## Abstract

We present results from a new XMM–Newton observation of the high-redshift quasar RXJ1028.6 – 0844 at a redshift of 4.276. The soft X-ray spectral flattening, as reported by a previous study with ASCA, is confirmed to be present, with, however, a reduced column density when modelled by absorption. The inferred column density for absorption intrinsic to the quasar is 2.1(+0.4−0.3) ×1022 cm−2 for cold matter, and higher for ionized gas. The spectral flattening shows remarkable similarity with that of two similar object, namely GB1428 + 4217 and PMNJ0525 − 3343. The results improve upon those obtained from a previous short-exposure observation for RXJ1028.6 – 0844 with XMM–Newton. A comparative study of the two XMM–Newton observations reveals a change in the power-law photon index from Γ≃ 1.3 to 1.5 on time-scales of about one year. A tentative excess emission feature in the rest-frame 5–10keV band is suggested, which is similar to that marginally suggested for GB1428 + 4217.

## Introduction

High-redshift quasars are powerful cosmological probes to study the evolution of massive black holes and quasar environments in the early Universe. Previous X-ray observations have suggested the presence of soft X-ray spectral flattening1 in some radio-loud quasars at redshifts z= 2–3 (Wilkes et al. 1992; Elvis et al. 1994; Cappi et al. 1997; Fiore et al. 1998; Yuan & Brinkmann 1999; Reeves & Turner 2000). This result is strengthened and extended to higher redshifts by its detection in a few extremely X-ray/radio-loud quasars at z > 4, namely RXJ1028.6 – 0844 (Yuan et al. 2000), GB1428 + 4217 (Boller et al. 2000; Fabian et al. 2001b), and PMNJ0525 − 3343 (Fabian et al. 2001a), with ROSAT, ASCA, and BeppoSAX. These object seem to have characteristics typical of blazars (Fabian et al. 1997, 1998; Zickgraf et al. 1997; Moran & Helfand 1997; Hook & McMahon 1998). The most plausible explanation for this effect is photoelectric absorption of soft X-rays by an associated medium with column densities of 1022–23 cm−2, although intrinsic spectral flattening cannot be excluded. The physical implication of this effect has been discussed extensively in the literature in terms of excess absorption (Elvis et al. 1998; Yuan et al. 2000; Fabian et al. 2001a,b) and intrinsic breaks in the X-ray spectra of blazars (Fabian et al. 2001a,b).

The contemporary X-ray observatories XMM–Newton and Chandra should be able to test these previous findings. Indeed, soft X-ray spectral flattening has been confirmed to be present in GB 1428 + 4217 and PMNJ0525 − 3343 (XMM–Newton, Worsley et al. 2004a,b), and in some other object at lower redshifts, for example PKS2126 − 0158 at z= 3.27 (XMM–Newton, Ferrero & Brinkmann 2003; BeppoSAX, Fiore et al. 2003). Tentative evidence was also found in the combined spectra of several z > 4, moderately radio-loud quasars (Chandra, Bassett et al. 2004).

RXJ1028.6 – 0844 was first detected as an X-ray source in the ROSAT All-Sky Survey (RASS), and was identified as a quasar at z= 4.276 (Zickgraf et al. 1997). It is also a radio source (PKSB1026-084) with a flux density of 220mJy at 5GHz (Otrupcek & Wright 1991) and a flat radio spectrum. Its X-ray colours in the ROSAT energy band imply a hard spectrum (Zickgraf et al. 1997). Its first X-ray spectrum, as obtained with a long ASCA observation made in 1999, flattens substantially towards soft-X-ray energies (Yuan et al. 2000); an excess (cold) absorption model required a column density of ∼2 × 1022 cm−2 for a local absorber or ∼2 × 1023 cm−2 for an absorber intrinsic to the quasar. A later short-exposure XMM–Newton observation found, however, only marginal evidence for excess absorption (Grupe et al. 2004). In this paper we report on a new XMM–Newton observation of RXJ1028.6 – 0844 with an exposure much longer than the previous observation. The measured X-ray spectrum – with substantially improved photon statistics – confirms the presence of the soft X-ray spectral flattening as detected by ASCA. The observations and data reduction are described in Section 2. We present the spectral analysis in Section 3, including a re-analysis of the previous XMM–Newton observation. Discussion of the results is given in Section 4, including comparisons with previous observations and with other similar object. Conclusions are summarized in Section 5. We adopt H0= 71kms−1Mpc−1, ΩΛ= 0.73, and Ωm= 0.27. The Galactic column density in the direction of RXJ1028.6 – 0844 is NGalH= 4.59 × 1020 cm−2 (Dickey & Lockman 1990). Errors are quoted at the 1σ level for one parameter of interest unless stated otherwise.

## Observation and Data Reduction

The quasar RXJ1028.6 – 0844 was observed with XMM–Newton on 2003 June 13 during satellite revolution 643 (observation ID 0153290101). The EPIC (European Photon Imaging Camera) MOS1, MOS2, and PN cameras were operated in the ‘primary full window’ imaging mode, and a thin filter was used to screen out optical/UV light. The observational log is shown in Table 1. The XMM–Newton Science Analysis System (SAS, v.6.0) and the most up-to-date calibrations (2004 August) were used for data reduction. Standard data reduction and screening procedures were followed. A fraction of the observation period suffered from high flaring background caused by soft protons. By inspecting the light curve of single events with energy E > 10 keV in the whole field of view, these periods were identified as having count rates higher than 1 cts−1 and 0.35 cts−1 for the PN and MOS detectors, respectively, as recommended by the XMM–Newton Science Operation Centre (SOC).

