© The Authors 2026

1. Introduction

Active galactic nuclei (AGNs) are powered by the accretion of gas onto super massive black holes (SMBHs). AGNs release huge amounts of radiation that can strongly influence the nuclear environment, the SMBHs’ growth, and the evolution of the host galaxy, as witnessed by scaling relations between the SMBH mass and galactic properties (Kormendy & Richstone 1995; Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000; Kormendy & Ho 2013). AGN feedback, which is believed to be driven by powerful outflows from the accretion disk (Silk & Rees 1998; Zubovas & King 2012), could be a promising mechanism to account for some of the most significant unresolved issues in astronomy, such as the underlying reasons for the strong correlation between SMBH mass and the stellar bulge velocity dispersion of the host galaxy (the M_BH − σ relation) and the evolution of the quasar luminosity function over cosmic timescales across different wavelengths.

Laha et al. (2021) identify ionized outflows as an essential element of AGN feedback. Ultra-fast outflows (UFOs) are highly ionized (log ξ = 3 − 6 erg s⁻¹ cm, with $\xi =\frac{L_{\mathrm{ion}}}{n_H{r}^2}$ being the ionization parameter) outflows observed as strongly blueshifted absorption features in the hard X-ray band associated with iron (multiplet FeXXV and doublet FeXXVI, for which we assume centroid rest-frame energies at E_FeXXV = 6.7 keV and E_FeXXVI = 6.97 keV, respectively, which is sufficient at CCD resolution); they have mildly relativistic velocities up to v_out = 0.2 − 0.4c (Chartas et al. 2002; Reeves & Braito 2019; Laurenti et al. 2021). They are now routinely observed in ∼30% of local Seyferts (Tombesi et al. 2010, hereafter T10) and quasars (QSOs) (Matzeu et al. 2023, hereafter M23).

Due to their high velocity, high ionization, and extreme variability, these near-relativistic outflows are most likely launched from the inner part of the accretion disk in the deepest part of the SMBH gravitational potential. Variability on timescales as short as 1 day has been reported; therefore, the typical location has been estimated to be within 0.01–0.1 pc from the central SMBH (Kraemer et al. 2018; Fukumura et al. 2022). Some cases (Chartas & Canas 2018; Giustini et al. 2011) even report variability on intra-observation timescales (10⁴ − 10⁵ s), placing the origin of UFOs even closer to the nuclear region.

Strong correlations between the bolometric luminosity of the AGNs and both the kinetic power (Fiore et al. 2017) and outflow velocity (Matzeu et al. 2023; Xu et al. 2025) of UFOs have been detected. These correlations suggest that faster winds are preferentially driven in high-luminosity or high-accreting AGNs. While the origin of these ionized outflows is not fully understood, three primary mechanisms have been proposed: radiation driving, magnetic driving, and thermal driving.

Determining which mechanism dominates is challenging. Although radiation likely plays a crucial role, it faces significant physical limitations. UFOs appear to consist of dust-free, highly ionized material; this lack of dust and opacity in UV or X-ray lines renders standard radiatively driven flow models inefficient. While radiation scattering on free electrons contributes (Tombesi et al. 2013), it may be insufficient to accelerate the fastest winds (Luminari et al. 2020). Consequently, magneto-hydrodynamical (MHD) effects in the accretion disk are increasingly viewed as a necessary component for generating these high-velocity outflows (Laha et al. 2021).

To distinguish between these mechanisms and assess their impact on host galaxies, it is essential to systematically detect the fastest wind components. This is critical because the potential feedback impact of outflows scales steeply with velocity (${\dot{E}}_{\mathrm{out}}\propto {v_{\text{out}}}^3$) (King & Pounds 2015). While individual cases of v_out > 0.3c UFOs have been reported in local AGNs through targeted, multi-observatory campaigns (Reeves et al. 2018b,a; Luminari et al. 2023) and such velocities have been observed in high-z lensed QSOs where the features are redshifted into the standard band (Chartas et al. 2002, C21), a systematic exploration of the 10 − 12 keV regime in a flux-limited sample of local sources has never been attempted.

The specific aim of this work is to perform the first systematic, blind search for UFO absorption features in the 7 − 12 keV rest-frame band in a flux-limited sample of bright local AGN observed with XMM-Newton, employing a homogeneous approach in both spectral and statistical analysis. By extending beyond the 10 keV boundary adopted by all previous systematic UFO surveys (e.g., T10, Gofford et al. 2013; Igo et al. 2020), we aim to determine whether the extreme velocities (v_out > 0.3c) inferred from individual source studies and high-z observations are a common property of local AGN winds.

We also aim to understand whether the presence of such UFOs depends on the accretion properties of the sources within our parent sample, which Svoboda et al. (2017) characterized self-consistently using simultaneous UV and X-ray data, as is the case for nuclear winds in stellar-mass black hole binaries (BHBs). In fact, in the case of BHBs, strong connections between X-ray winds and the source’s accretion state have been found: winds appear mostly in sources in the soft state, characterized by a high accretion rate (Ponti et al. 2012).

The paper is organized as follows: Sect. 2 presents the aims of the study and shows the analyzed sample. In Sect. 3, we describe the data reduction process. In Sect. 4, we present the methodology for data analysis and evaluate the statistical relevance and properties of the detected absorption. In Sect. 5, we present and discuss the most significant results of the analysis we performed. In Sect. 6, a summary of the results is provided. Throughout this paper, the following cosmological constants are assumed: H₀ = 70 km s⁻¹ Mpc⁻¹, Ω_Λ = 0.73, and Ω_M = 0.27.

2. Sample selection

We opted for the Svoboda et al. (2017) catalog updated with observations up to 2024, which contains numerous sources for which the UV and X-ray emission have been measured through simultaneous coverage by the OM and EPIC instruments on board XMM-Newton. Since a high number of spectral counts is required to detect UFOs, the following selection criterion was applied to the catalog to obtain our sample: F_{2 − 12 keV} ≥ 1.5 × 10⁻¹¹ erg s⁻¹ cm⁻².

