The return to the hard state of GX 339-4 as seen by Suzaku
P.-O. Petrucci, C. Cabanac, S. Corbel, E. Koerding, R. Fender
AAstronomy & Astrophysics manuscript no. gx339petrucciV3revised c (cid:13)
ESO 2018November 1, 2018
The return to the hard state of GX 339-4 as seen by Suzaku
P.-O. Petrucci , C. Cabanac , S. Corbel , E. Koerding , and R. Fender , UJF-Grenoble 1 / CNRS-INSU, Institut de Plan´etologie et d’Astrophysique de Grenoble (IPAG) UMR 5274, Grenoble,F-38041, France Universit´e de Toulouse, UPS-OMP, IRAP, Toulouse, France CNRS, IRAP, 9, av du Colonel Roche, BP 44346, F-31028Toulouse Cedex 4, France Laboratoire AIM (CEA/IRFU - CNRS/INSU - Universit´e Paris Diderot), CEA DSM/IRFU/SAp, F-91191 Gif-sur-Yvette, France Department of Astrophysics/IMAPP, Radboud University Nijmegen, P.O. Box 9010;6500 GL Nijmegen;TheNetherlands University of Oxford, Department of Physics, Astrophysics, Denys Wilkinson Building, Keble Road, OX1 3RH, Oxford,UK School of Physics and Astronomy, University of Southampton, Highfield, Southampton, SO17 1BJ, UKReceived .../Accepted ...
ABSTRACT
The reality of the disk recession is an information of prime importance to understand the physics of the state transitionsin X-ray binaries. The microquasar GX 339-4 was observed by Suzaku five times, spaced by a few days, during itstransition back to the hard state at the end of its 2010-2011 outburst. The 2-10 keV source flux decreases by a factor ∼
10 between the beginning and the end of the monitoring. Simultaneous radio and OIR observations highlightedthe re-ignition of the radio emission just before the beginning of the campaign, the maximum radio emission beingreached between the two first Suzaku pointings, while the IR peaked a few weeks latter. A fluorescent iron line is alwayssignificantly detected. Fits with a gaussian or Laor profiles give statistically equivalent results. In the case of a Laorprofile, fits of the five data sets simultaneously agree with a disk inclination angle of ∼
20 degrees. The disk inner radiusis < − R g in the first two observations but almost unconstrained in the last three due to the lower statistics. Asoft X-ray excess is also present in these two first observations. Fits with a multicolor disk component give disk innerradii in qualitative agreement with those obtained with the iron line fits. The use of a physically more realistic model,including a blurred reflection component and a comptonization continuum, give some hints of the increase of the diskinner radius but the significances are always weak (and model dependent) preventing any clear conclusion concerningdisk recession during this campaign. Interestingly, the addition of warm absorption significantly improves the fit ofOBS1 while it is not needed in the other observations. The radio-jet re-ignition occurring between OBS1 and OBS2,these absorption features may indicate the natural evolution of the accretion outflows transiting from a disk wind, anubiquitous characteristic of soft states, and a jet, signature of the hard states. The comparison with a long 2008 Suzakuobservation of GX 339-4 in a persistent faint hard state (similar in flux to OBS5) where a narrow iron line clearlyindicates a disk recession, is discussed.