Table 1

Summary of the XMM observations and data reduction information for each EPIC detector.

Table 1

Summary of the XMM observations and data reduction information for each EPIC detector.

The quasar was detected at a sky position away from the position of its radio counterpart (Simbad data base) and 2.1 arcsec from that of its optical counterpart (Zickgraf et al. 1997). Source X-ray events were extracted from a circle of 32-arcsec radius, which corresponds to the ≃87 per cent encircled energy radius. Background events were extracted from source-free regions using a concentric annulus of 52/128 arcsec radii for the MOS detectors, and circles of 32-arcsec radius at the same CCD read-out column as the source position for the PN detector. X-ray images, light curves, and spectra were generated from the extracted, cleaned events for the source and background. Parameters for data screening and source extraction are listed in Table 1. No photon ‘pile-up’ problem was found, as expected, nor any effect of low-energy noise above 0.2keV. We used the spectral range of 0.2–10keV for PN and 0.3–10keV for MOS. The EPIC response files (rmf and arf) were generated using the source information on the detectors. The EPIC spectra were rebinned to have a minimum of 30 counts in each bin.

The source profile in the 0.3–2keV band was compared with the point spread function of the detectors (FWHM of 5 arcsec for MOS and 6 arcsec for PN) and was found to be consistent with a point-like source. There was no significant variability found during the 40-ks exposure, although a ≃10 per cent drop in count rates (averaged over ∼5ks) was marginally detected, from 0.316 ± 0.09 cts−1 at the beginning to 0.271 ± 0.08 cts−1 towards the end of the observation.

## X-ray Spectral Analysis

Owing to an increase in the surface charge loss properties of the CCDs, which degrades the energy resolution, there has been a time-dependent significant change in the low-energy redistribution properties of the MOS cameras. This effect has been taken into account in the most up-to-date MOS calibration files; however, some small systematic uncertainty may still remain at below 0.5keV (Kirsch 2004; Kirsch et al. 2004). To minimize possible biases induced in the results, we treated the MOS spectra in two ways and compare the results. The first was simply to omit the spectral range below 0.5keV; the second was to use the spectral range down to 0.3keV and introduce a systematic error of 2 per cent in the 0.3–0.5keV band (as recommended in Kirsch 2004; Kirsch et al. 2004). As seen in Table 2, these two methods yielded statistically consistent results. We thus formally quote the results obtained using the 0.3–10keV band. xspec (v.11.3) was used for spectral fitting.

Table 2

Results of X-ray spectral fits.

Table 2

Results of X-ray spectral fits.

### Soft X-ray spectral flattening

#### Power-law model with local cold absorption

We began by fitting the spectra of each detector individually with a single power-law model2 modified by neutral absorption with the column density NH as a free parameter. This model gave acceptable3 fits to the PN and the MOS1 spectra (see Table 2), but not to the MOS2 spectrum [reduced χ2= 1.4 for 53 degrees of freedom (d.o.f.), i.e. a null hypothesis probability of Pnull= 0.02 only]. Inspection of the χ2 residuals of the fit singled out one energy bin at 1.82keV (a width of 80eV), which contributed 11 out of the total χ2 of 76. The feature can be fitted with either a narrow notch feature at E= 1.83 ± 0.02 keV (a width of 49+55−19eV and a covering fraction 0.99+0.01−0.53) or a Gaussian absorption line (E= 1.83 keV) of infinitely small width. This energy corresponds to 9.7keV in the quasar's rest frame, at which no known physical absorption feature is present. On the other hand, it is coincident with the instrumental absorption feature at 1.84keV resulting from the silicon edge. Furthermore, it appears in neither the PN nor the MOS1 spectrum. Exclusion of the 1.82-keV bin reduced the χ2 by 12, and resulted in an acceptable fit (χ2= 1.2 for 52 d.o.f.), with fitted parameters in good agreement with those for MOS1 within 1σ errors. We therefore consider the disagreement between MOS2 and MOS1/PN to be due to inadequate MOS2 calibration around the silicon edge, and ignore this energy bin hereafter. Parameters (including normalizations) for the two MOS spectra were tied together to be the same in joint fitting (MOS1+2). The fitted NH of ∼1.1 ×1021 cm−2 is significantly higher than NGalH (Table 2). The photon index is now Γ≃ 1.55, typical of blazars.

A power law with fixed Galactic absorption (4.59 ×1020 cm−2) yielded an unacceptable fit and a flat photon index (Γ= 1.3–1.4, Table 2). The improvement in χ2 for the fit with freely fitted NH over that with fixed NH=NGalH is substantial: χ2 was reduced by 36 for PN and by 31 for a joint MOS1+2 fit for one additional free parameter. Applying the F-test4 (Bevington & Robinson 1992) gives a chance probability ≪0.001.

