Redshifted Iron Emission and Absorption Lines in the Chandra X-Ray Spectrum of Centaurus A

Cen A hosts the closest active galactic nucleus to the Milky Way, which makes it an ideal target for investigating the dynamical processes in the vicinity of accreting supermassive black holes. In this paper, we present 14 Chandra HETGS spectra of the nucleus of Cen A that were observed throughout 2022. We compared them with each other, and contrasted them against the two previous Chandra HETGS spectra from 2001. This enabled an investigation into the spectral changes occurring on timescales of months and 21 years. All Chandra spectra could be well fitted by an absorbed power law with a strong and narrow Fe Kα line, a leaked power-law feature at low energies, and Si and S Kα lines that could not be associated with the central engine. The flux of the continuum varied by a factor of 2.74 ± 0.05 over the course of the observations, whereas the Fe line only varied by 18.8% ± 8.8%. The photon index increased over 21 years, and the hydrogen column density varied significantly within a few months as well. The Fe Kα line was found at a lower energy than expected from the Cen A redshift, amounting to an excess velocity of 326−94+84kms−1 relative to Cen A. We investigated warped accretion disks, bulk motion, and outflows as possible explanations of this shift. The spectra also featured ionized absorption lines from Fe xxv and Fe xxvi, describing a variable inflow.


INTRODUCTION
The distance to the galaxy Centaurus A (commonly abbreviated as Cen A, and also known as NGC 5128) is merely 3.8 ± 0.1 Mpc (Harris et al. 2010).It hosts the nearest active galactic nucleus (AGN) of type Seyfert 2 (Beckmann et al. 2011), which is powered by a su-permassive black hole (SMBH) with a mass of 5.5 ± 3.0 × 10 7 M ⊙ (Cappellari et al. 2009;Koss et al. 2022).Bowyer et al. (1970) detected X-ray emission from Cen A for the first time, and it has subsequently played a crucial role in the developing scientific understanding of AGNs.
AGN X-ray spectra are commonly described by an absorbed power law, with a strong Fe Kα emission line, and reflection features such as the Compton hump.Other features include a soft excess, and additional, but weaker emission or absorption lines, which can trace the composition, ionization, and dynamics of the system.
To better understand the processes at work in the central engine of an AGN, we decided to investigate how X-ray spectral features vary over the course of months and decades in the AGN of Cen A. AGNs feature large flux variability on these timescales, which is mostly associated with changes in the accretion rate, the absorbing column density (Risaliti et al. 2002), or the power law slope (Connolly et al. 2016).Some AGNs have strongly variable column densities, which change between Compton-thin and Compton-thick regimes, and are known as changing-look AGNs (e.g.Ricci et al. 2016).Correlation between variations in photon index and AGN luminosity have been observed in several AGNs (Fanali et al. 2013).For bright AGNs, there is a positive correlation between the two parameters.However, low luminosity AGNs, with 2 − 10 keV luminosities below 10 −3 L Edd were found to exhibit the opposite correlation (Gu & Cao 2009;Yang et al. 2015).The parameter L Edd refers to the Eddington luminosity.
The variability of X-ray spectral lines in AGNs is, however, less well established.The Fe K band generally shows a lesser degree of variability than other components of the spectrum (Markowitz et al. 2003).However, it has been possible to detect reverberation lags between the continuum and the Fe Kα line (Zoghbi et al. 2012).
The shape of spectral lines can further be used to investigate the structure and kinematics of the accretion disk.For instance, the shape, and width of the line are indicative of the inner radius of the emitting region.The centroid energy of a spectral line can also reflect bulk motion properties.For example, Doppler shifted emission and absorption lines can indicate the presence of outflows from the disk (Marziani et al. 2017;Waters et al. 2021;Mizumoto et al. 2021).
The X-ray spectrum of Cen A has been observed, and studied over five decades with many different telescopes.Over this interval, the main spectral shape has not been observed to strongly vary, despite significant luminosity variability (Rothschild et al. 2011).In this entire interval, the power law photon index has been found at Γ ≈ 1.8 (Culhane 1978;Grandi et al. 2003;Rothschild et al. 2006Rothschild et al. , 2011;;Fürst et al. 2016).Rothschild et al. (2011) analysed RXTE observations in the interval from 1996 to 2009, and found a consistent Γ = 1.822 ± 0.004.However, some spectra were also observed to be shallower than this.For instance, Mushotzky et al. (1978) found Γ = 1.66 ± 0.03, and Baity et al. (1981) observed it to vary from 1.68 ± 0.03 to 1.62 ± 0.03 over the course of six months.
The power law component extends unbroken to high energies, with a consistent slope.However, there is disagreement regarding the energy at which a break to a steeper power law occurs.Baity et al. (1981) detected a break to Γ ≈ 2.0 ± 0.2 at an energy of E break = 140 keV, using broadband observations from HEAO 1.In contrast, Kinzer et al. (1995) found a break between 300 − 700 keV with CGRO.A break of ≈ 180 keV was found by Miyazaki et al. (1996) with Ginga.Grandi et al. (2003) fitted Beppo-SAX data, and estimated a folding energy of E fold ≈ 600 keV.Rothschild et al. (2006Rothschild et al. ( , 2011) ) found a lower limit for the break in the power law of E break > 2 MeV.A NuSTAR spectrum provided a lower limit of: E fold > 1 MeV (Fürst et al. 2016).
The γ-ray emission from Cen A has also been detected, and studied with Fermi and H.E.S.S (Aharonian et al. 2009;Abdo et al. 2010;H. E. S. S. Collaboration et al. 2018).The spectrum was observed to follow a steepening power law, with Γ = 2.52 ± 0.13 stat ± 0.20 sys .No significant variability was detected over the course of eight years of observation.
The Fe Kα line was prominently observed in all Xray spectra since it was first detected by Mushotzky et al. (1978).There is disagreement over whether the line varies significantly.Rothschild et al. (2006Rothschild et al. ( , 2011) ) found the flux of the line to have a consistent value of 4.55 ± 0.14 × 10 −4 photons cm −2 s −1 .In contrast, Fukazawa et al. (2011) detected a 20 − 30% variation, and Andonie et al. (2022) found it to vary by a factor of 10.
Most spectral analyses did not detect any significant reflection features like the Compton hump.Rothschild et al. (2011) found an upper limit on the reflection fraction of R < 0.005.Fürst et al. (2016) described R < 0.01, and Markowitz et al. (2007) found that the spectra were best fit with R = 0.This is contrasted by Fukazawa et al. (2011), who detected a significant Compton hump, with a reflection fraction of R = 0.19.Evans et al. (2004) analysed two Chandra-HETGS spectra of Cen A that were observed in 2001.Their results are mostly consistent with the results of other spectral analyses of Cen A, except that they fitted a comparatively shallow power law slope of Γ = 1.64 ± 0.07.They detected excess X-ray emission at ≈ 2 keV, which they fitted by including a second power law with a dif-ferent absorption, and a photon index of Γ = 2 in their spectral model.Si and S Kα lines were also detected at 1.74 keV and 2.30 keV, respectively.They further discussed that the 20 eV width of the Fe Kα line indicates that it originates from a cold, neutral medium far from the SMBH.Andonie et al. (2022) analysed archival non-grating Chandra ACIS spectra of Cen A. They also estimated the radius of the region emitting the Fe Kα line (0.10 ± 0.05 pc), as well as the dust sublimation radius (0.04 ± 0.02 pc).
The redshift of Cen A had been independently, and consistently measured by different groups, using different methods.Graham (1978) first measured a heliocentric redshift for the entire galaxy of 1.825 ± 0.017 × 10 −3 from optical emission and absorption lines.Wilkinson et al. (1983), andSkrutskie et al. (2006) later measured systemic velocities corresponding to redshifts of z = 1.73 × 10 −3 , and z = 1.826 ± 0.017 × 10 −3 , respectively.Hui et al. (1995) and Walsh et al. (2015) studied the kinematics of planetary nebulae in Cen A, and found z = 1.805 ± 0.023 × 10 −3 , and z = 1.798 × 10 −3 .Woodley et al. (2007) investigated the globular clusters in Cen A, and measured a systemic velocity corresponding to a redshift of 1.821 ± 0.023 × 10 −3 .The average of the redshift measurements with known uncertainties is 1.819 ± 0.010 × 10 −3 .We will henceforth be using this value for the Cen A redshift.
This paper is structured as follows.Section 2 describes the observations that were analysed in this paper, as well as the data reduction, and methodology that was used.The spectral analysis utilizing two different models is described in Section 3. Section 4 discusses the results of this analysis, and investigates different interpretations of them.Finally, section 5 summarises and concludes the paper.

