Open Access
Issue
A&A
Volume 670, February 2023
Article Number A182
Number of page(s) 28
Section Extragalactic astronomy
DOI https://doi.org/10.1051/0004-6361/202245036
Published online 28 February 2023

© The Authors 2023

Licence Creative CommonsOpen Access article, published by EDP Sciences, under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This article is published in open access under the Subscribe to Open model. Subscribe to A&A to support open access publication.

1. Introduction

It is widely accepted that supermassive black holes (SMBHs; 106–1010M; e.g. Salpeter 1964; Magorrian et al. 1998) are hosted at the centre of virtually every known massive galaxy. The observed tight correlations between the host galaxy and the SMBH properties (see Kormendy & Ho 2013, for a review) strongly suggest that their formation and evolution are profoundly coupled with each other. Some physical mechanisms must have therefore linked the regions where the SMBH gravitational field dominates to the larger scales, where its direct influence is negligible. At this stage, the key underlying ingredients at play in the co-evolution paradigms of active galactic nuclei (AGNs) and galaxies still need to be understood. It has been proposed that highly ionised gas outflows could play a pivotal role in this process (e.g. King 2003, 2005; Gaspari & Sądowski 2017). The presence of such powerful winds is expected to regulate accretion of material onto (and ejection from) compact objects.

Through their mechanical power, ultra-fast outflows (UFOs) are accelerated at velocities greater than 10 000 km s−1 and up to a few dozen times the speed of light. For these reasons, UFOs are also able to inject momentum and energy over wide spatial scales via interaction with the interstellar medium (ISM) in the host galaxy. This process is expected to promote an efficient feedback mechanism (e.g. Murray et al. 2005; Di Matteo et al. 2005; Ostriker et al. 2010; Torrey et al. 2020), which is needed to reproduce the observed properties in galaxies, for example the scaling relations (King et al. 2011), and to regulate their overall mass-size ecosystem (e.g. Fabian 2012; King & Pounds 2015; Heckman & Best 2014).

Ultra-fast outflows are routinely detected in the X-ray spectra of 30–40% of local (z ≲ 0.1) AGNs (Tombesi et al. 2010; Gofford et al. 2013; Igo et al. 2020, hereafter T10; G13; Igo20), and in a handful of sources at intermediate to high redshifts (up to z ∼ 3; e.g. Chartas et al. 2002; Lanzuisi et al. 2012; Chartas et al. 2021, hereafter C21). They manifest themselves as absorption troughs associated with Fe XXV Heα and Fe XXVI Lyα transitions (Erest = 6.7–6.97 keV) blueshifted at energies Erest > 7 keV (all the line energies will be given in the source rest frame throughout this work). The degree of blueshift translates into the range of extreme outflow velocities observed from vout ∼ −0.03c up to −0.5c, for gas column densities and ionisations of NH ∼ 1023−24 cm−2 and log(ξ/erg cm s−1) ≳ 3, respectively (e.g. Reeves et al. 2003, 2018a; Pounds & Reeves 2009; Tombesi et al. 2011; Matzeu et al. 2017; Parker et al. 2018; Braito et al. 2018). The frequent detection of these features, supported by a detailed modelling of the high energy spectra of the most powerful local quasar (QSO) to host X-ray winds, PDS 456, indicates that UFOs arise in wide angle outflows, in turn implying that a significant amount of kinetic power is involved (Nardini et al. 2015; Luminari et al. 2018).

Evidence for low-ionisation UFO components has also been reported in the soft X-ray spectra in the Erest = 0.3–2 keV band (e.g. Braito et al. 2014; Longinotti et al. 2015; Reeves et al. 2016, 2018b, 2020; Serafinelli et al. 2019; Krongold et al. 2021), usually observed as blueshifted oxygen and neon ions. Similar high-velocity outflows, arising directly from the accretion disc region, have also been found in the UV spectra via prominent blueshifted, ionised absorption, and emission features in broad absorption line QSOs, typically between λrest = 50 and 2000 Å (e.g. Gaskell 1982; Wilkes & Elvis 1987; Richards et al. 2011). They are associated with lower-ionisation metal ions (C IV, Al II, Fe II, etc.; e.g. Crenshaw et al. 2003; Green et al. 2012; Hamann et al. 2018; Kriss et al. 2018, 2019; Mehdipour et al. 2022; Vietri et al. 2022). It has been shown that at least some of the UV-absorbing outflows in sub-Eddington systems can be driven by radiation pressure on spectral lines (e.g. Murray et al. 1995; Proga et al. 2000). We refer readers to the recent review by Giustini & Proga (2019) for more details.

Outflowing material at considerably lower velocities (typically within a vout of −5000 km s−1) and less ionised than UFOs at log(ξ/erg cm s−1) ∼ 1−3 (e.g. Sako et al. 2000; Parker et al. 2019), known as a warm absorber (WA), is also detected through absorption features and edges from He- and/or H-like ions of C, O, N, Ne, Mg, Al, Si, and S in X-rays (Halpern 1984; Mathur et al. 1997, 1998; Blustin et al. 2005; Reeves et al. 2013; Kaastra et al. 2014; Laha et al. 2014, 2016). Warm absorbers are detected in a substantial fraction of AGNs: ∼65% (Reynolds 1997; Piconcelli et al. 2005; McKernan et al. 2007). It was suggested by Tombesi et al. (2013) that, despite their physical distinction, UFOs and WAs might be somehow connected as part of the same wind, but originating from different locations (see Laha et al. 2021 for a comprehensive review of ionised outflows).

Finally, outflowing gas is also routinely observed at host-galaxy scales, in the ionised, neutral and molecular phases. These outflowing components observed at kiloparsec scales or beyond are now traced with modern sensitive optical/far-IR/millimetre/radio facilities (e.g. Morganti et al. 2005; Feruglio et al. 2010, 2017; Harrison et al. 2014; Brusa et al. 2015, 2018; Maiolino et al. 2012; Cresci et al. 2015; Bischetti et al. 2019a) and show lower velocities with respect to the accretion disc winds (vout ∼ −500 to −2000 km s−1, depending on the phase) and considerably higher mass outflow rates of up to 100–1000 M yr−1 (see Cicone et al. 2018).

Some models predict that the fast outflowing gas is accelerated by the radiation pressure caused by highly accreting black holes approaching the Eddington limit (e.g. Zubovas & King 2012). Subsequently, the energy deposited via shocks by the UFO into the galaxy ISM generates the galaxy-wide outflows observed in lower-ionisation gas (see Fabian 2012; King & Pounds 2015). Alternatively, massive sub-relativistic outflows are also expected in systems accreting at a lower Eddington ratio, due to magnetic (e.g. Fukumura et al. 2010, 2017; Kraemer et al. 2018) and/or thermal driving (e.g. Woods et al. 1996; Mizumoto et al. 2019; Waters et al. 2021). The global AGN feeding-feedback self-regulated framework has been supported by three-dimensional hydrodynamical simulations that unify the micro and macro properties of the AGN environment (e.g. Gaspari et al. 2013, 2020; Sądowski & Gaspari 2017; Yang et al. 2019; Wittor & Gaspari 2020); these simulations have, in turn, been corroborated by several multi-wavelength observations (e.g. Maccagni et al. 2021; Eckert et al. 2021; McKinley et al. 2022; Temi et al. 2022).

Multi-phase tracers would therefore allow us to probe galactic outflows in their full extent, that is, from the nuclear (< 1 pc) to the largest scales (> 10 kpc), and at the same time provide us with a comprehensive view of their driving mechanism. Fiore et al. (2017) reported a correlation between the velocity of the wind (for both UFOs and large-scale components) and the bolometric luminosity (), in agreement with that predicted by Costa et al. (2014) for energy-conserving outflows. However, the statistics of this work for the UFO sample are still limited (∼20 AGNs with UFOs), with only 50% at Lbol ≳ 1045 erg s−1.

A comparison between the momentum rates (i.e. out = outvout) observed over a range of spatial scales can be used to disentangle the wind propagation mechanisms: energy-conserving (large-scale) versus momentum-conserving (small-scale: UFO). The first reported cases of molecular outflows in systems that host a UFO were IRAS F11119+3257 (Tombesi et al. 2015), Mrk 231 (Feruglio et al. 2015), IRAS 17020+4544 (Longinotti et al. 2015, 2018), and APM 08279+5255 (Feruglio et al. 2017). These examples support the energy-driven wind scenario, deemed as the smoking gun for large-scale feedback. Indeed, conserving energy is crucial for achieving an effective macro-scale feedback to quench cooling flows and star formation (e.g. Gaspari et al. 2019). However, via the analysis of more sources, it emerged that not all outflows supported this scenario. Further molecular outflows observed in UFOs that host sources, such as PDS 456 (Bischetti et al. 2019b), MCG–03–58–007 (Sirressi et al. 2019), I Zwicky 1 (Cicone et al. 2014; Reeves & Braito 2019), and Mrk 509 (Zanchettin et al. 2021), revealed momentum rates at least two orders of magnitude below the expected value for an energy-conserving wind (see also Marasco et al. 2020; Tozzi et al. 2021). These results suggest a more complex physical mechanism and range of efficiencies in transferring the nuclear wind out to the large-scale galaxy structure, or significant AGN variability over the lifetime of the flow (Nardini & Zubovas 2018; Zubovas & Nardini 2020), as supported by chaotic cold accretion simulations (Gaspari et al. 2017). Related to this, an important parameter that is needed to constrain wind models is the UFO duty cycle, that is, how persistent accretion disc winds are. Indeed, the derivation of the energy injection rate by UFOs into the galaxy ISM must take this factor into account, with implications for the timescale and efficiency of propagation through the host galaxy (Zubovas & King 2016). The UFO duty cycle can be inferred from the fraction of AGNs in which they are observed, but it is highly degenerate with the opening angle. Hence, as of today, it is virtually unconstrained for sources above Lbol ∼ 1045 erg s−1. Observing large samples of sources at luminosities above the break L ≳ L of the AGN luminosity function (e.g. Aird et al. 2015), with a range of Eddington ratios and enough statistics to constrain the wind duty cycle, has become crucial for overcoming the limitations described above (e.g. Bertola et al. 2020).

This work is the first in a series of Supermassive Black Hole Winds in the X-rays (SUBWAYS) publications. The SUBWAYS programme has the dedicated goals of investigating the various manifestations of UFOs emanating from the environments of SMBHs in AGNs. This includes gaining significant advances in our understanding of the detection rate and physical properties of UFOs, their connection with WA features, and their role in providing a macro-scale feedback, as well as mapping the physical properties of the outflows across different galaxy scales and gas phases at different wavelengths and ionisation states. In this paper we present our first results of the SUBWAYS campaign, specifically designed to provide a solid detection of blueshifted absorption features in the Fe K band in the context of robust statistical grounds for high signal-to-noise ratio (S/N) sources at L ≳ 1045 erg s−1.

The paper is organised as follows: Sect. 2 presents the SUBWAYS sample and the target selection. In Sect. 3 we describe the reduction of the X-ray Multi-Mirror mission (XMM-Newton) data. In Sect. 4 we present all the details of the spectral analysis of the European Photon Imaging Camera (EPIC) data, including the continuum modelling, the procedure we adopted to search for Fe K emission and absorption features, the modelling of the Fe K band, and the Monte Carlo (ℳ𝒞) simulations we used to assign a robust significance level to the detections. In Sect. 5 we present our main results, namely the line detection rate as inferred from our spectral modelling, and in Sect. 6 we discuss our findings in light of recent results at both lower and higher redshifts; Sect. 7 summarises our results. Cosmological values of H0 = 70 km s−1 Mpc−1, ΩΛ0 = 0.73, and ΩM = 0.27 are assumed throughout this paper, and errors are quoted at the 90% confidence level or a difference in C-statistic (i.e. Δ𝒞 = 2.71; Cash 1979) for one parameter of interest, unless otherwise stated. The cosmic abundances are set to solar throughout the paper.

2. The SUBWAYS campaign

So far, the characterisation of the fastest components of accretion disc winds has mainly been carried out through studies based on inhomogeneous archival data, restricted to two distinct cosmic epochs and luminosity regimes, merely for practical reasons: (i) at z < 0.1, objects are close enough that it is relatively easy to collect > 5 − 10 × 103 counts in the 4–10 keV band in large samples (∼50 objects), but mainly limited to Seyfert luminosities (Lbol ≲ 1045 erg s−1; e.g. T10; G13; Igo20); (ii) at z ≳ 1.5, on small and sparse samples (< 10 objects) at Lbol ≳ 1046 erg s−1, which are mostly composed of gravitationally lensed objects (e.g. Chartas et al. 2009; Vignali et al. 2015; Dadina et al. 2016, C21). The distribution of these two samples in the Lz plane is shown in the upper panel of Fig. 1 (blue and magenta points, respectively).

thumbnail Fig. 1.

Luminosity (upper panel) and rest-frame 4–10 keV counts (lower panel) plotted against redshift for the objects in the SUBWAYS sample and the comparison samples (T10; C21), as labelled. In the lower panel we also mark the sources of the 3XMM sample, used to select the SUBWAYS targets, with small empty circles.