For fixed NH=NGalH, an acceptable fit was obtained only within the restricted 1–10keV range. We plot in Fig. 1 the data, the best-fit model (for a joint PN and MOS1+2 spectral fit) and its extrapolation down to the low-energy end of the detectors, and the data-to-model ratio as residuals. A systematic deficit of photons below 1keV is clearly indicated. Fig. 2 shows the confidence contours for the free-fitted total NH (Galactic plus excess) and Γ for PN and MOS1+2, respectively. Absorption in excess of NGalH is evident. It is noted that PN gives systematically lower NH values than MOS. We regard the results from MOS as more reliable than those from PN in consideration of the EPIC cross-calibration uncertainties as discussed in detail in Appendix A.

Figure 1

The spectra of the PN (circles), MOS1 (squares), and MOS2 (stars) cameras and the residuals as data-to-model ratio. The model is the best-fit power law (with Galactic absorption) to the joint PN+MOS1+MOS2 spectra within the restricted 1–10keV energy band and is extrapolated down to low energies. A systematic deviation from a power-law model with Galactic absorption towards low energies (below 1keV) is clearly indicated.

Figure 1

The spectra of the PN (circles), MOS1 (squares), and MOS2 (stars) cameras and the residuals as data-to-model ratio. The model is the best-fit power law (with Galactic absorption) to the joint PN+MOS1+MOS2 spectra within the restricted 1–10keV energy band and is extrapolated down to low energies. A systematic deviation from a power-law model with Galactic absorption towards low energies (below 1keV) is clearly indicated.

Figure 2

Confidence contours of fitted total column density and photon index for the model of a power law with local neutral absorption. The contours are at the 68, 90, and 99 per cent confidence levels, respectively, for two interesting parameters. Solid contours: the joint MOS spectra in the range 0.3–10keV; dashed contours: the PN spectrum. Also indicated by lines are the Galactic column density (dashed) and its conservative 30 per cent uncertainty range (dotted). Absorption in excess of the Galactic value is evident for such a model.

Figure 2

Confidence contours of fitted total column density and photon index for the model of a power law with local neutral absorption. The contours are at the 68, 90, and 99 per cent confidence levels, respectively, for two interesting parameters. Solid contours: the joint MOS spectra in the range 0.3–10keV; dashed contours: the PN spectrum. Also indicated by lines are the Galactic column density (dashed) and its conservative 30 per cent uncertainty range (dotted). Absorption in excess of the Galactic value is evident for such a model.

#### Absorption intrinsic to the quasar

This is the most plausible postulate in consideration of the statistical argument presented in previous work (see Introduction for references). For a neutral absorber, NH≃ 2 × 1022 cm−2 was found assuming cosmic abundances. The confidence contours for the excess NH versus Γ are shown in Fig. 3 for PN (dashed) and MOS1+2. If the metallicity of the absorber is lower than the cosmic value, which is not unexpected at such a high redshift (it could be about 10 per cent or less: e.g. Lu et al. 1996; Pettini et al. 1997; Prochaska & Wolfe 2000), the NH will be correspondingly higher than the value give here. The absorption-corrected luminosity in the quasar rest frame is 9.2 × 1046 ergs−1 in 1–10keV and 2.64 × 1047 ergs−1 in 1–50keV, respectively.

Figure 3

Column density of the intrinsic absorber versus power-law photon index as measured with MOS (solid) and PN (dashed) detectors. The absorber is assumed to be at the quasar redshift 4.276 and to have cosmic abundances. The contours correspond to confidence levels of 68, 90, and 99 per cent. The results from the current (2003) and the previous XMM observations (2002) are plotted. The variation in the spectral slope is significant, while there is no change in the absorption column density.

Figure 3

Column density of the intrinsic absorber versus power-law photon index as measured with MOS (solid) and PN (dashed) detectors. The absorber is assumed to be at the quasar redshift 4.276 and to have cosmic abundances. The contours correspond to confidence levels of 68, 90, and 99 per cent. The results from the current (2003) and the previous XMM observations (2002) are plotted. The variation in the spectral slope is significant, while there is no change in the absorption column density.

If the absorber is close enough to the central source, the gas is likely to be ionized. Indeed, the optical spectra of RXJ1028.6 – 0844 taken by Péroux et al. (2001) and Zickgraf et al. (1997) show no significant Lyman-limit absorption at 912Å. An estimate of the optical on the neutral hydrogen column density of NHi≲ 10−17 cm−2 along the line of sight to the optical–UV emission region (Yuan et al. 2000). If the X-ray absorber also intercepts the optical–UV light, at least moderate ionization of the gas is required. The lack of strong optical–UV extinction can be explained by an ionized, dust-free absorber. It is interesting to note that such an optical–UV property seems to be common with this type of object (Yuan et al. 2000; Fabian et al. 2001a,b). We tried to model the excess absorption with ionized absorption (absori in xspec). The ionization parameter (as defined in Done et al. 1992), however, could not be constrained with the current data ranging from an almost neutral to highly ionized absorber. The results of such an analysis are given in Fig. 4, where the data from the two XMM–Newton observations are combined in order to achieve a better constraint on the parameters (see Section 3.2.2). With increasing ionization state, the required column density increases from ∼2 × 1022 to ∼1 × 1023 cm−2.

Figure 4

Confidence contours (at the 68, 90, and 99 per cent levels) for the ionization parameter and column density of the ionized absorber model. The result is obtained from a joint spectral fit to the MOS spectra from the two XMM–Newton observations in 2003 and 2002 (see Section 3.2.2).

Figure 4

Confidence contours (at the 68, 90, and 99 per cent levels) for the ionization parameter and column density of the ionized absorber model. The result is obtained from a joint spectral fit to the MOS spectra from the two XMM–Newton observations in 2003 and 2002 (see Section 3.2.2).