OBSERVATIONS AND DATA ANALYSIS
We performed 14 observations of Cen A with the Chandra X-ray Observatory (Chandra; Weisskopf et al. 2000), using the Advanced CCD Imaging Spectrometer (ACIS; Garmire et al. 2003) optimized for spectroscopy (ACIS-S), and the High Energy Transmission Grating Spectrometer (HETGS; Canizares et al. 2005), from January to September 2022.The observations were performed with a reduced sub-array size, to further reduce pileup of the bright central source.
These results were compared with the two previous Chandra-HETGS observations of Cen A that were performed in 2001, which have already been described by Evans et al. (2004).These observations also used the ACIS-S, but with the full array size.properties of all the 16 Chandra observations analysed in this paper.The set of Chandra datasets used can be found in the Chandra Data Collection (CDC) 167.
We did not find any evidence that the different observing modes affected any of the results discussed in the following sections.We selected consistent energy ranges to minimise the impact of the different array sizes on the spectral analysis, as will be discussed in Section 3.
For all of these observations, we generated type-2, first order Chandra HETGS spectra1 , using CIAO version 4.14.0, and HEASOFT version 6.30.1.First, we identified the position of the zeroth order image, and created region files for the grating source and background spectrum sky boundaries, by running tg detect with the default parameters, followed by tg create mask, with an HETG width factor of 18. Next, we assigned grating events to spectral orders using tg resolve events, with a pixel randomization half-width of 0.5.We created +1 and −1 grating order type II PHA spectral files for the source and background using tgextract, with the default parameters.We generated RMF and ARF files for the spectra with mkgrmf and fullgarf, using the default parameters.Finally, we combined the positive and negative orders of individual, or groups of observations using combine grating spectra.We kept the HEG and MEG spectra separate.
Fig. 1 shows the Swift/BAT light curve of Cen A, with overplotted Chandra fluxes at the times of the observations.Cen A was brighter in 2001 than in 2022, but it had been even brighter during the intervening interval.
We used XSPEC (Arnaud 1996) version 12.12.1 to fit all Chandra spectra.We assumed solar abundances, as described by Wilms et al. (2000), and the cross-sections defined by Verner et al. (1996).The spectra were not rebinned prior to fitting, but were subsequently rebinned for visual clarity in the following figures.The best fits were found by minimising the C-statistic (Cash 1979).1.The list of Chandra observations of Cen A that we used.T denotes the total exposure time of each observation, and F2−10 refers to the absorbed 2−10 keV HEG flux, and is listed in units of 10 −10 erg cm −2 s.The parameter C refers to the sum of the total number of source counts of the MEG from 1.65 − 3.5 keV, and the HEG from 3.5 − 10.0 keV.The horizontal lines distinguish between the 2001, early 2022, and late 2022 groups of observations.