In order to gain significant advances in our understanding of the physical properties of UFOs in the QSO-like regime, a systematic approach is needed. The SUBWAYS sample consists of a total of 22 radio-quiet X-ray AGNs, mostly Type 1 and QSOs, where 17 sources have been observed with XMM-Newton (Jansen et al. 2001) between May 2019 and June 2020 (see Table 1) as part of a large programme (1.45 Ms, PI: M. Brusa) awarded in 2018 (cycle AO18 LP). In addition, the sample includes the data of five sources that meet the Lz–counts selection criteria (see below) already available in the archive1. A companion SUBWAYS paper, Paper II (Mehdipour et al. 2023), is primarily focused on the UV outflow spectroscopic analysis of Cosmic Origins Spectrograph (Green et al. 2012) data as part of a large complementary SUBWAYS observational campaign carried out with the Hubble Space Telescope.

Table 1.

Target properties of the large SUBWAYS campaign.

The SUBWAYS selection criteria are based on the following requirements. Firstly, The presence of the source in the 3XMM-DR7 catalogue2, matched to the SDSS-DR14 catalogue3, or to the Palomar-Green Bright QSO catalogue (PGQSO; Schmidt & Green 1983). Secondly, an intermediate redshift in the range, z = 0.1–0.4. This condition ensures that both WAs and UFOs can be studied at the same time, and provides the possibility to characterise the continuum up to 10 keV. Indeed, in order to recognise faint absorption features, it is key to achieve a good handling of the continuum in spectra with high counting statistics up to 10 keV. Thirdly, a count rate larger than ∼0.12 cts/s, in order to ensure counts of the order of ∼104 in the 4–10 keV band in the EPIC-pn spectra, obtained within a single XMM-Newton orbit. A by-product of this requirement also implies that our targets are QSOs (Lbol ≳ 1045 erg s−1; star points in Fig. 1), complementing the data already available in the archives for this kind of studies. Lastly, we discarded narrow-line Seyfert 1 galaxies (NLSy1s) due to the highly variable EPIC count rate, and QSOs in clusters/radio loud systems, in order to avoid contamination by processes other than AGN accretion and UFOs.

The lower panel of Fig. 1 shows the currently available rest-frame 4–10 keV counts for the SUBWAYS sample (large stars), compared to those in 3XMM-SDSS, 3XMM-PGQSO, and local and high-z QSO UFO samples (see the labels and caption for details). In this paper we focus specifically on the detection and characterisation of blueshifted absorption profiles in the Fe K band in the 17 newly observed sources plus the additional 5 from previous observations (up to AO18 cycle), for a total of 22 targets. The properties of the targets are listed in Table 1.

3. Data reduction

In this work we focus on the EPIC-pn (Strüder et al. 2001), EPIC-MOS 1, and MOS 2 (Turner et al. 2001) data. They were processed and cleaned by adopting the Science Analysis System SAS v18 (Gabriel et al. 2004) and the up-to-date calibration files. We initially checked for Cu instrumental emission in the EPIC-pn CCDs, between 7–8.5 keV and 7.8–8.2 keV, for the source extraction and subsequent high-background screening. We followed the Piconcelli et al. (2004) optimised procedure aimed at maximising the S/N in the 4–10 keV band (in the EPIC-pn), rather than using the conservative criterion based on the fiducial rejection of time-intervals of high-background count rates (i.e. between 10–12 keV). The S/N optimisation procedure is necessary to identify any absorption feature that would otherwise be diluted (e.g. Nardini et al. 2019), but insufficient if this does not also correspond to the optimal compromise between S/N and number of counts (see Fig. 2).

thumbnail Fig. 2.

Black hole mass and Eddington ratio distributions of our targets (see Table 1) and the comparison samples. The colour scheme for the samples is the same as in Fig. 1.

Given the relatively small EPIC-MOS collecting area at Eobs ≥ 4 keV, a 4–10 keV band optimisation would remove too many counts; we then optimised the filtering on the entire 0.3–10 keV band. Apart from the different reference bands, the applied method is the same for the pn and MOS instruments. We selected a background region free of instrumental features. These regions have dimensions of 40 or 50 arcsec depending on the possibility, for each observation, to find source-free regions on the detectors. In order to define the source regions different extraction radii were tested and for each radius we calculated the maximum level of background that can be tolerated in order to find the optimal S/N. Following Piconcelli et al. (2004), we define this level of background as the max background (see Appendix A for more details). The EPIC source spectra were individually inspected for the possible presence of photon pile-up by using the SAS task EPATPLOT. The ratios of singles to double pixel events were found to be within 1% of the expected nominal values, and thus no significant pile-up is present. The response files were subsequently generated with the SAS tasks RMFGEN and ARFGEN with the calibration EPIC files version v3.12. In Table B.1 we show a summary of the individual observations of the 22 SUBWAYS targets that were selected adopting a threshold of ≥1500 EPIC-pn net counts in the 4–10 keV band.

4. Spectral analysis

The pioneering UFO studies were conducted on large archival samples of AGNs. More specifically, T10 carried out a systematic hard-band (i.e. 3.5–10.5 keV) analysis on a sample of 42 sources (for a total of 101 observations), drawn from the archival XMM-Newton EPIC data, to carry out a blind search of Fe XXV Heα and Fe XXVI Lyα absorption lines. By analysing the data of 51 AGNs, obtained with the Suzaku observatory, G13 constructed broadband spectral models over the entire band pass (i.e. 0.6–10 keV).

For a robust analysis, we chose, as per G13, the entire EPIC band pass, where additional spectral complexities like WAs and/or strong soft excesses can also be taken into account. In this way we ensure that all our models accurately describe the overall continuum. We focus on the XMM-Newton EPIC-pn, MOS 1, and MOS 2 data in the 0.3–10 keV range. We applied a blind-search procedure in each of the 41 observations by adopting four spectral binning methods (for a total of 164 blind-searches; see Appendix C for details). These binnings are grpmin1, SN5, and OS3grp20, using the SAS routine SPECGROUP, and the optimal binning of Kaastra & Bleeker (2016, 𝒦ℬ hereafter), using the HEASOFT routine FTGROUPPHA.

We find that the spectral resolution delivered by the SN5 and OS3grp20 binning methods is too degraded compared to the grpmin1 and 𝒦ℬ ones, and therefore not suitable for the detection of faint, narrow absorption and emission features. The grpmin1 and 𝒦ℬ criteria produce nearly identical results, in terms of detection rate and statistical significance of the features. While grpmin1 would certainly be a more conservative option, we finally chose the 𝒦ℬ binning, which is specifically developed to optimise the S/N in narrow, unresolved spectral features whilst maintaining the necessary spectral resolution. Such a choice provided the right compromise between these binning methods. In other words, Kaastra & Bleeker (2016) worked out a binning scheme based on the resolution of the detector and the available number of photons at the energy of interest. Such a method allowed us to maximise the information provided by the Fe K absorption lines, and in this framework we adopted a maximum likelihood statistic as 𝒞−stat.

4.1. Continuum modelling

All the spectra were initially fitted with a power law and their corresponding Galactic absorption, modelled with Tbabs (Wilms et al. 2000), with column densities obtained from the HI4PI Collaboration (2016) survey. In order to accurately parameterise the properties of the underlying continuum, additional model components were required in the process such as warm or neutral absorption, or a soft excess, as outlined in Sects. 4.1.1 and 4.1.2.

We note that statistically identical results could have been achieved by modelling the continuum with distant and/or ionised Compton reflection models such as xillver (García et al. 2013) or relxill (Dauser et al. 2014; García et al. 2014). Since the XMM-Newton band pass is 0.3–10 keV, the contribution from the Compton reflection continuum is not well constrained; therefore, for simplicity, here we adopt a power-law plus blackbody (when required) parameterisation of the continuum. A similar approach was also adopted in the Catalogue of AGN in the XMM-Newton Archive (CAIXA) sample by Bianchi et al. (2009). Nonetheless, a thorough investigation using Compton and relativistic reflection models is addressed in a forthcoming companion paper, where we take advantage of the Nuclear Spectroscopic Telescope Array mission (NuSTAR) follow-up obtained in 2020 (PI: Bianchi).

Our baseline phenomenological model can be described as

(1)

where Tbabs represents the absorption due to our Galaxy, powerlaw accounts for the primary emission component, and zbbody1, 2 are two layers of blackbody emission to account for the soft X-ray excess4. This parameterisation of the soft excess is only phenomenological and hence the corresponding temperatures are not meaningful. XABS (when required) corresponds to a mildly ionised warm absorption component (see Sect. 4.1.1).

Once the best fit of the 0.3–10 keV continuum spectra of each of the 41 observations was reached, we performed a systematic search for iron K emission and absorption profiles between 4–10 keV using the following two methods: (i) blind-line search via energy–intensity plane contours plots (Sect. 4.2) and (ii) extensive ℳ𝒞 (e.g. Protassov et al. 2002) simulations (Sect. 4.4). The modelling approach of each individual spectral component is described below and the detailed continuum and absorption parameters are tabulated in Table D.1

4.1.1. Intrinsic absorption

Neutral or lowly ionised absorption is typically constituted by a distant (≳pc scales), less ionised and denser circumnuclear material compared to UFOs, generally outflowing at velocities in the range of ∼ − 100 to −1000 km s−1 (e.g. Kaastra et al. 2000; Kaspi et al. 2000a; Blustin et al. 2005). These absorbers are detected in the soft X-ray part of the spectrum at energies ≲2–3 keV and, depending on their properties, they can add a significant curvature to the spectra below 10 keV (e.g. Matzeu et al. 2016; Boller et al. 2021), and hence affect the overall continuum and line parameters in our broadband models. In the literature, the fraction of sources with reported WAs is > 60% (Tombesi et al. 2013; G13).

In 2MASS J105144+3539, an intrinsic neutral absorption column of NH = 6.7 ± 0.4 × 1021 cm−2 and two emission lines in the soft band, likely associated with collisionally ionised gas, are required. 2MASS J165315+2349 is classified as a Seyfert 2 and so requires a different model construction in the soft X-rays, with the presence of an intrinsic neutral absorber and emission arising from a distant scattered component. The continuum model differs from Eq. (1) as

(2)

where apec1, 2 (Smith et al. 2001) are thermal models accounting for two regions of emitting collisionally ionised plasma and powerlawscatt reproduces the distant scattered component. powerlawintr accounts for the primary continuum, which is absorbed by fully covering neutral material (phabs) with a column density of .

In this work we model the presence of fully covering mildly ionised absorption with a specifically generated XABS model from SPEX (Kaastra et al. 1996; Steenbrugge et al. 2003) converted to an XSPEC (Arnaud 1996) table5 (see the appendix of Parker et al. 2019). More specifically the XABS table covers the following ranges in the parameter space: column density in the range 1020NH/cm−2 ≤ 1022 with 10 logarithmic steps, and ionisation in the range −4 ≤ log(ξ/erg cm s−1) ≤ 6 with 20 linear steps, making this table well suited for investigating a wide range of absorbers in AGNs such as WAs. Although the presence of WAs could have a minimal effect on the Fe K region, we find it essential to include them as the most reliable continuum level must be determined.

The XABS table was generated by assuming the default SPEX setting where the spectral energy distribution (SED) of NGC 5548 was used as representative of a standard AGN input spectrum6 (see Steenbrugge et al. 2005, for more details). Another parameter in the model is the two-dimensional root mean square velocity (vrms), which gives a measure of the velocity dispersion of the line profile7. At the energy resolution of the EPIC CCDs (see Appendix C), the individual soft X-ray absorption lines are indeed unresolved, so in the fitting procedure the RMS velocity broadening was frozen to its default value, vrms = 100 km s−1 (unless specified otherwise), and the systemic velocities were set to 0 km s−1.

Here we also consider the possibility of partial covering along the line of sight. In this regime a fraction corresponding to fcov < 1 of the total flux is indeed absorbed, while a portion 1 − fcov leaks through the absorbing layer. This can also have a dramatic effect on the emerging continuum by imprinting a prominent spectral curvature at energies < 10 keV (e.g. Matzeu et al. 2016; Boller et al. 2021). On this basis a partial covering fraction was also added to the list of free parameters in our XABS table (i.e. 0 ≤ fcov ≤ 1). A more comprehensive physical analysis of low- and high-ionisation outflows is presented in a companion paper.

4.1.2. The soft excess

The soft X-ray excess is described as a strong featureless emission component that is often observed in unabsorbed AGNs below ∼ 2 keV. The physical mechanism responsible for this emission is still the subject of active debates; it may be a dual-coronal system (e.g. Done et al. 2012; Petrucci et al. 2013, 2018; Różańska et al. 2015; Middei et al. 2018; Ursini et al. 2020; Ballantyne & Xiang 2020; Porquet et al. 2021) or relativistic blurred reflection (e.g. Ross & Fabian 2005; Nardini et al. 2011; Wilkins & Fabian 2012; Walton et al. 2013; García et al. 2019; Jiang et al. 2019; Xu et al. 2021; Mallick et al. 2022).

