Extreme ultra-soft X-ray variability in an eROSITA observation of the Narrow-Line Seyfert 1 Galaxy 1H 0707-495

The ultra-soft Narrow-Line Seyfert 1 Galaxy 1H 0707-495 is a well-known and highly variable AGN, with a complex, steep X-ray spectrum, which has been the subject of extensive study with XMM-Newton in the past. 1H 0707-495 has been observed with eROSITA as one of the first CAL/PV observations on October 11, 2019 for about 60,000 seconds. The eROSITA light curves show significant variability in the form of a flux decrease by a factor of 58 with a 1 sigma error confidence interval between 31 and 235. This variability is primarily in the soft band, and is much less extreme in the hard band. No strong ultraviolet variability has been detected in simultaneous XMM-Newton Optical Monitor observations. The UV emission is about 10^44 erg s^-1, close to the Eddington limit. 1H 0707-495 entered the lowest hard flux state of all 20 years of XMM-Newton observations. In the eROSITA All-Sky Survey (eRASS) observations taken in April 2020 the X-ray light curve is still more variable in the ultra-soft band, but with increased soft and hard band count rates more similar to previously observed flux states. A model including relativistic reflection and a variable partial covering absorber can fit the spectra and provides a possible explanation for the extreme light curve behaviour. The absorber is probably ionised and thus more transparent to soft X-rays. This leaks soft X-rays in varying amounts, leading to large-amplitude soft X-ray variability.


Introduction
All previous and present X-ray missions have shown many narrow-line Seyfert 1 galaxies (NLS1), see Osterbrock & Pogge (1985) and Goodrich (1989), to have remarkable X-ray properties compared to Seyfert 1 galaxies with broader Balmer lines. NLS1 are generally characterised by steep soft X-ray spectra with photon indices of up to about 5 from simple power law fits. Detailed spectral modelling shows that NLS1 often have very strong soft X-ray excess components compared to their hard X-ray tails. A clear anti-correlation is found between the ROSAT spectral softness and the Hβ full-width at half-maximum intensity (FWHM) in type 1 Seyferts (Boller et al. 1996) and quasars (Laor et al. 1997). This is remarkable as the X-ray emission from most Seyfert 1 type galaxies originates predominantly from within a few to a few tens of Schwarzschild radii of their black holes, while Seyfert optical permitted lines are formed in a separate and significantly larger region. It appears that the anti-correlation between Hβ FWHM and ROSAT spectral softness is part of a more general set of relations which involve the Boroson & Green (1992) primary eigenvector, and it has been suggested that NLS1 may be those Seyfert 1s which are accreting at relatively high fractions of the Eddington rate (Tanaka et al. 2005). NLS1 often show sharp spectral cut-offs in the high energy spectrum, an open question which is still con-E-mail: bol@mpe.mpg.de troversially discussed (see Miller &Turner 2013 andRisaliti 2013). NLS1 can also show remarkably rapid, large-amplitude X-ray variability. One spectacular object, the radio-quiet, ultrasoft NLS1 IRAS 13224−3809, shows persistent giant-amplitude variability events by factors of 35-60 on timescales of just a few days, most likely due to strong relativistic effects (Boller et al. 1997). The ROSAT HRI light curve of IRAS 13224−3809 is nonlinear in character, suggesting that the X-ray emission regions on the accretion disc interact non-linearly or are affected by nonlinear flux amplification. Dramatic flux and spectral variability has also been seen in many other NLS1s, e.g. for some of the early ROSAT and ASCA publications see, Zwicky 159.034 (Brandt et al. 1995), WPVS007 (Grupe et al. 1995), 1H 0707-495 (Hayashida 1997), RE J1237+264 (Brandt et al. 1995), PHL 1092 (Forster & Halpern 1996), Mrk 766 (Leighly et al. 1996), and Ark 564 (Brandt et al. 1994).
1H 0707−495 has been observed with XMM-Newton over a time baseline of more than 20 years. In this paper we report on the eROSITA discovery of an extreme ultra-soft X-ray spectral state. The light curve is dominated by changes in the ultra-soft band, with much less pronounced variability in the hard X-ray band and no significant ultraviolet variability. We describe our data analysis in Sect. 2, discuss the light curve of the source in Sect. 3 and then perform flux resolved spectroscopy in Sect. 4, where we show that the variability can be explained by a temporally variable, ionised absorber (Sect. 5).
Article number, page 1 of 13 arXiv:2011.03307v1 [astro-ph.HE] 6 Nov 2020 A&A proofs: manuscript no. aandabol Large amplitude flux changes of about a factor > 50 are detected in the total and soft X-ray light curves, with a normalised excess value of 34.8 and 44.6 σ, respectively. The hard X-ray light curves and XMM OM light curve are much less variable, with normalised excess values of 2.1 and 1.7 σ, respectively (c.f. Sect. 3.2 and 3.3). The XMM OM light curve is shown in the fourth panel. The corresponding hardness ratios for the X-ray light curves are shown at the bottom. During the brightening the hardness ratio becomes softer and during low count rate intervals the hardness ratio is harder. Three count rate states referred to as high, medium, and low are marked with light red, yellow, and green colours (Section 2.1). The X-ray total-band light curves have a bin size of 400s; the soft and hard bands have a bin size of 600s.

