The Lively Accretion Disk in NGC 2992. III. Tentative Evidence of Rapid Ultrafast Outflow Variability

NGC 2992 was monitored with Swift-XRT throughout the year 2019 (from March 26 to December 14), to trigger a deep, high-flux XMM-Newton observation of the source. The triggering flux threshold was met on 2019 May 6, with a 2–10 keV flux F_2–10 = 7.0 ± 0.4 × 10⁻¹¹ erg cm⁻² s⁻¹ (Middei et al. 2022) and XMM-Newton promptly started its 250 ks pointing on the following day, for two consecutive orbits. NuSTAR observed NGC 2992 on 2019 May 10 for 120 ks, simultaneously to the second XMM-Newton orbit, i.e., after ∼185 ks from the beginning of the pointing. In this paper, we consider the same data set presented in Paper I and we adopt the same binning strategy and nomenclature for each spectral slice. We report a journal of the observations in Table 1.

For the XMM data, we oversample the instrumental resolution by at least a factor of 3 and require having no less than 30 counts in each background-subtracted spectral channel. The first XMM orbit is divided into bins of 5 ks each, while the second one, together with the NuSTAR observation, in bins of 5.8 ks. Different background regions do not affect the outcomes of the spectral analysis of the XMM-Newton time-averaged spectra, as discussed in Appendix A. NuSTAR spectra are binned in order to oversample the instrumental resolution by at least a factor of 2.5 and to have a signal-to-noise ratio greater than 3σ in each spectral channel. We adopt the cosmological parameters H₀ = 70 km s⁻¹ Mpc⁻¹, Ω_Λ = 0.73, and Ω_m = 0.27, i.e., the default ones in Xspec 12.11.1 (Arnaud 1996). Errors correspond to the 90% confidence level for one interesting parameter (Δχ² = 2.7), if not stated otherwise.

We fit the time-averaged spectra of the two XMM-Newton orbits between 2 and 12 keV with a model composed of an absorbed power law (zwabs×pow in XSPEC) multiplied by a Galactic absorption component (TBabs) with N_H ≡ 4.8 × 10²⁰ cm⁻² (Kalberla et al. 2005) and removing the energy range dominated by the Fe K lines (5–8 keV). The ratios between the time-averaged data and the best-fitting continuum models are plotted in Figure 1 (top panel), once the 58 keV band is included, and clear absorption features above 9 keV can be seen. We add to the absorbed power-law baseline model (tbabs×zwabs×pow) five narrow Gaussian lines, of which two reproduce the neutral Fe Kα and Kβ and the other three are associated with the ∼5.5, ∼6.7, and ∼7.0 keV transient emission lines, indicated as red flare, blue flare I, and blue flare II, respectively, in Paper I. No strong residuals are present below 9 keV and we obtain best-fit statistics χ²/dof = 200/154 and χ²/dof = 216/154 for the first and second orbit spectra, respectively.

Zoom In Zoom Out Reset image size

To blindly search for any absorption signature we add an additional Gaussian line left free to vary in the 6–12 keV range, with a normalization that can take both positive and negative values. The Gaussian width is fixed to zero to represent a narrow, unresolved line. We fit the two time-averaged XMM-Newton orbits, leaving the baseline parameters free to vary as well. Figure 1 (bottom panel) shows the corresponding contour plots between the normalization and the line energy centroid. For the first orbit (black data points and contour plots in Figure 1) the inclusion of a Gaussian line leads to a fit improvement of Δχ² = −18.48 for two additional degrees of freedom. The best-fit energy of the line is ${11.78}_{-0.15}^{+0.08}$ keV, with a normalization N = −1.6 ± 0.6 × 10⁻⁵ ph cm⁻² s⁻¹.

To evaluate the statistical significance of narrow, unresolved emission/absorption lines standard likelihood ratio tests could lead to inaccurate results (Protassov et al. 2002). Following the procedures described in Tombesi et al. (2010) and Walton et al. (2016), we create Monte Carlo routines to retrieve the statistical significance of the absorption line. We create 10,000 fake data sets of the XMM-Newton EPIC-pn spectrum using the fakeit command in Xspec with responses, background files, exposure times, and energy binning of the real data according to the following procedure. We first load the best-fitting model, composed of an absorbed power law and five Gaussian lines, and produce a new list of free parameters (i.e., column density of the cold absorber, power-law photon index and normalization, energy, and normalization of the emission lines) by drawing from a multivariate Normal distribution based on the covariance matrix via the Xspec command tclout simpars. This allows us to take into account uncertainties in the continuum model, too. Then, this new continuum, without absorption lines and with randomly sampled free parameters, is used to simulate a fake spectrum. Finally, the fake spectrum is fitted, first with the input model (i.e., without absorption lines), and then, including an additional unresolved Gaussian line to the model, with normalization and energy centroid left free to vary in the range [−1.0:+1.0] × 10⁻⁴ ph cm⁻² s⁻¹ and [6:12] keV, respectively. Being N the number of spectra in which the fit improvement is equal or higher than that of our observed line (i.e., Δχ² = −18.48) and S the total number of simulated spectra, then the estimated statistical significance of the detection is 1 − N/S. We obtain five spurious detections out of the total 10,000 trials, implying a statistical significance of 3.5σ.

We then apply the same procedure to the XMM spectrum of the second orbit (red data points and contour plots in Figure 1). We find two absorption lines at 9.28 ± 0.07 keV, with a normalization N = −8.5 ± 4.0 × 10⁻⁶ ph cm⁻² s⁻¹ (Δχ² = −10.18) and at 11.75 ± 0.15 keV, with a normalization N = −1.5 ± 0.6 × 10⁻⁵ ph cm⁻² s⁻¹ (Δχ² = −17.65). The significance of these two detections, estimated as above, is 2.2σ and 3.2σ, respectively. We note that any broadening of such unresolved absorption lines is likely due to the superposition of different spectral features from several time intervals, as we will show in the next sections.