Regardless of the physical origin of the soft excess, in this work we take a completely empirical approach by fitting its profile with one/two layers of blackbody emission, when required at the Δ𝒞/Δν ≳ 9.3/2 threshold (i.e. ≳99% confidence level for each blackbody component) in XSPEC (e.g. Porquet et al. 2004a; Piconcelli et al. 2005; Bianchi et al. 2009, G13). Although our phenomenological model largely ignores the detailed physics involved in the system, it allows us to fit and compare uniformly the underlying continua in our sample so that we can concentrate on the Fe K band. There might be some degeneracies between the soft excess and partial covering components during fitting; however, this is not an issue for our absorption line detections. We could have modelled equally well the soft excess with a reflection component and the final result would not change, as discussed above in Sect. 4.1.

4.2. Search for Fe K emission and absorption

In Fig. 3 (top) we report, as an example, the background-subtracted XMM-Newton spectra (EPIC-pn, black; EPIC-MOS 1, red; EPIC-MOS 2, green) and their corresponding background spectra of PG 0052+251. This source is one of the most luminous in the sample and is on the bright side of the luminosity/counts distribution in Type 1 AGNs (see Fig. D.1).

thumbnail Fig. 3.

Broadband EPIC data and best-fit continuum of PG 0052+251 between the rest-frame energy of 0.3–10 keV. Top: background-subtracted XMM-Newton spectra (EPIC-pn in black, EPIC-MOS 1 in red, EPIC-MOS 2 in green) and their corresponding background levels (shaded areas). The broadband best-fit continuum model (solid red) is shown alongside the individual model components: absorbed power-law (cyan), low-temperature (blue), and high-temperature blackbody (magenta). Bottom: corresponding residuals of the data compared to the best-fitting model.

The 0.3–10 keV (rest-frame) best-fitting broadband continuum model (excluding iron K emission and/or absorption lines) is indicated in solid-red. The residuals are shown in the bottom panel. Our baseline continuum models include the absorbed power-law component and the two blackbody components, with high and low temperatures (kT). The continua are well reproduced with our baseline model (see the model statistics reported in Table D.1). Two exceptions were found in the sample: when the soft X-ray emission was parameterised with one or two regions of collisionally ionised plasma modelled with apec in 2MASS J105144+3539 and 2MASS J165315+2349, respectively. The presence of strong residuals from neutral Fe Kα core emission at Erest = 6.4 keV from distant material is ubiquitous in the SUBWAYS sample (e.g. see Fig. 3).

In some observations (see Fig. D.1) we observed strong absorption residuals (also in the Fe K band) likely associated with Fe XXV–Fe XXVI transitions. Therefore, we ran a blind search, simultaneously in both EPIC-pn and EPIC-MOS spectra for every observation in the sample, in order to have a first assessment of energy, strength, shape and significance of any absorption or emission line relative to the underlying continuum model. We performed an inspection of the deviation in |Δ𝒞| from the best-fitting continuum model by generating two-dimensional energy–intensity contours plots. This method was adopted by T10 and G13, and extra details are described in Miniutti & Fabian (2006).

The search was performed with our baseline continuum model (Eqs. (1) or (2)) plus a narrow Gaussian line (with the velocity width fixed at σwidth = 10 eV). We also let the power-law photon index and normalisation free to vary during the search. In adopting our broadband ‘multi-component’ continuum model above, we ensure a better reproduction accuracy of the continuum level compared to a simpler two-component power law plus Gaussian line model restricted on the Fe K band. For this routine we freeze all the soft X-rays parameters from the broadband continuum to their best-fit values, re-fit, and run the scan along with the 5–10 keV, rest-frame energy band. The blind search method adopted here is carried out based on the following steps: (i) We have a baseline continuum model between 0.3–10 keV (described above) plus the unresolved Gaussian line required for the scan. The Fe Kα emission line at 6.4 keV was not included in the baseline model. For each run, the energy of the Gaussian is scanning the three EPIC spectra simultaneously between 5–10 keV in intervals of ΔE = 50 eV. The normalisation of the Gaussian component probes the intensity of the spectral line and is free to vary between negative and positive values in 250 steps. (ii) Each individual step in the energy–intensity plane with respect to the baseline model was recorded into a file including the corresponding Δ𝒞. (iii) The resulting confidence contours are plotted according to a mapped Δ𝒞 deviation of −2.3, −4.61, −9.21, −13.82 and −18.52 for 2 parameters of interest corresponding to the nominal 68%, 90%, 99%, 99.9% and 99.99% confidence levels. (iv) We inspect the contour plots to check whether there is any evidence of emission and/or absorption in the spectrum (see the text below).

If an emission or absorption line is detected, it is parameterised by using a Gaussian profile. All the key Gaussian absorption and emission parameters accounting for the detected lines, using the 𝒦ℬ binning, are tabulated in Table 2. The Δ𝒞 mapping provided by the energy–intensity contours is a powerful tool to detect emission or absorption profiles by visually assessing the location and strength of the line relative to the underlying continuum model. The spectral complexity within the 5–10 keV band can be enhanced by a number of atomic features such as ionised emission lines from Fe XXV and Fe XXVI, at 6.7 keV and 6.97 keV, respectively. As we have learned from previous work, in ultra-fast wind systems the ionised emission due to scattered photons from the outflowing material can be as important as the absorption (e.g. Sim et al. 2008; Nardini et al. 2015; Luminari et al. 2018; Matzeu et al. 2022). In several SUBWAYS spectra, the shape of the emission and/or absorption profiles is indeed complex/broad, which suggests a superposition/blending of multiple ionised lines. Steps (i)–(iv) were carried out in each fitted EPIC spectrum of the sample.

Table 2.

Gaussian emission and absorption parameters in the Fe K band corresponding to a total of 41 observations.

Examples of this method are shown in Fig. 4. In the top panel we show the corresponding residual of EPIC-pn spectrum (MOS 1 and MOS 2 are not included for clarity) without the emission and absorption components. The vertical dashed lines denote the position of the laboratory energy transition of Fe Kα (lime green), Fe XXV (magenta) and Fe XXVI (magenta).

thumbnail Fig. 4.

Plot of the residuals and the corresponding closed confidence contours plot from the blind scan of the type 1 AGN LBQS 1338−0038. Top: data/model ratio plot for the EPIC-pn spectrum (MOS 1 and MOS 2 are not included for clarity), showing the residuals in the 5–10 keV band. A broad emission and absorption profiles are seen at ∼ 6.6 keV and ∼ 8 keV, respectively, suggesting a P Cygni-like profile. The vertical lines indicate the Fe Kα laboratory transition at Elab = 6.40 keV (lime green) and Fe XXV and Fe XXVI at Elab = 6.70 keV and 6.97 keV, respectively (magenta). The outer to inner closed contours correspond to a Δ𝒞 significance of (−2.3 = 68%), (−4.61 = 90%), (−9.21 = 99%), and (−13.82 = 99.9%) relative to the best-fitting continuum. Bottom: corresponding blind-search contours, showing the evidence of strong emission profiles at high confidence level (see the colour bar on the right).

In Fig. 4 we show the result of the same blind-search procedure applied to the type 1 AGN LBQS 1338−0038. The energy–intensity contour map is showing the presence of highly significant emission and absorption profiles at centroid rest-frame energies of Eem ∼ 6.6 keV and Eabs ∼ 8 keV, respectively. What differs from 2MASS J165315+2349 is that both the emission and absorption have comparable width and the emission (as well as the absorption) seems to be originating from ionised material. Such features are highly reminiscent of the well-established P Cygni-like profile detected in PDS 456 (Nardini et al. 2015), where the emission component arises from photons scattered back into our line of sight from the same outflowing ionised outflow, averaged over all the viewing angles. The complete set of blind-search line contours of all the remaining observations in the sample are plotted in Fig. E.1. The visual inspection of the residuals makes the Fe K emission/absorption profiles detections largely qualitative at this stage. Nonetheless, we now have a strong basis to carry out a systematic identification of iron K absorption features that might be arising from a UFO.

4.3. Modelling the Fe K band

In this paper, we only adopt phenomenological models that are homogeneously fitted across the whole sample. All the Fe K emission and absorption residuals arising from the line blind-search (see Fig. E.1) are fitted with a cosmologically redshifted Gaussian model (zgauss in XSPEC) added to Eqs. (1) and/or (2). Modelling the Fe K features this way allows us to characterise the significance and intrinsic properties of the lines, namely the centroid rest energy, the line width, and the overall strength with respect to the underlying continuum. In some cases, when the feature is broader, we fix their line widths (for simplicity) between σwidth = 10 eV–400 eV (depending on which produces a larger improvement to the fit). When the lines are broad/resolved, the line width is left as a free parameter (see Table 2). A strong emission profile corresponding to the neutral Fe Kα core at ∼ 6.4 keV is present in almost every target. In some observations, of PG 0953+414, PG 1425+267, PG 1626+554, PG 1352+183, PG 1216+069, PG 0947+396_ obs1, 2MASS J105144+3539 and PG 0804+761, the Fe K emission lines are more complex (see Fig. E.1). Their Fe K emissions are broader and the centroid energies correspond to highly ionised iron emission in the Fe XXV and Fe XXVI domain. In the context of UFOs, broader iron K emission can arise from scattered photons from outflowing ionised material and can have crucial implications in determining the covering fraction of the outflowing gas (Sim et al. 2008, 2010; Nardini et al. 2015; Matzeu et al. 2016; Reeves & Braito 2019).

For the absorption features we estimated the outflow velocity by assuming that iron K absorption observed at energy Eabs is associated with H-like iron (Fe XXVI Lyα) gas, with laboratory rest-frame energy of E0 = 6.97 keV. Assuming E0 = 6.7 keV (i.e. Fe XXV Heα) would correspond in the above calculations to a faster vout, given the larger degree of blueshift. Our choice of E0 can be considered as a conservative choice for the outflow velocity estimate, based on the reference energy assumed. For a more appropriate identification of line transitions (i.e. Fe XXV Heα, Fe XXVI Lyα, or Fe K-shell edges), we need to carry out a photoionisation modelling of the Fe K features (with XABS and XSTAR models), which can yield an accurate description of the physical conditions of the gas, for example its ionisation state. A comprehensive physically motivated analysis of the emission/absorption profiles of the entire sample is the subject of a forthcoming companion paper.

For quantifying the statistical significance of the Gaussian lines in our modelling, we first adopt the Δ𝒞 improvement as used for our blind-search method. More specifically, we compute the PF significance derived by first obtaining the Δ𝒞/Δν between the best fit with and without a specific Gaussian line and subsequently compute the F-test probability. We detected blueshifted Fe K absorption lines in 11 out of 22 sources at PF ≳ 95%, where 9/22 have PF ≳ 99%.

Numerous authors in the literature (e.g. Vaughan et al. 2003; Porquet et al. 2004b; Markowitz et al. 2006; T10; G13; Igo20, Parker et al. 2020) have established that to obtain an adequate statistical test when determining the significance of a detection of atomic lines in rather complex spectra, an extensive ℳ𝒞 approach is required. In an absorption-line search framework we might detect unexpected lines at a specific energy, without any prior justification, (e.g. Protassov et al. 2002) over an arbitrary energy range. Indeed, as discussed in the next section, we find that our PF improvements over-predict the detection probabilities, as opposed to a more robust ℳ𝒞 simulation approach.

4.4. Monte Carlo approach

The ℳ𝒞 simulation method has now been extensively adopted in the literature (e.g. Porquet et al. 2004b; Miniutti & Fabian 2006; Tombesi et al. 2010; Gofford et al. 2013; Nardini et al. 2019; Parker et al. 2020; Middei et al. 2020) in order to achieve a robust determination of the significance of a spectral line independently from the spectral noise and the quality of the detector. The ℳ𝒞 approach overcomes the limitations of the often used F-test, which can sometimes over-predict the statistical significance of the line detection when compared to extensive ℳ𝒞 simulations. In this paper the ℳ𝒞 approach is focused on the Fe K absorption lines detected in 11 sources on the basis of PF ≳ 95%. We report the results on the detection probability based on ℳ𝒞 in Table 2.

This process was carried out by following these steps. (i) The continuum baseline null-hypothesis model (𝒩ℋℳ) is our final best-fitting 0.3–10 keV model (see Table D.1) re-adjusted after removing the Gaussian absorption component. For each test, we simulated 1000 EPIC-pn, 1000 EPIC-MOS 1 and 1000 EPIC-MOS 2 source and background spectra, by using the fakeit command in XSPEC. The simulated spectra were generated with the same exposure times and response files from the original data and grouped accordingly. We adopt the 𝒦ℬ binning (as described in Appendix C) with FTGROUPPHA. (ii) Our 0.3–10 keV simulated spectra are then fitted with the 𝒩ℋℳ, which takes into account the associated uncertainties from the 𝒩ℋℳ itself. During this procedure, we fixed the line width of any broad Gaussian emission features present in the spectra (both in the soft and hard X-ray band) at their corresponding best-fit energy values from the real data and we let their intensities free to vary. The dual blackbody temperature, normalisations and any Galactic, intrinsic neutral/warm absorptions were frozen to their best-fit values of the real data reported in Table D.1. (iii) A narrow Gaussian profile, with line width fixed at zero, was then added to the 𝒩ℋℳ, with normalisation also set to zero, but free to fluctuate between negative and positive values, in order to probe both absorption and emission features. The rest-energy centroid of the Gaussian line was stepped between 5 and 10 kev in ΔE = 25 eV increments with the STEPPAR command in XSPEC. Additionally, to prevent a local minimum during fitting, we also used the SHAKEFIT procedure developed by Simon Vaughan (see Sect. 3.2.2 in Hurkett et al. 2008). This process maps the Δ𝒞 variations relative to 𝒞null, which are recorded after each step as |Δ𝒞noise|. The degrees of freedom corresponding to both models are also recorded. (iv) The above steps were repeated through S = 1000 iterations for each test, which produced a |Δ𝒞noise| distribution under the null-hypothesis by mapping the statistical significance of any deviations from 𝒩ℋℳ due to random photon noise in the spectra. (v) The initial significance of the line derived from the real data |Δ𝒞line| was compared to the |Δ𝒞noise| distribution so that the number N of simulated spectra with a random noise fluctuation larger than the observed one can be evaluated. In case when the N simulated spectra have |Δ𝒞noise|≥|Δ𝒞line|, the ℳ𝒞 statistical significance (Pℳ𝒞) of the absorption line detection can be calculated as and reported in Table 2.