Data Extraction
2.1. eROSITA eROSITA (Predehl 2012;Merloni et al. 2012;Predehl et al. 2020) is the primary instrument on the Russian Spektrum-Roentgen-Gamma (SRG) mission. After a performance and verification phase eROSITA is presently performing 4 years in scanning mode to create all-sky survey maps, superseding the ROSAT all-sky survey (Trümper 1984;Voges et al. 1999;Boller Fig. 2. eROSITA light curves in the very-soft energy bands (0.2-0.5) keV (orange) and (0.5-0.8) keV (blue). The light curves are very similar.  Fig. 1 and Tab. 1, eROSITA started the observation slightly before XMM-Newton and finished before it. Telescope modules (TM) 5, 6 and 7 were active during the entire observation whereas TM 1 and TM 2 were only activated for the last 16 ks and 10 ks of the observation, respectively, when the source was essentially static. 1H 0707−495 was also observed during the Article number, page 2 of 13 Th. Boller, et al.: Ultra-soft X-ray variability, strong relativistic reflection and changing partial covering fraction first eROSITA all-sky survey (eRASS1) between April 26th and 29th in 2020, for a total exposure of 407 s. All cameras were active and the total number of counts is ∼ 400.
Prepared event data were retrieved from the C945 version of the standard processing for eROSITA products. srctool version 1.49 from the eROSITA Science Analysis Software System (eSASS) was used to extract light curves, spectra, and the necessary auxiliary files for data analysis (Brunner et al. 2018;Brunner, H. et al 2020). For the PV observation, data products were extracted with a source extraction circle of radius 60 and a background extraction annulus with inner and outer radii of 140 and 240 , respectively, excluding nearby contaminating sources. For eRASS1 data, the background annulus radii were extended to 230 and 595 to increase the counts statistics, albeit still excluding contaminating sources.
As shown in Fig. 1, for further characterisation we divide the PV observations into five count rate selected time intervals using 6 time points t 0 . . . t 5 -the [t 0 , t 1 ] section is selected as the high count rate state, the [t 1 , t 2 ] and [t 3 , t 4 ] sections are referred to as medium count rate state, the [t 2 , t 3 ] and [t 4 , t 5 ] sections are low count rate state, where t 0 = 58767.361111, t 1 = 58767.453704, t 2 = 58767.494213, t 3 = 58767.563657, t 4 = 58767.615741, t 5 = 58768.020833(MJD). Flux resolved spectral analysis is commonly used for highly variable objects as e.g. applied by Kammoun et al. (2019) for NGC 4395. The time sequence applied for 1H 0707−495 in this paper is high-mediumlow-medium-low.

XMM-Newton
The observation of 1H 0707−495 by XMM-Newton (Jansen et al. 2001) started on October 11, 2019 and lasted for 60700 seconds until October 12 (ObsID 0853000101). Extraction of the data was performed using the XMM Science Analysis System (SAS) version 18.0.0. For EPIC-pn (Strüder et al. 2001) the source and background photons were taken from circular regions with radii of 35 and 106 , respectively. The background area was chosen on the same CCD-chip as the source, and was chosen to be empty of other sources and exclude gaps in the CCD. The same applies for data taken from MOS2 (Turner et al. 2001), where source and background regions of 17 and 55 in radius were used. MOS1 did not deliver any science products during the observation. The source signal is too weak for an extraction from the Reflection Grating Spectrometer (RGS) (den Herder et al. 2001).
The XMM-Newton Optical Monitor (OM; Mason et al. 2001) covered the entire joint observation with eleven UVW1 exposures in Image mode, ten of which were also taken in Fast mode. We processed the OM data using the tasks omichain and omfchain of SAS version 18.0.0. The standard and recommended procedure was adopted and the output products were checked following the list of known caveats and visual tests advised in the guides 1 . We compared the surface brightness of 1H 0707−495 with two sources with high proper motion in the field, taken with the same aperture and with a similar count rate. The radial emission profiles were found to be very similar, thus the source can be considered close to point-like and with minor host contamination (see also Leighly & Moore 2004), also validating the automated coincidence loss correction in omichain.

Analysis Methods
The Interactive Spectral Interpretation System (ISIS) in combination with HEAsoft version 6.27.2 was used for the analysis of spectra and light curves. Cash (1979) statistics have been used for spectral fitting, as the signal to noise ratio is insufficient for χ 2 statistics.

Light curve analysis
In the following we describe the unique X-ray properties detected in the eROSITA observations. The results obtained from the simultaneous XMM-Newton observations are also discussed. We are also able to report on the first eROSITA all-sky survey observations performed in April 2020 in this Section.