As a further step, we take into account the phenomenological model adopted in Paper I, in which the EPIC-pn spectra from the two orbits (250 ks in total) are divided into 50 bins ranging between 2 and 12 keV. NuSTAR FPMA/B spectra, when available, are simultaneously fitted in the energy interval between 3 and 79 keV. We fit the 50 EPIC-pn spectra leaving the baseline parameters free to vary. Following the same procedure for the blind search of absorption lines described above, we show in Figure 2 the contour plots between the normalization and the energy centroid of the Gaussian line. We adopt the best-fitting models of Paper I, and in addition, we leave the energy centroid of the absorption line free to vary between 6 and 12 keV. We only show the time intervals where we find a detection at a confidence level greater than 99% (Δχ² improvement larger than 9.21, for two parameters of interest). The Gaussian width is fixed to 0 keV to represent a narrow, unresolved line. Solid, dashed, and dotted magenta lines indicate 99%, 90%, and 68% confidence levels, corresponding to Δχ² = −9.21, −4.61, and −2.3, respectively. We detect an absorption line at a confidence level greater than 99% in 13 out of 50 spectra (24%). Best-fit values of the baseline parameters are instead fully consistent with the values in Paper I, and thus, we do not report them here. The same procedure is then repeated for the 20 time intervals in which simultaneous EPIC-pn and FPMA/B spectra are available, including cross-calibration constants between the three detectors and leaving the Gaussian energy centroid to vary between 6 and 15 keV. Figure 3 reports the contour plots for the four out of 20 spectra (20%) showing an absorption line with a confidence level >99%.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 2 reports the best-fitting energies and fluxes of the absorption lines, the Δχ² improvement, and the overall χ²/ν value. For the XMM+NuSTAR spectra corresponding to 220 ks and 278 ks we include also the absorption lines detected with XMM-Newton alone, despite their lower statistical significance (at ${11.24}_{-0.10}^{+0.12}$ and ${10.71}_{-0.11}^{+0.08}$ keV, respectively). The non-detection of these lines in NuSTAR spectra can be explained in terms of their shorter exposure times and lower spectral resolution at these energies with respect to XMM-Newton. As a consistency check, we fit the unbinned pn spectra together with the FPMA/B spectra with a fixed 200 eV energy binning, leaving the normalization of the lines free to vary between the three detectors and using the Cash statistics (Cash 1976). The inferred normalizations and upper limits are always consistent with each other.

Table 2. Best-fit Parameters of the Time-resolved XMM-Newton and XMM-Newton+NuSTAR Analysis (Top and Bottom Boxes, Respectively)

Time	Energy	Normalization	EW	Δχ2	vout/c	vout/c	MC Sign.	χ2/ν
	(keV)	(10−5 ph cm−2 s−1)	(eV)		(Fe xxv He-α)	(Fexxvi Ly-α)	σ
XMM-Newton only
10 ks	${11.40}_{-0.10}^{+0.08}$	−6.5 ± 3.2	−165 ± 80	−11.79	${0.486}_{-0.007}^{+0.005}$	${0.456}_{-0.007}^{+0.005}$	2.11	126/131
35 ks	${11.15}_{-0.13}^{+0.08}$	−5.1 ± 2.6	−155 ± 80	−11.10	${0.469}_{-0.009}^{+0.006}$	${0.438}_{-0.009}^{+0.006}$	2.05	100/123
60 ks	10.15 ± 0.15	−4.1 ± 2.1	−90 ± 50	−9.62	0.39 ± 0.01	0.360 ± 0.013	1.76	119/129
70 ks	${11.25}_{-0.10}^{+0.12}$	−5.6 ± 2.8	−130 ± 70	−10.09	${0.476}_{-0.007}^{+0.008}$	${0.446}_{-0.007}^{+0.008}$	1.86	105/128
100 ks	11.05 ± 0.10	−5.6 ± 5.0	−130 ± 70	−9.59	0.462 ± 0.007	0.431 ± 0.007	1.75	131/126
130 ks	${9.90}_{-0.06}^{+0.07}$	−4.5 ± 2.0	−100 ± 40	−14.40	${0.372}_{-0.005}^{+0.006}$	${0.337}_{-0.005}^{+0.006}$	2.54	137/126
191 ks	10.15 ± 0.06	−4.3 ± 1.8	−110 ± 45	−15.83	0.393 ± 0.005	0.359 ± 0.005	2.43	151/127
202 ks	${11.73}_{-0.28}^{+0.07}$	−6.5 ± 3.5	−220 ± 120	−13.13	${0.508}_{-0.018}^{+0.004}$	${0.478}_{-0.019}^{+0.004}$	2.17	151/125
220 ks	${11.35}_{-0.18}^{+0.07}$	−5.5 ± 2.7	−170 ± 80	−14.15	${0.483}_{-0.012}^{+0.005}$	${0.453}_{-0.013}^{+0.005}$	2.22	107/124
237 ks	${9.32}_{-0.08}^{+0.05}$	−3.1 ± 1.7	−75 ± 40	−10.03	${0.319}_{-0.008}^{+0.005}$	${0.283}_{-0.008}^{+0.005}$	1.87	118/125
278 ks	10.70 ± 0.08	−4.9 ± 2.0	−125 ± 55	−15.76	0.437 ± 0.006	0.405 ± 0.006	2.70	117/128
283 ks	${9.96}_{-0.08}^{+0.10}$	−3.7 ± 1.8	−90 ± 50	−9.94	${0.377}_{-0.007}^{+0.008}$	${0.343}_{-0.007}^{+0.009}$	1.83	122/128
295 ks	${11.75}_{-0.11}^{+0.10}$	−5.4 ± 2.5	−165 ± 80	−12.58	${0.509}_{-0.006}^{+0.007}$	${0.480}_{-0.006}^{+0.007}$	2.37	160/141
XMM-Newton+NuSTAR
202 ks	${11.73}_{-0.12}^{+0.10}$	−3.7 ± 1.7	−130 ± 60	−10.78	${0.508}_{-0.008}^{+0.006}$	${0.478}_{-0.008}^{+0.006}$	2.02	430/375
220 ks	${11.24}_{-0.10}^{+0.12}$	−2.7 ± 1.4	−90 ± 50	−8.48	${0.476}_{-0.007}^{+0.008}$	${0.445}_{-0.007}^{+0.008}$	2.07	384/386
	${14.15}_{-0.14}^{+0.08}$	−3.6 ± 1.7	−170 ± 80	−12.20	${0.634}_{-0.006}^{+0.004}$	${0.610}_{-0.006}^{+0.004}$
278 ks	${10.71}_{-0.11}^{+0.08}$	−2.7 ± 1.6	−70 ± 45	−8.87	${0.437}_{-0.008}^{+0.006}$	${0.405}_{-0.008}^{+0.006}$	2.10	369/376
	11.46 ± 0.13	$-{3.8}_{-1.3}^{+1.6}$	$-{120}_{-30}^{+50}$	−13.19	${0.490}_{-0.009}^{+0.008}$	${0.460}_{-0.009}^{+0.008}$
295 ks	${11.65}_{-0.05}^{+0.08}$	−4.6 ± 1.5	−150 ± 50	−22.99	${0.503}_{-0.005}^{+0.006}$	${0.473}_{-0.005}^{+0.007}$	3.00	461/409

Note. Energies are in kiloelectronvolts and in the rest frame of the source (z = 0.00771). The statistical significance of each absorption line is determined via Monte Carlo simulations, see the text for details.