1. Introduction
Multi-wavelength observation campaigns of microquasars,like those done in X-ray and Radio in the last 10 years,were crucial to bring to light the strikingly link betweenthe ejection phenomena (mostly observed in radio) and theinner accretion flow, whose radiation seems to be the domi-nant component in the X-rays (e.g. Corbel et al. 2000, 2003;Coriat et al. 2011; Gallo et al. 2003). For instance, strongradio emission, interpreted by the presence of persistentjets (directly observed in a few cases, e.g. Dhawan et al.2000; Stirling et al. 2001), is generally detected when theX-ray emission peaks at a few tens of keV, in the so-calledhard state (e.g. Corbel et al. 2004; Fender et al. 2004).This X-ray emission is commonly believed to originate viainverse Compton process from a plasma of hot electrons(the so-called corona) scattering off UV/soft X-ray pho-tons produced by the cooler part of the accretion flow. Onthe other hand, in the so-called soft state the radio emis-sion is quenched (Fender et al., 1999; Corbel et al., 2000) and the X-ray data are spectrally dominated by soft X-rayemission. This emission is generally interpreted as signa-ture of a multi-color accretion disk component down to thelast stable orbit R ISCO .In the past ten years, high energy resolution observa-tions of several microquasars showed also the presence ofhighly ionized absorption features in their X-ray spectra.These features were interpreted as signature of ionized gazin the close environment of the black hole, their blueshiftsbeing indication of outflows or winds (e.g. Miller et al. 2004hereafter M04, Miller et al. 2006c). It has been realized thatthese features were more specifically observed in the softstate (e.g. Ponti et al. 2012; Diaz Trigo et al. 2011; DiazTrigo & Boirin 2012). These results suggest that, duringoutbursts, X-ray binaries may transit back and forth be-tween disk-jet (in the hard state) and disk-wind (in the softstate) configurations (e.g. Neilsen & Lee 2009). However theexact interplay between ejection and accretion phenomenaand the origin of the transition from one state to the others a r X i v : . [ a s t r o - ph . H E ] F e b .-O. Petrucci et al.: The return to the hard state of GX 339-4 as seen by SuzakuObs name Obs ID MJD XIS03 Exp. XIS 0-3 XIS 0-3 HXD/PIN HXD/GSOks 0.7-2 keV ( s − ) 2-10 keV ( s − ) 20-70 keV ( s − ) 70-200 keV ( s − )OBS1 405063010 55603.7 44.2 11.90 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Table 1.
Log of the 5 observations with their ID number, the corresponding date (in MJD), the sum of the XIS0 andXIS3 exposure time (each exposure being ∼
20 ks, the total exposure of the XIS0 and XIS3 instruments is about twicethis time) and the count rates in the different Suzaku instruments. .is still poorly known (see however the recent study done byKalemci et al. 2013).The commonly adopted toy-picture of the centralregions of microquasars plays on the relative importanceof the accretion disk and hot corona emission along theoutburst (e.g. Esin et al. 1997; Done et al. 2007). Theaccretion disk is assumed to be present in between anouter and inner radius R out and R in while the hot coronais localized in between R in and R ISCO . In the hard state, R in (cid:29) R ISCO and the hot corona dominates the observedemission. The inner part of the accretion disk (close to R in ) is then cold, explaining its poor detection in thesoft X-rays in this state. Reversely, in the soft states R in ∼ R ISCO , i.e. the hot corona is no more present andthe spectra are dominated by the accretion disk emission.If this picture is correct, variations of the disk innerradius R in should occur during the state transitions, firstdecreasing during the hard-to-soft transition but then in-creasing during the soft-to-hard one. This interpretation isapparently supported by the observations, in hard states,of weak reflection components (e.g. Gierlinski et al. 1997;Barret et al. 2000; Miller et al. 2002; Zycki et al. 1998;Joinet et al. 2007), the absence of relativistic broadening ofthe iron line in a few cases (e.g. Tomsick et al. 2009; Plantet al. 2013) and the absence of obvious thermal components(e.g. Poutanen et al. 1997; Dove et al. 1997; Remillard &McClintock 2006; Done et al. 2007; Dunn et al. 2010), allpotential signatures of small and remote reflecting area.The variable disk inner radius during state transitionsthen becomes a natural key ingredient in most theoreticalmodels, controlling or resulting from the spectral andtiming evolution of microquasars during the outburst(e.g. Esin et al. 1997; Meyer et al. 2000; Belloni et al.2005; Remillard & McClintock 2006; Ferreira et al. 2006;Petrucci et al. 2008).Recent XMM observations, much more sensitive, espe-cially in the soft X-rays (i.e. below 2 keV), compared tothe other existing X-ray missions, apparently ruled out thepresence of a recessed disk in the hard states (e.g. Milleret al. 2006a,b). The long XMM observation monitoredduring the 2004 outburst of GX339-4 did not agree witha simple power-law although the system was in a typical,although bright ( ∼
10% L
Edd ), hard state. According tothe authors, a very strong soft excess as well as a broademission feature around 6.4 keV were also present and wellfitted by multicolor disk and a relativistically broadenedemission line respectively. This suggests the presence ofan accretion disk extending towards the vicinity of theblack hole (but see below). SWIFT/XRT observations of
Fig. 1. Top:
RXTE/PCA 3-20 keV count rate and hard-ness ratio light curves of the complete 2010-2011 outburstof GX 339-4.