Detection of large amplitude flux changes
During the eROSITA observations 1H 0707−495 showed a dramatic flux drop in about one day (see Fig. 1). The source is brightest at the beginning of the observations, with rapid fluctuations in count rate, followed by a subsequent decline in count rate going down close to the background level. The highest count rate detected in the eROSITA (0.2-7.0) keV light curve is 1.112 ± 0.064 counts s −1 . The corresponding lowest count rate is 0.019 ± 0.014 counts s −1 . The resulting mean amplitude variability is a factor of 58, with a 1 σ error confidence interval with factors between 31 and 235. Similar large amplitude count rate changes are deduced from the EPIC-pn and XMM-MOS2 light curves, where the lowest count rate values are consistent with the corresponding background values.  Fig. 3 for motivation for these energy band selections). The soft variability appears similar to the total band variability, with count rate changes by a factor of larger than 50. In the hard band the variability amplitude is about a factor of 10, obtained from the XMM MOS2 light curve. The normalised excess variance (NEV) is a powerful and commonly used method to test whether a time series is significantly variable above a certain threshold (e.g. Nandra et al. 1997;Vaughan et al. 2003;Ponti et al. 2004). NEV values have been calculated for the total, soft and hard band eROSITA light curves based on Eq. 1 and 2 of Boller et al. (2016). Both the total and soft band eROSITA light curves are highly variable with NEV values of 34.8 and 44.6 σ, respectively. The NEV value for the hard eROSITA light curve is 2.1σ, quantifying that we see higher amplitude variability in the soft and total bands compared to the hard energy band.

Energy-dependence of the variability
We have further analysed the soft X-ray light curve in the energy bands (0.2-0.5) keV and (0.5-0.8) keV (see Fig. 2, top panel, for the eROSITA light curves). Interestingly, they appear almost identical. Both are significantly variable with NEV values of 31.3 and 21.8 σ for the (0.2-0.5) keV and (0.5-0.8) keV, respectively. Above 0.8 keV the variability abruptly declines up to the highest energies probed (c.f. Fig. A.1).
NEV values are then computed in energy-resolved bins to created NEV spectra for each detector. The results are shown in Fig. 3. Data from eROSITA are shown in red and computed using 400s light curve bins, data from EPIC-PN are shown in black and computed using 400s bins, and data from EPIC-MOS2 are shown in green using 600s bins. The larger time and energy bins are required for MOS2 given the lower number of counts. NEV values may differ slightly between instruments due to varying exposures/bin size. However, all NEV spectra reveal dramatic variability below 0.8 keV, with a striking drop off between 0.8-2.0 keV. NEV values above 2.0 keV could not be computed with eROSITA due to high background and very low variability, but for EPIC-PN and EPIC-MOS2, the variability increases slightly from 2-4 keV before dropping again from 4-8 keV.
The strong soft X-ray variability is extreme in relation to the weak hard X-ray variability and the lack of ultraviolet variability (next section). Such extreme ultra-soft and large-amplitude flux variability been detected for the first time with eROSITA observations in active galactic nuclei. Extremely large-amplitude variability has been observed in the past in objects such as IRAS 13224−3809 (Boller et al. 1997), GSN 069 (Miniutti et al. 2019) and RX J1301.9+2747 (Giustini et al. 2020). What distinguishes the discovery here is the presence of such variability in the soft X-ray band, with simultaneous observations that show the absence of such variations in the hard X-rays (above 0.8 keV).

The absence of strong UV variability -XMM-Newton OM observations
The source remained quite constant in the UV based on an NEV value of 1.7σ. This is indicated both by photometry from imaging, and by count rates from the timing light curve. The omichain photometry indicates a count rate level around ∼ 12.56 cts s −1 and an AB magnitude of ∼ 15.8. No reddening correction has been applied and the OM UWW1 data indicate the observed count rate. The omfchain light curve is shown in Fig. 1 with a bin time of 400 seconds. The light curve does not show significant variability during the ∼ 60 ks observation, in contrast to the highly variable soft X-ray light curve. Indication for one low-amplitude count rate increase in the first OM exposure is seen, but not of the order seen in soft X-rays. It is well known that in NLS1s the optical-UV emission varies less than X-rays (e.g., Ai et al. 2013)  The largest difference is 10 per cent in the UVW1. In particular, the OM rates were observed at ≈ 11 and ≈ 14 cts s −1 in those two epochs, respectively (Robertson et al. 2015), similarly to our 2019 observations. Moreover, Robertson et al. (2015) found no evidence of strong correlations between UV and X-rays on timescales of less than a week (but see Pawar et al. 2017). The flux at the effective wavelength of the UVW1 filter (i.e. 2910 Angstrom) was computed from the omichain count rates using the correction factors listed in the OM_COLORTRANS_0010.CCF calibration file. The average UVW1 flux throughout the whole observation is 1.83 × 10 −11 erg cm −2 s −1 . It is marginally higher with (1.86 ± 0.01) × 10 −11 erg cm −2 s −1 in the high flux state. For the low flux state the UVW1 flux is consistent with the average flux with (1.82 ± 0.01) × 10 −11 erg cm −2 s −1 .
In Fig. 4 we show the OM-UVW1 and (EPIC-pn) 2 keV luminosity of these two epochs (orange and green stars for the high and low count rate states, respectively), compared to other NLS1s (e.g., Gallo 2006) and to broad-line AGN (e.g. Liu et al. 2016)  From this comparison, it is clear that even the brighter state observed in our joint eROSITA/XMM-Newton observation is under-luminous in X-rays, with respect to typical NLS1s and to past 1H 0707−495 observations as well (Gallo 2006;Fabian et al. 2009). This indicates that we did observe an unusually Xray weak state of 1H 0707−495, especially when compared to other NLS1s given their UV emission. Remarkably, the UV level of 1H 0707−495 remained within comparable values for the last ∼ 20 years (e.g. Robertson et al. 2015;Done & Jin 2016, and references therein).