Download table as:  ASCIITypeset image

The statistical significance of each detected absorption line listed in Table 2 is then estimated via Monte Carlo simulations, using 1000 fake data sets for each spectrum and the procedure outlined above. We report the inferred significances in Table 2, ranging between 1.76σ and 3.00σ, in Figures 2 and 3 for the XMM-Newton and joint XMM-Newton+NuSTAR observations, respectively.

We show in the top panels of Figure 4 the eight EPIC-pn spectra with an absorption line with a significance >2σ together with the residuals with respect to a spectral model without and with the absorption line, respectively. Similarly, in the four bottom panels, we show the fits to the joint XMM-Newton+NuSTAR observations with >2σ significance absorption lines, indicating in red and blue the residuals due to absorption lines in the EPIC-pn and FPMA/B data sets, respectively. For visual clarity, we plot the combined FPMA/B spectra (setplot group command in Xspec). As a summary of our findings, we plot in Figure 4 (bottom panel) the eight time intervals in which an absorption line is detected on top of the 2–10 keV EPIC-pn light curve.

Zoom In Zoom Out Reset image size

To estimate the statistical significance of the absorption lines in the global set of spectra, rather than in every single one, we simulate the entire set of 50 XMM-Newton time slices via Monte Carlo routines, using the same procedure outlined above. For each time slice we set input parameters drawn from a normal distribution around the best-fit values excluding the Gaussian absorption component. Then, we fit again all the time slices and check how many spurious absorption components are detected. We simulate the entire set 1000 times and we find that, on average, 2.4 spurious absorption lines are detected at a significance level equal to or higher than our Monte Carlo derived 2σ threshold, corresponding to a fit improvement Δχ² < −11 (see Table 2). We note that 5 out of 8 detections in the observed spectra have a significance Δχ² < −13, while for such significance only 1.2 spurious lines, on average, are detected in the simulated set. Figure 9 (Appendix B) reports the full set of results and the average number of spurious detections for differentΔχ² thresholds.

Most of the variable absorption lines are detected above 10 keV, an energy range in which the effective area of the EPIC-pn detector significantly decreases. To assess the impact of possible calibration effects we analyze the EPIC-pn spectrum of the Blazar 3C 273, which is well known for showing a simple, featureless continuum (see Appendix A). The spectrum does not show any deviation from a simple power-law continuum or absorption features, further demonstrating that the absorption features in NGC 2992 cannot be ascribed to instrumental effects.

To get a complete characterization of the outflow we apply WINE to all the time intervals with UFO signatures detected with a significance >2σ.

WINE is a self-consistent, physically motivated model for wind absorption and emission profiles. Here we provide a brief overview of the model and we refer to Laurenti et al. 2021 and A. L. Luminari et al. 2023, in preparation  for a comprehensive description. We represent the wind as a series of thin slabs and we start radiative transfer from the innermost one using the XSTAR photoionization code (Kallman & Bautista 2001) and providing the required parameters, i.e., the inner radius r₀, the slab column density δ N_H and ξ₀, the ionization parameter at r₀. The density profile of the wind n(r) is parameterized through the exponent α such that $n{(r)={n}_{0}({r}_{0}/r)}^{\alpha }$, while n₀ can be derived by inverting the definition of the ionization parameter:

$$\begin{eqnarray}&&{\xi }_{0}\equiv {L}_{\mathrm{ion}}^{{\prime} }/({n}_{0}{r}_{0}^{2}),\end{eqnarray} \tag{ 1 }$$

where ${L}_{\mathrm{ion}}^{{\prime} }$ is the incident ionizing luminosity in the energy range between 13.6 and 13.6 keV in the gas reference frame, which together with the incident spectrum, is a proxy of the AGN luminosity in XSTAR. Similarly to the density, we implement a power-law behavior for the wind velocity as $v{(r)={{\rm{v}}}_{0}({r}_{0}/r)}^{\zeta }$, where ζ is a free parameter of the model. Then, we propagate the simulation to the second slab, using the transmitted spectrum and luminosity as the incident ones and calculating analytically ξ, r, n, v. We iterate the procedure up to the kth slab, so that the total wind column density N_H = k · δ N_H is reached.

This slicing allows us to reproduce the scaling of the wind properties, including its velocity profile. Given that XSTAR, as well as many other photoionization codes (e.g., Cloudy, Ferland et al. 2017 and SPEX, Kaastra et al. 1996), assumes a null gas outflowing velocity v, we implemented a procedure to take into account v in the radiative transfer calculations. This procedure is carried out in a fully special relativity framework and represents a major novelty of WINE. Given the mildly relativistic velocities usually displayed by UFOs, relativistic effects lead to sizeable effects on the appearance of both emission and absorption profiles, which must be properly accounted for to correctly estimate the wind properties, particularly its N_H (see Luminari et al. 2020 for a detailed description). As a result of these effects, ${L}_{\mathrm{ion}}^{{\prime} }$ will be in general different (i.e., lower) than the rest-frame measured luminosity, L_ion. In particular, in the case of a power-law incident spectrum with photon index Γ, the luminosity in the gas frame can be written as ${L}_{\mathrm{ion}}^{{\prime} }\,=\,{\left(\tfrac{1-v}{1+v}\right)}^{\tfrac{2+{\rm{\Gamma }}}{2}}\cdot {L}_{\mathrm{ion}}$, where v is in units of c. Moreover, line emissivities calculated by XSTAR are convolved for each slab with Monte Carlo profiles to accurately represent the wind emission spectrum as a function of its geometry, as well as its ionization, column density, and velocity. This allows us to constrain the presence of wind emission components with higher accuracy with respect to XSTAR or other general-purpose photoionization codes. However, we find that the inclusion of emission features is not statistically supported in any of the time slices, possibly due to a small wind covering factor and/or the limited signal-to-noise ratio of the spectra. For the same reason we do not implement a detailed velocity profile, but we rather set ζ ≡ 0, i.e., a constant velocity. As a result, the free parameters of the WINE model, which will then be constrained by fitting the model to the data, are

• 1.  
  ξ₀, the ionization parameter at the inner boundary of the wind
• 2.  
  N_H, the wind column density
• 3.  
  v₀, the outflowing velocity

The remaining free parameters of the model, i.e., r₀, α, are fixed to r₀ = 5 r_S , α = 0, as explained in Section 3.2. Finally, the AGN ionizing spectral energy distribution (SED) is described through a power law with Γ = 1.7 and 2–10 keV luminosity L_2–10 = 1.0 × 10⁴³ erg s⁻¹ (which implies L_ion = 2.67 × 10⁴³ erg s⁻¹), which are the average values for the time slices analyzed (see Paper I).