Svoboda et al. (2017) draw an analogy between stellar mass black holes in different spectral states and populations of AGNs. They show that AGNs occupy analogous regions of the hardness-flux (or accretion) diagram when the AGN energy output is measured in a way that separates the thermal disk from the hard X-ray corona. For each source, the bolometric output is decomposed into disk luminosity and hot corona emission. The redshift and Galactic extinction corrected UV flux were used to estimate the disk luminosity L_D at 2910 Å (see Svoboda et al. 2017 for details), which can be defined as

$$\begin{array}{c}\hfill L_D=A·4\pi D_L^2\lambda F_{\lambda ,2910\mathrm{\AA }},\end{array}$$

where D_L is the luminosity distance constrained from the redshift measurement and A is an arbitrary factor, which can be chosen so that the sum of the disk and power-law luminosity, L_tot, roughly corresponds to the bolometric luminosity L_Bol. The corona (power-law) luminosity (L_P) can be defined from the X-ray luminosity by extrapolating it to the energy interval 0.1 − 100 keV

$$\begin{array}{c}\hfill L_P=4\pi D_L^2{F}_{0.1-100\mathrm{keV}},\end{array}$$

where F_{0.1 − 100 keV} is the X-ray flux at the 0.1 − 100 keV energy range calculated by an extrapolation of the observed 2–12 keV flux. In the end, we can assume L_tot ≃ L_Bol = L_D + L_P. The spectral hardness parameter H is defined as:

$$\begin{array}{c}\hfill H=\frac{L_P}{L_P+L_D}·\end{array}$$

This highlights the importance of having simultaneous measures in UV and X-ray bands: not only does UV data lead to a precise estimation of the accretion-disk luminosity, but is also crucial both to reduce the effect of variability in the measurements of the disk and power-law luminosities, and to constrain the exact value of H, and thus evaluate the ionization state of the source. The results of Svoboda et al. (2017) show that radio loudness increases on average with H, and with decreasing Eddington Ratio (${\lambda }_{\mathrm{Edd}}=\frac{L_{\mathrm{Bol}}}{L_{\mathrm{Edd}}}$, that is a measure of the accretion rate of the SMBH), suggesting that the jet emission occurs in the low accretion states, consistent with BHBs (Moravec et al. 2022). The original catalog contained multiple observations for several sources; we decided to analyze one observation per AGN. The longest observation for each source in the XMM-Newton public archive was selected to ensure good spectral statistical quality while avoiding source-dependent bias in the sample. This led to the analysis of 33 XMM-Newton observations for the AGN listed in Table A.1, while Table A.2 summarizes the details of the selected observations. Figure 1 shows the distribution of L_Bol versus redshift for the sources of our sample.

Fig. 1. Bolometric luminosity of the 33 AGNs considered in the sample (see Table A.1) plotted against their redshift.

L_Bol and the H information are taken from the Svoboda et al. (2017) catalog. We used more recent observations than those from 2017 in 9 cases (out of 33) where significantly longer exposures were available. We observe the typical long-term X-ray variability of up to 50%. However, given the relative contribution of L_D and L_P, the impact on L_Bol and H is limited (up to ∼15%), therefore, we keep the values from Svoboda et al. (2017).

For the masses of the selected black holes, we report the values inferred by recent studies using the reverberation mapping method or the single-epoch method, based on Balmer line properties, in Table A.1. Finally, from L_Bol and BH masses, we calculated the Eddington ratios.

3. Data reduction

The XMM-Newton data reduction was carried out with the Science Analysis System (SAS) software package 21.0.0, specifically designed to reduce and analyze data collected by the XMM-Newton observatory (Schartel et al. 2022). In this work, we have studied only the spectra extracted from the EPIC-pn camera, as it has the highest collecting area, especially in the hard band. We started from the EPIC observation data files (ODFs) available in the XMM-Newton archive, and performed standard data reduction. We removed noise events and periods dominated by soft proton flares, extracting a single event, high-energy (10–12 keV) light curve to identify intervals of flaring background; a threshold in counts/seconds was chosen to define the “low background” intervals (0.4 for EPIC-pn). Then, we obtained the net exposure time (see Table A.2) for each observation, representing the time available for analysis. For most of them, around 60–70% of the nominal duration was usable.

During the creation of the spectra, we visually checked if the observations were affected by pile-up, which can occur for bright sources (F_{0.2 − 12 keV} ≳ 10⁻¹¹ erg s⁻¹ cm⁻² for EPIC pn, depending on the observation mode). To overcome the pile-up in SBS 1301+540, whose observations were taken in full-window mode, we repeated the extraction process, excluding the PSF center (see Schartel et al. 2022) over an area of radius 200–250 pixels and verifying that pile-up was negligible in that case. The distribution of the net counts collected in the 2–12 keV band (see Table A.2) for the pn camera is shown in Fig. 2, with a median value of ∼1.4 × 10⁵ counts. We then extracted the spectra for source and background, and finally created the RMF and ARF (“response matrices”) using the standard SAS tools ARFGEN and RMFGEN.

Fig. 2. Distribution of the net counts in the 2–12 keV band for the pn camera (see Table A.2) after removing periods dominated by soft proton flares.

As we extend the spectral band for the analysis to 12 keV, a regime where the effective area of the XMM-Newton telescopes decreases significantly, calibration uncertainties increase, and particle background dominates the overall instrument background, we applied three different prescriptions to estimate the background spectrum:

• the observation background: we extracted spectra from a spatial region close to the source extraction region on the same pn CCD, and at the same approximate distance from the readout node in pn to ensure that similar corrections for Charge Transfer Efficiency apply. Since the vast majority of the observations were taken in small window mode, this standard technique was considered not sufficient alone, because the small chip area used in this mode is almost entirely dominated by the source emission;

• the blank sky (BS) background: we extracted background spectra from black field event lists (Carter & Read 2007) for the full window mode observation, using the same spatial region in detector coordinates as the source;

• the filter wheel closed (FWC) background: we extracted spectra from exposures where a filter is used to occult the optical path to the sky, available for all pn observing modes. These exposures enable a direct estimate of the instrumental unfocused particle background. As in the case of the blank sky background, spectra were extracted using the same spatial region in detector coordinates as the source.

All the sources had their spectra extracted first with the observational background method. The spectra that hinted at the presence of a nuclear wind were then re-extracted with both the BS background, where possible, and with the FWC background in all cases.

To properly bin the spectra, we adopted the optimal binning method outlined by Kaastra & Bleeker (2016) (KB), which accounts for source and background levels, as well as instrument resolution. This scheme, implemented via ftgrouppha in heasoft¹, provides a variable bin size matched to the energy-dependent resolution of the EPIC-pn CCD and the spectral SN, and is specifically designed to maximize the information content for narrow, unresolved spectral features. M23 systematically compared this method against three other commonly used binning prescriptions (grpmin1, SN5, and OS3grp20) in the context of Fe K absorption line searches in XMM-Newton spectra of the SUBWAYS sample. They found that the KB and grpmin1 binning produce nearly identical results in terms of detection rate and line significance, with KB making the narrow features more easily identifiable in the spectra, while coarser schemes can dilute faint absorption features, making the KB binning the optimal choice for this type of analysis.

Since the KB optimal binning does not impose a minimum number of counts per bin, some spectral bins may contain too few counts for the χ² approximation to be valid. We therefore adopted the Cash Statistic or C_stat (Cash 1979) throughout the spectral analysis, which provides unbiased parameter estimates and confidence intervals in the Poissonian regime, as demonstrated by Humphrey et al. (2009) and Kaastra (2017), even when the number of counts per bin is low.

4. Data analysis

In this chapter, we outline the methodology used to fit the observed spectra, to search for UFO absorption features, and to assess the significance of the detected winds.