SPECTRAL ANALYSIS
Initially, we fitted the spectra of individual observations.As they have a limited sensitivity, we fitted them with a comparatively simple phenomenological model that only described the three most important features of the spectrum; the power law, the Fe Kα line, and the Hydrogen absorption.In XSPEC notation, the model we used is: constant*ztbabs*(powerlaw+gauss).
The HETGS MEG and HEG spectra were fitted jointly for each observation.The constant component was used to account for slight normalization differences between them, with a fixed value of 1.0 for the HEG spectrum.The normalizations of the HEG and MEG spectra differed by at most 7% in individual observations.This agrees well with previous results that found approximately 8% difference between the spectral normalizations of HEG and MEG 2 .The ztbabs component describes the total absorption, featuring contributions from Cen A and the Milky Way.Since we expect most of the absorption to take place in Cen A, we set the redshift in ztbabs to the value for the host galaxy.The component gauss is used to describe the Fe Kα line, on top of the powerlaw continuum.Other emission or absorption lines were not resolved in the spectra of most of the individual observations.
Individual spectra lacked the sensitivity to constrain various spectral parameters, investigate weaker features, and fit more complex physical models.We found that the main difference between the spectral fits of individual observations was the normalization of the power law.This caused most of the variation in the 2 − 10 keV flux shown in Table 1.Most other spectral parameters had consistent values for spectra of individual observations.Furthermore, we did not detect any indication of significant variation in the spectra obtained within a few weeks or even months of each other.
Therefore, to better investigate the Cen A spectra, we merged the data from all observations into one of three groups, corresponding to the two observations in 2001, the first five observations in 2022, and the subsequent nine observations in the same year.The observations in 2022 were divided into two groups to investigate changes occurring on timescales of a few months.The selection of observations to include in the early and late 2022 grouped spectra was based on the close proximity in time of the final nine observations, as well as the lower flux detected in most of the earlier five observations (see Fig. 1).
To investigate the differences occurring between 2001 and 2022 as best as possible, we also created a spectrum that combined all observations from 2022.Finally, to obtain the best constraints on some spectral properties that were found to be consistent over this 21-year interval, we also analyzed the spectrum created by grouping all data together.We will subsequently refer to these five groups of merged spectra as the 2001, early 2022, late 2022, 2022, and total spectra.The total exposure time of these five groups of observations is: 98.36 ks, 109.30ks, 170.03 ks, 279.33 ks, and 377.69 ks, respectively.The total number of counts from 1.65 − 3.5 keV in the MEG, and 3.5 − 10.0 keV in the HEG is: 85771, 47536, 98392, 145928, and 231699, respectively.The three non-overlapping groups of observations are indicated via different colors in Fig. 1.Their merged spectra are shown in Fig. 2, indicating their qualitatively similar spectral shapes.This is essential for fitting the total spectrum.
The background spectrum is mostly flat across the energy range for both the MEG and HEG.It is consistently at least one order of magnitude below the source spectrum between 2−10 keV.Even at the lowest energy we investigated, of 1.65 keV, the background spectrum is still fainter than the source spectrum by a factor of at least 3.The part of the spectrum below 2.5 keV is brighter than expected from extrapolating the higher energy spectral shape.This feature was first described by Turner et al. (1997), and studied in greater detail in the 2001 Chandra observations by Evans et al. (2004).To accurately describe it, we included a leaked power law component that is only weakly absorbed.This is still a nuclear emission which is either the result of a leaky absorber or emission from the innermost part of the jet.Therefore, all of the following spectral fits include the additional terms constant*tbabs*powerlaw.The constant was set to a low initial value, in order to only describe the slight discrepancies observed at low energies, rather than the main spectral shape.The Hydrogen column density in tbabs was set equal to the weighted average of the Milky Way absorption in the direction of Cen A, of N H = 2.35 × 10 20 cm −2 (HI4PI Collaboration et al. 2016).Finally, the slope and normalization of the powerlaw component were set equal to those of the power law describing the main part of the spectrum.
Nevertheless, the MEG spectrum is too faint to accurately describe below about 1.65 keV.At lower energies, the source is also comparable to the background level.Therefore, we only investigated the spectra at greater energies.The smaller size of the sub-array used for the 2022 observations limits the HEG energy range to above about 2.6 keV3 .Furthermore, the +1 and −1 arms of all the HEG spectra significantly deviated at about 3.3 keV, with the appearance of an apparent emission line in the -1 arm.This was caused by a large drop in the response efficiency, possibly due to a chip gap.Above 3.5 keV, the HEG spectra have a greater sensitivity than the MEG spectra, but the two are consistent with each other, when correcting for their slightly different normalizations.Therefore, we selected an energy range of 1.65 − 3.5 keV for the co-added MEG spectra, and 3.5 − 10.0 keV for the co-added HEG spectra.
The low energy part of the spectra also exhibit several peaks away from the continuum shape.Evans et al. (2004) detected Si and S Kα lines, which have rest frame energies of 1.740 keV, and 2.307 keV, respectively.In the merged spectrum of all observations, we saw other features that might be interpreted as emission lines.
In order to determine the statistical significance of these lines, we fitted a segment of the total spectrum in the energy range around each line with both ztbabs*powerlaw, and ztbabs*(powerlaw+gauss).We calculated the Bayesian Information Criterion (BIC; Schwarz 1978) of the two fits, and only selected lines as statistically significant if the addition of the gauss component resulted in a lower BIC.Of all the deviations from the spectrum that we observed, only the Si and S Kα lines satisfied this condition, so we will subsequently only describe them.
We fitted the five groups of Chandra spectra with two main types of models, that each include the features described above.Figs. 3, 4, as well as Figs. 8 and 9 in Appendix A show the best fits to the five spectra, using models A and B.