Comparison with 20 years of XMM-Newton observations and eROSITA all-sky survey observations
The analysis and comparison with 20 years of XMM-Newton observations from 2000 (Boller et al. 2002) and 2019 (this paper) reveals that 1H 0707−495 entered a historical low hard flux band emission, first detected in simultaneous eROSITA XMM-Newton observations (see times (depending on the location of the source in the sky, the number increasing towards high ecliptic latitudes) for ∼ 40 s every ∼ 4 hours. For 1H 0707−495, the net exposure is 407 s with 392 counts observed in total in the 0.2−7.0 keV band. To convert counts to rates, we have applied PSF-loss and vignetting corrections since the source enters the FoV in each passage at different offset angles. From the survey data we have extracted the light curves in the soft and hard energy bands. Fig. 6 shows the comparison between the eROSITA PV and eROSITA eRASS1 observations for the soft (0.2-0.8 keV) and hard (0.8-7.0 keV) energy band on a logarithmic scale. The soft light curve count rate has increased again during the eRASS1 observations, with less amplitude variability compared to the PV observations and the hard band count rate has also increased. We also report the related soft and hard band fluxes in Fig. 5.

Spectral Analysis
In the previous Section we have shown that the soft-band light curve displays extreme and significant X-ray variability while the hard-band light curve is less variable. Fig. 7 shows the full eROSITA and XMM-Newton EPIC pn spectrum of the 2019 observation. For comparison, previously observed spectra of the highest and lowest flux state from 2008 (327 ks, Fabian et al. 2009) and 2011 (80 ks, Fabian et al. 2012), respectively, are shown. As has already been seen in the overall flux evolution of 1H 0707−495 (Fig. 5), the source in 2019 was caught in a very low flux state, with a flux even lower to that observed in 2011 (Fabian et al. 2012).  The spectral shape of 1H0707−495 is characterised by a strong soft component, then an almost flat part, and then a strong drop at around 7 keV first reported by Boller et al. (2002). A number of alternative models have been discussed to explain this shape, see Fabian et al. (2012) for a discussion. These models generally explain the spectrum by a combination of relativistic reflection (e.g., Hagino et al. 2016) together with a strong soft excess, as well as superimposed absorption features caused by a strong wind (Dauser et al. 2012). In the following we use these earlier studies to guide our spectral analysis, concentrating on the cause for the spectral variability seen here. We note that other models based on in-homogeneous accretion flows (e.g. (Merloni et al. 2006) have also been proposed to explain the complex spectral and timing properties of NLS1s with near-Eddington accretion flows.  Fig. 8. eROSITA, EPIC-pn and EPIC-MOS2 spectra of the entire observation, including the best fit relativistic reflection model (upper panel, brown, solid line). The middle panel shows the residuals of the bestfitting empirical model with a blackbody used to describe the soft excess and the lower panel shows the residuals for the relativistic reflection model. The spectra have been re-binned for plotting purposes only, for visual clarity. The spectra of EPIC-MOS2 and eROSITA were re-scaled to the flux normalisation of the EPIC-pn, using the best fit detector constants.

Relativistic Reflection Model
Due to the spectral similarity to the 2011 observation (see Fig. 7), guided by Fabian et al. (2012) and in agreement with analyses of the higher flux states (e.g., Fabian et al. 2009;Zoghbi et al. 2010;Dauser et al. 2012) we describe the combined 0.5-10 keV data with a relativistic reflection model. For this analysis all spectra were optimally binned according to Kaastra & Bleeker (2016) and modelled using the Cash (1979) statistic.
Foreground absorption is accounted for using tbnew (Wilms et al. 2000) with abundances from Wilms et al. (2000), fixing the equivalent hydrogen column density to the Galactic 21 cm equivalent width of 4.02 × 10 20 cm −2 (HI4PI Collaboration et al.