4.1. Continuum modeling

To model the 2–12 keV rest-frame spectra, we used a Baseline Model (BM) made of a combination of emission components due to direct continuum and reflection characteristic of AGN emission, as a starting point to reproduce the standard features present in AGNs, which we can write as: ZTBABS*(MYTORUS_S+MYTORUS_L+ZPOW). The components of the BM are the following: the power-law represents the continuum generated by Inverse Compton scattering of UV photons from the accretion disk and semi-relativistic electrons in the hot corona. ZTBABS is a multiplicative component at the source redshift, used to reproduce the effect of cold absorption due to the material along the line of sight to the observer. The Galactic column density of the Milky Way was not considered since we are fitting at energies > 2 keV where its contribution is negligible for our sources.

MYTORUS (Murphy & Yaqoob 2009) is a model that reproduces the effects of reflection by the torus of the primary power-law emission, that is the Compton-scattered continuum (MYTorus_S) plus the associated fluorescent lines (MYTorus_L). The inclination angle and relative intensity are fixed to their default values, while the column density is set to be the same for the continuum and fluorescent lines components (coupled mode), with typical best-fit values around N_H ∼ 10²³ cm⁻².

We adopted MYTorus to model the reprocessed emission from distant circumnuclear material, as it self-consistently accounts for the Compton-scattered continuum and the associated fluorescent Fe K lines. This physically motivated model has become the standard for spectral fitting of local Seyfert galaxies, superseding older infinite-slab approximations such as pexrav (Magdziarz & Zdziarski 1995) or pexmon (Nandra et al. 2007) (see Yaqoob 2012; Baloković et al. 2018 for detailed comparisons).

Some spectra required additional components to adequately describe the continuum and other emission features. When necessary, the model above was modified, by including the presence of a second power-law to reproduce the soft excess and/or Gaussian emission lines to model the broad component from Fe Kα with KDBLUR2, a convolution model applied to the Gaussian emission line to mimic relativistic effects from an accretion disk around the SMBH.

In some cases, it was also necessary to add Gaussian absorption lines at energies < 6 keV to obtain a good model for the spectra. These features may represent UFO absorption from other elements (e.g., Si XIV, SXV, ArXVIII, or Ca XIX) or possible inflows. To distinguish between the two possibilities, a full modeling of the ionized absorbers with photoionization models such as Xstar is needed. However, we defer the analysis of such features to a forthcoming paper, as we focus on detecting UFOs through Fe band features. We illustrate the BM model applied to sources that exhibit a UFO in Appendix B.

We verified that none of the X-ray spectra of the radio-loud AGNs in the sample were dominated by the radio-jet contamination; in particular Madsen et al. (2015) shows how in 3C 273, the most radio-loud source in the sample, with R ≫ 100, the jet contributes significantly in the emission only over 20 − 30 keV, which it is outside the analyzed range, therefore it does not affect our modeling.

4.2. Search for Fe K absorption

To blindly search for any absorption signature, we added a Gaussian absorption line (zgauss) to the best-fit continuum model, with its energy free to vary in the 7–12 keV rest-frame range, negative normalization, and width also free to vary up to a maximum of 1 keV, to avoid fitting continuum fluctuations. The fit identifies the strongest absorption feature in this band and produces the Best Fit Model (BFM). This procedure follows the same logic as the blind-search method of Miniutti & Fabian (2006), adopted by T10 and M23. The main difference is that, in those works, the line width is frozen, leaving 2 free parameters, and the Δχ² or C_stat improvement is mapped onto a two-dimensional energy–intensity grid.

In our case, the additional free parameter (line width) makes the construction of such maps impractical; instead, the fit converges to the strongest feature, and its significance is evaluated via the ΔC_stat improvement for 3 degrees of freedom. We adopted ΔC_stat > 7.815 as the detection threshold, corresponding to 95% confidence for 3 degrees of freedom (Lampton et al. 1976). For comparison, T10 and M23 used ΔC_stat > 9.21, corresponding to 99% confidence for 2 degrees of freedom, since they froze the width parameter. Features exceeding our threshold were then subjected to Monte Carlo (MC) simulations to robustly assess their significance, as described in Sect. 4.2.1.

In Table E.1 we list the absorption lines found in the blind search, which show a ΔC_stat < 7.815 and are therefore considered not significant. Several of these features correspond to UFOs detected in previous works (e.g., Gofford et al. 2013, T10), but were discarded due to our conservative approach.

4.2.1. Statistical analysis and robustness of results

Following a robustly conservative approach, in the analysis here presented, we select candidate absorption features only if they meet the statistical criteria for detection with all three background subtraction methods outlined in Sect. 3, and we report significance results relative to the set up that led to the smallest improvement in the Cash statistic between the model with an absorption line and the one without it.

For the features considered significant on the basis of the ΔC_stat value, we implemented MC simulations to estimate more accurately their statistical significance:

• We simulated 1000 EPIC-pn spectra using the FAKEIT command in XSPEC, assuming the BM and the same exposure times, background levels, and response matrices of the real observation;

• We apply and fit both BM and BFM on the fake, line-less spectra, evaluating for every simulation the ΔC_stat between the two models, varying the line energy over the entire 2–12 keV band and looking for the strongest random fluctuations in the simulated line-less spectra;

• We compare the values of ΔC_stat of each simulation with the difference in C_stat evaluated on the real spectrum of the source and calculate the fraction of simulations where the ΔC_stat(sim) >ΔC_stat(real).

This procedure aims to estimate the level of significance of a detected absorption line: if more than α = 5% of the 1000 simulations show a Cash statistic improvement greater than the observed ΔC_stat, we reject, in a conservative approach, the tentative absorption feature, as it could be a possible noise statistical fluctuation. Figure 3 shows the result of this test applied to the absorption feature detected in NGC 2617, which shows a significance (1 − α) over 99%. The significance values associated with the absorption lines detected are shown in Table 1.

Fig. 3. Result of the MC simulations for the spectrum of NGC 2617. The blue histogram shows the distribution of ΔCstat obtained by fitting the BM and BFM models to the simulated line-less spectra. The red line shows the observed ΔCstat for this source. α represents the fraction of MC simulations showing a ΔCstat greater than the observed one. In this case, the absorption feature is detected with a confidence 1 − α > 99%.

4.2.2. Uncertainties of the absorption parameters

In this section, we describe how we estimated the statistical uncertainties on the best-fit parameters of the absorption lines using MC simulations. For each source, we ran a MC simulation that generated 1000 spectra based on the BFM and fit them with it, leaving the absorption line parameters free. We define the confidence interval for each parameter as the interval that encompasses 95% of the distribution of best-fit values from the 1000 simulations.