The specific set of Xspec tables used in this work has been tailored to the (rather extreme) properties of the outflow in NGC 2992. In order to fully understand the behavior of the wind as a function of its ξ₀, N_H, v₀ it is instructive to examine the wind radial thickness. We define the thickness as ${{\rm{\Delta }}}_{r}=\tfrac{r({N}_{{\rm{H}}})-{r}_{0}}{{r}_{0}}$, i.e., the difference between the radius enclosing a column density N_H and r₀, normalized by r₀. According to α, Δ_r can be written as a function of the initial parameters as follows:

$$\begin{eqnarray}{{\rm{\Delta }}}_{r}=\left\{\begin{array}{ll}{\left(1-\tfrac{{N}_{{\rm{H}}}}{{n}_{0}{r}_{0}(\alpha -1)}\right)}^{\tfrac{1}{1-\alpha }}-1 & \alpha \ne 1\\ \exp \left(\tfrac{{N}_{{\rm{H}}}}{{n}_{0}{r}_{0}}\right)-1 & \alpha \,=\,1\end{array}\right..\end{eqnarray} \tag{ 2 }$$

It can be seen that, as expected, Δ_r increases for increasing α. In particular, for a constant density profile (α = 0), the thickness can be written as

$$\begin{eqnarray}\begin{array}{rcl}\displaystyle \frac{{N}_{{\rm{H}}}}{{n}_{0}{r}_{0}} & = & \displaystyle \frac{{N}_{{\rm{H}}}{\xi }_{0}{r}_{0}}{{L}_{\mathrm{ion}}^{{\prime} }}\,=\,\displaystyle \frac{{N}_{{\rm{H}}}{\xi }_{0}{r}_{0}}{{\left(\tfrac{1-\beta }{1+\beta }\right)}^{\tfrac{2+{\rm{\Gamma }}}{2}}{L}_{\mathrm{ion}}}\\ & \approx & 2.35\,\times \,{10}^{-33}{\left(\displaystyle \frac{1+\beta }{1-\beta }\right)}^{\tfrac{2+{\rm{\Gamma }}}{2}}{N}_{{\rm{H}}}{\xi }_{0}{r}_{0,S}\ \displaystyle \frac{1}{{\lambda }_{\mathrm{ion}}}\\ & = & 0.17{\left(\displaystyle \frac{1+\beta }{1-\beta }\right)}^{\tfrac{2+{\rm{\Gamma }}}{2}}\cdot \displaystyle \frac{{N}_{{\rm{H}}}}{{10}^{24}}\cdot \displaystyle \frac{{\xi }_{0}}{{10}^{5}}\cdot \displaystyle \frac{{r}_{0,S}}{5},\end{array}\end{eqnarray} \tag{ 3 }$$

where ${L}_{\mathrm{ion}}^{{\prime} }$ is the relativistic-corrected luminosity in the wind reference frame (see Section 3.1), r_0,S is the launching radius in units of the Schwarzschild radius r_S = 2 r_g = 2 G M_BH/c², β = v₀/c and N_H, ξ₀ are expressed in units of cm⁻², erg cm s⁻¹, respectively. In the last step, we assume λ_ion = 7 × 10⁻³, as appropriate for NGC 2992 (see Paper I). The relativistic correction is expressed by the term ${\left(\tfrac{1+\beta }{1-\beta }\right)}^{\tfrac{2+{\rm{\Gamma }}}{2}}$, which for v₀ = 0.4c, Γ = 1.7 corresponds to 4.8.

In Figure 5 we show Δ_r and ξ(r) (top and bottom panels, respectively) as a function of N_H, up to 7.5 × 10²⁴ cm⁻², for v₀ = 0.4c, $\mathrm{log}\left(\tfrac{{\xi }_{0}}{\mathrm{erg}\ \mathrm{cm}\ {{\rm{s}}}^{-1}}\right)=4.0,5.0;$ these values are representative of those reported in Table 1 and of the best-fit values obtained with WINE (see later). From left to right r₀ = 5,10,50 r_S . The very low normalized luminosity of NGC 2992, λ_ion = 7 × 10⁻³, together with the high N_H and v₀, contribute to significantly increasing the radial extension of the flow. This, in turn, produces a rapidly decreasing ionization profile through the wind column, due to geometric dilution of the incident radiation flux. As a result, to reproduce the observed high ξ, N_H we need a very low r₀, of the order of 5 r_S . Although this low value may rise questions about its physical meaning, it must be primarily intended as a numerical strategy to represent a very geometrically thin wind, and thus to minimize the decrease of the ionization parameter and reproduce the high ξ₀ of the observations. We discuss this point further in Section D using the results from the fits described in Section 4.2 below. We also note that the wind properties are variable between one time slice and the following one; the observations have a duration of ≈5 ks, which can be translated in a dynamical length, for a velocity = 0.4c, of around 5 r_S (using M_BH = 3 × 10⁷ M_⊙, as estimated in Paper I), in agreement with our r₀. We also note that an isothermal density profile (α = 2), as would be expected for a wind expanding in spherical symmetry, would not be able to reproduce such high column densities, since the maximum value would be limited to ${N}_{{\rm{H}}}\,=\,2{n}_{0}{r}_{0}=2{L}_{\mathrm{ion}}^{{\prime} }/({\xi }_{0}{r}_{0})=3.2\,\times \,{10}^{24}$ cm⁻² for ξ₀ = 10⁵ erg cm s⁻¹, v₀ = 0.4c, r₀ = 5r_S . The parameter space of the Xspec tables is as follows:

• 1.  
  N_H = [0.5,20.0]·10²⁴ cm⁻² with a 0.5 × 10²⁴ cm⁻² step
• 2.  
  $\mathrm{log}({\xi }_{0}/\mathrm{erg}\ \mathrm{cm}\ {{\rm{s}}}^{-1})$ = [3.00,6.50] with a 0.25 step
• 3.  
  v₀ = [0.15,0.55] c with a 0.05c step

Finally, we tried several values for the turbulent broadening σ_v of the absorption lines. Although the data are weakly sensitive to σ_v , we find that σ_v = 2500 km s⁻¹ best reproduces the observed lines and we use this value hereafter.