Model A
Model A is a phenomenological model describing the spectrum as a set of absorbed power laws, absorption features, and emission lines.
The parameter constant 1 describes slight normalization differences between the HEG and MEG spectra.The Milky Way absorption is parameterized by tbabs, with a fixed column density of N H = 2.35 × 10 20 cm −2 (HI4PI Collaboration et al. 2016).The Cen A absorption in the line of sight is described by ztbabs, with a redshift of 1.819 × 10 −3 , and a column density that is free to vary.The diskline model is a physically accurate description of an emission line profile in an accretion disk, which we used to fit the Fe Kα line.The two gauss components represent the emission lines of Si, and S Kα.The parameters of components with identical names were linked.Components with different numbers describe different features, so were not linked.Table 2 lists the best fitting parameters for model A applied to the five groups of spectra, as well as their 2 − 10 keV fluxes.The results are also depicted in Fig. 5.
The merged 2001 spectrum was the brightest, and featured a lower Hydrogen column density and power law index than both the early and late spectra from 2022.The 2 − 10 keV flux dropped by a factor of 1.89 ± 0.01 from 2001 to early 2022.Meanwhile, the Hydrogen column density increased by 1.59 +0.07 −0.08 × 10 22 cm −2 from 2001 to late 2022.The fits also found a slight difference of 0.46 +0.09 −0.10 × 10 22 cm −2 between the column density for the early and late 2022 spectra.The photon index of the power law also increased by 0.165 ± 0.003 from 2001 to late 2022.Even though the increase in the Hydrogen column density and the power law index also contributed to the reduction in flux, that was predominantly caused by a decrease in the power law normalization.
There is a degeneracy between these two parameters, so we sought to determine whether the observed varia-tion in the photon index could instead be attributed to variations in the Hydrogen column density, and other fit parameters.For this purpose, we fitted all five spectra again with model A, but with the photon index set to the fixed value of Γ = 1.815.That is the photon index found by Fürst et al. (2016) for a NuSTAR spectrum, which also agrees well with the results of several other X-ray spectral analyses of Cen A (Culhane 1978;Grandi et al. 2003;Rothschild et al. 2006Rothschild et al. , 2011)).
Comparing the results of the two spectral fits, we found that the three merged spectra that were generated from the 2022 observations all had a lower BIC when the photon index was fixed at Γ = 1.815.The reason for this, is that fits with a free Γ already found it to be close to 1.815.However, the best fit photon indices for the late 2022, and entire 2022 spectra with a free Γ are inconsistent with a value of 1.815, within 1σ errors.This indicates that the errors in the parameter may be underestimated.
In contrast, the BIC increased by 35.36 for the 2001 spectrum when freezing the photon index at Γ = 1.815.This demonstrates that the best fit value for it, of 1.646± 0.002 is indeed inconsistent with Γ = 1.815.Therefore, the photon index did significantly change from 2001 to 2022.The variation of N H and Γ between 2001 and 2022 cause the best fit values for these two parameters in the total spectrum to be unreliable.
The leaked power law component has a variable strength, but nevertheless remains close to c 2 ≈ 2.8 × 10 −3 in all spectra.It can vary on short timescales, as it was found to decrease from 3.06 ± 0.27 × 10 −3 to 2.54 +0.18  −0.16 ×10 −3 from early to late 2022.This may, however, be due to a degeneracy between the leaked power law strength and the Hydrogen column density.
As Fig. 2 shows, the Fe Kα line is narrow, which means that either the inclination of the disk is low, or its inner radius is large, or both.The inclination of the AGN is not known.Although Neumayer et al. (2007) found a mean inclination of the warped gas disk of ≈ 34 • , inconsistent values have been found for the inclination of the jet.For instance, Dufour et al. (1979) found an inclination of 72 ± 3 • , and Skibo et al. (1994) measured it to be 61 ± 5 • .However, Hardcastle et al. (2003) argued for an inclination of ≈ 15 • , and Müller et al. (2014) found it to be 12 − 45 • .
A too small inclination is at odds with the identification of Cen A as a Seyfert 2 galaxy (Beckmann et al. 2011), given the AGN unification model (Antonucci 1993).Furthermore, the galaxy has a high inclination, and both the jet and counterjet from the AGN are visible, even though the counterjet appears noticeably weaker.The parameter NH describes the Hydrogen column density.The power law component is parameterized by the index (Γ) and its normalization (NPL).The diskline model describing the Fe Kα line has an energy of E l , an emissivity power law index of q, an inner accretion disk radius of Rin, and a normalization of NL.The flux of the line is denoted as F l , and its equivalent width is EWL.The two other emission lines are described by Gaussian functions with a standard deviation of σ.The redshift, z, describes the shift of the centroid of the emission line relative to its rest frame energy.The strength of the leaked component of the power law is described by the constant, c2.F2−10 denotes the absorbed 2 − 10 keV flux of the spectrum.The Cash statistic (C), and its corresponding BIC are listed at the bottom of the table.For all of these fits, there are 1713 bins, and 1699 degrees of freedom.
We investigated whether Chandra spectra could distinguish between different inclinations, as the shape of the emission line described by diskline is inclination dependent.For this purpose, we jointly fitted the spectra of all individual observations simultaneously with model A. The inclination in each of the fits was fixed to a particular value between 10 • and 90 • , in steps of 10 • .We allowed the inner radius and emissivity to vary.In all cases, comparably good spectral fits were found, so we concluded that these Chandra spectra were insensitive to different inclinations.
An inclination of 60 • was assumed in previous studies of the Cen A X-ray spectrum (Fukazawa et al. 2011;Fürst et al. 2016), as it is equally likely to find a higher or a lower inclination, given a uniform, isotropic distribution of angles.As we were unable to constrain the inclination, we will also assume it to be 60 • .This choice of inclination requires a large inner and outer radius of the disk for an accurate description of the shape of the Fe Kα line.The fits can only constrain the outer radius to within an order of magnitude, so we set it to a value of 10 6 r g , where r g = GM/c 2 is the gravitational radius.The best fit values found for the inner radius depend on the choice of inclination for the fit.At a lower inclination, comparable fits are found with smaller inner and outer disk radii.The errors quoted for the inner radii do not incorporate the uncertainty of the inclination of the system.The redshift of the Fe Kα line is unaffected by the inclination selected for the fits.
We allowed the emissivity index, q, to vary freely.The resulting fits were significantly better than those obtained by freezing it at a value of q = −3.
We compared the measured centroid energies of the three fluorescent emission lines with their rest frame energies.The Kα 1 and Kα 2 lines are so close in energy, that they merge to form a single emission line in the observed spectra.The laboratory-measured rest frame energies are found by calculating the weighted average of the Kα 1 and Kα 2 lines, using a 2:1 ratio of intensities.The resulting Fe, Si, and S Kα rest frame energies are 6.3996796 ± 0.0000074, 1.739788 ± 0.000017, and 2.307490 ± 0.000026 keV, respectively4 .The redshifts associated with the difference between the measured centroid energy, and these rest frame energies have also been listed in Table 2.
The Fe Kα line is consistently found at a higher redshift that is inconsistent with that of the galaxy as a whole, as is shown in Fig. 6.The redshifts found in the 2022 spectra are larger than the one from the 2001 spec-trum, but they are nevertheless all still consistent with each other, within 1σ errors.The best fit parameters of the Fe Kα line remained mostly consistent between 2001 and 2022.This means that the total spectrum provides the best estimate of its properties.The 1, 2, and 3σ errors of the Fe Kα line energy for the total spectrum correspond to redshifts of: 2.95 +0.28 −0.31 × 10 −3 , ±0.60 × 10 −3 , and ±0.90 × 10 −3 , respectively.This is also depicted in Fig. 6.We found that the spectra could not be well fitted by setting the centroid energy to the value expected from the Cen A redshift.Despite the decrease in fit parameters, the BIC increased by 6.8 for the total spectrum, which further demonstrates the inconsistency between the Cen A redshift, and that of the Fe Kα line.In contrast, the Si Kα line in the total spectrum was fitted with a centroid energy that is redshifted by 1.68 +0.43  −0.19 × 10 −3 relative to its rest frame energy.All the measured reshifts of this line are consistent with this value, and also with the Cen A redshift.However, the redshifts of the Si Kα line are inconsistent with those found for Fe Kα.The width of the Si Kα line remained constant, within errors, throughout all observations.It increased in amplitude, from a normalization of 33.9 ± 4.7 × 10 −4 photons cm −2 s −1 in 2001, to 62.0 +9.7 −9.0 × 10 −4 photons cm −2 s −1 in late 2022.Of the three fluorescent emission lines, the S Kα line is the least well constrained in the spectra.Its centroid energy was fit with a wide range of different redshifts, some of which were not consistent with each other.The total spectrum was best fit with a Gaussian function that was redshifted by z = 1.40 +0.71  −0.92 × 10 −3 .This value is consistent with the Cen A redshift, the redshift of the Si Kα line, but inconsistent with the redshift of the Fe Kα line.There is also significant variation in the best fit width of the S Kα line.However, it is unclear if this is caused by a variation in the line, or instead reflects the complex shape of the spectrum at these energies.The amplitude remained consistent within errors.
We investigated if the Si Kα and S Kα lines could be self-consistently described by an AGN fluorescent emission line spectrum.Regardless of the ionization degree, it was not possible to get an accurate fit to the spectral shape that included these lines, especially not when also including the Fe Kα line.Furthermore, the Si and S lines are too bright compared to the expected ratio of their fluxes relative to Fe (Rahin & Behar 2020).Rather than the expected 0.104 : 0.069 : 1.0 ratio of the Si to S to Fe fluxes, we found ratios of 5.38 ± 0.56 : 0.506 ± 0.095 : 1.0 for the total spectrum.Therefore, we conclude that the Si Kα and S Kα lines in these spectra are not produced by the central engine around the SMBH.This is further supported by the Si and S lines having redshifts that are inconsistent with those measured for Fe Kα.In contrast, the two lines are consistent with the Cen A redshift, which indicates an accurate energy calibration of the Chandra instruments.Therefore, the excess redshift of the Fe Kα line is unlikely to have been caused by an error in the energy calibration.
This could also explain the variation in the unabsorbed Si Kα flux, which correlated with the Hydrogen column density.It might have been caused by the wrong assumption that the line is absorbed by the same column density as the rest of the spectrum.When fitting the Si line without the ztbabs absorption component, its flux was found to remain consistent across all spectra.Using a different absorption for the Si and S lines affects their amplitude, but the measured width remains consistent within errors.The lines are shifted to slightly higher energies, which slightly reduces their measured redshifts, but does not affect any of the above discussion of their properties, as they are still consistent with the Cen A redshift.
The normalization of the Fe Kα line was found to vary slightly between 2001 and 2022.This corresponds to a flux decrease of 18.8 ± 8.8%.In contrast, the continuum flux varied significantly more, by 47.2 ± 0.2%.As a result of this, the equivalent width of the line is largest when the continuum flux is lowest, so in the early 2022 grouped spectrum.The normalization of the Fe Kα line remained stable between early and late 2022, indicating that it varies more slowly, and possibly to a lesser degree than the continuum flux.
We also fitted spectral model A, but with the diskline component replaced with gauss, a Gaussian line profile.The results of these fits are very similar to those shown in Table 2.In particular, the redshifts for the Fe Kα line were identical, within errors, and were also all inconsistent with the Cen A redshift.In these fits, we found the width of the line, as parameterized through the standard deviation of the Gaussian to increase from σ = 18.8 +4.6  −2.8 eV in 2001, to σ = 28.9+2.9 −2.3 eV in 2022.This corresponds to a velocity dispersion for the half width at half maximum of between 1.04 +0.25 −0.16 ×10 6 m s −1 and 1.59 +0.16 −0.13 ×10 6 m s −1 .This is broader than the energy resolution of the HETG, and can therefore not be purely attributed to it.The shift, and varying width of the Fe Kα line can be seen in Fig. 7, which zooms into the 6.15 − 7.15 keV interval of the spectra.It also shows the asymmetry of the line, which is expected.
The variable width of the Fe Kα line is described by a variable emissivity, when fit with diskline.The emissivity power law index decreased from −1.96 ± 0.15 in 2001, to −2.40 +0.14 −0.09 in 2022.The inner radius of the accretion disk was found to be consistent within errors from 2001 to 2022, possibly due to a degeneracy with the emissivity.
Motivated by the energy shift of the Fe Kα line, we also investigated whether the redshift used by ztbabs differs from the Cen A value.In so doing, we investigated to what extent the absorption edges were shifted in energy.Allowing this redshift to vary freely only slightly reduced the C statistic, but always increased the BIC.Although the best fit redshifts of ztbabs were slightly larger than the Cen A redshift, they had large errors, which made them consistent with it.Therefore, we kept the ztbabs redshift frozen at a value of 1.819×10 −3 in the best fit results shown in Tables 2 and 3, as well as Fig. 5.