We interpreted the absorption lines as blueshifted Fe XXVI, which has a rest-frame energy of 6.97 keV. From the comparison between the observed energy E_obs, estimated as the median of the distribution of the simulations, and the rest-frame energy of Fe XXVI, we calculated the outflow velocity from this formula, corrected for cosmological redshift: $\frac{E_{\mathrm{obs}}}{E_{\mathrm{FeXXVI}}}=\sqrt{(1-\frac{v_{\text{out}}}{\mathrm{c}})/(1+\frac{v_{\text{out}}}{\mathrm{c}})}$.

Table 1 reports the properties of the absorption lines detected in our sample and the estimation of v_out from the shift in energy. We also compared our results with those obtained using the Bayesian X-ray Analysis tool (BXA; Buchner et al. 2014). The comparison of the results from the two methods is shown in Appendix C. The results on UFO properties are perfectly consistent for all the sources (see Fig. C.1).

4.3. Photoionization modeling of Fe K features

In order to characterize the physical properties of the absorbers and confirm the nature of UFOs, we replaced the Gaussian absorption profiles detected with > 95% confidence level with Xstar photoionization tables (Bautista & Kallman 2001), which provide a first-order physical measurements of the absorbing medium of the outflows. In particular, we are able to quantify the ionization parameter ξ, the column density of the medium N_H, and the systemic redshift of the material relative to the observed one, which translates into the outflow velocity v_out.

For each source, we used Xstar grids covering turbulent velocities in the range σ_turb in the span 300 − 5000 km/s, choosing the table that minimized the residuals of the fitting, since different values of σ_turb are necessary to properly reproduce the observed widths of the Fe K absorption lines.

The best fit values obtained from the fit with Xstar tables are shown in Table 2. For each source, we first found the value of v_turb that best reproduced the detected absorption lines, using a grid that treated v_turb as a free parameter; then we selected the tables with the closest fixed values. The Xstar modeling reproduces results comparable with the Gaussian absorption line component of the BFM, identifying the features as blueshifted Fe XXVI absorption line, or a blend of Fe XXV and Fe XXVI. For almost all sources, the Xstar modeling led to an improvement in the statistic relative to the Baseline model, with ΔC_stat > 7.815. In some cases, the Xstar tables fit the residuals better than a Gaussian absorption line model, suggesting that secondary features are modeled together with the main feature. In one case, instead, PG 2304+042, the improvement in the fit with the Xstar table is ΔC_stat < 7.815. This source, as shown in Fig. B.3, exhibits a very deep, narrow absorption feature, with EQW > 100 eV, and even the grids with the smallest v_turb (i.e., 300 km/s) struggle to fit the observed residuals entirely. The narrow, deep profile in PG 2304+042 may require a multi-zone absorber or a higher-resolution grid to be fully reproduced; nonetheless, the Gaussian parameters are fully consistent with blueshifted Fe XXV/XXVI at v_out ∼ 0.33c.

5. Results and discussion

5.1. Extending the high end of the UFO velocity distribution

Table 1 presents the best-fit parameter values of the analyzed absorption lines, along with their corresponding velocities and uncertainties, at 95% significance level. Features classified as UFO based on their velocity are marked with * symbol. Although we detected absorption lines with high significance in 8 out of 33 sources, we classified as UFO only absorption features blueshifted by v_out > 10 000 km s⁻¹, as in the work of T10, so NGC 985 (which has an inflow), NGC 4051, and one of the lines in NGC 5548 are excluded. In the following, we compare our work with that of T10, as both utilize the highest-quality spectra available in the XMM-Newton archive. Both samples are predominantly composed of nearby Seyferts, despite our sample spanning a larger luminosity and redshift range.

We note that comprehensive analyses of UFO properties across larger, heterogeneous compilations have been presented in recent literature. Gianolli et al. (2024) (G24 hereafter), combining three independent subsamples, performed a comprehensive population study of UFO properties across a wide range of redshift and luminosity; we compare our results with their correlations in Sect. 5.2. Yamada et al. (2024) assembled a legacy database of 583 X-ray winds in 132 AGN at z = 0 − 4, classifying them into UFOs, low-ionization UFOs, and warm absorbers. Laurenti et al. (2026) studied 122 robust UFO detections, exploring for the first time the dependence of UFO properties on AGN classes. In this work, instead, we perform a homogeneous blind search in a previously unexplored energy range (10 − 12 keV) within a single flux-limited sample.

We report the presence of UFOs in six AGN corresponding to a fraction of $f_{\mathrm{UFO}}=(18.2_{-7.3}^{+11.0})\%$, where the uncertainties are computed with the Gehrels formulae (Gehrels 1986) for low number statistics. We note that analyzing a single observation per source, while avoiding source-dependent biases, reduces the probability of catching a transient UFO in any given target. To quantify this effect, we compare with the strategy of T10, which analyzed an average of ∼2.4 observations per source. If we restrict the T10 sample to UFOs detected only in the longest observation of each source – mirroring our selection – the detection rate drops from ∼35% to ∼14% (6 UFOs over 42 sources), fully consistent with our measured fraction. Rescaling by the average number of observations recovers the original T10 rate (14%×2.4 ∼ 34%), supporting the interpretation that the per-observation detection probability is comparable between the two studies, and that the difference in the observed UFO fraction is driven by the number of independent epochs sampled rather than by differences in sensitivity or sample properties. This also underscores the impact of UFO variability and duty cycle on single-epoch detection fractions, and motivates the multi-epoch extension of this work discussed in Sect. 6.

Figure 4 (left) shows the comparison between the velocity distribution obtained from our sample and the one found by T10. It is evident that our distribution is shifted toward higher velocities, thanks to the extension of the UFO search up to 12 keV. A KS test between the two distributions yields a p-value of 0.028, corresponding to a confidence level of 97.2% that they are statistically different. It is important to note that we would not have been able to detect such features if our analysis had been limited to 10 keV, as done in T10 and Gofford et al. (2013). In fact, Mkn 766, NGC 5548, NGC 7469, and PG 2304+042 show UFOs with velocities higher than 0.3c, which are absorption features detected above 10 keV.

Fig. 4. Left: Comparison between the UFO velocity distributions of our sample (magenta) with the T10 sample (purple). Right: Same for the EQW distribution.

As previously discussed, revealing UFOs with v_out > 0.3c (i.e., v_out > 90 000 km/s), is difficult as the absorption features are highly blueshifted and, therefore, are located at energies where the effective area of the operational X-ray telescopes decreases. Furthermore, it is well known that the EPIC-pn camera calibration is less constrained in this high-energy regime. To validate the robustness of our detections above 10 keV against potential instrumental artifacts, we cross-checked our results using simultaneous NuSTAR observations, which offer good sensitivity and calibration in the hard X-ray band, albeit with lower spectral resolution. The results of this validation are detailed in Appendix D.