2016
). The redshift to the source is set to z = 0.04057 (Jones et al. 2009).
In order to account for potential differences in gain of the data due to cross-calibration between instruments, multiplicative constants (detector constants) for eROSITA (C eROSITA ) and MOS2 (C MOS2 ) with respect to pn were introduced into the models.
The relativistic reflection is described with the relxill model (Dauser et al. 2010García et al. 2014), which calculates the relativistically smeared spectrum reflected from the innermost regions of an ionised accretion disc. relxill is based on the xillver model (García et al. 2013) for non-relativistic reflection. Based on previous results, which suggest a very compact primary source of radiation (Dauser et al. 2012;Fabian et al. 2012) we use relxill's lamp-post flavor, relxilllp, which assumes that the primary source of the X-ray radiation is compact and located above the black hole on its rotational axis. The incident radiation from this source, the so-called "corona", takes the form of a power law with an exponential cutoff fixed at E cut = 300 keV. The strength of the reflection component is parameterized by the source intrinsic reflection fraction, f refl . It is defined in the frame of the primary source as the fraction of photons emitted towards the disc compared to emitted towards the observer (see, Dauser et al. 2016, for a detailed definition). As detailed in Dauser et al. (2014a), in the case of a low source height the strong light bending effects would lead to most photons being focused on the disk and therefore easily to a reflection fraction of 10 and larger.
Applying this relativistic reflection model to the eROSITA, EPIC-pn, and EPIC-MOS spectra provides a good description of the data (with statistic/dof = 1.21). The spectra and the corresponding model are shown in Fig. 8 and the best fit parameters are listed in Table 2. We emphasise that no additional empirical black body component is necessary to achieve a good fit when applying this relativistic reflection model to the data. Adding an additional low temperature (kT ∼ 0.1keV) black body component does not improve the fit statistics. For completeness, the comparison to a simple power law plus black body model is also shown in Fig. 8.
The best-fit parameters of this model are in good agreement with previous results on relativistic reflection modelling of 1H 0707−495. Similarly to previous studies (Fabian et al. 2012(Fabian et al. , 2009(Fabian et al. , 2004Kara et al. 2015), iron is highly overabundant, with an abundance of A Fe = 10.0 +0.0 −1.5 being consistent with the upper limit allowed by the reflection model. With Γ = 2.64 +0.04 −0.08 , the recovered photon-index of the incident power law is also in agreement with these earlier studies. The spin parameter is well constrained and with a value of a = 0.9960 +0.0013 −0.0030 close to maximal spin, while the height of the primary source, 1.39 +0.023 −0.142 r g , implies a very compact X-ray source that is extremely close to the black hole. These values are also consistent with earlier studies employing the lamp post geometry (Fabian et al. 2012;Dauser et al. 2012;Kara et al. 2015). While these parameters tend to be consistent between the different earlier observations, the inclination of the accretion disc was found to vary widely, ranging from 23 •  up to 78 • (Dauser et al. 2012). The value found in the present analysis, θ = 73 • .1 +1.8 −1.6 , is at the upper end of this range. We emphasise, however, that the self-similarity of reflection spectra in the lamp-post geometry results in a degeneracy between inclination and lamp-post height h (Dauser et al. 2012), which might be the reason for the large spread of observed inclinations. Recently, Szanecki et al. (2020) apply their newly developed relativistic reflection model for an Article number, page 6 of 13 Th. Boller, et al.: Ultra-soft X-ray variability, strong relativistic reflection and changing partial covering fraction extended lamp post source and confirm the compact nature of the corona in agreement with the interpretation presented in our work.
The reflection fraction is determined to be very high with f refl = 46 +13 −10 , implying that most of the radiation emitted from the primary source is reflected on the disk and only a minor fraction is directly observed. This result is in agreement with previous observations starting with Fabian et al. (2002), all consistently finding that 1H 0707−495 is extremely reflection dominated 3 (see, e.g., Kara et al. 2015). Calculating the expected reflection fraction for such a point-like lamp post source close to a very rapidly rotating black hole leads to values of f LP refl =12-20 (see Dauser et al. 2014a). This is in rough agreement with the high values we find, but still suggests a certain difference of the primary source in 1H 0707−495 to the standard lamp post source.
Inspecting the residuals of the relativistic model in Fig. 8 in more detail reveals that the drop in flux around 6 to 7 keV is not entirely correctly modelled. We note, however, that a fast absorption by an ionised outflow as discovered by Done et al. (2007); Dauser et al. (2012) might explain the model over-predicting the flux around 7 keV. Tailoring a disc wind model to the parameters of the 1H 0707−495 system, Hagino et al. (2016) were able to partly explain this drop by ionised absorption seen under different velocities due to a wind cone emitted between 45 • and 56 • that is intercepting the line of sight. Detailed analysis of all available data by Kosec et al. (2018) seem to consolidate the existence of an ultra-fast stratified outflow in 1H 0707−495.

Spectra at high, medium, and low count rates
In order to investigate the effect of the strong flux variability during the observation, we created three flux-resolved spectra, selected based on count rate segments highlighted in Fig. 1. The specific times of the selection are given in Sect. 2. As already seen by the detailed analysis of the light curves in different bands (see Sect. 3), the majority of the flux variability is detected below 1 keV. Figure 9 compares the eROSITA and EPIC-pn spectra in the three selected count rate intervals. In the following sections, we explore the time evolution of the spectra with a partial covering model with relativistic reflection.

A Changing Partial Coverer
Considering that X-ray absorption affects the soft energies the strongest, we now check whether varying absorption can explain the large changes observed in the soft flux of 1H 0707−495. To test this hypothesis we employed the partial covering model TBpcf (Wilms et al. 2000) to act as a changing absorption component in the line of sight towards the emission region. In order to test this scenario, we fit the spectra of all count rate states simultaneously, keeping all parameters of the continuum the same, including the column density of the partial coverer, N H . The only parameters that were allowed to vary between the two observations are the ionisation parameter of the reflection model as well 3 in case a relativistic reflection model is used to describe the data; see above for alternative explanations  Dauser et al. (2016) c reflection fraction, see Dauser et al. (2014a) d iron abundance with respect to solar values (Grevesse et al. 1996) e ionization parameter, defined as ξ = 4πF/n where F is the incident flux and where n is the particle density Table 3. Best fit parameters and confidence intervals for our best fit model fitting simultaneously the data of the three flux states. The model consists of relativistic reflection in combination with a changing partial coverer. If only one value is given per row, this means it was tied between all spectra. See Table 2 for an explanation of the symbols used.