Model B
In the interval between the Fe Kα line and the Fe edge, we detected ionized absorption features corresponding to Fe XXV and Fe XXVI (see Fig. 7).To model these absorption lines self-consistently, we used the "xs-tar2xspec" functionality within the XSTAR suite (Kallman & McCray 1982;Bautista & Kallman 2001) to create a table model of photoionized absorption spectra.
The input spectrum was assumed to be typical of an AGN, composed of a T = 25, 000 K blackbody and a Γ = 1.7 power law.The power law component was bent to zero flux below 0.3 keV to avoid infinite flux at low energy.Solar elemental abundances relative to Hydrogen and a gas turbulent velocity of v turb = 300 km s −1 were also assumed.The grid includes 6400 individual XSTAR realizations, with 100 grid points in the ionization range 1 ≤ log ξ ≤ 6 and 64 grid points in the column density range 1.0×10 21 cm −1 ≤ N H ≤ 6.0×10 23 cm −2 .XSPEC is able to interpolate between these models to derive the overall best fit model.The relatively high resolution of the grid ensures that important gas effects tied to ionization are not lost owing to coarse sampling.In the following, we will refer to this multiplicative model as xstar abs.
We also replaced the powerlaw and diskline components of model A with comparable MYTorus (Murphy & Yaqoob 2009) components.
The zeroth order power law continuum in model B is described by zpowerlw.
The scattered continuum is parameterized by constant s *mytorusS, and the fluorescent Fe Kα and Fe Kβ lines are described by constant l *rdblur*mytorusL.The component mytorusL includes the Compton shoulder, and therefore allows us to test its impact on the measured excess redshift of the Fe Kα line.
Therefore, model B is described as: constant 1 *(tbabs*ztbabs*(xstar abs*zpowerlaw+ constant s *mytorusS+constant l *rdblur*mytorusL+ gauss+gauss)+constant 2 *tbabs*zpowerlaw), in XSPEC notation.The parameters of the three MY-Torus models were linked, and the photon index, redshift, and normalization were set equal to the values in the zpowerlaw model.As in model A, the two gauss components represent the Si and S Kα lines, and constant 2 *tbabs*zpowerlaw describes the leaked power law.Components of this model with identical names had linked parameters, but different subscripts differentiate components with independent parameters.We again assumed an inclination of 60 • .
Table 3, and Fig. 5 show the best fit results when fitting the five spectra with model B. Identical parameters had mostly the same values, within errors, between models A and B. The best fit parameters of the Si and S Kα lines are consistent between the two models, and have not been repeated in Table 3.
In model B, we assumed that the ionized absorber, xstar abs, is located closer to the central engine than the origin of the reflection and fluorescence spectrum, and therefore only acts on the coronal spectrum, zpowerlaw*mytorusZ.To account for velocity differ- ences and emission and absorption at different radii, we allowed the redshift of the ionized absorber to vary.
The Fe XXV absorption line could not be constrained in the early 2022 spectrum (see Fig. 7), so the best fit identified a higher ionization degree for it than for the other spectra.Therefore, the ionization degree and ionized column density of the early 2022 spectrum might be overestimated.The ionization degrees of all other spectra are consistent, within errors.Due to significant variation in the properties of the ionized absorber from 2001 to 2022, its best fit parameters in the fit to the total spectrum are unreliable.
The redshifts of the zpowerlaw component, which are also used by all MYTorus components, are lower than the corresponding redshifts found for the Fe Kα line with diskline in model A. This may be the consequence of the use of this redshift to describe many spectral features, not just the Fe Kα line.Therefore, the redshifts found in model A are still the most reliable for estimating the shift of the Fe Kα line itself.In the fits with model B, the redshifts for the 2001 and early 2022 spectra are comparatively low and high, and have strongly asymmetric uncertainties.Nevertheless, the redshifts found by model B are still consistent with the values found by model A, and are inconsistent with the Cen A redshifts, for all five spectra.This further supports the notion that the Fe Kα line is indeed found at a lower energy than expected from the Cen A redshift, even when including a Compton shoulder.The components of the rdblur model are consistent, yet less well constrained than the comparable parameters of the diskline model from spectral model B.
The MYTorus N H describes the column density in the equatorial plane of the torus.At an inclination of 60 • , it does not contribute to the absorption seen along the line of sight, but instead affects the fluorescent and reflection features.That is why a mytorusZ component was not included, as it has no effect at this inclination.The MYTorus Hydrogen column density appeared to vary significantly from 2001 to 2022.However, the reason for this is that it could not be well constrained for the 2001 spectra, and was likely to have been significantly overestimated, as it only has a small impact on the spectra at an inclination of 60 • .
Of particular note is that the strength of the scattering component, parameterized by the constant s factor, was found to be 0 for all spectra.We could only place an upper limit on this reflection fraction, of c s < 0.027 for the total spectrum.The variation of the constant l factor is the result of changes in other parameters, and does not reflect the small variation in the Fe Kα line flux (see Table 2).
The spectral fits using model B have a lower C statistic than the ones using model A, but have a higher BIC.Model A is the simplest possible model to describe the main spectral shape.Model B is more complex, and uses more parameters to describe finer features of the spectra.