In order to compare with literature results, and following, for example, T10, we consider as robust detections only absorption lines with PF ≳ 99% and Pℳ𝒞 ≳ 95%. The discussion of the Fe K emission/absorption detection rate in our sample is presented in Sect. 5.1.

5. Results

A total of 14 absorption features with energies Erest ≳ 7.1 keV and PF ≳ 99% are found. Of these, eight are robustly detected with Pℳ𝒞 ≳ 95% while 6 have Pℳ𝒞 < 95% and are therefore considered non-detections. In PG 1114+445 (ObsID 0651330501) an absorption line at was detected at the PF > 99% confidence level. However, such a feature is likely consistent with a neutral iron K edge so no ℳ𝒞 test was applied here.

In Fig. 5 (top), we plot the unfolded EPIC-pn, MOS 1, and 2 data showing the 8 Fe K absorption lines detections with Pℳ𝒞 ≳ 95%. To avoid model and data convolution issues, the spectra in each panel are initially unfolded against a simple Γ = 2 power law (with normalisation of 1) and subsequently their corresponding best-fitting model are superimposed. In Fig. 5 (middle) the plot is in terms of the EPIC data counts normalised by the effective areas and Fig. 5 (bottom) are their corresponding residuals.

thumbnail Fig. 5.

XMM-Newton EPIC spectra and the corresponding best-fit models focused on the Fe K band. Top: unfolded EPIC-pn (black), EPIC-MOS 1 (dashed red), and EPIC-MOS 2 (dashed green) spectra, between 5–10 keV, of the eight observations where the Fe K absorption line was detected at Pℳ𝒞 ≳ 95%. The spectra were firstly unfolded against a power law of Γ = 2, and their corresponding best-fitting model (solid red) was overlaid afterwards. Middle: corresponding EPIC data counts and best-fitting model. Bottom: data/model ratio. The presumed absorption features at ∼ 9.2 keV and ∼ 9.5 keV present in, for example, 2MASSJ105144+3539 and 2MASS J165315+2349, respectively, are simply not significant enough to be considered as detections (see Figs. 4 and E.1), and hence they were not included in the best-fitting models.

We conservatively identify these Fe K absorption lines as highly ionised iron, specifically Fe XXVI Lyα K-shell transitions, all blueshifted with respect to their laboratory rest energies. Some of these lines can be a blend of both Fe XXV Heα and Fe XXVI Lyα resonant transitions and might be indistinguishable with the current EPIC energy resolution. In a forthcoming paper (Matzeu et al., in prep.), we will carry out a comprehensive physical modelling of these features where an accurate measurement of the ionisation balance, as well as density and velocity, of the outflowing gas can be achieved. A photoionisation analysis of the Fe K lines will also help in obtaining a quantitative identification of the absorption and emission features. As presented in Sect. 4.3, the corresponding outflow velocities were conservatively estimated by choosing Erest = 6.97 keV as a reference energy (see Table 2).

5.1. Line detection rate

Here, we quantify the probability of whether or not the detected absorption lines are caused by statistical fluctuation (‘shot noise’). This can be done by using the binomial distribution (e.g. T10; G13). For an event with a null probability, p, the likelihood of n detections after N trials is given by the expression

(3)

In this context, n is the number of absorption lines detected in N systems, and, depending on the latter quantity, we investigate two different cases where we take into account: case (i) all the individual targets, or N(i) = 22; and case (ii) all the individual observations (with total net counts of ≥ 1500 cts in the 4–10 keV band), or N(ii) = 41.

In case (i) we have n95 = 7 Fe K absorption line systems detected in N(i) = 22 observations at a significance of Pℳ𝒞 ≥ 95%. So the probability of one of these absorption profiles being due to fluctuating noise can be taken as p < 0.05. The probability of all of the observed absorption systems being associated with noise is then reasonably low, with P95, (ii) < 6.17 × 10−5 (≲0.006%). This suggests that the observed lines are unlikely to be associated with simple statistical fluctuations in the spectra.

In case (ii) we have a total of n95 = 8 detections out of N(ii) = 41 individual observations. Here we have P95, (ii) < 6.89 × 10−4 (≲0.07%).

5.2. Photoionisation modelling of Fe K features: Initial results

Although this paper is solely focused on UFO detection, we present a preliminary photoionisation analysis of the Fe K features and we provide first-order physical measurements of their properties. This analysis is carried out so that our SUBWAYS results can be compared with those previously obtained in Tombesi et al. (2011, T11 hereafter) T11, G13 and Igo20. The search for absorption features and the Gaussian modelling of the absorption profiles in the Fe K band described in Sect. 4.3 suggest they can be ascribed to outflowing and highly ionised material likely associated with Fe XXV Heα–Fe XXVI Lyα transitions.

In contrast with the phenomenological models used before, modelling the absorption features with XSTAR allows us to probe the physical properties of the absorbing medium. More specifically, we are able to quantify the ionisation state (ξ), the column density (NH) and the systemic redshift of the material relative to the observed one, which translates into the outflow velocity (vout; see below for more details). Through the photoionisation modelling approach it is also possible to infer the geometric properties, such as the radial distance from the ionising source, the covering factor and the resulting overall kinematics (e.g. Gofford et al. 2015; Matzeu et al. 2017).

We replaced the Gaussian absorption profiles, detected at the Pℳ𝒞 ≳ 95% confidence level, with XSTAR photoionisation models, generated with a power-law SED input spectrum of Γ = 2, by using the XSTAR suite v2.54a (Bautista & Kallman 2001; Kallman et al. 2004). We adopted various XSTAR grids with different turbulent velocity, defined as , so that an accurate description of the width of the Fe K absorption lines could be provided. Choosing a grid with a smaller σturb results in a smaller equivalent width of the profile in the data, and the absorption would saturate too quickly at lower NH. So for each source we adopted grids with σturb ranging between 1000–10 000 km s−1 (see Table 3).

Table 3.

XSTAR parameters for the Fe K absorption features detected in the SUBWAYS sample with Pℳ𝒞 ≳ 95%.

The measured column densities are ranging between 1023NH/cm−2 ≲ 1024 with a mean value of and a median of . We also report the measured ionisation distribution, which is found to extend between 3.5 ≲ log(ξ/erg cm s−1) ≲ 5.5, with mean/median values of and , respectively.

The outflow velocity distribution measured with XSTAR8 ranges between −0.3 ≲ vout/c ≲ −0.1 (see Table 3). The mean/median values are and , respectively. We also compare the outflow velocities measured with XSTAR and with the one measured from the Gaussian fitting. For the latter, the measured velocities are found to lie between −0.3 ≲ vout/c ≲ −0.05 (see Table 2) with mean/median values of and , respectively. We find that both the phenomenological and the physical modelling of the Fe K absorption features have the same distribution, as confirmed at the 99.7% confidence level by a Kologoromv-Smirnov test. The XSTAR-based approach returns a ∼10% higher mean velocity but a comparable median, within ∼3%.

In our sample a considerable fraction of the Fe K absorbers are characterised by material with high column density and highly ionised material, likely H-like iron. Such a result does not come as surprise when considering the hard average photon index () measured on the entire sample (see Fig. 6), which would over-ionise the outflowing material (see Matzeu et al. 2022, Figs. 2 and 3). A comprehensive photoionisation analysis with XSTAR (and other physically motivated wind models) will be presented in a companion paper, where customised photoionisation tables will be generated with more realistic optical/UV/X-ray SED inputs for each individual SUBWAYS source (e.g. Nardini et al. 2015; Matzeu et al. 2016).

thumbnail Fig. 6.

Histogram comparing the photon index (Γ) measured in SUBWAYS (green; this work, with a mean value of ; vertical line) with those measured in an X-ray-selected sample (CAIXA sample, 150 sources with ; Bianchi et al. 2009), the PGQSO sample (Piconcelli et al. 2005, 40 sources ), and local AGNs (T10; G13 with and , respectively).

6. Discussion

In this work we searched for Fe K absorption features in a sample of 22 targets (41 observations), of which 17 were observed as part of the XMM-Newton large programme carried out in AO18. Through a systematic blind line scan performed in all the observations and supported by a ℳ𝒞 procedure, we detected iron K absorption lines in 7/22 sources (i.e. ∼30%) at the Pℳ𝒞 ≳ 95% confidence level. Through our statistical approach, we have found 2 robust Fe K absorption line detections at Pℳ𝒞 ≳ 99% in 2MASS J165315+2349 and LBQS 1338−0038. The remaining 5 detections are still significant but with 95%≲Pℳ𝒞 ≲ 99%, in PG 1202+281, 2MASS J105144+3539, PG 1114+445, PG 0804+761 and PG 0947+396 (see Table 2).

Such absorption (and sometimes emission) profiles are associated with highly ionised He- and/or H-like iron, arising from the outflowing material, as their centroid energy is blueshifted with respect to the QSO systemic redshift. In this paper we only focused on the search and phenomenological analysis of such features. Thus, for the estimate of the outflow velocities, we assumed Fe XXVI Lyα at Erest = 6.97 keV as a reference energy for a conservative result, which corresponds to the lowest possible outflow velocity. Accordingly, we find that the average outflow velocity measured in our sample is , as shown in Fig. 7.

thumbnail Fig. 7.

Distributions and mean outflow velocity values, indicated by the vertical dashed lines, as measured in: this work (SUBWAYS) ; T10 ; G13 ; Igo20 ; and C21 .

For our search of Fe K absorption and emission features, an accurate parameterisation of the underlying broadband continuum (i.e. 0.3–10 keV) in each observation was required. We therefore summarise the phenomenological continuum findings of SUBWAYS. We found that 27 out of the 41 (∼65%) SUBWAYS observations are characterised by intrinsic soft X-ray absorption. More specifically, 20/27 systems can be identified as fully covering, mildly ionised (warm) absorbers, while 5/27 are partially covering the line of sight. We note that 2/27 spectra are modified by a fully covering neutral absorber, where in 2MASS J165315+2349 the spectral curvature at energies < 2 keV is caused by a column density NH ∼ 1023 cm−2, consistent with the values measured in Seyfert 2 galaxies. We find that a prominent soft excess, at a > 99% significance, is present in the vast majority of the spectra in our sample: 39/41; one zbbody component is required in 13/39 and two components in 26/39 sources.

In Fig. 6 we compare the primary continuum photon indices (Γ) with those measured in previous works in the literature: a purely X-ray-selected sample, CAIXA (Bianchi et al. 2009), an optically selected PGQSO sample (Piconcelli et al. 2005), and low-z AGNs analysed with XMM-Newton (T10) and Suzaku (G13). We find that the SUBWAYS sample tends to be characterised by Γ < 2 (), only slightly harder compared to the optically selected PGQSO sample () and largely consistent with the CAIXA (Bianchi et al. 2009) () and T10 () samples. The mean photon index of the Suzaku sample is softer, with ().

6.1. Sample comparisons

The SUBWAYS sample size is indeed small compared to those in T10; G13 and Igo20. With this in mind this selection of targets must be taken as an initial exploration of the intermediate-z population that is bridging the gap of UFOs studies between low- and high-z sources.

We find that our overall measurements seem to be skewed to higher values of NH and vout compared to (T10 and G13, whilst the ionisation state of the absorber is on the same order of magnitude (slightly lower). The latter parameter can be influenced by the SED input assumed when generating the photoionisation grids. We also find that the outflow velocity measured in PG0947+396 (Obs1), being the highest in the sample, is unlikely to be associated with outflowing, highly ionised material, but rather the result of an artefact of the EPIC CCD and/or some background issue. Although the absorption line is significantly detected at the PF > 99% (Gaussian modelling) and PMC ∼ 97% (ℳ𝒞 approach) confidence level, it is weakly detected with XSTAR at ∼90%. Furthermore, an outflow velocity of vout > 0.30c is generally considered on the high end of the scale of ultra fast winds and can carry a huge amount of kinetic power (e.g. Matzeu et al. 2017; Reeves et al. 2018a). These events are more likely to be present in highly accreting sources where Lbol/LEdd → 1 (or above) and Eddington fractions of Lbol/LEdd ∼ 10% (Bianchi et al. 2009) might be not enough to drive such strong outflows, although it cannot be ruled out as magneto-hydrodynamic (MHD) driving mechanisms could come into play (e.g. Fukumura et al. 2010, 2022; Kraemer et al. 2018; Luminari et al. 2021), especially in low-Eddington regimes.