Parameter
Low (1) Medium (2)  as the covering fraction, f pc , of the partial coverer. All parameters as determined from the best fit are listed in Table 3. A decomposition of the relativistic model for each flux state is shown in Fig. 10. The best fit in this configuration yields stat/dof = 1209.3/806 = 1.50, i.e., the overall very strong variability can be explained solely by a variation of the covering fraction of the absorber and by a variation of the ionisation of the reflector. Over Article number, page 7 of 13   Fig. 9. The count rate selected eROSITA and EPIC-pn spectra extracted from time windows highlighted in Fig. 1. The underlying model is the best-fitting reflection model absorbed by a partial coverer with varying covering fraction. The corresponding parameters are listed in Table 3. The data is strongly re-binned to facilitate visual inspection. Data of MOS2 are used in the spectral fits, but omitted in this plot to provide a clearer view. The lower panels show the residuals, belonging to each flux state. The spectra of eROSITA are scaled according to the fitted detector constant to match the EPIC-pn data. The wind is directly detected from the more prominent edge at 0.8 keV in the low flux state. the course of the observation the covering fraction f pc < 0.05 for the highest flux states and then increases to f pc = 0.28 +0.16 −0.22 in the medium flux spectrum and f pc = 0.75 ± 0.06 in the low flux spectrum. We note that the time sequence of flux states is highmedium-low-medium-low. Due to the low number statistics we had to merge the two medium and the two low states in order to derive constrained covering fractions. Therefore, the covering fraction evolution needs some care in the interpretation.
For the parameters tied between the flux selected spectra, a comparison with the best fit to the combined spectrum (see Tab. 2) shows that the photon index Γ, the reflection fraction f refl , the inclination θ, the lamp-post height h, and the spin a are consistent with the results from the combined spectra. Only the iron abundance is now reduced to a more reasonable value of Z Fe = 4.8 +3.8 −1.2 , possibly implying that the very high abundances found also in the earlier observations might be due to soft variability which was ignored in the analysis. Initial fits where we kept the ionisation parameter of the reflector linked between flux selected spectra did not yield a satisfactory description of the data, with significant residuals remaining in the 1 keV band, which we attribute to emission in the Fe L band. Therefore we allowed the ionisation parameter to vary between the three count rate selected spectra, which led to a good fit to the data. Note that we would not expect most other parameters such as the spin or the inclination to vary during the observation. The only parameter which has been suggested to change (see, e.g., Kara et al. 2015) is the height of the primary source. In our case a constant height satisfactorily describes the data and a potential additional change of the height can not be  Fig. 10. Decomposition of the model for each count rate state into the radiation of the corona which can be directly observed (dotted), the radiation reflected by the disk reaching the observer (dashed), the sum of both (dashed-dotted) and this sum partially absorbed (solid). The low count rate state is shown in the upper panel, the medium one in the middle panel, and the high count rate state at the bottom. Since the covering fraction in the high regime is zero, there is no absorption of the emitted primary and reflected spectra. The galactic foreground absorption is not shown in this plot.
detected. This can also partly be attributed to the lower signal to noise in these spectra.
The ionisation parameter changes from a consistent and fairly low ionisation for the high and medium flux selected spectra (log ξ = 0.68 +0.10 −0.21 and log ξ = 0.64 +0.17 −0.24 , respectively) to a larger values of log ξ = 1.74 +0.05 −0.04 in the low flux spectrum for the high and medium state respectively.
As we discuss below, this change is likely not a physical change of the reflection component but the fit compensating for un-modelled ionisation in the absorber. In order to test a possible ionised partial coverer we used the model zxipcf (Reeves et al. 2008). Unfortunately, due to the short observations, the data is not able to constrain the ionisation of the absorber. Even freezing all relxill parameters and only allowing the normalisation, reflection fraction, and partial coverer fractions to vary freely, the ionisation in the medium flux state is completely unconstrained. In the low state it is slightly constrained to log ξ < 2. However, due to the low signal to noise, this value has to be treated with care. Longer observations would be required to analyse the ionisation of the partial coverer. The low count rate statistics does not allow us to constrain the ionisation of the absorber directly from the observations.

The extreme and varying UV to X-ray flux ratio
One important new observational result is that within less than one day the ratio of the UV to the X-ray emission shows large variations. The UV emission is rather constant with L UV ≈ 10 44 erg s −1 , similar to the values reported by (Done & Jin 2016), before applying bolometric corrections, which is close to the Eddington limit. The X-rays emission however drops in amplitude by more than a factor of 50 (c.f. Fig. 1). On time scales shorter than one day a strongly varying X-ray flux during a constant UV flux has been detected. Buisson et al. (2017) have analysed a sample of 21 active galactic nuclei using data from the Swift satellite to study the variability properties of the population in the X-ray, UV and optical band. For nine out of their 21 sources the UV is lagging the X-rays. For 1H 0707−495 the authors did not find strong correlations between the X-ray and the UV, similar to the results reported in this paper. Buisson et al. (2017) found 1H 0707−495 in a low flux state during their Swift observations and argue that in such cases the source height of the illuminating corona is low, similar to the values reported in Table 2 in this paper, which makes it difficult to detect UV-X-ray lags. In the previous subsection, we inferred that the X-ray variations are primarily due to varying covering fraction of a partial absorber. This seems not to affect the UV. This implies that these are caused by independent physical processes. 1H0707−495 is extremely under-luminous in the X-ray compared to other NLS1 and BLS1 at this epoch (c.f. Fig. 4) as well as to the 1H 0707−495 highstate flux observations from (Fabian et al. 2009). This supports the interpretation that the X-rays are suppressed in this observation, hence possibly absorbed, and hence absorption-related changes could explain the variability.