No redshifted K-shell absorption edges from iron ions (Eedge= 7.1–9.3 keV in the rest frame from FeI to Fexxvi) are detected. Assuming the cosmic abundance of iron (3.4 × 10−5), a K-shell edge with optical depth τ∼ 0.02–0.1 is predicted from the above NH range for neutral–highly ionized iron. This is consistent with the upper limit on τ estimated from the joint PN and MOS1+2 spectra, which ranges from 0.1 to 0.2 (90 per cent level) for ions from Fei to Fexxvi (absorption cross-section from 3.8 × 10−20 to 3.3 × 10−20cm2atom−1).

#### Intrinsic spectral break

We also considered the possibility that the soft X-ray spectral flattening is an intrinsic feature. The physical implication of a break in the intrinsic X-ray spectra of blazars has been discussed in Fabian et al. (2001a,b) in the context of a cut-off in the energy distribution of electrons or of soft seed photons for Compton scattering. We modelled the spectral flattening with, as an approximation, a broken power law modified by local absorption. Acceptable fits could be obtained for both PN and MOS1+2. However, NH and the low-energy photon index Γlow-E could not be constrained because of strong coupling. We thus fixed NH=NGalH. The fitted high-energy photon indices Γhigh-E are 1.44 ± 0.04 (PN) and 1.49 ± 0.06 (MOS1+2), and the break energy Ebreak and low-energy index Γlow-E are listed in Table 2. This model gave acceptable fits, which are statistically indistinguishable from the models of power law with either local or intrinsic absorption.

### Comparison with a previous observation and spectral variability

#### A previous XMM–Newton observation

RXJ1028.6 – 0844 was previously observed with XMM–Newton in revolution 445 with a short exposure of 7ks in 2002 May (PI: S. Mathur; observation ID: 0093160701). The results, as published in Grupe et al. (2004), gave absorption NH values similar to what we find here, but a flatter spectral slope of Γ≃ 1.3. In order to achieve a self-consistent comparison of the two observations – free from the effect introduced by different versions of the evolving calibration and data-processing software – we analysed the data from that observation. The data were taken from the XMM–Newton science archive. The observation is described in Grupe et al. (2004). We used exactly the same data-screening criteria and source/background extraction regions as used for the current observation (see Section 2 and Table 1). The good exposure and source count rate are 3.7ks and 0.39cts−1 for PN (0.2–10keV), and 6.9/7.0ks and 0.12/0.13cts−1 for MOS1/2 (0.3–10keV). A comparison with Table 1 reveals that the broad-band count rates were higher in the 2002 observation than in 2003 by about 30 per cent.

The PN and MOS spectra were binned to have a minimum of 25 and 20 counts in each bin, respectively. The results of the spectral fits are in good agreement with those obtained by Grupe et al. (2004). For an absorbed power-law model, the total local absorption NH is 10.1(+2.0−1.2)× 1020 cm−2 for joint MOS1 and MOS2 (MOS1+2) spectra and 6.0(+1.3−1.0)× 1020 cm−2 for PN. Again, the fitted NH is systematically lower for PN than for MOS, as found in the 2003 observation. While the absorption NH is in good agreement between the two observations, the photon indices are not. The photon indices obtained for the 2002 observation are Γ2002= 1.30 ± 0.06 and 1.27 ± 0.05 for MOS1+2 and PN, respectively; that is, the spectrum was steeper during the observation of 2003 (Γ2003= 1.53 ± 0.03). This result can be seen clearly in Fig. 3, where excess (neutral) absorption NH is plotted versus Γ for the 2002 observation, assuming that absorption is intrinsic to the quasar at z= 4.276. The spectral steepening remains even when only the hard-band 2–10keV spectra are considered (Γ2002= 1.33+0.08−0.14 and 1.23+0.08−0.13 for MOS1+2 and PN, respectively). The Galactic absorption-corrected flux in the 1–10keV band is 1.9 × 10−12ergs−1cm−2 (averaged MOS1 and MOS2 value).

#### Joint spectral fit of the two observations

We quantified the spectral variability by fitting jointly the spectra of the two observations. We used the MOS spectra only, in consideration of possible PN calibration uncertainties (see Appendix A). For MOS1 and MOS2 spectra from the same observation, all parameters (including normalization) were tied together. The results are summarized in Table 3.

Table 3

Joint fits to the current (2003) and a previous (2002) XMM-Newton observation. Only the MOS spectra were used. The absorption NH is assumed to be the same in the two observations and the photon index is freely fitted.

Table 3

Joint fits to the current (2003) and a previous (2002) XMM-Newton observation. Only the MOS spectra were used. The absorption NH is assumed to be the same in the two observations and the photon index is freely fitted.