Neutral absorption, power law
Comparing the different spectral fits, we investigated how variable individual components were over the course of months and decades.The main spectral shape remained mostly consistent throughout all observations, and the largest variation was observed in the amplitude of the power law.
The Hydrogen column density was observed to vary from early to late 2022, and featured even larger differences when compared over a 21-year interval.This result is expected, and consistent with previous findings.The ionized absorber model, xstar abs, has parameters of NI, the ionized column density, and ξ, the ionization degree of the accretion disk.The rdblur model has an emissivity power law index of q, and an inner accretion disk radius of Rin.All the MYTorus models use the same Hydrogen column density, NH2, as well as the parameters of the zpowerlaw component.The remaining parameters are described in Table 2.The parameters of the two gauss components are equivalent to those listed in Table 2.For all of these fits, there are 1713 bins, and 1694 degrees of freedom.
The photon index of the power law was found to be consistent at a value of ≈ 1.81 throughout 2022.This is also consistent with many previous spectral analyses that measured a similar photon index (Culhane 1978;Grandi et al. 2003;Rothschild et al. 2006Rothschild et al. , 2011;;Fürst et al. 2016).However, the merged 2001 spectrum was found to be significantly shallower, with Γ = 1.646±0.002,using model A. This result is in agreement with that of Evans et al. (2004), and corresponds to a similarly shallow slope as the one found by Mushotzky et al. (1978) and Baity et al. (1981).The inability to fit all Chandra spectra from 2001 and 2022 with the same photon index provides an indication that it is variable, and can become significantly shallower than it usually is, albeit on timescales of years or decades.
Furthermore, we detected an anticorrelation between the photon index and the luminosity of Cen A. The 2001 spectrum was both the hardest, with a photon index of Γ = 1.646 ± 0.002 (for model A), and the brightest, with an unabsorbed 2 − 10 keV luminosity of 1.40 ± 0.26 × 10 −4 L Edd .In contrast, the 2022 spectrum was best fit with Γ = 1.803 ± 0.002, and an unabsorbed luminosity of 9.7 ± 1.8 × 10 −5 L Edd .This anticorrelation agrees with the results of Yang et al. (2015) and Connolly et al. (2016), who found a decreasing photon index with an increasing luminosity for AGN accreting with a 2 − 10 keV luminosity between 10 −6.5 − 10 −3 L Edd .A possible explanation for this effect is that the synchrotron power law from the jet becomes stronger than that from the advection-dominated accretion flow (Yuan & Cui 2005).