Given the difficulty, we can study these highly blueshifted features only with an accurate characterization of the background, and if the signal-to-noise ratio of the observation is very high. Remarkably, Luminari et al. (2023) presented the discovery of an extreme UFO at E ∼ 11 keV with 3σ confidence level in the nearby Seyfert galaxy NGC 2992 caught during a high-flux X-ray state. It is worth noting that winds with v_out > 0.3c have been observed in high-z QSOs (see Chartas et al. 2002, 2021, C21 hereafter), since UFOs absorption features fall, due to the redshift of the sources, into the observability range of XMM-Newton and Chandra.

We detected a UFO in NGC 5548 and NGC 7469 for the first time. Both sources reveal a highly complex, cold, and low-ionized absorption in the soft band (Kaastra et al. 2014, Behar et al. 2017), and the presence of a UFO lends support to a multiphase nuclear outflow scenario. Moreover, NGC 5548 exhibits two absorption lines, indicating that the detected UFOs are likely composed of different components at various velocities, thereby validating the hypothesis that AGN outflows are inherently multiphase.

Figure 4 (right) presents the distribution of equivalent widths (EQW) for UFOs in our sample compared with T10. Even if the difference between the two samples is not statistically significant (the KS test yields a significance of only 70%), the EQWs obtained in this work are larger than 40–50 eV, while the T10 distribution goes down to 10–20 eV, and ∼40 eV is close to the median EQW. The higher EQWs found in this work can possibly be explained by Fig. 5. Here we show the EQWs of the absorption lines versus the velocity of the nuclear winds of both samples. We divided the plane into three bins, with limits of v_out < 0.03c and v_out > 0.3c, separating the low-velocity UFOs, those with velocities in the typical range, and those detected at high velocity via the 12 keV analysis extension. We calculated the global mean value of v_out and EQWs for UFOs in each bin and tested the linear correlation between the two quantities, first with the Pearson test, which reported a confidence level over 99%, and then with LINMIX (Kelly 2007), a Bayesian tool for linear regression that takes into account errors on both the x and y axes. The red line represents the linear best fit, while the shaded area represents its 95% confidence interval. The latter shows a weak positive correlation between outflow velocity and EQW. Having 2/3 of the detected UFOs in the high-velocity regime means that we are also shifting the EQW distribution toward higher values. Indeed, Table 1 shows that we are sensitive to weak features with EQW ∼10 − 20 eV, but these are all in the low velocity regime.

Fig. 5. Comparison between the UFOs’ equivalent width and outflow velocity distributions of our sample (in red) with the sample from T10 (in blue). The dashed plum line marks the threshold of 0.03c, while the dashed turquoise line marks 0.3c. The green, orange, and purple points are the mean of the velocity and EQW in the three velocity bins. The red line and shaded area show the linear best fit and its 95% confidence interval computed with LINMIX, taking into account errors on both vout and EQW.

Figure 6, adapted from Luminari et al. (2021), shows the outflow velocity of UFOs versus λ_Edd of the AGN for our sample (in black) and compared to Fiore et al. (2017), Gofford et al. (2015) and several individual sources (see Appendix E of Luminari et al. 2021 for the complete list of references). While it appears that the v_out increases with the accretion rate of the AGNs, our sources appear to have faster UFOs than local sources with higher λ_Edd. The figure also shows the allowed velocity ranges in a radiative driving framework, including special relativistic effects (Luminari et al. 2020), for the terminal velocity v_t and the maximum velocity v_max. The latter represents an upper limit for the observable velocity, for a nuclear wind launched at r ≥ 50 R_G, where R_G is the gravitational radius, assumed in Luminari et al. (2021) as the lower bound for the UFO launching radius. The high velocities of the UFOs in our sample can hardly be explained uniquely with a radiative driving model, suggesting the involvement of other launching mechanisms, such as MHD (Laha et al. 2021).

Fig. 6. Comparison between UFO velocities from the literature (see label) and the results of the present work (red circles) as a function of λEdd (from Luminari et al. 2021). The theoretical limit, according to the terminal (maximum) velocity vt (vmax), for R0 ≥ 50 rG is shown in the green (light-blue) shaded area.

Most of the sources in our sample were previously analyzed by other works in search of UFOs. However, the six UFOs identified in our analysis were never detected before (but see below). In Table F.1, we present a comparison of our results with those published in the literature for the same sources. In some cases (5 out of 16), our results differ from those published on the same observations, either because we identify absorption features that were not originally discovered or vice versa.

There are several possible explanations for this discrepancy. (1) we performed the fit on a larger energy range, enabling an extension of the velocity parameter space under study and, in some cases, a better determination of the underlying continuum; (2) we have investigated the detection and properties of the absorption features against all different available methods to determine the instrument particle background, which constitute the dominant background EPIC-pn component in hard X-rays; (3) we have left the line width as a free parameter and used an initial threshold of 95% confidence for three parameters (ΔC_stat > 7.815), while the typical approach is to adopt a narrow, unresolved line, and (4) finally, we are using state-of-the art spectral analysis methods, which have been perfected in the last few years such as the proper statistical treatment of the Poissonian maximum likelihood (Kaastra 2017), optimal binning (Kaastra & Bleeker 2016), Monte Carlo and Bayesian methods to estimate the posterior absorption feature parameter distributions (Buchner et al. 2014).

An interesting example is Mkn 766. Several low-velocity features have been reported in the past for this source, based on observations different from the one analyzed here (v_out = 0.44 ± 0.007c, 0.091 ± 0.004c in T10 and v_out = 0.061 ± 0.008c, 0.017 ± 0.004c in Gofford et al. 2013). On the contrary, in the same observation analyzed here, a significant feature was identified at the same energies (∼9.5 keV) both in T10 and in Pounds et al. (2003) but interpreted as a 9.28 keV H-like Iron edge outflowing at ∼ 0.05c and linked to a transient X-ray flare, rather than a high-velocity UFO. This highlights the importance of an accurate continuum modeling above the absorption feature, which allows us to distinguish between the two scenarios.

It is worth noting also that in some of the sources (e.g., Mrk 509) in which previous studies found UFOs, we actually detect UFO-related absorption lines, but with insufficient significance (see Sect. 4.2.1). They are shown in Table E.1 in the Appendix.

Finally, by examining the literature, it is possible to find cases in which different results are obtained when comparing observations of the same source taken at different epochs. For example, 3C 120 has a UFO with v_out = 0.161c (Tombesi et al. 2014), and NGC 4051 has a faster UFO with v_out = 0.150c (T10), both detected in observations not analyzed in this work. This is expected and highlights the transient nature of UFOs.

5.2. Physical properties of the UFOs

Using the results of the fit with Xstar photoionization tables, we can place the physical properties of the UFOs detected in this work in the broader context of AGN wind studies, by updating the correlations between the observed outflow parameters (ξ, N_H and v_out) presented by G24, who combined the T10 sample with results from SUBWAYS (M23), and the high-z sample from C21. Figure 7 shows the three correlation planes from their Fig. 4, now including our new detections (red circles). The best-fit relations reported by G24 are shown as black lines, while the updated fits incorporating our data are shown in red, with the shaded area representing the 68% confidence interval.