Zoom In Zoom Out Reset image size

In order to get a zeroth-order characterization of the wind features we first apply WINE to the time-averaged data from the first and second orbit, ranging respectively from 0–125 ks and from 175–300 ks. As before, the energy range is 2-12(3-79) keV for the XMM-Newton(NuSTAR) data sets. The model in Xspec reads as

$$\begin{eqnarray*}&&{\rm\small{CONST}}\times \mathrm{TB}{\rm\small{ABS}}\,\times (\mathrm{zwabs}\times {\mathrm{WINE}}_{\mathrm{abs}}\times \mathrm{powerlaw}+5\mathrm{zgauss})\end{eqnarray*}$$

where the constant component (const) accounts for the cross-calibration factor between pn and FPMA/B spectra, when present. In order to check the accuracy of WINE we also fit the data using the same model and replacing WINE with XSTAR tables. These tables are built using the same initial conditions (i.e., σ_v , r₀, L_ion, Γ) and spanning the same parameter range for ξ₀, N_H. Following the standard procedure, we use the XSTAR table redshift as a proxy for the wind velocity, again spanning the same range of velocities as the v₀ parameter in the WINE tables. We report in Table 3 the best-fit values using both WINE and XSTAR. For ease of comparison with the WINE values, we report the relativistically corrected N_H for XSTAR, obtained by correcting the best-fit column density (and associated error) according to the best-fit redshift-derived velocity (see Luminari et al. 2020 for more details). For each orbit, we obtain a reasonable agreement between the WINE and XSTAR fits, even though the overall fit statistic is better for the WINE fits.

Table 3. Best-fit values for the Time-averaged Spectra, Corresponding to the First XMM-Newton Orbit (Columns 2 and 3) and the Joint Second XMM-Newton Orbit+NuSTAR (Columns 4 and 5)

	First Orbit		Second Orbit
zwabs
NH (1022 cm−2)	0.87 ± 0.04	0.84 ± 0.2	1.02 ± 0.03	1.04 ± 0.03
powerlaw
Γ	1.67 ± 0.01	1.663 ± 0.008	1.655 ± 0.007	1.658 ± 0.007
norm (10−2)	2.17 ± 0.04	2.15 ± 0.03	1.81 ± 0.02	1.82 ± 0.02
	WINE abs	XSTAR	WINE abs	XSTAR
$\mathrm{log}\left(\tfrac{{\xi }_{0}}{\mathrm{erg}\ \mathrm{cm}\ {{\rm{s}}}^{-1}}\right)$	${4.1}_{-0.2}^{+0.6}$	${4.4}_{-0.6}^{+0.4}$	4.5 ± 0.4	${4.02}_{-0.08}^{+0.28}$
v0 (c)	${0.337}_{-0.006}^{+0.007}$	0.350 ± 0.006	${0.372}_{-0.008}^{+0.007}$	0.415 ± 0.006
NH (1024 cm−2)	3.1 ± 0.8	${4.0}_{-2.5}^{+4.9}$	${3.8}_{-1.6}^{+1.9}$	${1.2}_{-0.4}^{+1.1}$
χ2/dof	190/151	201/151	692/466	707/466

Note. Columns 2 and 4 correspond to the fits using WINE, while Columns 3 and 5 to those using XSTAR.

Download table as:  ASCIITypeset image

We then apply the same WINE fitting model to the time slices showing absorption lines with a significance >2σ (estimated via Monte Carlo simulations). The analyzed data sets are 10, 35, 130, and 191 ks (XMM-Newton) and 202, 220, 278, and 295 ks (XMM-Newton+NuSTAR). We also include the XMM-Newton coadded spectra corresponding to 60 and 191 ks, since their absorption features have consistent line energy according to the fit in Section 2.2. For the 191 ks observation, we obtain a reduced chi-squared ${\chi }_{\nu }^{2}\,=\,1.14$, which improves when stacking with the 60 ks one to ${\chi }_{\nu }^{2}\,=\,0.93$. Best-fit values are consistent between the 191 and the stacked 60 + 191 fit. Table 4 reports the best-fit parameters. We also indicate the wind density n₀, derived inverting Equation (1) and using the best-fit values for ξ₀, v₀, Γ, and L_ion, r₀ from Section 3. We note that, for a gas in photoionization equilibrium, its ionic population, and thus the emerging spectrum, is mainly determined by the value of ξ₀. Observationally, once L_ion, ξ₀ are measured, it is possible to derive an estimate of ${n}_{0}{r}_{0}^{2}$, but the two parameters cannot be disentangled, and thus, they both remain mostly unknown for the majority of UFOs (for further discussion see Nicastro et al. 1999; Krongold et al. 2007; Luminari et al. 2022). However, thanks to our estimate of r₀, we are able to break this degeneracy, and thus, to provide a value for the wind density. Figure 6 shows the contour plots for all the analyzed time slices, while best-fit spectra and corresponding theoretical model are plotted in Figure 10 in Appendix C, where we also discuss a further absorption line detected in the 220 ks NuSTAR spectrum at E ≈ 14 keV. We note that the reduced χ² are similar to those in Table 2, in which absorption features are fitted with Gaussian lines. However, fit statistic shows a significant improvement with respect to the time-averaged WINE fits for the first and second orbits reported in Table 3, since we are now able to resolve the variable wind features on a 5 ks timescale and analyze them one by one. The averaged spectra are instead 125 ks long and therefore only allow for a characterization of the average wind features.

Zoom In Zoom Out Reset image size

Table 4. Top: best-Fit Values for the WINE Fits and Derived Number Density of the Wind n₀