First, since the fitted NH of the two observations are in good agreement, we assumed that there was no variability in the absorption. Thus the NH values for the two observations were tied together in the fitting. Absorbed power-law models were used. As a test, we tied Γ20022003 together in the fitting, which resulted in a fit only marginally acceptable (χ2= 242 for 219 d.o.f.). Setting the two indices as independent parameters improved the fit significantly, reducing χ2 by 24 for one additional free parameter; the F-test [Bevington & Robinson (1992); see Footnote 4 for the argument for the validity of the F-test used here regarding the boundary condition warning discussed by Protassov et al. (2002)] gives a chance probability ≪0.001 for Γ2002 and Γ2003 being the same. The fit is good (χ2= 217 for 218 d.o.f.), indicating that the spectra of the two observations can be fitted well with two different continua (in slope and normalization) attenuated by the same amount of absorption. The fluxes in the 0.2–1keV band were comparable in the two observations, while in the 1–10keV band the flux was higher by a factor of 2 in 2002 May compared with 2003 June (Table 3). The two power-law continua, before attenuation by any absorption, cross over at ≃0.4 keV in the observer's frame, i.e. ≃2 keV in the quasar rest frame.

The combined data set improves the spectral photon statistics and gives a better constraint on the excess absorption. For a neutral absorber intrinsic to the quasar, NH= 2.1(+0.4−0.3) × 1022 cm−2, i.e. excess absorption is evident. An ionized absorber model also gives a good fit, suggesting that the ionization status is unconstrained. The confidence contours for the ionization parameter and NH are shown in Fig. 4.

Secondly, we tested whether the variation in spectral slope could result from variability in the absorption, such as from the ionization parameter. We fitted the ionized absorber model with independent ionization parameters and NH, but tied the indices together (Γ20032002). No satisfactory result could be obtained for this model; the best fit was significantly worse than that with the above variable-Γ model (Δχ2= 14). This is not surprising, as the difference in Γ arises primarily in the hard 1–10keV band, which corresponds to 5.3–53keV in the quasar rest frame. At such high energies the amount of absorption of X-ray photons is decreasing dramatically.

We also fitted a broken power law to the data, with absorption fixed at NGalH. There is a marginal indication of a higher break energy of 1.6 ± 0.2 keV for the flatter spectrum in 2002 May than for the steeper spectrum in 2003 June; however, the significance is low. The low-energy photon indices and the normalizations at 1keV (below or close to the break energies) are comparable in the 2002 and 2003 observations.

No spectral variability is detected within a 40-ks duration in the 2003 June observation with XMM–Newton.

## Discussion

### The presence of soft X-ray spectral flattening

#### XMM–Newton results

We have shown the presence of soft X-ray spectral flattening in the z= 4.276 quasar RXJ1028.6 – 0844 using an observation made with XMM–Newton. The result confirms the previous report based on ASCA data (Yuan et al. 2000). In the excess absorption scenario, the derived values of the absorption NH from our analysis are consistent with those obtained by Grupe et al. (2004) from a previous short XMM–Newton observation. In that study the authors suggested that strong excess absorption was marginal. This is not surprising given the lower signal-to-noise ratio of their data and consequently the weaker constraints on NH compared with this work, which benefits from a much longer exposure.

We note that PN tends to give systematically lower NH values than MOS. This is probably due to discrepancies in the calibration below 1keV between PN and MOS, as reported in the most recent XMM calibration status (Kirsch et al. 2004; see also XMM–Newton SOC XMM-SOC-CAL-TN-0018.5) Although a complete solution is yet to be reached, preliminary indications suggest that PN is probably the cause of the problem. A detailed discussion on this issue and a quantitative PN–MOS comparison taking the most up-to-date calibration into account is given in Appendix A. In summary, we consider the NH values derived from MOS spectra to be more reliable. Furthermore, it may also be the case that both PN and MOS yield systematically higher fluxes at low energies than XMM–Newton RGS and Chandra. If this turns out to be the case, the true column in the absorption scenario could be even higher than what is reported here.

#### Comparison with previous results

The total NH assuming local absorption is ≃(0.11 ± 0.01) × 1022 cm−2 measured from the XMM–Newton observations (joint MOS from the two observations). This value is a factor of 2–3 times smaller than the value measured from ASCA (Yuan et al. 2000). For intrinsic absorption at z= 4.286, the NH inferred by XMM–Newton (a few times 1022 cm−2) becomes about 10 times smaller than that obtained by ASCA. Intrinsic variability in the absorption cannot be ruled out. However, we speculate that a systematic difference in the instrumental calibration of the two missions might play at least a partial role. This is because a similar trend was also found for GB1428 + 4217 and PMNJ0525 − 3343 (Worsley et al. 2004a,b). It is not clear which instrument causes the difference. In the cases of GB1428 + 4217 and PMNJ0525 − 3343, the previous results before XMM–Newton were obtained by a joint fit of both the ASCA and BeppoSAX spectra (Fabian 2001a,b). This fact, together with the aforementioned XMM EPIC calibration issue, suggests that perhaps both missions, rather than merely XMM–Newton, may be the cause. Improved XMM EPIC calibration and independent investigations by other instruments (Chandra or XMM–Newton RGS) are needed to resolve this problem.

Comparing the observations of XMM–Newton in 2003 with those of ASCA in 1999, the spectral photon indices are consistent within their mutual 1σ errors; no significant flux variability (>10 per cent) is detected in the 1–10keV band.