In Fig. 7, we show the vout distributions, and their mean values, measured with XMM-Newton in previous works in the literature, such as in T10 ; Igo20 ; C21 . An interesting trend is shown in Fig. 7. By looking at all the measurements, the UFO outflow velocities seem to increase with redshift. Although the statistical footing of this trend is beyond the scope of this paper, we can recognise that such behaviour does arise from a high-Lbol selection bias expected in sources at progressively higher redshift as the feeding becomes stronger (e.g. Di Matteo et al. 2005), in particular due to the larger inflow of cold gas mass triggering chaotic cold accretion and boosting accretion rates by a few orders of magnitude compared with quiescent hot modes (e.g. Gaspari et al. 2017). Another way to interpret this trend is simply realise that the outflow velocity seems to increase with the luminosity (e.g. Saez & Chartas 2011; Matzeu et al. 2017; Chartas & Canas 2018, C21), as shown below in Fig. 8, and on the other hand the most luminous sources are observed at higher z. Additionally, another possible bias that is involved at higher z is that higher velocity shifts become more detectable with increasing redshift.

thumbnail Fig. 8.

Outflow velocity of the ionised absorber for our SUBWAYS targets (red filled stars) plotted against their bolometric luminosity. Also plotted are the same quantities as derived in the analysis of the low-z AGN samples of Tombesi et al. (2012, red empty squares) and G13 (blue empty triangles) and in the high-z AGN sample of C21 (black filled circles). The power-law least-squares fit to the combined samples (a total of 49 individual objects and 63 observations) is shown with a solid line. The shaded area represents the uncertainty of the slopes of our fits to the data. Here the outflow velocity is simply described as the absolute value of vout.

In Fig. 8 we show the wind velocities, as measured from our X-ray spectral fits with Gaussian lines and tabulated in Table 2, plotted against the bolometric luminosity of the SUBWAYS targets (see Table 1) shown as red stars. We added and recomputed the fit of the correlation of vwind versus Lbol of the low-z (T10; G13, red squares and blue triangles, respectively) and high-z samples already presented in C21, including also our SUBWAYS measurements. For all samples, bolometric luminosities are consistently computed from the 2–10 keV luminosities assuming a luminosity-dependent bolometric correction (Duras et al. 2020). The best fit parameters of a linear relation in log space of log(vout) = A(Lbol)B are A = −3.88 ± 1.40 and B = 0.19 ± 0.03. Overall, the SUBWAYS data fit right between the low- and the high-z data. We find a Kendall’s (rank) correlation coefficient of τ = 0.45 with a null probability of pnull = 1.8 × 10−7. The strength and slope of this correlation is partially driven by the six data points in C21, with log(Lbol/erg s−1) ≳ 48, that may be affected by additional uncertainties associated with the magnification factor due to lensing.

The low- and high-z fit correlation in C21 (see their Table 10) returned a slope of B = 0.20 ± 0.03, a correlation coefficient of τ = 0.51 with a null probability pnull = 6.0 × 10−8. Their slope is consistent with our measurement, whereas their coefficient is about 10% higher, which suggests that our correlation, with all the four samples included, is slightly weaker than in C21. A similar correlation was also observed by Matzeu et al. (2017) in the luminous QSO PDS 456 between the XMM-Newton, NuSTAR, and Suzaku observation from 2001–2014. The slope of the correlation in Matzeu et al. (2017, i.e. 0.22 ± 0.04) is largely consistent with what we have found here.

Overall, with our result we can conclude that there is a correlation between the outflow velocities and bolometric luminosities within the overall low- intermediate- high-z samples. A positive correlation with a slope of 0.5 between outflow velocities and the luminosities of the AGNs is what it would be expected in a radiatively driven wind scenario as the radiation pressure plays a key role in driving the outflows.

The fact that the observed slope is lower than the expected value can be explained in several ways. One possible explanation, already suggested in C21, is that we did not include outflows with velocities ≲10 000 km s−1, as in T10.

Another plausible explanation is that, as the luminosity keep increasing, the inner part of the UFOs detected in SUBWAYS might be over-ionised, with weaker absorption features (e.g. Parker et al. 2017; Pinto et al. 2018) leading to their observability being pushed to the outer streamlines. Within this regime the observed velocities, due to their radial dependence, would appear slightly slower and such a physical condition leads to an overall flattening of the slope (Matzeu et al. 2017) from the nominal value of 0.5. The outflows shown in Fig. 8 have a range of mass outflow rates and black hole masses, so it is not clear that a simple scaling of velocity with luminosity is likely. If instead we assume that all the systems are close to their Eddington luminosities LEdd and the outflows have the Eddington momenta LEdd/c (see e.g. King 2003; King & Pounds 2015, and references therein) one finds that the velocities should be of order 0.1 c, as observed (see also Fig. 7).

In reality, the driving (and launching) mechanism responsible for the observed UFOs are likely the result of a complex interaction between radiation pressure and MHD driving. Indeed, it was previously found, in Matzeu et al. (2016) that in the powerful disc wind observed in PDS 456 in 2013, the radiation pressure alone, imparted from a strong flare, could have not deposit enough kinetic power on the outflowing material and hence suggesting that an additional launching mechanism, such as MHD, was also involved. Decoupling and assessment of each individual contribution remains a challenging subject in disc wind physics with the current CCD detectors.

6.2. Strong features around rest energies of 9 keV

In our SUBWAYS sample, the Fe K absorption line with the highest degree of blueshift was detected in PG 0947+396 (Obs1) at Erest = 9.5 ± 0.1 keV (Eobs ∼ 7.9 keV).

Despite its reasonable significance (see Table 2), there are a few caveats that could rule out its UFO identification. The energy shift from the H-like iron rest energy is rather large (even larger if He-like Fe is considered), and corresponds to an outflow velocity of vout = −0.30 ± 0.01c, which is, by far, the fastest of the sample. Such a result could be (in principle) at odds considering the relatively low Eddington fraction of this source (i.e. Lbol/LEdd ∼ 0.06). These kinds of outflow velocities are more common in sources that are accreting near or above their Eddington limit, for example PDS 456. In this campaign, a second ∼ 60 ks observation (PG 0947+396 Obs2) was carried out about 5 months later and no Fe K absorption line was detected in the spectra (see Fig. E.1). The same applies for the first 20 ksXMM-Newton observation in 2001. Having said that, it is not impossible to have such a powerful UFO in a low-Eddington regime as other driving mechanisms, such as MHD, could play a key role (e.g. Fukumura et al. 2017). Indeed, further monitoring of this source will shed some light on the presence of a UFO.

Strong residuals in emission at Erest ∼ 9 keV are detected in 2MASS J165315+2349 and PG1114+445 (0651330801) at the PF > 99% confidence level, which might be associated with a possible high-order iron K transition or perhaps associated with an instrumental calibration artefact. The origin of this feature could be associated with Ni XXVII Heβ 1s → 3p, or a blend thereof; however, no lower transitions are observed in the spectra. Such a feature is also observed in PG 1114+445, as blueshifted Fe XXVI Lyβ (8.25 keV). The dominant emission line in the pn background is the Cu Kα line at 8.04 keV. Weaker surrounding lines include Ni Kα (7.47 keV), Zn Kα (8.63 keV) and Cu Kβ (8.90 keV).

So another possible origin of these spurious features might arise from background subtraction issues during the data reduction process, or even from the detector itself such as in PG 1114+445. After a careful check we confirm that the line detections are genuine as no such issues were found. A thorough characterisation of the physical properties of the winds responsible for the detected Fe K lines will be carried out by using physically motivated models such as XSTAR, XABS, WINE (Luminari et al. 2021), and XRADE (Matzeu et al. 2022), and will be presented in a forthcoming SUBWAYS paper.

A second absorption line at Erest ∼ 11 keV (Eobs ∼ 9 keV) is detected in LBQS 1338−0038 at the PF = 99.3% confidence level however, following a ℳ𝒞 procedure the significance of the line drops considerably to Pℳ𝒞 = 87.2%. This discrepancy in the detection can be likely attributed to the lack of data at energies Eobs > 10 keV being at the edge of the XMM-Newton band pass, and hence extra care is needed. Nonetheless, this high energy absorption feature is also present during a ∼ 50 ksNuSTAR exposure one year later in 2020 (PI Bianchi). A joint analysis focused on of the XMM-Newton and NuSTAR data of LBQS 1338−0038 will be presented in a companion letter (Matzeu et al., in prep.).

7. Summary and conclusions

We have carried out a systematic search focused on absorption features in the Fe K band in a sample of 22 (41 observations) luminous (2 × 1045Lbol/erg s−1 ≲ 2 × 1046) AGNs at intermediate redshifts (0.1 ≲ z ≲ 0.4) as part of the large XMM-Newton programme SUBWAYS. For each XMM-Newton observation, the data reduction was performed by optimising the level of background in order to increase of the S/N within the 4–10 keV band. Afterwards, an additional and crucial step was the appropriate choice of an optimal spectral binning to avoid a loss of information for any search of weak features, such as UFOs. We applied the four spectral binnings that are most used in the literature, and subsequently, after cross-checking them, all our results are based on the 𝒦ℬ binning.

The main results are summarised below:

  • We carried out an XMM-Newton broadband analysis between 0.3 and 10 keV. We find that in 27 out of 41 observations (∼65%) our targets have intrinsic absorption in the soft X-rays, of which ∼70% can be identified as fully covering, mildly ionised (warm) absorbers and 5 out of 27 (∼30%) partially cover the line of sight.

  • We then analysed the EPIC spectra of each of the 41 observations by first performing a series of blind-search line scans, in both emission and absorption, focused on the 5–10 keV band. For the overall Fe K emissions, the energies are in the range 6.2 ≲ Erest/keV ≲ 9, ∼90% are consistent with the Fe Kα core (detected in 20 out of 22 targets), and ∼36% are consistent with Fe XXV–XXVI (detected in 8 out of 22 targets). Their equivalent widths are in the range of 10 ≲ EW/eV ≲ 468, and the high-end values are due to complex emission features such as in PG 1352+183, likely arising from a blend between Fe Kα[β] and Fe XXV.

  • For the Fe K absorption features, we detected 14 absorption lines at the PF ≳ 99% confidence level with energies in the range 7 ≲ Erest/keV ≲ 11; ∼85% are consistent with Fe XXV–XXVI, and one detection is consistent with an iron K edge. Their equivalent widths are in the range −200 ≲ EW/eV ≲ −40, which is in line with what is expected for relatively narrow Fe K absorption profiles.

  • Thanks to extensive ℳ𝒞 simulations, we confirmed absorption lines corresponding to highly ionised iron in 7 out of 22 sources. These findings yield a UFO detection fraction of ∼30% of the total sample, at a Pℳ𝒞 ≳ 95% significance level. These features likely correspond to Fe XXV Heα and/or Fe XXVI Lyα. By using the Fe XXVI lab transition as the reference energy, we measured outflow velocities in the range −0.3 ≲ vout/c ≲ −0.05 with average and median velocities of and .

  • In this work we also present preliminary results of photoionisation modelling of the iron K features detected at the Pℳ𝒞 ≳ 95% confidence level with XSTAR. We find median values of and for the column densities and ionisation parameter, respectively.

  • The measured outflow velocities with XSTAR are in the range −0.3 ≲ vout/c ≲ −0.1, where the mean and median values are and , respectively. Such a distribution is largely comparable with the outflow velocities measured with the phenomenological (Gaussian) modelling. Such results confirm that the absorption detected in the Fe K band arises from fast, highly ionised material with a high column density, as typically observed in UFOs.

  • By comparing our results with previous work, we computed a power-law least-squares fit to the low-z (T10; G13), intermediate-z (SUBWAYS), and high-z (C21) data, which show a positive correlation between outflow velocity and bolometric luminosity within the overall low-, intermediate-, and high-z samples, with slope 0.19 ± 0.03. Such a voutLbol correlation is also observed in Matzeu et al. (2017) with a slope of 0.22 ± 0.04.

The outcome of this work independently provides further support for the existence of highly ionised matter propagating at mildly relativistic speeds, which is expected to play a key role in the self-regulated AGN feeding-feedback loop that shapes galaxies, as shown by hydrodynamical multi-phase simulations (Gaspari et al. 2020, for a review). These results suggest that the likely dominant driving mechanism of UFOs is radiation pressure arising in high-accretion regimes. It is important to note that MHD also plays a key role in the driving and launching mechanism of disc winds, and future observations at micro-calorimeter resolution will contribute towards distinguishing each component.

An alternative scenario that has been put forward is that the origin of Fe K absorption features can be attributed to a layer of hot gas located at the surface of the accretion disc rather than from an outflowing wind (Gallo et al. 2013). Thus, the prominent and blueshifted absorption lines are the result of a strong relativistic reflection component that dominates the hard X-ray continuum rather than the primary emission. This model was successfully applied to the NLSy1 IRAS13224−3809 by Fabian et al. (2020). The unprecedented spectral micro-calorimeter resolution from future UFO observations, such as X-Ray Imaging Spectroscopy Mission (XRISM/Resolve; Tashiro et al. 2020) and Advanced Telescope for High-ENergy Astrophysics/X-ray Integral Field Unit (Athena/X-IFU; Barret et al. 2018), will greatly contribute towards disentangling each of these scenarios, including the disc wind’s launching and driving physical mechanism (e.g. Giustini & Proga 2012; Fukumura et al. 2022, Dadina et al., in prep., Matzeu et al., in prep.).