Time Frame (ks)	10	35	130	60 + 191	202	220	278	295
zwabs
NH (1022 cm−2)	1.1 ± 0.2	1.0 ± 0.2	0.8 ± 0.2	0.9 ± 0.1	1.1 ± 0.1	1.2 ± 0.1	1.1 ± 0.1	1.1 ± 0.1
powerlaw
Γ	1.81 ± 0.04	1.69 ± 0.06	1.63 ± 0.06	1.64 ± 0.04	1.70 ± 0.03	1.69 ± 0.03	${1.67}_{-0.02}^{+0.03}$	1.68 ± 0.03
norm a	3.2 ± 0.2	1.9 ± 0.2	1.9 ± 0.2	1.9 ± 0.1	1.88 ± 0.09	1.83 ± 0.09	${1.91}_{-0.09}^{+0.010}$	1.94 ± 0.09
WINE abs
$\mathrm{log}\left(\tfrac{{\xi }_{0}}{\mathrm{erg}\ \mathrm{cm}\ {{\rm{s}}}^{-1}}\right)$	4.6 ± 0.6	4.2 ${}_{-0.4}^{+0.7}$	${4.7}_{-0.6}^{+0.4}$	3.75 ± 0.2	${4.5}_{-0.5}^{+0.3}$	${4.7}_{-0.8}^{+0.5}$	>4.5 (6.5) b	${4.5}_{-0.4}^{+0.3}$
v0 (c)	${0.45}_{-0.02}^{+0.03}$	0.32 ${}_{-0.03}^{+0.02}$	${0.21}_{-0.03}^{+0.01}$	0.37 ± 0.01	${0.35}_{-0.02}^{+0.03}$	${0.27}_{-0.03}^{+0.02}$	${0.43}_{-0.01}^{+0.02}$	0.35 ± 0.01
NH (1024 cm−2)	${8.2}_{-2.7}^{+2.8}$	${7.8}_{-5.2}^{+4.2}$	${6.6}_{-3.3}^{+4.8}$	${5.1}_{-2.3}^{+3.8}$	${4.0}_{-2.8}^{+2.7}$	${5.1}_{-4.2}^{+2.4}$	${5.9}_{-2.0}^{+2.9}$	${5.8}_{-2.5}^{+1.8}$
$\mathrm{log}\left(\tfrac{{n}_{0}}{{\mathrm{cm}}^{-3}}\right)$	10.7 ± 0.6	${11.35}_{-0.7}^{+0.4}$	${11.0}_{-0.4}^{+0.6}$	11.8 ± 0.2	${11.0}_{-0.3}^{+0.5}$	${11.0}_{-0.5}^{+0.8}$	<10.9 (8.9) b	${11.0}_{{}_{0}.3}^{+0.4}$
χ2/dof	126/135	102/127	131/130	177/190	432/379	399/392	378/382	448/401
${\dot{M}}_{\mathrm{out}}$	${5.0}_{-1.6}^{+1.7}$	${3.3}_{-2.2}^{+1.8}$	${1.9}_{-1.0}^{+1.4}$	${2.5}_{-1.1}^{+1.9}$	${1.9}_{-1.3}^{+1.7}$	${1.8}_{-1.5}^{+1.0}$	${3.4}_{-1.2}^{+1.7}$	${2.8}_{-1.2}^{+0.9}$
(1025 g s−1)
${\dot{p}}_{\mathrm{out}}$	${6.7}_{-2.3}^{+2.4}$	${3.2}_{-2.1}^{+1.7}$	${1.2}_{-0.6}^{+0.9}$	${2.7}_{-1.2}^{+2.1}$	${2.0}_{-1.4}^{+1.2}$	${1.5}_{-1.2}^{+0.8}$	${4.4}_{-1.5}^{+2.2}$	${2.9}_{-1.2}^{+0.9}$
(1035 g cm s−2)
${\dot{E}}_{\mathrm{out}}$	${5.3}_{-1.8}^{+2.0}$	${1.6}_{-1.1}^{+0.9}$	${0.4}_{-0.2}^{+0.3}$	${1.7}_{-0.8}^{+1.3}$	${1.2}_{-0.8}^{+0.7}$	${0.6}_{-0.5}^{+0.3}$	${3.3}_{-1.1}^{+1.6}$	${1.7}_{-0.7}^{+0.6}$
(1045 erg s−1)
${\dot{M}}_{\mathrm{out}}$ (M⊙ yr−1)	0.8 ± 0.3	0.5 ± 0.3	${0.3}_{-0.1}^{+0.2}$	${0.4}_{-0.2}^{+0.3}$	0.3 ± 0.2	${0.3}_{-0.2}^{+0.1}$	${0.5}_{-0.2}^{+0.3}$	${0.4}_{-0.2}^{+0.1}$
${\dot{p}}_{\mathrm{out}}$ (Lbol/c)	${89}_{-30}^{+32}$	${62}_{-41}^{+33}$	${20}_{-11}^{+15}$	${51}_{-23}^{+38}$	${37}_{-26}^{+23}$	${28}_{-23}^{+15}$	${77}_{-26}^{+38}$	${51}_{-22}^{+16}$
${\dot{E}}_{\mathrm{out}}$ (Lbol)	${23.5}_{-8.1}^{+8.9}$	${10.6}_{-7.3}^{+5.8}$	${2.2}_{-1.3}^{+1.6}$	${10.3}_{-4.7}^{+7.7}$	${7.2}_{-5.1}^{+4.6}$	${3.9}_{-3.3}^{+2.1}$	${19.0}_{-6.6}^{+9.5}$	${9.8}_{-4.2}^{+3.3}$

Note. Bottom: mass, momentum, and energy transfer rates (see Section 5), both in cgs and normalized units. Errors are reported at the 90% c.l. (a) In units of 10⁻² ph. keV⁻¹ cm⁻² s⁻¹ at 1 keV. (b) Values in parentheses indicate the upper/lower bounds. The 1σ best-fit values are $\mathrm{log}\left(\tfrac{{\xi }_{0}}{\mathrm{erg}\ \mathrm{cm}\ {{\rm{s}}}^{-1}}\right)={5.2}_{-0.3}^{+0.8},\mathrm{log}\left(\tfrac{{n}_{0}}{{\mathrm{cm}}^{-3}}\right)={10.1}_{-0.8}^{+0.3}$.

Download table as:  ASCIITypeset image

We verified that a different model component accounting for the neutral absorption along the line of sight (ztbabs instead of zwabs) does not affect the best-fitting parameters and the associated statistics.

The mass outflow rates associated with the UFO features can be calculated through the formula from Crenshaw & Kraemer (2012):

$$\begin{eqnarray}&&{\dot{M}}_{\mathrm{out}}\,=\,4\pi {r}_{0}{N}_{{\rm{H}}}\mu {m}_{p}{C}_{f}{v}_{0},\end{eqnarray} \tag{ 4 }$$

where μ, m_p are the mean atomic mass per proton (set to 1.2, see Gofford et al. 2015) and the proton mass, respectively, and we set r₀ = 5r_S following Section 3.2. Due to the weakness of the UFO emission features, we are not able to directly constrain the covering factor C_f from the observations, so we assume a mean value of 0.4 from the detection occurrence of UFOs in AGN samples (Tombesi et al. 2010; Igo et al. 2020; Matzeu et al. 2022). Then, we calculate the momentum transfer rate as ${\dot{P}}_{\mathrm{out}}\,=\,{\dot{M}}_{\mathrm{out}}{v}_{0}$ and the kinetic energy according to the special relativity formula as in Laurenti et al. (2021):

$$\begin{eqnarray}&&{\dot{E}}_{\mathrm{out}}\,=\,(\gamma -1)\cdot {\dot{M}}_{\mathrm{out}}{c}^{2},\end{eqnarray} \tag{ 5 }$$