#### Comparison with other object

We compare the spectral shape of RXJ1028.6 – 0844 with those of GB1428 + 4217 and PMNJ0525 − 3343 in the quasar's rest frame. Following Worsley et al. (2004b), we produced the data-to-model ratio for RXJ1028.6 – 0844 where the model is the best-fit power law with Galactic absorption in the restricted 1–10keV band. The result is plotted in Fig. 5, together with those for GB1428 + 4217 and PMNJ0525 − 3343 from Worsley et al. (2004b, their figs2 and 5). It should be noted that the data-to-model ratio is free from the effect of instrument response, Galactic absorption, and redshift. It can be seen that the three spectra agree remarkably well in terms of the break energy and the shape of the spectral cut-off. The NH values of intrinsic (cold) absorbers of a few times 1022 cm−2 as measured by XMM–Newton in these three object are in good agreement. The striking spectral similarity shared by these object at different redshifts argues for a real soft X-ray flattening against instrumental effect, and suggests a common nature to this phenomenon.

Figure 5

Data-to-model ratio in the rest frame for RXJ1028.6 – 0844 (stars), where the model is the best-fit power law with Galactic absorption in the restricted 1–10keV spectral range. Only the PN spectrum (2003 May observation) is plotted, which is rebinned for demonstration. Also plotted are the data-to-model ratios for GB1428 + 4217 (filled squares) and PMNJ0525 − 3343 (open circles) for comparison, taken from fig.5 in Worsley et al. (2004b). The spectra are plotted in their respective rest-frame energies.

Figure 5

Data-to-model ratio in the rest frame for RXJ1028.6 – 0844 (stars), where the model is the best-fit power law with Galactic absorption in the restricted 1–10keV spectral range. Only the PN spectrum (2003 May observation) is plotted, which is rebinned for demonstration. Also plotted are the data-to-model ratios for GB1428 + 4217 (filled squares) and PMNJ0525 − 3343 (open circles) for comparison, taken from fig.5 in Worsley et al. (2004b). The spectra are plotted in their respective rest-frame energies.

Another similarity lies in their optical–UV properties, which argue for a highly ionized, dust-free absorber model (Yuan et al. 2000; Fabian et al. 2001a,b; Worsley et al. 2004a,b). It is interesting to note that excess absorption in several z > 4, moderately radio-loud quasars, as tentatively suggested by their combined Chandra spectra, has also similar NH values of a few times 1022 cm−2 (Bassett et al. 2004).

### X-ray spectral variability

It is worth stressing that, while spectral variability is indeed typical of flat-spectrum quasars, the extremely flat spectrum in the 2–10keV band during the 2002 observation is consistent only within the 2σ uncertainty range with the limiting value of Γ≃ 1.5 for a relativistic distribution of particles emitting via synchrotron and inverse Compton (in the simplest hypothesis). If the production of such flat high-energy spectra is confirmed, revision of the widely accepted emission scenarios will be required.

Although flux variability over short time-scales is a distinctive characteristic of blazar emission, no significant variations have been detected within a single observation. It should be noted, however, that, because of the high redshift of the source, the intrinsic time-scale sampled by the observation (∼2 h) might be too short to detect significant variations. (For powerful quasars, doubling time-scales of the order of ∼several hours to a day might be more typical, for example 3C 279, Wehrle et al. 1998.)

### Excess emission around 5–10keV?

Worsley et al. (2004b) pointed out possible excess emission at energies around 5–10keV in the quasar rest-frame spectrum for GB1428 + 4217. The evidence is only marginal. Interestingly, the same spectral structure also appears in RXJ1028.6 – 0844, as can be seen in Fig. 5. The similarity of the energy position of this feature in two object at different redshifts is remarkable. If this feature is real, it may come from an additional spectral component that is peaked around 5–10keV in the rest frame. We modelled the XMM–Newton spectra taken in 2003 May by adding a steep power-law component in the above spectral models, following Worsley et al. (2004b). Both the PN and MOS1+2 spectra were fitted jointly to improve the statistics. A steep photon index is yielded for the second power law, as Γ= 2.3–4.4 for a model with local absorption and Γ= 1.9–7.2 for intrinsic absorption at z= 4.276 (90 per cent confidence range for one interesting parameter). These values are consistent with that obtained for GB1428 + 4217, Γ∼ 1.8–2.6 (Worsley et al. 2004b). The model does improve the fit in the 5–10keV (rest-frame) band, although the statistical significance is not high (Δχ2≃−7 for three additional free parameters).

If this excess emission feature proves to be real, it might be the first evidence for the presence of emission originating from bulk Comptonization on the soft photon field through which the relativistic jet propagates (Begelman & Sikora 1987). The detection of such a feature could carry key clues to the amount of (cold) leptons flowing in the jet (Sikora & Madejski 2000). The lack of its detection in the majority of object so far remains a puzzle. However, in most cases the X-ray emission might be dominated by the non-thermal emission from relativistic particles, and thus the possibility of detecting such a component could be limited to cases of particularly high jet Lorentz factors (which would shift its peak up to high energies) and/or low states/steep power law of the non-thermal relativistic component (the latter case could of course be tested, in principle). Observations with even higher signal-to-noise ratios than the present ones or stacking spectra from different sources might be a way to clarify the issue.