1

In this analysis we discarded all observations with a net count threshold of ≲1500 in the 4–10 keV band.

4

In some SUBWAYS targets only one blackbody component is required (see Sect. 4.1.2), while no blackbody component is needed for the absorbed sources 2MASS J105144+3539 and 2MASS J165315+2349.

6

We are aware that our XABS table is built based on the ion balance calculated for a typical Seyfert, while our sample consists of QSOs of higher luminosity. By testing the same data with an XSTAR grid with turbulent velocity of vturb = 300 km s−1 and a power-law SED input with a photon index of Γ = 2, we get ionisation parameters that are slightly higher but consistent within the errors. For this reason, we allowed ξ to explore the wide range of ionisation reported above.

8

The systemic redshift of the absorber obtained from fitting with XSTAR is given in the observer’s rest-frame zabs and related to vzabs = [(1+zabs)2−1]/[(1+zabs)2+1], and correcting for the systemic velocities of the sources u we obtain vout/c = (uvzabs)/[1−(uvzabs)].

Acknowledgments

GAM and all the italian co-authors acknowledge support and fundings from Accordo Attuativo ASI-INAF n. 2017-14-H.0. MB is supported by the European Union’s Horizon 2020 research and innovation programme Marie Skłodowska-Curie grant No 860744 (BID4BEST). MG acknowledges partial support by HST GO-15890.020/023-A, the BlackHoleWeather program, and NASA HEC Pleiades (SMD-1726). BDM acknowledges support via Ramón y Cajal Fellowship RYC2018-025950-I. SM is grateful for the NASA ADAP grant 80NSSC20K0438. AL acknowledges support from the HORIZON-2020 grant “Integrated Activities for the High Energy Astrophysics Domain” (AHEAD-2020), G.A. 871158. SRON is supported financially by NWO, the Netherlands Organization for Scientific Research. M.Gi. is supported by the “Programa de Atracción de Talento” of the Comunidad de Madrid, grant number 2018-T1/TIC-11733. We warmly thank Katia Gkimisi and Raffaella Morganti for useful discussions.