Fig. 7. Correlations between observed UFO parameters: ionization parameter log(ξ) vs. outflow velocity log(vout) (panel a), column density log(NH) vs. log(vout) (panel b), and log(NH) vs. log(ξ) (panel c). Blue diamonds, green triangles, and black squares represent the T10, M23, and C21 samples from G24, respectively; red circles are the detections from this work. The black solid lines show the original best-fit relations, while the red solid lines and shaded areas show the updated fits and their 68% confidence intervals after including our new detections.

The most striking change is in panel (a): the log(ξ)– log(v_out) correlation flattens significantly with slope $0.{22}_{-0.22}^{+0.20}$ (with respect to $0.{71}_{-0.13}^{+0.19}$) and is no longer statistically significant. This is driven by the fact that several of our UFOs have high velocities (v_out > 0.3c) but moderate ionization parameters (log(ξ)∼3.8 − 4.2), populating a previously unexplored region of the parameter space. This suggests that the outflow velocity is not primarily governed by the gas ionization state, as would be expected in radiative-driving scenarios, lending further support to MHD-driven acceleration models (e.g., Fukumura et al. 2017, 2022).

In panel (b), the log(N_H)−log(v_out) relation remains positive but becomes slightly flatter (slope $1.{20}_{-0.15}^{+0.19}$ versus $2.1_{-0.13}^{+0.15}$), consistent with the fact that our high-velocity detections, enabled by the extension to 12 keV, do not require the very large column densities (log(N_H/cm⁻²) > 24) typical of the C21 high-redshift quasars. The log(N_H)−log(ξ) correlation in panel (c) remains essentially unchanged (slope $0.{54}_{-0.13}^{+0.15}$ versus $0.{52}_{-0.18}^{+0.15}$), as our sources populate the same ranges as the T10 and M23 samples in this plane.

It is worth noting that the original G24 correlations in the log(ξ)−log(v_out) and log(N_H)−log(v_out) planes are largely driven by the high-redshift quasars sample, whose absorber parameters are subject to both significant systematic uncertainties, being strongly model-dependent, and observationally selection biased, as the lower spectral quality of high-z data restricts detections to the strongest absorption features, favoring intrinsically high N_H and ξ values. The flattening we observe when our local, high-quality detections are included highlights the need for a homogeneous reanalysis of the UFO population across redshift.

Finally, the outflow velocity measured by Gaussian fitting is comparable with the findings of Xstar tables, confirming that the observed features are due to either Fe XXV or XXVI Kα lines, or a blend of the two. A Kolmogorov-Smirnov test applied to the velocity results shows that the photoionization and the Gaussian modeling of the Fe K absorption features have the same distribution, with a 96.3% confidence level.

5.3. UFO versus AGN properties

One of the goals of our analysis consist to understand if the incidence of UFOs depends on the AGN properties, such as the mass of the SMBHs, as L_bol, λ_Edd, and H. We divided our sample into two groups: one with only sources showing UFO detections and the other with sources that did not. Then we studied the distributions of the two groups with respect to accretion properties (as shown in Fig. 8) of AGN and verified whether there are significant differences between the two groups using a KS test.

Fig. 8. Distribution of AGN with (orange) and without UFOs (green) with respect to their bolometric luminosity, BH mass, Eddington ratio, and Hhardnessparameter (see Table A.1).

We emphasize that such a comparison between UFO-hosting and non-UFO-hosting sources is restricted to the 33 AGN in our parent sample, for which accretion properties (L_bol, H, λ_Edd) were derived self-consistently from simultaneous XMM-Newton OM and EPIC data by Svoboda et al. (2017). Our sample is not designed to provide a comprehensive census of the UFO population, and our results on UFO incidence versus AGN properties should not be extrapolated to the general AGN population. For broader analyses of UFO–accretion correlations across heterogeneous compilations, we refer the reader to G24, Yamada et al. (2024) and Laurenti et al. (2026), for example. Within this limited scope, we investigate whether the incidence of UFOs in our sample depends on the AGN properties.

The KS lends to the following confidence levels: 1 − p = 73% for L_bol, 1 − p = 87% for the mass of the SMBHs, 1 − p = 30% for the λ_Edd, and 1 − p = 4% for H values. Therefore, in none of the cases can we state that the two samples come from different distributions, as none of the values exceeds the 95% confidence level. It is also important to highlight the small size of the UFO sample in this work, making it challenging to derive firm statistical conclusions.

These results conflict with what has been found for stellar BHBs: our study, in fact, indicates that the launch of UFOs in AGNs requires physical conditions of the accretion disk and corona that differ from those underpinning the production of outflows in stellar black holes, indicating a dependence more on the duty cycle of the sources than on their accretion properties.

Although our analysis found no UFOs in AGNs with M_BH > 10⁸, the work of M23 revealed several winds in this mass range, suggesting this is just a small sample size effect and there is no real dependence between this AGN property and the presence of UFOs.

6. Conclusions

We have performed an investigation on the presence and properties of UFOs, the most powerful, highly ionized nuclear AGN winds, in a sample of bright local Seyfert galaxies. We collected high-quality XMM-Newton pn spectra for 33 sources. The sources (reported in Table A.1) were selected from the same catalog studied by Svoboda et al. (2017), choosing a threshold in flux of F_2−12 keV = 1.5 × 10⁻¹¹ erg s⁻¹ cm⁻². We searched for absorption features between 7 − 12 keV rest-frame, characteristic of UFOs, by adding Gaussian absorption lines to the spectral continuum models. Furthermore, we implemented MC simulations to evaluate the statistical significance of these absorptions, as well as the confidence levels for the energy, width, and strength of each line. The main results of this study are summarized below:

1. We report highly significant (95% confidence level) detections of UFOs in six sources of the sample, corresponding to a fraction of $f_{\mathrm{UFO}}=(18.2_{-7.3}^{+11.0})\%$. We highlight that these winds were previously undetected.

2. The extension to E > 10 keV in the search for UFO-related absorption features has led to the detection of four UFOs with very high velocities (v_out > 0.3c) in Mkn 766, NGC 5548, NGC 7469, and PG 2304+042. The reliability of the highest-velocity detection methodology was successfully validated against simultaneous NuSTAR data for NGC 7469 (Appendix D).

3. The photoionization modeling of Fe K features through Xstar tables agrees with the results obtained with the Gaussian modeling, confirming our results.

4. The common presence of UFOs of such high velocity indicates that several distinct acceleration mechanisms may be operating in AGNs. This also suggests a broader velocity distribution for X-ray winds, which may affect the role of UFOs in AGN feedback processes.

5. The comparison of our results with the outcomes of previous studies on different observations allowed us to monitor the intrinsic variability UFOs in AGNs over time, confirming that the presence and velocity of these highly ionized nuclear winds change on long timescales.