Changing partial covering fractions causing large amplitude and ultra-soft count rate variations
The most important result of the analysis is that the major source of variability observed in the spectrum can be explained by a variation of the absorber's covering fraction.
Our spectral analysis showed that the variation of the Xray spectrum is consistent with changes induced by a partial absorber of varying covering factor and constant column density in front of the X-ray emitting corona and accretion disc. As expected the covering fraction is increasing significantly with decreasing flux of the source. With N H = 12 +6 −4 × 10 22 cm −2 , the equivalent hydrogen column density of the partial coverer is consistent with that seen in typical AGN absorption events. Markowitz et al. (2014) find peak N H column densities of 4-26 ×10 22 cm −2 in the largest sample of cloud obscuration events. Studying the long-term X-ray spectral variability of a sample of 20 Compton-thin type II galaxies, Laha et al. (2020) find 11 sources that require a partial-covering, obscuring component in all or some of the observations. Not only are the N H ranges quoted in both studies fully consistent with our derived value but also the presence of a varying partial cover seems to be present in a fair fraction of AGN.
We note that there has been a controversial discussion if a partial coverer in 1H0707−495 can explain the strong 7 keV edge (e.g., Fabian et al. 2004;Gallo et al. 2004;Done et al. 2007). In our model, however, the partial coverer does not explain the 7 keV edge. While this edge is mainly modelled by relativistically smeared reflection from the accretion disk, our partial covering model describes the varying absorption in the soft X-rays. In this paper we combine relativistic reflection very close to the black hole, a few R G with partial covering occurring at larger distance up to a few hundred R G . From analyses of much longer observations (Dauser et al. 2012;Kosec et al. 2018), it is known that a strongly ionised wind is present in 1H0707−495. In the low state observation, the absorption feature around 0.8 keV is evidence that this outflow is also present in our observation (c.f. Fig. 9). The wind is not detected in the higher flux states, as the outflowing winds are strongly flux dependent as shown by e.g. Parker et al. (2017) and Reeves et al. (2018). As the existence of such an ultra fast outflow (UFO) has been brought in direct connection with the observed partial covering in other sources (e.g., PDS 456, Reeves et al. 2018), it is possible that the observed partial covering in the soft X-rays is connected to these previously detected UFOs. This absorption is likely connected or even directly caused by the UFO detected previously. The UFO will also affect the the Fe-K region around 7 keV (Kosec et al. 2018), but could not be detected in our observations due to the lower signal to noise.
The change in partial covering fractions combined with UFO features may also explain the observed shape of the NEV spectra. On short timescales, the absorber is likely driving the variability; likely attributable to small variations in ionisation and covering fraction as the material passes along the line of sight. As seen in Fig. 10, the absorber seems to affect the spectral shape between 0.3-4 keV, which explains why these energy bins have higher NEV values. In particular, most of the variability is seen below 0.8keV, in agreement with what is seen in the light curve.
The NEV spectra also reveal very little variability in the 0.8-2.0 keV and 4-8 keV bands. This may be explained by the presence of UFO features in these energy bands. The outflow may be more stable on short timescales, instead varying on longer timescales. This behaviour would suppress the variability on short timescales in these energy ranges, explaining the drops in the NEV (c.f. Sect. 5.3 for a more detailed discussion on the connection between outflowing winds and partial covering).
At the same time, however, we also measure a change of the ionisation parameter of the relativistic reflection component. We consider it unlikely that this change of ionisation is indicative of changes in the accretion disc, and rather caused by the simplified (neutral) absorber model. As discussed in Sect. 5.1, the data do not allow to constrain the ionisation of the absorber.
Due to lack of additional information, such as the ionisation of the absorber, it is not easy to estimate a distance and size of the obscuring cloud. Given the fast timescale of the putative absorption event and the strong change of the covering fraction within 20-40 ks, the absorber will probably much be closer to the X-ray source than the BLR (see Sect. 5.3). This close distance makes it very likely that it will be partly ionised.
Ionised absorbers, however, are more transparent in the soft X-rays than neutral absorbers and thus show leakage effects in the soft X-rays. The change in log ξ of the reflector seen here mainly affects the soft X-rays, and thus might mimic this effect of ionised absorption. We note that longer observations of such a partly obscured state would be necessary to constrain more detailed ionised absorption models for the partial coverer 4 .
An illustration of the changing partial coverer scenario with relativistic reflection is shown in Fig. 11. Due to gravitational light bending, the majority of the photons emitted from the corona are approximately in equal parts bent towards the black hole and onto the accretion disk (c.f. Fig. 1 and 2 of Dauser et al. A&A proofs: manuscript no. aandabol Fig. 11. Illustration of the considered scenario. Above a spinning black hole, X-rays are emitted isotropically. Due to the compact corona very close to the black hole, the majority of the photons either hit the accretion disk or fall into the black hole. From the high to the low flux state, a partial coverer is obscuring and increasing part of the emitted X-ray radiation. 2014b)). While in the high flux state we have an unobscured view onto the inner parts of the accretion disk, partially covering clouds absorb the reflected spectrum in the lower flux states with increasing covering fraction for a decreasing observed soft X-ray flux.