References

  1. Aird, J., Coil, A. L., Georgakakis, A., et al. 2015, MNRAS, 451, 1892 [Google Scholar]
  2. Arnaud, K. A. 1996, ASP Conf. Ser., 101, 17 [Google Scholar]
  3. Ballantyne, D. R., & Xiang, X. 2020, MNRAS, 496, 4255 [NASA ADS] [CrossRef] [Google Scholar]
  4. Barret, D., Lam Trong, T., den Herder, J.-W., et al. 2018, Proc. SPIE, 10699, 106991G [Google Scholar]
  5. Bautista, M. A., & Kallman, T. R. 2001, ApJS, 134, 139 [Google Scholar]
  6. Bertola, E., Dadina, M., Cappi, M., et al. 2020, A&A, 638, A136 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  7. Bianchi, S., Guainazzi, M., Matt, G., Fonseca Bonilla, N., & Ponti, G. 2009, A&A, 495, 421 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  8. Bischetti, M., Maiolino, R., Carniani, S., et al. 2019a, A&A, 630, A59 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  9. Bischetti, M., Piconcelli, E., Feruglio, C., et al. 2019b, A&A, 628, A118 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  10. Blustin, A. J., Page, M. J., Fuerst, S. V., Branduardi-Raymont, G., & Ashton, C. E. 2005, A&A, 431, 111 [CrossRef] [EDP Sciences] [Google Scholar]
  11. Boller, T., Liu, T., Weber, P., et al. 2021, A&A, 647, A6 [EDP Sciences] [Google Scholar]
  12. Braito, V., Reeves, J. N., Gofford, J., et al. 2014, ApJ, 795, 87 [NASA ADS] [CrossRef] [Google Scholar]
  13. Braito, V., Reeves, J. N., Matzeu, G. A., et al. 2018, MNRAS, 479, 3592 [Google Scholar]
  14. Brusa, M., Bongiorno, A., Cresci, G., et al. 2015, MNRAS, 446, 2394 [NASA ADS] [CrossRef] [Google Scholar]
  15. Brusa, M., Cresci, G., Daddi, E., et al. 2018, A&A, 612, A29 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  16. Cash, W. 1979, ApJ, 228, 939 [Google Scholar]
  17. Chartas, G., & Canas, M. H. 2018, ApJ, 867, 103 [NASA ADS] [CrossRef] [Google Scholar]
  18. Chartas, G., Brandt, W. N., Gallagher, S. C., & Garmire, G. P. 2002, ApJ, 579, 169 [Google Scholar]
  19. Chartas, G., Charlton, J., Eracleous, M., et al. 2009, New A Rev, 53, 128 [NASA ADS] [CrossRef] [Google Scholar]
  20. Chartas, G., Cappi, M., Vignali, C., et al. 2021, ApJ, 920, 24 [NASA ADS] [CrossRef] [Google Scholar]
  21. Cicone, C., Maiolino, R., Sturm, E., et al. 2014, A&A, 562, A21 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  22. Cicone, C., Severgnini, P., Papadopoulos, P. P., et al. 2018, ApJ, 863, 143 [NASA ADS] [CrossRef] [Google Scholar]
  23. Costa, T., Sijacki, D., & Haehnelt, M. G. 2014, MNRAS, 444, 2355 [Google Scholar]
  24. Crenshaw, D. M., Kraemer, S. B., & George, I. M. 2003, ARA&A, 41, 117 [NASA ADS] [CrossRef] [Google Scholar]
  25. Cresci, G., Mainieri, V., Brusa, M., et al. 2015, ApJ, 799, 82 [Google Scholar]
  26. Dadina, M., Vignali, C., Cappi, M., et al. 2016, A&A, 592, A104 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  27. Dauser, T., García, J., Parker, M. L., Fabian, A. C., & Wilms, J. 2014, MNRAS, 444, L100 [NASA ADS] [CrossRef] [Google Scholar]
  28. Di Matteo, T., Springel, V., & Hernquist, L. 2005, Nature, 433, 604 [NASA ADS] [CrossRef] [Google Scholar]
  29. Done, C., Davis, S. W., Jin, C., Blaes, O., & Ward, M. 2012, MNRAS, 420, 1848 [Google Scholar]
  30. Duras, F., Bongiorno, A., Ricci, F., et al. 2020, A&A, 636, A73 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  31. Eckert, D., Gaspari, M., Gastaldello, F., Le Brun, A. M. C., & O’Sullivan, E. 2021, Universe, 7, 142 [NASA ADS] [CrossRef] [Google Scholar]
  32. Fabian, A. C. 2012, ARA&A, 50, 455 [Google Scholar]
  33. Fabian, A. C., Reynolds, C. S., Jiang, J., et al. 2020, MNRAS, 493, 2518 [NASA ADS] [CrossRef] [Google Scholar]
  34. Feruglio, C., Maiolino, R., Piconcelli, E., et al. 2010, A&A, 518, L155 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  35. Feruglio, C., Fiore, F., Carniani, S., et al. 2015, A&A, 583, A99 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  36. Feruglio, C., Ferrara, A., Bischetti, M., et al. 2017, A&A, 608, A30 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  37. Fiore, F., Feruglio, C., Shankar, F., et al. 2017, A&A, 601, A143 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  38. Fukumura, K., Kazanas, D., Contopoulos, I., & Behar, E. 2010, ApJ, 715, 636 [NASA ADS] [CrossRef] [Google Scholar]
  39. Fukumura, K., Kazanas, D., Shrader, C., et al. 2017, Nat. Astron., 1, 0062 [NASA ADS] [CrossRef] [Google Scholar]
  40. Fukumura, K., Dadina, M., Matzeu, G., et al. 2022, ApJ, 940, 6 [NASA ADS] [CrossRef] [Google Scholar]
  41. Gabriel, C., Denby, M., Fyfe, D. J., et al. 2004, ASP Conf. Ser., 314, 159 [NASA ADS] [Google Scholar]
  42. Gallo, L. C., Fabian, A. C., Grupe, D., et al. 2013, MNRAS, 428, 1191 [NASA ADS] [CrossRef] [Google Scholar]
  43. García, J., Dauser, T., Reynolds, C. S., et al. 2013, ApJ, 768, 146 [Google Scholar]
  44. García, J., Dauser, T., Lohfink, A., et al. 2014, ApJ, 782, 76 [Google Scholar]
  45. García, J. A., Kara, E., Walton, D., et al. 2019, ApJ, 871, 88 [CrossRef] [Google Scholar]
  46. Gaskell, C. M. 1982, ApJ, 263, 79 [NASA ADS] [CrossRef] [Google Scholar]
  47. Gaspari, M., & Sądowski, A. 2017, ApJ, 837, 149 [NASA ADS] [CrossRef] [Google Scholar]
  48. Gaspari, M., Brighenti, F., & Ruszkowski, M. 2013, Astron. Nachr., 334, 394 [NASA ADS] [Google Scholar]
  49. Gaspari, M., Temi, P., & Brighenti, F. 2017, MNRAS, 466, 677 [Google Scholar]
  50. Gaspari, M., Eckert, D., Ettori, S., et al. 2019, ApJ, 884, 169 [NASA ADS] [CrossRef] [Google Scholar]
  51. Gaspari, M., Tombesi, F., & Cappi, M. 2020, Nat. Astron., 4, 10 [Google Scholar]
  52. Giustini, M., & Proga, D. 2012, ApJ, 758, 70 [NASA ADS] [CrossRef] [Google Scholar]
  53. Giustini, M., & Proga, D. 2019, A&A, 630, A94 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  54. Gofford, J., Reeves, J. N., Tombesi, F., et al. 2013, MNRAS, 430, 60 [Google Scholar]
  55. Gofford, J., Reeves, J. N., McLaughlin, D. E., et al. 2015, MNRAS, 451, 4169 [Google Scholar]
  56. Green, J. C., Froning, C. S., Osterman, S., et al. 2012, ApJ, 744, 60 [NASA ADS] [CrossRef] [Google Scholar]
  57. Halpern, J. P. 1984, ApJ, 281, 90 [NASA ADS] [CrossRef] [Google Scholar]
  58. Hamann, F., Chartas, G., Reeves, J., & Nardini, E. 2018, MNRAS, 476, 943 [NASA ADS] [CrossRef] [Google Scholar]
  59. Harrison, C. M., Alexander, D. M., Mullaney, J. R., & Swinbank, A. M. 2014, MNRAS, 441, 3306 [Google Scholar]
  60. Heckman, T. M., & Best, P. N. 2014, ARA&A, 52, 589 [Google Scholar]
  61. HI4PI Collaboration (Ben Bekhti, N., et al.) 2016, A&A, 594, A116 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  62. Hurkett, C. P., Vaughan, S., Osborne, J. P., et al. 2008, ApJ, 679, 587 [NASA ADS] [CrossRef] [Google Scholar]
  63. Igo, Z., Parker, M. L., Matzeu, G. A., et al. 2020, MNRAS, 493, 1088 [NASA ADS] [CrossRef] [Google Scholar]
  64. Jansen, F., Lumb, D., Altieri, B., et al. 2001, A&A, 365, L1 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  65. Jiang, J., Fabian, A. C., Dauser, T., et al. 2019, MNRAS, 489, 3436 [NASA ADS] [CrossRef] [Google Scholar]
  66. Kaastra, J. S., & Bleeker, J. A. M. 2016, A&A, 587, A151 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  67. Kaastra, J. S., Mewe, R., & Nieuwenhuijzen, H. 1996, in 11th Colloquium on UV and X-ray Spectroscopy of Astrophysical and Laboratory Plasmas, 411 [Google Scholar]
  68. Kaastra, J. S., Mewe, R., Liedahl, D. A., Komossa, S., & Brinkman, A. C. 2000, A&A, 354, L83 [NASA ADS] [Google Scholar]
  69. Kaastra, J. S., Kriss, G. A., Cappi, M., et al. 2014, Science, 345, 64 [Google Scholar]
  70. Kallman, T. R., Palmeri, P., Bautista, M. A., Mendoza, C., & Krolik, J. H. 2004, ApJS, 155, 675 [Google Scholar]
  71. Kaspi, S., Brandt, W. N., Netzer, H., et al. 2000a, ApJ, 535, L17 [NASA ADS] [CrossRef] [Google Scholar]
  72. Kaspi, S., Smith, P. S., Netzer, H., et al. 2000b, ApJ, 533, 631 [Google Scholar]
  73. King, A. 2003, ApJ, 596, L27 [Google Scholar]
  74. King, A. 2005, ApJ, 635, L121 [Google Scholar]
  75. King, A., & Pounds, K. 2015, ARA&A, 53, 115 [NASA ADS] [CrossRef] [Google Scholar]
  76. King, A. R., Zubovas, K., & Power, C. 2011, MNRAS, 415, L6 [NASA ADS] [CrossRef] [Google Scholar]
  77. Kormendy, J., & Ho, L. C. 2013, ARA&A, 51, 511 [Google Scholar]
  78. Kraemer, S. B., Tombesi, F., & Bottorff, M. C. 2018, ApJ, 852, 35 [NASA ADS] [CrossRef] [Google Scholar]
  79. Kriss, G. A., Mehdipour, M., Kaastra, J. S., et al. 2019, A&A, 621, A12 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  80. Krongold, Y., Longinotti, A. L., Santos-Lleó, M., et al. 2021, ApJ, 917, 39 [NASA ADS] [CrossRef] [Google Scholar]
  81. Laha, S., Guainazzi, M., Dewangan, G. C., Chakravorty, S., & Kembhavi, A. K. 2014, MNRAS, 441, 2613 [NASA ADS] [CrossRef] [Google Scholar]
  82. Laha, S., Guainazzi, M., Chakravorty, S., Dewangan, G. C., & Kembhavi, A. K. 2016, MNRAS, 457, 3896 [NASA ADS] [CrossRef] [Google Scholar]
  83. Laha, S., Reynolds, C. S., Reeves, J., et al. 2021, Nat. Astron., 5, 13 [NASA ADS] [CrossRef] [Google Scholar]
  84. Lanzuisi, G., Giustini, M., Cappi, M., et al. 2012, A&A, 544, A2 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  85. Longinotti, A. L., Krongold, Y., Guainazzi, M., et al. 2015, ApJ, 813, L39 [Google Scholar]
  86. Longinotti, A. L., Vega, O., Krongold, Y., et al. 2018, ApJ, 867, L11 [CrossRef] [Google Scholar]
  87. Luminari, A., Piconcelli, E., Tombesi, F., et al. 2018, A&A, 619, A149 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  88. Luminari, A., Nicastro, F., Elvis, M., et al. 2021, A&A, 646, A111 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  89. Maccagni, F. M., Serra, P., Gaspari, M., et al. 2021, A&A, 656, A45 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  90. Magorrian, J., Tremaine, S., Richstone, D., et al. 1998, AJ, 115, 2285 [Google Scholar]
  91. Maiolino, R., Gallerani, S., Neri, R., et al. 2012, MNRAS, 425, L66 [Google Scholar]
  92. Mallick, L., Fabian, A. C., García, J. A., et al. 2022, MNRAS, 513, 4361 [NASA ADS] [CrossRef] [Google Scholar]
  93. Marasco, A., Cresci, G., Nardini, E., et al. 2020, A&A, 644, A15 [EDP Sciences] [Google Scholar]
  94. Markowitz, A., Reeves, J. N., & Braito, V. 2006, ApJ, 646, 783 [NASA ADS] [CrossRef] [Google Scholar]
  95. Mathur, S., Wilkes, B. J., & Aldcroft, T. 1997, ApJ, 478, 182 [NASA ADS] [CrossRef] [Google Scholar]
  96. Mathur, S., Wilkes, B., & Elvis, M. 1998, ApJ, 503, L23 [NASA ADS] [CrossRef] [Google Scholar]
  97. Matzeu, G. A., Reeves, J. N., Nardini, E., et al. 2016, MNRAS, 458, 1311 [Google Scholar]
  98. Matzeu, G. A., Reeves, J. N., Braito, V., et al. 2017, MNRAS, 472, L15 [NASA ADS] [CrossRef] [Google Scholar]
  99. Matzeu, G. A., Lieu, M., Costa, M. T., et al. 2022, MNRAS, 515, 6172 [NASA ADS] [CrossRef] [Google Scholar]
  100. McKernan, B., Yaqoob, T., & Reynolds, C. S. 2007, MNRAS, 379, 1359 [NASA ADS] [CrossRef] [Google Scholar]
  101. McKinley, B., Tingay, S. J., Gaspari, M., et al. 2022, Nat. Astron., 6, 109 [Google Scholar]
  102. Mehdipour, M., Kriss, G. A., Brusa, M., et al. 2023, A&A, 670, A183 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  103. Middei, R., Bianchi, S., Cappi, M., et al. 2018, A&A, 615, A163 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  104. Middei, R., Tombesi, F., Vagnetti, F., et al. 2020, A&A, 635, A18 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  105. Miniutti, G., & Fabian, A. C. 2006, MNRAS, 366, 115 [NASA ADS] [Google Scholar]
  106. Mizumoto, M., Done, C., Tomaru, R., & Edwards, I. 2019, MNRAS, 489, 1152 [NASA ADS] [CrossRef] [Google Scholar]
  107. Morganti, R., Tadhunter, C. N., & Oosterloo, T. A. 2005, A&A, 444, L9 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  108. Murphy, E. M., Lockman, F. J., Laor, A., & Elvis, M. 1996, ApJS, 105, 369 [NASA ADS] [CrossRef] [Google Scholar]
  109. Murray, N., Chiang, J., Grossman, S. A., & Voit, G. M. 1995, ApJ, 451, 498 [NASA ADS] [CrossRef] [Google Scholar]
  110. Murray, N., Quataert, E., & Thompson, T. A. 2005, ApJ, 618, 569 [NASA ADS] [CrossRef] [Google Scholar]
  111. Nardini, E., & Zubovas, K. 2018, MNRAS, 478, 2274 [Google Scholar]
  112. Nardini, E., Fabian, A. C., Reis, R. C., & Walton, D. J. 2011, MNRAS, 410, 1251 [NASA ADS] [CrossRef] [Google Scholar]
  113. Nardini, E., Reeves, J. N., Gofford, J., et al. 2015, Science, 347, 860 [Google Scholar]
  114. Nardini, E., Lusso, E., & Bisogni, S. 2019, MNRAS, 482, L134 [Google Scholar]
  115. Ostriker, J. P., Choi, E., Ciotti, L., Novak, G. S., & Proga, D. 2010, ApJ, 722, 642 [NASA ADS] [CrossRef] [Google Scholar]
  116. Parker, M. L., Pinto, C., Fabian, A. C., et al. 2017, Nature, 543, 83 [Google Scholar]
  117. Parker, M. L., Matzeu, G. A., Guainazzi, M., et al. 2018, MNRAS, 480, 2365 [NASA ADS] [CrossRef] [Google Scholar]
  118. Parker, M. L., Longinotti, A. L., Schartel, N., et al. 2019, MNRAS, 490, 683 [NASA ADS] [CrossRef] [Google Scholar]
  119. Parker, M. L., Alston, W. N., Igo, Z., & Fabian, A. C. 2020, MNRAS, 492, 1363 [NASA ADS] [CrossRef] [Google Scholar]
  120. Perna, M., Lanzuisi, G., Brusa, M., Mignoli, M., & Cresci, G. 2017, A&A, 603, A99 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  121. Petrucci, P. O., Paltani, S., Malzac, J., et al. 2013, A&A, 549, A73 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  122. Petrucci, P.-O., Ursini, F., De Rosa, A., et al. 2018, A&A, 611, A59 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  123. Piconcelli, E., Jimenez-Bailón, E., Guainazzi, M., et al. 2004, MNRAS, 351, 161 [Google Scholar]
  124. Piconcelli, E., Jimenez-Bailón, E., Guainazzi, M., et al. 2005, A&A, 432, 15 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  125. Pinto, C., Alston, W., Parker, M. L., et al. 2018, MNRAS, 476, 1021 [NASA ADS] [CrossRef] [Google Scholar]
  126. Porquet, D., Kaastra, J. S., Page, K. L., et al. 2004a, A&A, 413, 913 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  127. Porquet, D., Reeves, J. N., Uttley, P., & Turner, T. J. 2004b, A&A, 427, 101 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  128. Porquet, D., Reeves, J. N., Grosso, N., Braito, V., & Lobban, A. 2021, A&A, 654, A89 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  129. Pounds, K. A., & Reeves, J. N. 2009, MNRAS, 397, 249 [NASA ADS] [CrossRef] [Google Scholar]
  130. Proga, D., Stone, J. M., & Kallman, T. R. 2000, ApJ, 543, 686 [Google Scholar]
  131. Protassov, R., van Dyk, D. A., Connors, A., Kashyap, V. L., & Siemiginowska, A. 2002, ApJ, 571, 545 [NASA ADS] [CrossRef] [Google Scholar]
  132. Reeves, J. N., & Braito, V. 2019, ApJ, 884, 80 [NASA ADS] [CrossRef] [Google Scholar]
  133. Reeves, J. N., O’Brien, P. T., & Ward, M. J. 2003, ApJ, 593, L65 [Google Scholar]
  134. Reeves, J. N., Porquet, D., Braito, V., et al. 2013, ApJ, 776, 99 [NASA ADS] [CrossRef] [Google Scholar]
  135. Reeves, J. N., Braito, V., Nardini, E., et al. 2016, ApJ, 824, 20 [NASA ADS] [CrossRef] [Google Scholar]
  136. Reeves, J. N., Braito, V., Nardini, E., et al. 2018a, ApJ, 854, L8 [NASA ADS] [CrossRef] [Google Scholar]
  137. Reeves, J. N., Lobban, A., & Pounds, K. A. 2018b, ApJ, 854, 28 [Google Scholar]
  138. Reeves, J. N., Braito, V., Chartas, G., et al. 2020, ApJ, 895, 37 [Google Scholar]
  139. Reynolds, C. S. 1997, MNRAS, 286, 513 [NASA ADS] [CrossRef] [Google Scholar]
  140. Richards, G. T., Kruczek, N. E., Gallagher, S. C., et al. 2011, AJ, 141, 167 [Google Scholar]
  141. Ross, R. R., & Fabian, A. C. 2005, MNRAS, 358, 211 [CrossRef] [Google Scholar]
  142. Różańska, A., Malzac, J., Belmont, R., Czerny, B., & Petrucci, P. O. 2015, A&A, 580, A77 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  143. Saez, C., & Chartas, G. 2011, ApJ, 737, 91 [NASA ADS] [CrossRef] [Google Scholar]
  144. Sako, M., Kahn, S. M., Paerels, F., & Liedahl, D. A. 2000, ApJ, 543, L115 [Google Scholar]
  145. Salpeter, E. E. 1964, ApJ, 140, 796 [NASA ADS] [CrossRef] [Google Scholar]
  146. Schmidt, M., & Green, R. F. 1983, ApJ, 269, 352 [NASA ADS] [CrossRef] [Google Scholar]
  147. Serafinelli, R., Tombesi, F., Vagnetti, F., et al. 2019, A&A, 627, A121 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  148. Sim, S. A., Long, K. S., Miller, L., & Turner, T. J. 2008, MNRAS, 388, 611 [CrossRef] [Google Scholar]
  149. Sim, S. A., Miller, L., Long, K. S., Turner, T. J., & Reeves, J. N. 2010, MNRAS, 404, 1369 [NASA ADS] [Google Scholar]
  150. Sirressi, M., Cicone, C., Severgnini, P., et al. 2019, MNRAS, 489, 1927 [NASA ADS] [CrossRef] [Google Scholar]
  151. Sądowski, A., & Gaspari, M. 2017, MNRAS, 468, 1398 [CrossRef] [Google Scholar]
  152. Smith, R. K., Brickhouse, N. S., Liedahl, D. A., & Raymond, J. C. 2001, ApJ, 556, L91 [Google Scholar]
  153. Steenbrugge, K. C., Kaastra, J. S., de Vries, C. P., & Edelson, R. 2003, A&A, 402, 477 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  154. Steenbrugge, K. C., Kaastra, J. S., Crenshaw, D. M., et al. 2005, A&A, 434, 569 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  155. Strüder, L., Briel, U., Dennerl, K., et al. 2001, A&A, 365, L18 [Google Scholar]
  156. Tashiro, M., Maejima, H., Toda, K., et al. 2020, SPIE Conf. Ser., 11444, 1144422 [Google Scholar]
  157. Temi, P., Gaspari, M., Brighenti, F., et al. 2022, ApJ, 928, 150 [NASA ADS] [CrossRef] [Google Scholar]
  158. Tombesi, F., Cappi, M., Reeves, J. N., et al. 2010, A&A, 521, A57 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  159. Tombesi, F., Cappi, M., Reeves, J. N., et al. 2011, ApJ, 742, 44 [Google Scholar]
  160. Tombesi, F., Cappi, M., Reeves, J. N., & Braito, V. 2012, MNRAS, 422, L1 [Google Scholar]
  161. Tombesi, F., Cappi, M., Reeves, J. N., et al. 2013, MNRAS, 430, 1102 [Google Scholar]
  162. Tombesi, F., Meléndez, M., Veilleux, S., et al. 2015, Nature, 519, 436 [NASA ADS] [CrossRef] [Google Scholar]
  163. Torrey, P., Hopkins, P. F., Faucher-Giguère, C.-A., et al. 2020, MNRAS, 497, 5292 [NASA ADS] [CrossRef] [Google Scholar]
  164. Tozzi, G., Cresci, G., Marasco, A., et al. 2021, A&A, 648, A99 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  165. Turner, M. J. L., Abbey, A., Arnaud, M., et al. 2001, A&A, 365, L27 [CrossRef] [EDP Sciences] [Google Scholar]
  166. Ursini, F., Petrucci, P. O., Bianchi, S., et al. 2020, A&A, 634, A92 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  167. Vaughan, S., Edelson, R., Warwick, R. S., & Uttley, P. 2003, MNRAS, 345, 1271 [Google Scholar]
  168. Vietri, G., Misawa, T., Piconcelli, E., et al. 2022, A&A, 668, A87 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  169. Vignali, C., Iwasawa, K., Comastri, A., et al. 2015, A&A, 583, A141 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  170. Walton, D. J., Nardini, E., Fabian, A. C., Gallo, L. C., & Reis, R. C. 2013, MNRAS, 428, 2901 [NASA ADS] [CrossRef] [Google Scholar]
  171. Waters, T., Proga, D., & Dannen, R. 2021, ApJ, 914, 62 [NASA ADS] [CrossRef] [Google Scholar]
  172. Wilkes, B. J., & Elvis, M. 1987, ApJ, 323, 243 [Google Scholar]
  173. Wilkins, D. R., & Fabian, A. C. 2012, MNRAS, 424, 1284 [NASA ADS] [CrossRef] [Google Scholar]
  174. Wilms, J., Allen, A., & McCray, R. 2000, ApJ, 542, 914 [Google Scholar]
  175. Wittor, D., & Gaspari, M. 2020, MNRAS, 498, 4983 [NASA ADS] [CrossRef] [Google Scholar]
  176. Woods, D. T., Klein, R. I., Castor, J. I., McKee, C. F., & Bell, J. B. 1996, ApJ, 461, 767 [NASA ADS] [CrossRef] [Google Scholar]
  177. Xie, Y., Li, A., & Hao, L. 2017, ApJS, 228, 6 [Google Scholar]
  178. Xu, Y., García, J. A., Walton, D. J., et al. 2021, ApJ, 913, 13 [NASA ADS] [CrossRef] [Google Scholar]
  179. Yang, H. Y. K., Gaspari, M., & Marlow, C. 2019, ApJ, 871, 6 [CrossRef] [Google Scholar]
  180. Zanchettin, M. V., Feruglio, C., Bischetti, M., et al. 2021, A&A, 655, A25 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  181. Zubovas, K., & King, A. 2012, ApJ, 745, L34 [Google Scholar]
  182. Zubovas, K., & King, A. 2016, MNRAS, 462, 4055 [NASA ADS] [CrossRef] [Google Scholar]
  183. Zubovas, K., & Nardini, E. 2020, MNRAS, 498, 3633 [Google Scholar]