6. The comparison of the physical properties of our UFO detections with the population study of G24 shows that the addition of high-velocity, moderate-ionization UFOs from our sample significantly flattens the log(ξ)−log(v_out) correlation. The log(N_H)−log(v_out) relation is also moderately flatter, while the log(N_H)−log(ξ) correlation remains unchanged.

7. We investigated the possible correlations between the presence and characteristics of the detected UFOs and AGN properties. Our analysis shows no particular distinctions between AGNs that exhibit UFOs and those that do not, in our sample, on the basis of the mass of SMBHs, L_Bol, λ_Edd, and H, in contrast with what was found for X-ray nuclear winds in stellar BHs. This could suggest that all AGNs may host UFOs detectable only in specific epochs. Their identification could also depend on the circumnuclear environment or the inclination with respect to the line of sight, which we did not investigate in this work.

In summary, this study represents the first exploratory, systematic search for very high-velocity (v_out > 0.3c) UFOs in the 10 − 12 keV band of local AGNs, using the deepest available CCD-resolution spectra from the XMM-Newton archive. Our results demonstrate that such extreme velocity components are present in a non-negligible fraction of local bright AGNs, motivating dedicated follow-up at higher spectral resolution. The XRISM Resolve micro-calorimeter, with its well-calibrated response up to ∼11.5 keV and spectral resolution superior to EPIC-pn by more than an order of magnitude, is ideally suited to confirm and characterize these features. The XRISM archive is rapidly growing with long observations of several bright Seyferts in our sample, and a systematic search for > 10 keV UFOs using Resolve data is an ongoing project that will be presented in a forthcoming paper.

A complementary approach is to target high-redshift QSOs, where the cosmological redshift brings the > 10 keV absorption features into the standard observing band of current X-ray telescopes. This strategy is being pursued by the WISSHFUL XMM-Newton Heritage Programme² (Lanzuisi et al. 2026, Borrelli et al. in preparation), which is obtaining very deep observations of luminous QSOs at z = 2 − 4, at both BH growth and star-formation peaks, providing an independent route to systematically characterize the fastest components of AGN winds at Cosmic Noon.

Moreover, we plan to extend this work by analyzing multiple high-quality observations of bright AGNs available in the XMM-Newton archive, systematically extending the UFO search up to 12 keV to identify the most powerful UFOs and study their variability and duty cycle over time.

Acknowledgments

LB and GL acknowledge support from the Large Program “DELUX” of the “Ricerca Fondamentale 2024” INAF program. EP acknowledges funding from the European Union – Next Generation EU, PRIN/MUR 2022 (2022K9N5B4). JS acknowledges GACR support from the project 26-226145.

References

Appendix A: Sample properties

Appendix B: Continuum modeling

In this section, we present the best-fit models used to reproduce AGN emission in the spectra of sources with UFOs. The BM (described in Sect 4.1) is made of a combination of emission components due to direct continuum and reflection. Some of the sources needed additional components in order to achieve a good description of the spectra:

• in Mkn 766 (B.1, left), we added a Gaussian emission line to model the broad component from Kα, taking into account relativistic effects;

• MRK 110 (B.1, right), we added a Gaussian emission line to model the broad component from Kα, taking into account relativistic effects, a second power-law to reproduce the soft-excess;

• NGC 2617 (B.2, left), we added a second power-law to reproduce the soft-excess;

• NGC 5548 (B.2, right), the BM was sufficient to model the spectrum;

• NGC 7469 (B.3), we added a Gaussian emission line to model the broad component from Kα, taking into account relativistic effects, a second power-law to reproduce the soft-excess;

• PG 2304 (B.3), we added a Gaussian emission line to model the broad component from Kα, taking into account relativistic effects.

Fig. B.1. Left: Best-fit Baseline Model applied to Mkn 766 data (top) and residuals (bottom). Right: Best-fit Baseline Model applied to Mrk 110 data (top) and residuals (bottom). The UFO’s absorption features are highlighted with an arrow.

Fig. B.2. Left: Best-fit Baseline Model applied to NGC 2617 data (top) and residuals (bottom). Right: Best-fit Baseline Model applied to NGC 5548 data (top) and residuals (bottom). The UFO’s absorption features are highlighted with an arrow.

Fig. B.3. Left: Best-fit Baseline Model applied to NGC 7469 data (top) and residuals (bottom). Right: Best-fit Baseline Model applied to PG 2304 data (top) and residuals (bottom). The UFO’s absorption features are highlighted with an arrow.

Appendix C: Comparison to BXA

In this section, we compare the determination of the best-fit parameters described in Sect. 4.2 with the Bayesian X-ray Analysis (BXA; Buchner et al. (2014)). BXA employs the nested sampling algorithm UltraNest for Bayesian parameter estimation. We have used Gaussian prior distributions for line energies, width, and depth, with distributions equal to the 99.7% confidence error on the formal forward-folding best fit. As shown in Fig. C.1, the measurements with the two methods are consistent within the 90% level statistical uncertainties.

Fig. C.1. Comparison between the determination of the absorption line spectral parameters using a standard forward-folding approach and BXA. The red line is the bisector.

Appendix D: NuSTAR validation of UFO detections

As discussed in Sect. 5, the effective area and calibration of the XMM-Newton EPIC-pn camera become increasingly uncertain above 10 keV. To validate the robustness of our highest-energy UFO detections against potential instrumental artifacts, we cross-checked our results using simultaneous NuSTAR observations, which offer superior sensitivity and well-constrained calibration in the hard X-ray band, at the expense of a factor 3-4 worse energy resolution.

We verified the availability of strictly simultaneous NuSTAR data for all sources in our sample exhibiting UFOs in the 9-12 keV band (Mkn 766, NGC 5548, NGC 7469, and PG 2304+042). Among these, only NGC 7469 possesses contemporary XMM-Newton and NuSTAR observations, obtained during a coordinated campaign in 2015. While a comprehensive analysis of all available coordinated pointings from that campaign and the derivation of detailed variability constraints are beyond the scope of this paper, we use them here as a crucial diagnostic test for our > 10 keV detection methodology.

For this validation, we selected two specific epochs from the 2015 campaign: Epoch 1, the NuSTAR observation (OBS-ID 60101001014) strictly simultaneous with the primary XMM-Newton observation analyzed in the main text (OBSID 0760350801). Epoch 2, an additional set of simultaneous pointings taken between December 22-24 (XMM-Newton OBSID 0760350601 and NuSTAR OBSID 60101001008). In this second epoch, the fast UFO feature is similarly prominent, yielding an improvement in the Cash statistic of ΔC = 11.8 in the XMM-Newton data alone.

To assess the consistency of the absorption feature across both observatories, we restricted our spectral modeling to the 7 − 12 keV band. We applied the BM to both datasets and extracted the corresponding residuals. Furthermore, we performed a systematic line scan across this energy range by stepping a narrow (σ = 10 eV) Gaussian absorption line through the data. This scan was executed three times for each epoch: on the XMM-Newton data alone, on the NuSTAR data alone, and on the combined XMM-Newton and NuSTAR spectra.

