009), and it is an important result. Nevertheless, investigations of the LLAGN phenomenon should not be restricted to these types of sources only.

In an attempt to provide some more physical insights into the reality and significance of this new trend, we also explore here the possible correlations between various X-ray measures that might influence (if not artificially create) it. In particular, we scrutinize the way our measured Γ relates to the number of counts, the X-ray flux, and luminosity. Some studies showed that the photon index correlates with the X-ray luminosity, becoming softer in more luminous sources, whether measured in soft, 0.2–2 keV (Forster & Halpern 1996; Lu & Yu 1999; Gierlinski & Done 2004; Williams et al. 2004), or hard, 2– 10 keV (Dai et al. 2004; Porquet et al. 2004; Wang et al. 2004) bands only, or comprising the whole spectrum, and even as they vary (Chiang et al. 2000; Petrucci et al. 2000; Vaughan et al. 2001). However, because many others have not found such trends, the idea that the choice of the sample involved in these studies may contribute to producing the correlations has also been put forward. Interestingly, our ChaMP sample does not show this correlation either. Figure 7 illustrates the dependence of Γ measured in the 0.5–8 keV range as a function of the total number of counts in this energy range, the X-ray flux, and the X-ray luminosity. As before, we emphasize the different optical spectral classifications; the fact that, e.g., the various types of galaxies separate rather well from each other in diagrams like this suggests that the correlation is physical, and most probably related to the accretion physics in these nuclei, rather than an artificial effect of fitting various (simple) models to their X-ray spectra.

Zoom In Zoom Out Reset image size

The anticorrelation between Γ and L/L_edd (or simply, L_X/M_BH) that we find to characterize the nuclear emission of nearby galaxies is certainly surprising. This trend is opposite to what more luminous galaxy nuclei, i.e., quasars, exhibit: their X-ray spectra soften as they become more luminous. Figure 8 shows the Γ–L/L_edd anticorrelation followed by the low-luminosity galaxy nuclei along with measurements of Γ and L/L_edd for a sample of SDSS quasars with optical spectra that are ChaMP detected and X-ray analyzed in the same manner we handled our sample of nearby galaxy nuclei (Green et al. 2009). The quasar L/L_edd values are obtained using the average bolometric correction of L/L_X = 83 estimated by Ho (2008), with no luminosity dependence, and the black hole masses from Shen et al. (2008).

Zoom In Zoom Out Reset image size

To ease comparison with previous work on quasars, we use in this plot the 2–10 keV L_X luminosity, which is obtained by extrapolating the measured unobscured L_0.5-8keV value, using the best Γ measurements. Note that the anticorrelation between Γ and L/L_edd that nearby galaxy nuclei show is even more pronounced when the 2–10 keV L_X is plotted against Γ. This is expected, as softer objects will be less luminous in 2–10 keV than those in the 0.5–8 keV range, while the hard objects will be more luminous. The magnitude of this effect, i.e., the ratio of the two L_X values, is a function of the photon index Γ as given by

$$\begin{equation} L_X(2\hbox{--}10\ \mbox{keV}) = L_X(0.5\hbox{--}8\ \mbox{keV}) \times \frac{10^{2-\Gamma } - 2^{2-\Gamma }}{8^{2-\Gamma } - 0.5^{2-\Gamma }}. \end{equation} \tag{ 1 }$$

The largest difference, and thus the most significant effect on the shape of the Γ–L/L_edd trend, is for the softest (Γ ≳ 3) objects; however, it does not exceed ∼1 dex. Note that for these particular objects the measurement errors are also among the highest, and hence contributed the least weight to the best-fit Γ–L/L_edd relation.

The ChaMP quasars fall well within the locus of values expected based on previous work. The ChaMP quasars do not show, however, a clear Γ–L/L_edd positive correlation. The ChaMP quasars span a wide range in redshift and their L/L_edd measurements reflect a mix of BH mass estimates based on all Hβ, Mg ii, and C iv, which may add scatter to an underlying correlation (Shen et al. 2008). We show for comparison the results of linear regression fits of the Γ–L/L_edd correlation reported by Shemmer et al. (2008) and Kelly et al. (2008), with the results for the Hβ and C iv based estimates of the M_bh shown separately. We use these data only for the purpose of global comparison of the quasar properties with those of the nearby galaxy nuclei, and do not attempt to improve upon the previous work on the characterization or calibration of the Γ–L/L_edd relation for quasars.

Clearly, the quasar Γ's are not negatively correlated with their Eddington ratios, even with the shift in L_X produced by the energy conversion mentioned above, which might contribute to such trend (the quasar X-ray data are analyzed and measured via the same techniques employed for the nearby galaxy nuclei). Thus, over the whole L/L_edd range we explore here by putting together luminous and weak AGNs, there is clearly a break in the Γ–L/L_edd correlation, which seems to happen at L/L_edd ≈ 10⁻², and Γ ≳ 1.5. It is important to note that, for the samples of quasars investigated by both Kelly et al. (2008) and Shemmer et al. (2008), the scatter in the claimed correlation between Γ and L/L_edd is largest as the samples reach the weakest accretion rates (while hardly reaching the L/L_edd ≈ 10⁻² level), and the hardest values ever measured for the quasar photon index Γ ≳ 1.5; while suggestive of a break, the quasar data alone cannot, however, be used to single it out. Other studies that include measurements of lower luminosity managed to point out these types of deviations from the expected correlation (e.g., Zhang et al. 2008; Green et al. 2009); however, the data remained sparse at those levels of accretion, leaving the effect unquantified.