Fe Kα line
Several previous studies of the Cen A X-ray spectrum concluded that the disk generating the fluorescent Fe Kα line had to have a large extent to account for the apparent stability of its flux over intervals of several years, compared with a significant variability of the continuum (Rothschild et al. 2006(Rothschild et al. , 2011;;Fürst et al. 2016).We found the Fe Kα flux to only vary slightly, by 18.8 ± 8.8% between 2001 and 2022.This is comparable to the 20 − 30% variation found by Fukazawa et al. (2011).However, it should be noted that the flux we found for the line in the total spectrum, 2.08 +0.14 −0.12 × 10 −4 photons cm −2 s −1 , is less than half as large as the consistent 4.55 ± 0.14 × 10 −4 photons cm −2 s −1 flux found by Rothschild et al. (2011) using RXTE data from 1996 to 2009.
The lack of large variation in the Fe Kα line flux is contrasted by Andonie et al. (2022), who observed a variation of about an order of magnitude between two non-grating Chandra ACIS observations of Cen A. How-ever, we re-analysed the spectra from those two observations, and measured a consistent Fe Kα line flux.
We also found the width of the Fe Kα line to vary from 2001 to 2022, but remain consistent on short time scales.The increased standard deviation of the best fit gauss model corresponds to an increased emissivity and a slightly larger, albeit still consistent inner radius, when fitting with diskline.
The centroid energy of the Fe Kα line was observed to be significantly offset from its expected energy based on the known recession velocity of Cen A. In Model A, we measured a redshift of 2.95 +0.28  −0.31 × 10 −3 for the total spectrum, compared to the Cen A redshift of 1.819 ± 0.010 × 10 −3 .This excess redshift was observed by both spectral models A and B, as well as other models we investigated, involving gauss, and pexmon components.The excess redshift also does not depend on the assumption made in the spectral fits, such as the selection of an inclination of 60 • .The Fe Kα line energy was found to only vary within the respective errors from 2001 to 2022.
AGN spectral lines between 6 and 7 keV with significant redshifts relative to the systemic velocity of the galaxy, have been previously found by Chandra in M81 * (Young et al. 2007;Shi et al. 2021).However, these concerned the Fe XXVI Lyα emission line, and also featued a blueshifted line.
The absolute energy calibration of Chandra's HETGS has a systematic error on the scale of ≈ 100 km s −15 .Doppler shifts caused by velocities of as low as ≈ 50 km s −1 have been detected (Zhang et al. 2012).Furthermore, both the Si and S Kα lines, which probably do not originate in the accretion disk, were redshifted by an amount slightly smaller, but still consistent with the Cen A redshift, and inconsistent with the Fe Kα redshift.This leads us to conclude that the excess redshift of the Fe Kα line is likely not the result of an offset of the Chandra energy calibration.
The redshift to Cen A is measured relative to a heliocentric reference frame.The motion of Chandra relative to this reference frame might slightly offset the measured energies.We assume that the Doppler shift caused by the Chandra orbit around Earth, averaged over several observations, is small compared with the shift caused by the orbit of Earth around the Sun.The magnitude of the component of the orbital velocity of Earth in the direction of Cen A, as observed in the heliocentric reference frame, is at most 25.4 km s −1 .The average orbital velocity component in the direction of Cen A, weighted by the exposure time, was calculated to be 25.3, 15.8, 4.5, 7.9, and 13.2 km s −1 , for the 2001, early, and late 2022, 2022, and total observations, respectively.Subtracting these from the measured redshifts, we find the Fe Kα line to have an average radial velocity relative to the Cen A systemic velocity, of 250 +100 −180 , 450 ± 200, 430 +120 −190 , 460 +120 −130 , and 326 +84 −94 km s −1 , respectively.Using the XSPEC error command, we found the line to be offset from the Cen A redshift by 3.62σ, equivalent to a p-value of 0.0145%, for the total spectrum, fit with model A. We found a similar significance with model B as well.
These velocity shifts are still up to one order of magnitude smaller than the component of the orbital velocity in the line of sight at the inner radius.The width of the line is far greater than the shift of its center away from the Cen A systemic velocity.
Studies of the varying kinematics in the images of infrared and radio spectral lines of the circumnuclear disk within several hundred pc of the SMBH revealed complex structures that have been explained via a warped disk model (Quillen et al. 2006;Neumayer et al. 2007;Espada et al. 2017;McCoy et al. 2017).These studies traced red-, and blueshifted regions, but did not find any clear indication of large bulk motion relative to the Cen A systemic velocity.However, McCoy et al. (2017) found two absorption complexes, one of which moved at the systemic velocity of Cen A, the other was redshifted by 60 km s −1 .
One way to interpret the excess redshift of the Fe Kα line, is if the warped structure is still present at much smaller radii.For simulations of warped accretion disks around black holes, see e.g.Ogilvie (1999); Tremaine & Davis (2014); Liska et al. (2023a,b) In that case, the excess redshift could be caused by a greater visibility of the redshifted side of the disk.The blueshifted side would occupy a smaller solid angle, and parts of it may be blocked by the warp.Abarr & Krawczynski (2021) simulated the Fe Kα line profile for a warped disk, and found that it can appear to be shifted to lower energies.
In this model, we would expect the warp to propagate around the disk, which would change the size of the excess redshift identified.At a radius of 5 × 10 3 r g , the orbital period is 19 years.Most of the Fe line originates at greater distances from the SMBH, and the warp is expected to have a longer precession than orbital period (Inoue 2012).Therefore, this model can account for the consistency of the excess redshift over 21 years, but does require it to vary sinusoidally on longer timescales.
A theoretical study by Pringle (1996) argued that AGN disks would only show warps at radii of ≥ 10 6 r g .However, in a followup study, the critical radius above which disks can develop warps was set at r crit ≈ 4 × 10 3 r g (Pringle 1997).Nevertheless, large amplitude warps were only found to develop at radii orders of magnitude larger than the critical radius.The characteristic timescale of variability of the warp was calculated to be of order 10 6 yr.Therefore, it remains unclear if significant warps can develop at the required radii to account for the observed excess redshift.
A different class of explanations for the observed excess redshift can be found by arguing that it reflects the bulk motion of the central engine relative to the center of the galaxy.This motion might be in the form of an oscillation with a period that is many times longer than the span of the observations.Assuming a constant velocity at the value found from the total spectrum, the central engine would have travelled 7.0 +1.8 −2.0 × 10 −3 pc along the line of sight over the 21-year span of observations.This would be an unusually large velocity of the SMBH relative to its host galaxy.However, even larger velocities have been inferred in a number of systems with off-center SMBHs (Menezes et al. 2014(Menezes et al. , 2016;;Shen et al. 2019;Reines et al. 2020;Chu et al. 2023).Velocities like these can be caused from the merger of two SMBHs.Asymmetries in the spin and mass of merging SMBHs result in an asymmetric gravitational wave emission, which produces a recoil of the merged SMBH (Peres 1962;Gualandris & Merritt 2008;Blecha et al. 2011Blecha et al. , 2016)), with velocities of up to 4000 km s −1 (Campanelli et al. 2007).This recoil results in a damped oscillation around the galactic center that can last for more than 1 Gyr (Gualandris & Merritt 2008).The kinematics, metallicities, and halo features of Cen A suggest that it had a major merger ≈ 2 Gyr ago (Wang et al. 2020).For a large velocity of the SMBH relative to the galaxy to be maintained for 2 Gyr would require a large initial velocity, and weak dynamical friction.
The main drawback of this interpretation is that the AGN spectral lines observed at other wavelengths do not feature a comparably large offset from the Cen A systemic velocity.Additionally, it is unusual for an SMBH to have a large velocity relative to the galaxy, but still be found at its center.
One way of maintaining a large velocity at small radii, but only minimal bulk motion at distances of hundreds of parsecs, is if the velocity shift is caused by the orbit of the SMBH around another massive body.Given the consistency of the measured excess redshift, this would have to be a large orbit, with an orbital period much longer than 21 years.Given the size of the velocity shift, the secondary body would also have to be comparably massive, so it would have to be another SMBH.A close binary SMBH in Cen A was also suggested by Cosandey (2022) to explain peculiarities in the EHT image of the galactic center of Cen A (Janssen et al. 2021).However, Cosandey (2022) argued for orbital periods in the 10 −1 − 10 1 yrs order of magnitude range, which might be too small to justify the consistency of the excess redshift observed.
A third type of explanation for this shift, is that it reflects an outflow of material from the disk.Cen A hosts prominent jets that present one avenue for an outflow, albeit at a higher velocity (Tingay et al. 1998;Snios et al. 2019).If the excess redshift of the Fe Kα line is caused by an outflow, it would require the line emission to predominantly originate from the far side of the black hole.The emission from the near side would either have to be suppressed, or be from non-outflowing material.In this scenario, we might expect the excess redshift to vary significantly over the course of a few years.It remains unclear whether such a model could account for the consistency observed over 21 years.Asymmetric AGN outflows resulting in a redshifted Fe XXVI emission line have previously been described by Young et al. (2007).However, Shi et al. (2021) later observed both redshifted and blueshifted components of the line for the same source.Similarly, the excess redshift could alternatively represent a inflow of material towards the SMBH on the closer side.It is similarly unclear if such an inflow could remain consistent over 21 years.
Another possibility is that the Fe Kα line is produced by gas illuminated by the counterjet, and accelerated by it to the measured recession velocity.This could account for the excess redshift, the equivalent width of the line, and the consistency in time.However, it has difficulty explaining the velocity dispersion observed in the line, which is up to an order of magnitude larger than the shift.This model would also require that a similar region does not exist in the direction of the jet, or that the emission from it is strongly absorbed.This could be possible if it has a high column density, thereby obscuring the emission.
Further studies of the excess redshift of the Fe Kα line, and its variability over years and decades are required to distinguish between the different potential explanations that were discussed here.A greater spectral resolution could also help to identify possibly suppressed blueshifted wings of the line, and line profiles inconsistent with a planar disk.Investigating the cause of the excess redshift could help improve our understanding of the kinematics in accretion disks around AGNs, and expand the ways in which it can be probed.
The size of the disk emitting the Fe Kα line photons can be estimated from the spectral fits of the line, under the assumption of a particular inclination.When fixing it at a value of 60 • , and setting the outer radius to be 10 6 r g , we find consistent inner radii for all grouped spectra.The total spectrum was best fit with R in = 4.8 +0.9 −1.2 ×10 3 r g .Alternatively, when equating the half width at half maximum of the Fe Kα line with the Doppler shift of a stable orbit, for an inclination of 60 • , we find radii of 4.1 +1.0 −1.6 × 10 4 r g , and 2.66 +0.43 −0.54 × 10 4 r g for the 2001, and 2022 grouped spectra, respectively.
These redshifts are significantly larger than those found for the Fe Kα line.This could be indicative of an inflow of the ionized material.Assuming that the SMBH moves with the systemic velocity of Cen A, rather than with the velocity indicated by the excess redshift of the Fe Kα line, the inflow velocities equate to: 690 ± 460 km s −1 and 3950 +260 −220 km s −1 , for the 2001, and the 2022 spectra, respectively.If this redshift is caused by an inflow, the properties reflect the variable inflow.
The mass accretion rate can be estimated from the properties of the ionized absorber.We extrapolated from the Fe ion column density to describe the total composition of the inflowing material.Furthermore, we assumed an isotropic accretion, and assumed that the inflow starts at the inner radius of the disk.However, the mass accretion rate estimated in this way is six orders of magnitude too large for the measured luminosities.There is further inconsistency with this association due to the system being brighter at a time when a lower ionized absorber column density and inflow velocity was measured.This indicates that this description might be too simplistic.The redshift of the absorption lines may derive a component from the orbital motion, if the blueshifted components are blocked from view.
The spectral fits did not require the addition of a broad Fe line at 6.8 keV, as suggested by Grandi et al. (2003).There was no indication of a break in the power law.The best fit reflection strength, not including any fluorescent features, was 0 in all spectra, which agrees with the results of Markowitz et al. (2007); Rothschild et al. (2011);Fürst et al. (2016), but contrasts those of Fukazawa et al. (2011).It is unclear why these spectra lack reflection features, as the optical depth should be sufficient to produce these.