## Conclusions

We have presented a new X-ray spectroscopic study of the high-redshift (z= 4.276) quasar RXJ1028.6 – 0844 with XMM–Newton. The high signal-to-noise spectrum confirms the presence of the soft X-ray spectral flattening, which was reported previously with ASCA data (Yuan et al. 2000). This spectral feature can be modelled by either excess absorption of the quasar X-rays or an intrinsic break at ∼1keV in the X-ray spectra of the source. In the absorption scenario, the derived column density for a cold absorber intrinsic to the quasar is 2.1(+0.4−0.3) × 1022 cm−2. This value is comparable with those reported in two similar object, GB1428 + 4217 (z= 4.72) and PMNJ0525 − 3343 (z= 4.4) from XMM–Newton observations. The remarkable similarity in the shape of the spectral cut-off among these object at different redshifts argues against instrumental effect as an origin, but rather argues for a common nature to the soft X-ray flattening in high-z blazars. In terms of the soft X-ray spectral flattening, the results are consistent statistically with and improved upon those obtained from a previous short-exposure observation for RXJ1028.6 – 0844 with XMM–Newton (Grupe et al. 2004). A comparative study of the two XMM–Newton observations revealed a spectral steepening from Γ≃ 1.3 in 2002 to Γ≃ 1.5 in 2003, and a consequent drop in flux in the hard-energy band above ∼1keV.

The derived columns from XMM–Newton observations, however, are reduced when compared with the previous ASCA results (Yuan et al. 2000). We speculate that this might be due to systematic instrumental effect, probably inherent in both missions. Future improved XMM–Newton EPIC calibration and independent investigations by other instruments (such as Chandra) are needed to resolve this issue.

A tentative excess emission feature in the rest-frame 5–10keV band is suggested, which bares remarkable similarity to that marginally imprinted in the X-ray spectrum of GB1428 + 4217 (Worsley et al. 2004b).

## Acknowledgments

We thank Richard Saxton of the XMM–Newton calibration team for useful advice on the EPIC calibration issues. Matt Worsley is thanked for help in making the plot of Fig. 5. WY thanks Franz Bauer for comments and a careful reading of the manuscript. ACF thanks the Royal Society for their support. AC acknowledges MIUR and INAF for financial support. This research has made use of the NASA/IPAC Extragalactic Data base (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.

### Appendix

#### Appendix A: Effect of Calibration Uncertainty on the Results

In the most recently updated reports on the XMM–Newton calibration status, Kirsch et al. (2004, see also XMM–Newton SOC XMM-SOC-CAL-TN-00186) demonstrated significant discrepancies at low energies (below ∼1 keV) between the EPIC PN and MOS cameras, and the EPIC and the RGS,7 as well as overall differences in the XMM–Newton instruments and those of Chandra. While MOS and RGS agree with each other in general, PN gives higher fluxes below 0.7keV by 10–15 per cent with respect to MOS and by 20–30 per cent with respect to RGS (Kirsch et al. 2004). It appears that the problem arises mainly from the PN calibration at low energies, although it is not clear whether the redistribution matrix or the effective area is deficient. This means that PN spectra tend to give less absorption column than MOS and RGS, similar to what we find in this work (Tables 2 and 3), as well as the results obtained by Grupe et al. (2004). We thus argue that the NH values derived from MOS spectra are more reliable.

If the effect is purely due to an effective area problem of PN, we can test the PN–MOS consistency by taking into account the reported difference in the calibration. To simplify the treatment, we assumed that the PN effective area below 1keV was underestimated by a factor of f(E), which is energy-dependent. f(E) was estimated from fig.10 in Kirsch et al. (2004), in which the MOS/RGS normalization factors with respect to PN are plotted versus energy. We adopted conservative values of f(0.3keV) ∼ 1.4, f(0.6keV) ∼ 1.15, and f(1keV) ∼ 1.0. f(E) at any other energy E within 0.3–1.0keV was interpolated using a binomial function. We corrected the PN effective area (ARF) by multiplying by f(E) at the corresponding energy. The PN spectral fitting was repeated with this corrected ARF, and the results are listed in Table A1. The fitted total NH is (11.1 ± 0.8)1020 cm−2, in excellent agreement with the MOS results, while Γ remains unchanged. We therefore conclude that the NH fitted from the PN spectrum is likely to be underestimated. More quantitative and reliable estimation of NH in PN spectra must await the completion of the PN and MOS calibration at low energies (Kirsch et al. 2004).

Table A1

Absorbed power-law model fit to the PN spectrum using the corrected effective area ARF calibration.

Table A1

Absorbed power-law model fit to the PN spectrum using the corrected effective area ARF calibration.