We note here that the comparison of the nearby galaxies' central X-ray emission with that of the high redshift QSOs may be inadequate since X-rays from SF activity is not yet accounted for in the former. We addressed this issue by including in the modeling of the original X-ray spectrum (Section 2.1) a Γ = 2 component with the expected L_X(SF). A power law with Γ = 2 provides the best simple description of the mix of hot gas and high mass X-ray binaries (HMXBs) that comprise the SF activity (e.g., Kim et al. 1992a, 1992b; Nandra & Pounds 1994; Ptak et al. 1999; George et al. 2000; Colbert et al. 2004; Reddy & Steidel 2004; Lehmer et al. 2005). To estimate the potential L_X(SF) in each galaxy, we use star formation rates (SFR) calculated and available for these objects in the MPA/JHU catalog (Section 2; Brinchmann et al. 2004b), along with the L_X–(SFR) correlation quantified by Gilfanov et al. (2004). There are 83 objects with total L_X above the line describing the L_X–(SFR) correlation, for which we modeled the remaining—presumably AGN component. This reanalysis produces significant deviations from the Γ–L_X/L_edd correlation only for 6% of the sample (five out of the 83 objects involved in this analysis). The scatter and the error bars for both Γ and L_X are only slightly increased. The slope of the correlation flattens but remains consistent within the errors with our previous measurements.

The X-ray photon index Γ and the Eddington ratio L/L_edd show a double-sloped relation, with positive and negative correlations above and below L/L_edd ≈ 0.01 and Γ ≳ 1.5, respectively. We see now that a given AGN spectral index Γ may correspond to two different luminosity levels, with the luminosity difference greater for sources characterized by softer spectra (Figure 8). This break in the Γ–L/L_edd relation, when studied over a large range of accretion power, fits well into the theoretical ideas of a transition in the AGN accretion mode: a standard (Shakura & Syunyaev 1973) accretion disk/corona at high Eddington rates, i.e., quasar phase, and an ADAF (e.g., Narayan & Yi 1994) at low L/L_edd. This break we find in the Γ–L/L_edd relation may provide the best empirical evidence to date for such a transition.

The inflection point in the Γ–L/L_edd relation is dominated by Ts, consistent with the optical nature of these systems. While hypothesized to be the result of mixed AGN and SF ionization (e.g., Ho et al. 2003, A. Constantin 2010, in preparation), their AGN component could be either an S- or L-like. Ss appear to be the low L quasar-like, efficiently accreting systems, while the Ls are the ADAF objects, so T's locus at the break region is suggestive of a switch in the accretion mode. The optically defined Ts are then transition systems that map the X-ray inflection as well.

Note that this idea of transition in the accretion mode, from ADAF to standard-disk as the accretion rates increases, has been first proposed, and later developed, mainly based on results of investigations of smaller BH mass accretors, i.e., stellar mass black-hole X-ray binaries, or XRBs; see Narayan & McClintock (2008) for a recent review. Interestingly, the relation between Γ and L/L_edd that XRBs exhibit in the different phases of their temporal variability is also multivalued: a given spectral index Γ may correspond to two different luminosity levels, with the luminosity difference greater for sources characterized by softer spectra. That is, the XRBs show a positive Γ–L/L_edd correlation while in their high/soft states (e.g., Kubota & Makishima 2004) and an anticorrelation while in their low/hard states (e.g., Yamaoka et al. 2005). Several clear examples of such a turn (or convergence point) in the Γ–L/L_edd relation measured in XRBs are illustrated in, e.g., Yuan et al. (2007) and Wu & Gu (2008).

In terms of the physical phenomenology governing the black hole accretion process over more than 10 orders of magnitude in the Eddington ratio, the transition from an ADAF to a standard disk accretion seems to adequately account for both the XRB and AGN emission properties. The ADAF scenario qualitatively explains the anticorrelation via Comptonization of thermal synchrotron photons as the dominant cooling mechanism at low L/L_edd ratios: the Compton y-parameter increases with increasing of the optical depth, which is caused by increase of the accretion rate, the X-ray spectrum becomes harder (Esin et al. 1997). Further increase in the accretion rate would cause both an increase in the released energy and a decrease in the electron temperature, weakening the corona, and consequently the optical depth, reducing the y-parameter, and leading to softer spectra, and thus the positive Γ–L/L_edd correlation (e.g., Janiuk & Czerny 2000). The ADAF scenario has been relatively successful in explaining a variety of the LLAGN properties (Ho 2008), while the standard disk-corona model has been widely invoked to explain quasar emission. Our finding of a nonmonotonic Γ–L/L_edd relation seems to provide the first direct empirical link between these two different types of accretion in AGNs.

The AGN–XRB physical analogy has been discussed rather extensively (see Maccarone et al. 2005), and is particularly supported by the two "fundamental planes" of the BH activity on all mass scales, the L_radio − L_X − M_bh (Merloni et al. 2003; Falcke et al. 2004) and the L_bol − M_bh − T_break (McHardy et al. 2006). There was however not much evidence for existence of spectral states in massive black holes, similar to those of the stellar black holes. Our discovery of an anticorrelation between Γ and L/L_edd, and thus the discovery of a turning point in the Γ–L/L_edd relation for AGNs, provides this evidence, which constitutes an important empirical constraint to the idea that these systems are really the analogs of each other, in spite of the vast difference of scales.