The resulting best-fit spectra, residuals, and ΔC contour scans for both epochs are presented in Fig. D.1 and Fig. D.2. The presence of these highly blueshifted absorption features is confirmed in the independent NuSTAR data, with the lines detected at energies of ∼10.8 keV in Epoch 1 and ∼10.3 keV in Epoch 2. For each epoch, the line energies and depths are perfectly consistent between the XMM-Newton and NuSTAR observations. This cross-instrument agreement clearly demonstrates that the > 10 keV UFO detections identified in the EPIC-pn data are robust astrophysical features and are not driven by instrumental noise or by effective-area calibration drops. We acknowledge that the NuSTAR cross-validation presented here is limited to a single source, as it is the only target in our sample with strictly simultaneous hard X-ray coverage in the > 10 keV band where our new UFO detections lie. While this successfully demonstrates that our methodology does not produce spurious detections from EPIC-pn calibration artifacts, a comprehensive multi-instrument confirmation of all four v_out > 0.3c UFOs will require dedicated follow-up, which we plan to pursue with both archival NuSTAR data (where available, albeit nonsimultaneous) and upcoming XRISM/Resolve observations.

Fig. D.1. XMM-Newton pn and NuSTAR FPMA-B spectra (left) and line scan (right) of NGC 7469 in the 8-12 keV band for Epoch 1, alongside the best-fit Baseline Model and corresponding residuals. The high-velocity UFO absorption features are clearly visible in both instruments at ∼10.8 keV.

Fig. D.2. XMM-Newton pn and NuSTAR FPMA-B spectra (left) and line scan (right) of NGC 7469 in the 8-12 keV band for Epoch 2, alongside the best-fit Baseline Model and corresponding residuals. The high-velocity UFO absorption features are clearly visible in both instruments at ∼10.3 keV.

Appendix E: Discarded absorption lines

As discussed in Sect. 4.2.1, we considered an absorption line significantly detected if its addition in the model led to a statistical improvement from the continuum model of ΔC_stat > 7.815, corresponding to a significance of 95% for a χ² distribution with 3 degrees of freedom (Lampton et al. 1976). Table E.1 reports possible UFO detections that were discarded as the absorption lines were detected with significance below this threshold (i.e., ΔC_stat < 7.815).

We decided to include in this section an absorption feature detected with high confidence (> 98%) in OBS:0830540101 of ARK 564. The properties of the UFO are shown in Table E.2. While the feature is detected with high statistics, the proper modeling goes beyond the simplistic models used in this analysis due to the known Fe K relativistic emission line observed in this source (Kara et al. 2017). Therefore, we delay the modeling of the detected absorption line in a forthcoming paper.

Appendix F: Comparison with literature results

Table F.1 reports the comparison between our results with the outcomes of previous analysis on the same observations in our sample.

All Tables

All Figures

	Fig. 1. Bolometric luminosity of the 33 AGNs considered in the sample (see Table A.1) plotted against their redshift.
In the text

	Fig. 2. Distribution of the net counts in the 2–12 keV band for the pn camera (see Table A.2) after removing periods dominated by soft proton flares.
In the text

Fig. 3. Result of the MC simulations for the spectrum of NGC 2617. The blue histogram shows the distribution of ΔCstat obtained by fitting the BM and BFM models to the simulated line-less spectra. The red line shows the observed ΔCstat for this source. α represents the fraction of MC simulations showing a ΔCstat greater than the observed one. In this case, the absorption feature is detected with a confidence 1 − α > 99%.

In the text

	Fig. 4. Left: Comparison between the UFO velocity distributions of our sample (magenta) with the T10 sample (purple). Right: Same for the EQW distribution.
In the text

Fig. 5. Comparison between the UFOs’ equivalent width and outflow velocity distributions of our sample (in red) with the sample from T10 (in blue). The dashed plum line marks the threshold of 0.03c, while the dashed turquoise line marks 0.3c. The green, orange, and purple points are the mean of the velocity and EQW in the three velocity bins. The red line and shaded area show the linear best fit and its 95% confidence interval computed with LINMIX, taking into account errors on both vout and EQW.

In the text

	Fig. 6. Comparison between UFO velocities from the literature (see label) and the results of the present work (red circles) as a function of λEdd (from Luminari et al. 2021). The theoretical limit, according to the terminal (maximum) velocity vt (vmax), for R0 ≥ 50 rG is shown in the green (light-blue) shaded area.
In the text

Fig. 7. Correlations between observed UFO parameters: ionization parameter log(ξ) vs. outflow velocity log(vout) (panel a), column density log(NH) vs. log(vout) (panel b), and log(NH) vs. log(ξ) (panel c). Blue diamonds, green triangles, and black squares represent the T10, M23, and C21 samples from G24, respectively; red circles are the detections from this work. The black solid lines show the original best-fit relations, while the red solid lines and shaded areas show the updated fits and their 68% confidence intervals after including our new detections.

In the text

	Fig. 8. Distribution of AGN with (orange) and without UFOs (green) with respect to their bolometric luminosity, BH mass, Eddington ratio, and Hhardnessparameter (see Table A.1).
In the text

	Fig. B.1. Left: Best-fit Baseline Model applied to Mkn 766 data (top) and residuals (bottom). Right: Best-fit Baseline Model applied to Mrk 110 data (top) and residuals (bottom). The UFO’s absorption features are highlighted with an arrow.
In the text

	Fig. B.2. Left: Best-fit Baseline Model applied to NGC 2617 data (top) and residuals (bottom). Right: Best-fit Baseline Model applied to NGC 5548 data (top) and residuals (bottom). The UFO’s absorption features are highlighted with an arrow.
In the text

	Fig. B.3. Left: Best-fit Baseline Model applied to NGC 7469 data (top) and residuals (bottom). Right: Best-fit Baseline Model applied to PG 2304 data (top) and residuals (bottom). The UFO’s absorption features are highlighted with an arrow.
In the text

	Fig. C.1. Comparison between the determination of the absorption line spectral parameters using a standard forward-folding approach and BXA. The red line is the bisector.
In the text

	Fig. D.1. XMM-Newton pn and NuSTAR FPMA-B spectra (left) and line scan (right) of NGC 7469 in the 8-12 keV band for Epoch 1, alongside the best-fit Baseline Model and corresponding residuals. The high-velocity UFO absorption features are clearly visible in both instruments at ∼10.8 keV.
In the text

	Fig. D.2. XMM-Newton pn and NuSTAR FPMA-B spectra (left) and line scan (right) of NGC 7469 in the 8-12 keV band for Epoch 2, alongside the best-fit Baseline Model and corresponding residuals. The high-velocity UFO absorption features are clearly visible in both instruments at ∼10.3 keV.
In the text