Appendix A: Signal to noise optimisation method

By following the Piconcelli et al. (2004) optimisation method, we maximised the level of background that can be tolerated resulting into an increase of the S/N. This is carried out by testing, through an iterative process, different extraction source radii and for each radius the level of the background, defined as max background, is derived together with the corresponding S/N.

For observations that are not affected by background flares, the background level remains stable along the exposure and, consequently, there is no need to filter-out any particular time intervals. In this case the max background level is stable and independent from the extraction region radius, and the highest S/N is achieved at the maximum allowed extraction radius, which, in our case, is r = 40 arcsec. Differently, for observations affected by strong background flares, the dimension of the extraction regions regulates the dominance of the source signal over the background and thus, the max background level changes as a function of the extraction radius itself. For those cases, the S/N depends on the interplay between the relative source dominance – which tends to diminish as the source extraction radius increases, leading to a lower acceptable max background – and the overall source counts, which tend to increase as the extraction radius increases. This interplay, at the end, defines a couple of values for the extraction radius and the max background for which we have the maximum S/N for our data. Between the extreme cases, we have observations for which there are short and/or weak background flares. In these cases we can appreciate some changes of the max background, but the best extraction region has the dimensions of the maximum allowed radius. It is finally worth noting how the different effective areas among MOS and pn drive different instrument responses to background flares and, thus, different S/N and max background curves.

Appendix B: Summary of SUBWAYS observations

In Table B.1 we summarise the individual observation details from the 22 SUBWAYS targets such as their ID, net exposure time and the corresponding total number of counts for each EPIC detector. We also tabulated the maximum S/N obtained from our optimisation procedure. We discarded all the observations with a total count ≲1500 in the EPIC-pn.

Table B.1.

Optimised EPIC data reduction:4–10 keV band

Appendix C: Spectral binning

In this analysis we we were interested in searching narrow and strong absorption features in the EPIC spectra. After maximising the S/N, an additional and crucial step was to carefully include the appropriate choice of an optimal spectral binning in order to avoid the loss of information in the search of weak features like UFOs. We applied the four spectral binning methods most used in the literature (listed below) and subsequently cross-checked their results: grpmin1 - We over-sampled the EPIC-pn and EPIC-MOS resolution by imposing that each energy channel contains a minimum of 1 count through the SPECGROUP task within SAS (statistics: 𝒞stat). This resulted in a binning resolution of ΔE = 5 eV and 15 eV per energy bin throughout the 0.3–10 keV energy band for pn and MOS, respectively, for all the observations. SN5 - The data were binned to ensure a significance of at least 5σ per energy channel with SPECGROUP (statistics: χ2). For the pn, such binning resulted in a binning resolution of ΔE = 5 eV at 1 keV and ΔE ∼ 20 eV at 6.4 keV, whilst ΔE = 15 eV and ΔE ∼ 60 eV for MOS. OS3grp20 - This binning (also obtained with SPECGROUP) is commonly adopted in the literature and corresponds to an oversampling approximately 3× of the instrumental full width half maximum. Additionally, the data are grouped to a minimum of 20 counts per bin in order to use χ2 statistics. This binning prescription is a safe approach especially in cases where the level of significance of the measured signal is not known a priory. Such binning yielded a resolution of ΔE ∼ 40 eV, ∼45 eV at 1 keV and ΔE ∼ 75 eV, 80 eV at 6.4 keV in EPIC-pn and -MOS, respectively. Kaastra–Bleeker (𝒦ℬ) – The Kaastra & Bleeker (2016) optimal binning option in FTGROUPPHA task within the HEASOFT package (statistics: 𝒞stat) in order to maximise the signal in the spectra. In this sophisticated method, a variable binning scheme is followed so that each bin resolution matches the CCD full width half maximum energy resolution of the EPIC detectors (or at least is not smaller than 1/3). Depending on the spectra in question, the 𝒦ℬ binning produced a EPIC-pn data resolution ranging between ΔE ∼ 50–70 eV and ΔE ∼ 100–150 eV at 1 keV and 6.4 keV and a EPIC-MOS resolution between ΔE ∼ 40–60 eV and ΔE ∼ 100 − 140 eV at 1 and 6.4 keV.

Appendix D: Broadband continuum modelling

In this work, despite being mainly focused on the iron K band, we carefully modelled the 0.3 − 10 keV continuum with the simplest phenomenological solution (see Equation 1). The summary of the best-fitting continuum parameters and the overall statistics, including for each EPIC detector, are tabulated in Table D.1. In Figure D.1 we show the individual plots for each SUBWAYS observation as per in Figure 3. It is important to note that the final errors for the best-fit continuum models are recalculated and propagated to the final continuum plus iron K emission/absorption lines model.

thumbnail Fig. D.1.

Broadband fitting and Fe K residuals of the XMM-Newton observations in the SUBWAYS sample. The EPIC-pn (black), MOS 1 (red), and MOS 2 (green) spectra, the corresponding background spectra (shaded area), and the best-fit model (solid red) are plotted. The individual model components are: absorbed power-law (cyan), black-body low- kT (blue) and high-kT (magenta), soft X-ray Gaussian line (dark green), and scattered power-law (orange). EPIC-MOS are visually binned to 10σ for clarity.

thumbnail Fig. D.1.

continued.

thumbnail Fig. D.1.

continued.

Table D.1.

Summary of the phenomenological final best-fitting continuum model parameters of each observation of the SUBWAYS sample

Appendix E: Blind-line search results

As in Figure 4, we shown the residual (top) and blind-scan search contours (bottom) for the remaining 39 in Appendix E.1.

thumbnail Fig. E.1.

Blind line search results, as in Figure 4, for the SUBWAYS sample with the corresponding residuals of the EPIC-pn spectrum in the rest-frame energy between 5 and 10 keV.

thumbnail Fig. E.1.

continued.

thumbnail Fig. E.1.

continued.

All Tables

Table 1.

Target properties of the large SUBWAYS campaign.

Table 2.

Gaussian emission and absorption parameters in the Fe K band corresponding to a total of 41 observations.

Table 3.

XSTAR parameters for the Fe K absorption features detected in the SUBWAYS sample with Pℳ𝒞 ≳ 95%.

Table B.1.

Optimised EPIC data reduction:4–10 keV band

Table D.1.

Summary of the phenomenological final best-fitting continuum model parameters of each observation of the SUBWAYS sample

All Figures

thumbnail Fig. 1.

Luminosity (upper panel) and rest-frame 4–10 keV counts (lower panel) plotted against redshift for the objects in the SUBWAYS sample and the comparison samples (T10; C21), as labelled. In the lower panel we also mark the sources of the 3XMM sample, used to select the SUBWAYS targets, with small empty circles.

In the text
thumbnail Fig. 2.

Black hole mass and Eddington ratio distributions of our targets (see Table 1) and the comparison samples. The colour scheme for the samples is the same as in Fig. 1.

In the text
thumbnail Fig. 3.

Broadband EPIC data and best-fit continuum of PG 0052+251 between the rest-frame energy of 0.3–10 keV. Top: background-subtracted XMM-Newton spectra (EPIC-pn in black, EPIC-MOS 1 in red, EPIC-MOS 2 in green) and their corresponding background levels (shaded areas). The broadband best-fit continuum model (solid red) is shown alongside the individual model components: absorbed power-law (cyan), low-temperature (blue), and high-temperature blackbody (magenta). Bottom: corresponding residuals of the data compared to the best-fitting model.

In the text
thumbnail Fig. 4.

Plot of the residuals and the corresponding closed confidence contours plot from the blind scan of the type 1 AGN LBQS 1338−0038. Top: data/model ratio plot for the EPIC-pn spectrum (MOS 1 and MOS 2 are not included for clarity), showing the residuals in the 5–10 keV band. A broad emission and absorption profiles are seen at ∼ 6.6 keV and ∼ 8 keV, respectively, suggesting a P Cygni-like profile. The vertical lines indicate the Fe Kα laboratory transition at Elab = 6.40 keV (lime green) and Fe XXV and Fe XXVI at Elab = 6.70 keV and 6.97 keV, respectively (magenta). The outer to inner closed contours correspond to a Δ𝒞 significance of (−2.3 = 68%), (−4.61 = 90%), (−9.21 = 99%), and (−13.82 = 99.9%) relative to the best-fitting continuum. Bottom: corresponding blind-search contours, showing the evidence of strong emission profiles at high confidence level (see the colour bar on the right).

In the text
thumbnail Fig. 5.

XMM-Newton EPIC spectra and the corresponding best-fit models focused on the Fe K band. Top: unfolded EPIC-pn (black), EPIC-MOS 1 (dashed red), and EPIC-MOS 2 (dashed green) spectra, between 5–10 keV, of the eight observations where the Fe K absorption line was detected at Pℳ𝒞 ≳ 95%. The spectra were firstly unfolded against a power law of Γ = 2, and their corresponding best-fitting model (solid red) was overlaid afterwards. Middle: corresponding EPIC data counts and best-fitting model. Bottom: data/model ratio. The presumed absorption features at ∼ 9.2 keV and ∼ 9.5 keV present in, for example, 2MASSJ105144+3539 and 2MASS J165315+2349, respectively, are simply not significant enough to be considered as detections (see Figs. 4 and E.1), and hence they were not included in the best-fitting models.

In the text
thumbnail Fig. 6.

Histogram comparing the photon index (Γ) measured in SUBWAYS (green; this work, with a mean value of ; vertical line) with those measured in an X-ray-selected sample (CAIXA sample, 150 sources with ; Bianchi et al. 2009), the PGQSO sample (Piconcelli et al. 2005, 40 sources ), and local AGNs (T10; G13 with and , respectively).

In the text
thumbnail Fig. 7.

Distributions and mean outflow velocity values, indicated by the vertical dashed lines, as measured in: this work (SUBWAYS) ; T10 ; G13 ; Igo20 ; and C21 .

In the text
thumbnail Fig. 8.

Outflow velocity of the ionised absorber for our SUBWAYS targets (red filled stars) plotted against their bolometric luminosity. Also plotted are the same quantities as derived in the analysis of the low-z AGN samples of Tombesi et al. (2012, red empty squares) and G13 (blue empty triangles) and in the high-z AGN sample of C21 (black filled circles). The power-law least-squares fit to the combined samples (a total of 49 individual objects and 63 observations) is shown with a solid line. The shaded area represents the uncertainty of the slopes of our fits to the data. Here the outflow velocity is simply described as the absolute value of vout.

In the text
thumbnail Fig. D.1.

Broadband fitting and Fe K residuals of the XMM-Newton observations in the SUBWAYS sample. The EPIC-pn (black), MOS 1 (red), and MOS 2 (green) spectra, the corresponding background spectra (shaded area), and the best-fit model (solid red) are plotted. The individual model components are: absorbed power-law (cyan), black-body low- kT (blue) and high-kT (magenta), soft X-ray Gaussian line (dark green), and scattered power-law (orange). EPIC-MOS are visually binned to 10σ for clarity.

In the text
thumbnail Fig. E.1.

Blind line search results, as in Figure 4, for the SUBWAYS sample with the corresponding residuals of the EPIC-pn spectrum in the rest-frame energy between 5 and 10 keV.

In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.