## References

Bassett
L. C.
Brandt
W. N.
Schneider
D. P.
Vignali
C.
Chartas
G.
Garmire
G. P.
,
2004
,
AJ
,
128
,
523
Begelman
M. C.
Sikora
M.
,
1987
,
ApJ
,
322
,
650
Bevington
P. R.
Robinson
D. K.
,
1992
,
Data Reduction and Error Analysis for the Physical Sciences
,
2nd edn
.
McGraw-Hill
, New York
Boller
Th.
Fabian
A. C.
Brandt
W. N.
Freyberg
M. J.
,
2000
,
MNRAS
,
315
,
L23
Cappi
M.
Matsuoka
M.
Comastri
A.
Brinkmann
W.
Elvis
M.
Palumbo
G. G. C.
Vignali
C.
,
1997
,
ApJ
,
478
,
492
Dickey
J. M.
Lockman
F. J.
,
1990
,
ARA&A
,
28
,
215
Done
C.
Mulchaey
J. S.
Mushotzky
R. F.
Arnaud
K. A.
,
1992
,
ApJ
,
395
,
275
DOI:
Elvis
M.
Fiore
F.
Wilkes
B.
McDowell
J.
Bechtold
J.
,
1994
,
ApJ
,
422
,
60
DOI:
Elvis
M.
Fiore
F.
Giommi
P.
P.
,
1998
,
ApJ
,
492
,
91
DOI:
Fabian
A. C.
Brandt
W. N.
McMahon
R. G.
Hook
I. M.
,
1997
,
MNRAS
,
291
,
L5
Fabian
A. C.
Iwasawa
K.
Celotti
A.
Brandt
W. N.
Hook
M.
,
1998
,
MNRAS
,
295
,
L25
DOI:
Fabian
A. C.
Celotti
A.
Iwasawa
K.
McMahon
R. G.
Carilli
C. L.
Brandt
W. N.
Ghisellini
G.
Hook
I. M.
,
2001
,
MNRAS
,
323
,
373
DOI:
Fabian
A. C.
Celotti
A.
Iwasawa
K.
Ghisellini
G.
,
2001
,
MNRAS
,
324
,
628
DOI:
Ferrero
E.
Brinkmann
W.
,
2003
,
A&A
,
402
,
465
Fiore
F.
Elvis
M.
Giommi
P.
P.
,
1998
,
ApJ
,
492
,
79
DOI:
Fiore
F.
Elvis
M.
Maiolino
R.
Nicastro
F.
Siemiginowska
A.
Stratta
G.
D'Elia
V.
,
2003
,
A&A
,
409
,
57
Grupe
D.
Mathur
S.
Wilkes
B.
Elvis
M.
,
2004
,
AJ
,
127
,
1
DOI:
Hook
I. M.
McMahon
R. G.
,
1998
,
MNRAS
,
294
,
L7
DOI:
Kirsch
M.
,
2004
,
EPIC Status of Calibration and Data Analysis
, http://xmm.vilspa.esa.es/docs/documents/CAL-T-0018-2-3.pdf
Kirsch
M.
et al
,
2004
,
SPIE
,
5488
,
103
DOI:
Lu
L.
Sargent
W. L. W.
Barlow
T. A.
Churchill
C. W.
Vogt
S.
,
1996
,
ApJS
,
107
,
475
DOI:
Moran
E. C.
Helfand
D. J.
,
1997
,
ApJ
,
484
,
L95
DOI:
Otrupcek
R. E.
Wright
A. E.
,
1991
,
Pub. Astron. Soc. Australia
,
9
,
170
Péroux
C.
Storrie-Lombardi
L. J.
McMahon
R. G.
Irwin M., Hook
I. M.
,
2001
,
AJ
,
121
,
1799
DOI:
Pettini
M.
Smith
L. J.
King
D. L.
R. W.
,
1997
,
ApJ
,
486
,
665
DOI:
J. X.
Wolfe
A. W.
,
2000
,
ApJ
,
533
,
L5
DOI:
Protassov
R.
Van Dyk
D.
Connors
A.
Kashyap
V. L.
Siemiginowska
A.
,
2002
,
ApJ
,
571
,
545
DOI:
Reeves
J. N.
Turner
M. J. L.
,
2000
,
MNRAS
,
316
,
234
DOI:
Sikora
M.
G
,
2000
,
ApJ
,
534
,
109
DOI:
Wehrle
A. E.
et al
,
1998
,
ApJ
,
497
,
178
DOI:
Wilkes
B. J.
Elvis
M.
Fiore
F.
McDowell
J. C.
Tananbaum
H.
Lawrence
A.
,
1992
,
ApJ
,
393
,
L1
DOI:
Worsley
M. A.
Fabian
A. C.
Turner
A. K.
Celotti
A.
Iwasawa
K.
,
2004
,
MNRAS
,
350
,
207
DOI:
Worsley
M. A.
Fabian
A. C.
Celotti
A.
Iwasawa
K.
,
2004
,
MNRAS
,
350
,
L67
DOI:
Yuan
W.
Brinkmann
W.
,
1999
, in
Aschenbach
B.
Freyberg
M.
, eds,
MPE Report 272
,
Highlights in X-ray Astronomy
. Max-Planck-Institut für Extraterrestriche Physik, Garching, p.
240
Yuan
W.
Matsuoka
M.
Wang
T.
Ueno
S.
Kubo
H.
Mihara
T.
,
2000
,
ApJ
,
545
,
625
DOI:
Zickgraf
F.-J.
Voges
W.
Krautter
J.
Thiering
I.
Appenzeller
I.
Mujica
R.
Serrano
A.
,
1997
,
A&A
,
323
,
L21
1
This effect is commonly referred to as excess absorption in the literature. Here we use a general term in consideration of possible alternative explanations.
2
Fpho(E) ∝E−Γ, where Γ is the photon index.
3
We regard a model fit as acceptable if the null hypothesis probability derived from the fit is greater than 10 per cent (Pnull > 0.1), as is commonly quoted.
4
Recently, Protassov et al. (2002) have questioned the validity of using the F-test to test for the significance of adding an additional spectral component, as this involves testing a hypothesis that is on the boundary of the parameter space. We note that the F-test we used here is not affected by the boundary condition problem, because we are testing for the significance of various values of NH between NH=NGalH (null hypothesis) and the free-fitted value, rather than for the addition of a model component.
7
The Reflection Grating Spectrometer on-board XMM–Newton.

## Author notes

Present address: Yunnan Astronomical Observatory, Chinese Academy of Sciences, Phoenix Hill, PO Box 110, Kunming, Yunnan, 650011 China.