where $\gamma \,=\,\tfrac{1}{\sqrt{1-{\beta }^{2}}},\beta \,=\,\tfrac{{v}_{0}}{c}$. We report ${\dot{M}}_{\mathrm{out}},{\dot{P}}_{\mathrm{out}},{\dot{E}}_{\mathrm{out}}$ for each spectrum in Table 4. Errors are calculated using the standard linear propagation approximation. We do not include the error associated with the black hole mass ${M}_{\mathrm{BH}}\,=\,{3.0}_{-1.5}^{+5.5}\,\times \,{10}^{7}{M}_{\odot }$, whose normalized interval (i.e., the error interval divided by the mean value) is higher than those of the WINE parameters reported in Table 4. Including the uncertainty on M_BH would result in a factor of ≈2 and ≈4 increase of the lower and upper bounds of the energetic, respectively. We also note that the commonly used lower limit for r₀, built from the assumption that v₀ corresponds to the escape velocity of the flow, is equal to ${r}_{\min }=\,{4.9}_{-0.7}^{+0.4},{22.3}_{-2.2}^{+6.5}{r}_{S}$ for the fastest and slowest velocities reported in Table 4, i.e., v₀ = 0.45, 0.21c; using these values would result in generally higher values of the energetic, which linearly scales with r₀ (see Equation (4)).

Interestingly, the fact that ${r}_{\min }\geqslant {r}_{0}$ implies that the detected UFO velocities are always lower than the escape ones; therefore, the wind will need additional acceleration, e.g., through radiation or magnetocentrifugal forces (e.g., Blandford & Payne 1982; Proga et al. 2000; Cui & Yuan 2020, but see Section 6 for further discussion), in order to overcome the gravitational force of the central black hole. In case of insufficient acceleration, the outflow may turn into a so-called failed wind, which is ubiquitously expected in all the radiative driving scenarios as a result of the over-ionization of the gas closer to the black hole (Higginbottom et al. 2014; Dannen et al. 2019), and in turn, is fundamental for the shielding of the outer gas layers (see Zappacosta et al. 2020 and references therein). Given the short distance from the black hole, this outflow may also be linked to the dynamics of the X-ray corona, whose physical dimension is supposed to vary according to accretion rate variations (see, e.g., Kara et al. 2019; Alston et al. 2020); as discussed in Section 1 and 5.3, NGC 2992 is indeed a strongly variable source.

Our derived values for the energetic are extremely high with respect to the typical UFO ones reported in the literature; as an example, ${\dot{E}}_{\mathrm{out}}$ is usually found to be around 0.1–1 times L_bol (see, e.g., Fiore et al. 2017), while here it is in the range of 2–23 L_bol. This is due to the very high values found for v₀, N_H, i.e., ∼0.3c, 6 × 10²⁴ cm⁻², making NGC 2992 an outlier with respect to the nearby Seyferts, which typically show v ∼0.1c, N_H ∼ 10²³ cm⁻² (see, e.g., Tombesi et al. 2011). We note that, as a result of the high observed velocities, the relativistic reduction of the gas opacity (and then, of its observed column density) is particularly significant. This, together with the lower collecting area of XMM-Newton above 10 keV, makes the features with low N_H more difficult to detect, possibly resulting in a bias in our analysis toward high N_H, and in turn, in an underestimate of the wind activity. This bias may explain, at least partly, the high column densities reported in Table 4 with respect to the Tombesi et al. (2011) mean values. Moreover, as we will discuss below, NGC 2992 shows evidence of a changing-look activity, hinting at differences in the accretion-ejection dynamics with respect to canonical Seyfert galaxies. Interestingly, UFOs detected in quasars show higher v₀, N_H with respect to Seyfert galaxies; as an example, Chartas et al. (2021) concentrate on a sample of quasars at 1.4 ≤ z ≤ 3.9, finding an average v₀ ≈ 0.3 c and N_H ≈ 4 × 10²³ cm⁻² (which, once relativistically corrected for 0.3c, corresponds to ${N}_{{\rm{H}}}^{\mathrm{rel}}\sim 8\,\times \,{10}^{23}\,{\mathrm{cm}}^{-2}$). Nardini et al. (2015) and Tombesi et al. (2015) both found similar ${v}_{0}\approx 0.25c,{N}_{{\rm{H}}}^{\mathrm{rel}}\,\approx 1.2\,\times \,{10}^{24}\,{\mathrm{cm}}^{-2}$ for the UFOs in PDS 456 (z = 0.184) and in IRAS F1119+3257 (z = 0.189), respectively.

Thanks to our time-resolved analysis we are able to estimate the duty cycle of the wind, i.e., the fraction of time during which it is observed. Considering that the amount of spurious detection within the full set of time slices amounts to ≈2 for our 2σ significance threshold (see Section 2.2 and Appendix B), we can conservatively assume that at least six of the XMM-Newton UFO detections (out of a total of eight) are not due to noise fluctuations. For a total of 50 time slices, this translates into a lower limit for the duty cycle of 6/50 = 12%. However, we caution that the observing bias discussed above may likely result in an underestimate of the duty cycle. We can derive mass and energy outflow rates representative of the total observing time as the average of the values reported in Table 4 times the wind duty cycle, thus obtaining ${\dot{M}}_{\mathrm{out}}^{\mathrm{tot}}\,=\,0.05{M}_{\odot }\,{\mathrm{yr}}^{-1},{\dot{E}}_{\mathrm{out}}^{\mathrm{tot}}\,=\,1.3{L}_{\mathrm{bol}}$. We also note that, from a historical point of view, UFO features have been detected in high-flux observations only, which however occurred only ≈30% of the time from the first X-ray observation of NGC 2992 in 1978 (Marinucci et al. 2018). Thus, the inferred duty cycle may not be regarded as representative of the global AGN lifetime.