Partially covering absorbers and ultra-fast outflows
Outflowing winds launched from the accretion by radiation pressure or magnetic fields disc are considered as an important AGN feedback process. For radiation pressure dominated winds, outflows can reach velocities up to about 0.3 c and can drive substantial amounts of material into the interstellar medium. These winds have been discovered mainly based on XMM-Newton observations by e.g. (Pounds et al. 2003a,b,c;King & Pounds 2003;Reeves et al. 2003). Outflowing winds with such high velocities have been named as UFO by Tombesi et al. (2010) in a systematic study of bright XMM-Newton AGN.
Multiple outflow absorption lines have been detected in one of the most variable AGN IRAS 13224−3809 by (Parker et al. 2017). The authors argue that the X-ray emission from within a few gravitational radii of the black hole is ionising the disk winds up to hundreds R G . It was also shown that the outflow absorption lines are strongly flux dependent with strongest in the low flux state and weakest in the high flux state due to increasing ionisation towards higher flux values. When the ionisation becomes sufficiently high, the outflow may become "over-ionised" and may no longer be visible. Such a scenario was also discussed in Gallo et al. (2019), where absorption features were detected in the beginning of a flare in the NLS1 Mrk 335, but not in the brightest prolonged flare states.
Ionised outflowing winds have been connected to absorbing partial covering by e.g. Reeves et al. (2018) and references therein. The authors argue that the outflowing wind is inhomogeneous and complex than a simple homogeneous outflow, which is capable of partially covering the X-ray source. In this scenario the X-ray absorption depends on the ionisation state, the distance of the absorber and the covering fraction.
The two XMM-Newton observations of PDS 456 reported by Reeves et al. (2018) were taken over two consecutive satellite orbits. The authors argue that much of the spectral variability between the observations appears to be reproduced by the variability of the low-ionisation partial covering absorber, which is primarily driven by a change of the covering fraction. This appears consistent the the low flux states and and the varying covering fractions reported for PDS 456 and now for 1H 0707−495 in this paper.
Partial covering absorbers have been put into context with the ultra fast outflows and winds in several other papers. Reeves et al. (2020) provide a further exploration for the spectral shape and variability of PDS 456, noting in particular the significant differences in the soft band fit when using neutral and ionised partial covering components. There are many other works which analyse the soft-and hard-band emission and absorption features in PDS 456, concluding that an outflowing absorber can explain these features and the observed variability (e.g. Matzeu et al. 2016b,a;Parker et al. 2018).
A larger sample of Seyfert galaxies analysed in Tombesi et al. (2013) also reveals that many AGN which display UFO signatures also show evidence for warm absorption, and based on their observed properties, propose that these may actually be part of a single large-scale outflow. Simultaneous observations of absorption and outflowing components are also presented for individual sources, including Mrk 335 (Longinotti et al. 2019, but see also Gallo et al. 2019) and PG 1211+143 (Pounds et al. 2016). This lends further support to the idea that such components may be physically linked and appear simultaneously, as in the observations presented in this work.

Speculations on the partial coverer size and location
Because the absorption is only partial, we can place limits on the projected size of the absorber. For such an extreme configuration of compact corona and large black hole spin, most of the observed flux is due to reflection from within a radius of 5-10 R g around the black hole , suggesting that the absorbing structure is smaller than this scale.
From Fig. 1 we estimate that a first obscuration event is seen between times t 0 and t 3 for about 20000 seconds where the count rate decreases from the highest count rate state to the low count rate state. Between t 3 and t 4 the count rate increases again but probably with the covering fraction found in the medium flux state. A third obscuration event might be detected from t 4 until the end of the eROSITA observations, where the source is found in the lowest count rate state with the highest covering fraction.
To estimate the distance of the absorbing cloud we adopt Eq. 2 of Beuchert et al. (2017). Considering cloud number densities n H from 10 9 cm −3 to 10 10 cm −3 yields distances from 11 R G to 1100 R G for the first obscuration event, which assumes Keplerian orbits, corresponding to an orbital velocity of 0.2 c to 0.02 c. To change the covering fraction from the t 0 to t 3 from less than 0.1 to 0.73 within about 20000 seconds, the projected length of the absorber is in the range of 1.2 × 10 13 cm to 1.2 × 10 14 cm, or 0.03 to 0.3 light days. This seems reasonable but we avoid further speculation on distances and sizes of the absorber in order not to over-interpret the available data.

Summary
Large amplitude variability with a factor of more than 50 has been detected in the eROSITA light curves. The soft band (0.2-