XRISM
The Resolve instrument on the X-Ray Imaging and Spectroscopy Mission (XRISM ; XRISM Science Team 2020) will be able to determine the properties of the Cen A spectrum with far greater sensitivity than is possible with previous X-ray spectrometers.In particular, it will enable an investigation into the exact properties of the Compton shoulder of the Fe Kα line.We simulated a 100 ks XRISM spectrum based on the best fit to the total Chandra spectrum, using model B. Due to its higher energy resolution, and greater effective area, it will be able to constrain the redshift of the Fe Kα line, and the ionized absorber up to one order of magnitude better than was possible when combining 378 ks of Chandra HETGS exposure.Furthermore, it will be possible to constrain the inclination of the disk to within 10 • , by fitting the exact shape of the Fe Kα line at a greater energy resolution.These spectra will also be much more sensitive to the Hydrogen column density in the torus, and the inner radius of the disk.

CONCLUSIONS
The AGN at the center of Cen A was observed by the Chandra HETGS twice in 2001, and 14 times in 2022.All spectra were well described by an absorbed power law with a strong and narrow Fe Kα line.We grouped spectra from different observations together, to analyse weaker spectral features, and investigate spectral variability on timescales of months and years.Variation in the flux on short and long timescales was predominantly caused by a variation in the amplitude of the entire spectrum.
Within these observations, the AGN was brightest in 2001, and also had the hardest power law slope and the smallest Hydrogen column density.The Hydrogen column density was found to vary on timescales of months.The power law slope varied as well from 2001 to 2022, but remained consistent throughout 2022.
There was no indication of any reflection features, or any break in the power law.To fit the part of the spectra below 2 keV required the addition of a leaked power law component, with an amplitude of about 0.3% of the main power law feature.
The Fe Kα line increased in width from 2001 to early 2022, and became dimmer by 18.8 ± 8.8%, but remained consistent throughout 2022.Si and S Kα lines were also detected, but could not be associated with the accretion disk.
The energy of the Fe Kα line was measured to be lower than expected from the Cen A redshift.This excess redshift was consistently found in different spectral models of the line profile, and in all spectra.The total spectrum was best fit with an excess velocity of 326 +84 −94 km s −1 , and is inconsistent with the Cen A redshift with a significance of 3.62σ.
We interpret this result as possibly indicative of a warped accretion disk on sub-parsec scales, which enhances the redshifted, but reduces or obscures part of the blueshifted wing of the line emitting region.We also consider the possibility of motion of the SMBH relative to the center of the galaxy, as well as an outflow or inflow from the disk.
The spectra also featured absorption lines of Fe XXV and Fe XXVI.The properties of these lines varied significantly from 2001 to 2022, with a higher ionized column density, and a significantly higher redshift in the latest observations.These result may be interpreted as a variable inflow with velocities of 690 ± 460 km s −1 and 3950 +260 −220 km s −1 in 2001 and 2022.

Figure 1 .
Figure 1.Swift/BAT light curve of Cen A with overplotted 2 − 10 keV Chandra fluxes for each of the observations listed in Table 1.The inset shows the observations obtained throughout 2022.The Swift/BAT daily light curve was rebinned by a factor of 10, for display clarity.The colors of the Chandra datapoints indicate the three distinct groups into which the spectra were merged.The error bars for the Chandra fluxes are smaller than the size of the data points.

Figure 2 .
Figure 2. 2 − 10 keV spectra of the grouped 2001, early 2022, and late 2022 spectra.This figure contains both HEG and MEG spectra, which are depicted with the same color, unlike the following spectra.It also shows the best fit to the data, using model B. The spectra have been rebinned for visual clarity.

Figure 3 .
Figure 3.The best fit spectra, and the ratio of the data to the folded model, for the grouped 2001 spectrum.The first panel shows the spectrum, and the best fit using model B. The subsequent two panels depict the residuals normalized by the folded model, for spectral models A and B. The spectra were rebinned for visual clarity.

Figure 4 .
Figure 4.The best fit spectra, and the ratio of the data to the folded model, for the grouped 2022 spectrum.The layout of the spectra is identical to that of Fig. 3.

Figure 5 .
Figure 5. Variation of the best fit parameters of models A and B between the 2001, early 2022, late 2022, 2022, and total spectra.The value of the best fit of a parameter in the total spectrum is unreliable, if it varied significantly from 2001 to 2022.

Figure 6 .
Figure 6.Redshift of the Fe Kα line, as found from the best fits using model A to the 2001, early 2022, late 2022, 2022, and total spectra.The red dashed line indicates the redshift of the systemic velocity of Cen A. The redshift for the total spectrum is shown alongside its 1σ (black), 2σ (dark grey), and 3σ (light grey) error bars.

Figure 7 .
Figure 7. Zoomed in spectra of the 6.15 − 7.15 keV energy range, featuring the Fe Kα and Fe Kβ emission lines, the Fe XXV and Fe XXVI absorption lines, as well as the Fe edge.The best fits of model B to the spectra are also depicted as continuous lines through the data.The vertical blue line describes the expected centroid energy of the Fe Kα line, based on the Cen A systemic velocity.The spectra were rebinned for display clarity.The 2022 and total spectra were shifted upwards in this figure, to distinguish them from the other spectra.