UFO features were already detected in Marinucci et al. (2018) in two high-flux observations of NGC 2992, the 2003 XMM-Newton and the joint 2015 NuSTAR+Swift one, with 2–10 keV luminosities L_2–10 = 1.3 × 10⁴³, 7.6 × 10⁴² erg s⁻¹, respectively, comparable to that of our 2019 observations, L_2–10 = 1.0 × 10⁴³ erg s⁻¹. Our derived values for ${\dot{M}}_{\mathrm{out}},{\dot{E}}_{\mathrm{out}}$ are a factor >10 higher than those reported in Marinucci et al. (2018) due to the different wind properties: our best-fit values for N_H, v_out are in the range of 4–8 × 10²⁴ cm⁻² and 0.2–0.4c, while in 2003(2015) observations they found N_H = 2.2 × 10²³(1.8 × 10²²) cm⁻² and v₀ ≈ 0.2–0.3 c. We note that neglecting the relativistic reduction of the wind opacity would have led to a 35%–55% lower N_H (according to our range of v₀) with respect to the WINE-derived value. As an additional consistency check, we fit again the 2003 XMM-Newton observations using the same fitting model of Marinucci et al. (2018) and replacing the original wind tables, computed with the Cloudy code (Ferland et al. 2017), with our WINE tables. We obtain best-fit values consistent with those in Marinucci et al. (2018) and significantly lower than the present ones, further confirming the robustness of WINE when compared to different codes on one side, and on the other, the peculiarity of the wind features reported in this paper with respect to the archival ones. The high column densities inferred for the 2019 data set, easily exceeding the Compton-thickness threshold, suggest a scenario in which the observed features can be ascribed to independent, high-velocity clouds ejected from the disk, possibly embedded in a much lower N_H wind. We discuss such a scenario in the next subsection.

Equation (4) is calculated under the assumption of a spherical symmetric flow (Crenshaw et al. 2003; Crenshaw & Kraemer 2012), as commonly assumed for UFOs (see e.g., Tombesi et al. 2012; Nardini et al. 2015; Fiore et al. 2017; Chartas et al. 2021). However, the outflow in NGC 2992 shows a very low duty cycle and a short-term variation of its spectral appearance, equal to or lower than the 5 ks timescale of our observations. Moreover, we do not detect any emission feature associated with the UFO; therefore, we are not able to put constraints on its angular extension. Therefore, we also consider a complementary scenario, dubbed cloud scenario in which the UFO absorption features are due to outflowing (spherical) gas clouds passing through our line of sight for a time t ≈ 5 ks (see Bianchi et al. 2009 for a similar approach). For each time slice we derive the mass and energy outflows as follows. We compute the average radial distance of each cloud as

$$\begin{eqnarray}&&{r}_{\mathrm{avg}}\,=\,{r}_{0}+r({N}_{{\rm{H}}}/2)={r}_{0}+\displaystyle \frac{{N}_{{\rm{H}}}/2}{{n}_{0}},\end{eqnarray} \tag{ 6 }$$

where r(N_H/2) is the radius enclosing half of the wind column density and the last term is valid for a constant density wind (α = 0, see Equation (2)). Then, assuming that the cloud is rotating at a distance r_avg from the black hole with a Keplerian velocity v_rot, its dimension D can be estimated as $D=t\cdot {v}_{\mathrm{rot}}\,=\,t\cdot \sqrt{{{GM}}_{\mathrm{BH}}/{r}_{\mathrm{avg}}}$, where t = 5 ks. Finally, the mass of the cloud is given by

$$\begin{eqnarray}&&{M}_{\mathrm{out}}\,=\,\displaystyle \frac{4}{3}\pi {\left(D/2\right)}^{3}{n}_{0}.\end{eqnarray} \tag{ 7 }$$

We report M_out in Table 5, together with the associated energy E_out = (γ − 1)M_out c², where now we use the composition between outflowing and rotational velocity, i.e., ${\beta }_{+}\,=\sqrt{{{\rm{v}}}_{\mathrm{rot}}^{2}+{{\rm{v}}}_{0}^{2}}\ /{\rm{c}}$. We also calculate the average mass and energy rates as the ratio between the sum of M_out, E_out for all the observations divided by the total observing time, i.e., 250 ks. With respect to ${\dot{M}}_{\mathrm{out}}^{\mathrm{tot}},{\dot{E}}_{\mathrm{out}}^{\mathrm{tot}}$ computed assuming spherical symmetry, we obtain lower values by a factor of ≈100 and 2, respectively. Given the large number of assumptions, in Table 5 we only report the mean values, with the aim of demonstrating the importance of a proper description of the UFO geometry and dynamic for a reliable estimate of its energetic.

Table 5. Energetic According to the Cloud Scenario

Time Frame (ks)	Mout	Eout
	(10−6 M⊙)	(1047 erg)
10	0.65	1.71
35	7.55	13.46
130	2.71	2.75
60 + 191	27.7	67.6
202	3.61	7.35
220	2.82	3.78
278	0.06	0.12
295	3.06	5.94
Total	48.2	102.7
	${\dot{M}}_{\mathrm{out}}$	${\dot{E}}_{\mathrm{out}}$	${\dot{E}}_{\mathrm{out}}$
	(M⊙ yr−1)	(erg s−1)	(Lbol)
Time avg.	6.08 × 10−3	4.11 × 1043	0.23

Note. From top to bottom we report M_out and E_out (left to right) for the single observations and their sum. The last row shows the average mass and energy rates for the total observing time.

Download table as:  ASCIITypeset image

We are not able to find any correlation among the UFO parameters and between these and the properties of the continuum spectrum (e.g., photon index, normalization), both for the present and the Marinucci et al. (2018) observations. However, we note that the 2003 and 2015 UFO features were also accompanied by a broad component of the Fe Kα emission line, while instead both the wind and the emission are absent in the low-flux XMM-Newton observations from 2010–2013, which have L_2–10 < 4 × 10⁴² erg s⁻¹.

By analyzing all the optical and X-ray observations from 1978 up to 2021, Guolo et al. (2021) found evidence for a changing-look behavior of NGC 2992, which appears to be driven primarily by the intrinsic luminosity (i.e., the accretion rate), rather than by obscuration or transient events such as TDEs. Their main finding supporting this interpretation is the anti-correlation between L_2–10 and the FWHM of the Hα line; moreover, they also found a positive correlation between L_2–10 and (i) the flux of the Fe Kα line and (ii) the flux of the Hα line in the optical band. Hβ line was also detected, albeit with a lower confidence, for the brightest observations, showing a fairly constant Hα/Hβ flux ratio of ∼9. The luminosity threshold for the appearance of the Hα line is 2.6 × 10⁴² erg s⁻¹, corresponding to an Eddington ratio of ≈1%. Interestingly, they suggest the possibility that, due to the low accretion rate, the accretion disk could be thin, pressure dominated at large radii (as in the standard Shakura & Sunyaev 1973 picture) and radiatively inefficient in the innermost regions (Yuan & Narayan 2014), and the high-luminosity intervals are associated with instabilities at the boundary between these two regimes. UFOs have been always observed in high-luminosity states; unless this evidence is entirely due to the low constraining power of the low-flux spectra, it represents an indication of a link between the radiative efficiency of the inner region and the ejection of matter in the form of relativistic disk winds.