Bottom:
A zoom of the end of the outburst,with the date of the 5 Suzaku observations indicated by theblack dots and the vertical dotted lines.XTE J1817- 330, during its decline to the hard state, ledRykoff et al. (2007) to the same conclusions i.e. no diskrecession. From their SWIFT survey of stellar mass blackholes, Reynolds & Miller (2013) do not find evidence forlarge-scale truncation of the accretion disk in the hardstate either, at least for X-ray luminosities larger than10 − L Edd .The estimates of inner disk radii based on continuumspectroscopy are subject however to considerable uncer-tainties (e.g. Merloni et al. 2000; Zimmerman et al. 2005;
Cabanac et al. 2009 hereafter C09) and different data anal-ysis may give different conclusions. A recent re-analysis ofthe data used in Miller et al. (2006b) suggests that theobserved broad iron line may be an artifact due to an im-proper correction of the pile-up in the MOS data (Done& Diaz Trigo, 2010). Suzaku pointing on the same sourcein its hard state has also shown that the spectrum couldbe consistent with a truncated disk (Tomsick et al., 2009).More specifically, by re-analyzing the whole SWIFT/XRTdataset of XTE J1817-330, C09 does observe a slight in-crease of the disk radius. This occurs apparently when the2 -10 keV luminosity decreases below ∼ × − L Edd .These authors analyzed other sources in the same way andobtain similar, though less significant, results.The reality of the disk recession is clearly an infor-mation of prime importance that we crucially need if wewant to understand the physics of the state transitions.Confirming, and precisely measuring, this recession (ifany) should allow to constrain and refine most of thepresent theoretical models. On the other hand, the absenceof recession will imply to strongly reconsider our presentunderstanding of the microquasar phenomenon.We present in this paper a Suzaku campaign on the mi-croquasar GX 339-4 aiming at catching the recession, if any,of the accretion disk during a soft-to-hard state transition.Sect. 2 detailed the observation and data treatment andSect. 3 the data analysis. While the constraints on the diskinner radius, discussed in Sect. 4, prevent any clear con-clusions concerning its recession, the observation of ionizedabsorbing features in the soft X-rays may suggest a diskwind whose properties may evolve during the transitions.Theses results are discussed in Sect. 5 before concluding.
2. Observations and data treatment
GX 339-4 was observed five times ( ∼
20 ks each) by Suzakuat the end of its last outburst in February 2011, as soon asthe source became visible by the satellite. The 5 observa-tions were separated by a few days in order to follow thespectral evolution of the object all along its transition backto the hard state. The log of these observations is detailedin Tab. 1, with the corresponding dates. The RXTE/PCA3-20 keV light curve of the complete 2010-2011 outburst ofGX 339-4 is plotted at the top of Fig. 1 together with thehardness ratio . A zoom of the last part of the outburst,with the dates of the 5 Suzaku observations indicated bythe vertical dotted lines, is shown at the bottom of Fig. 1.For the data treatment we use the most up-to-date cal-ibration files and the HEASoft version 11.6.1. We run theSuzaku XIS/HXD aepipeline (V1.1.0) tools to reprocessthe data from scratch. We take care to pile-up effect in theXIS instrument by running first the aeattcor2 tool whichcorrects Suzaku attitude data for the effects of ”thermalwobbling” caused by thermal distortions of the satellitebodies . Then we run the pile-up fraction estimation tool Assuming a 10 solar masses black hole. The hardness ratio is defined as the ratio of the (5.7-9.5 keV)count rate over the (2.9-5.7 keV) count rate http://heasarc.gsfc.nasa.gov/ftools/caldb/help/aeattcor2.html pileest also released in the HEASoft package to estimatethe amount of pileup in the XIS images and disregardedregions with >
10% pileup fraction during the spectralextraction. Only the first two observations suffered frompileup ( ∼
19 and 10% for OBS1 and OBS2 respectivelyfor the most central pixels) and a circular region of ∼ .To create the background files in the XIS energy rangewe use the ftool xisnxbgen which estimates the non X-raybackground spectrum of the XIS instrument. While theCosmic X-ray Background is expected to be low in theXIS energy band, we take it into account following therecipe indicated in the ABC guide V4.0 (p. 81-82) butrenormalized to the XIS FOV corresponding to the 1/4window mode. The total X-ray background files (includingthe non-X-ray as well as the cosmic X-ray background) ofthe HXD/PIN instruments were computed using the ftools hxdpinxbpi . For the non X-ray background, we use the”tuned” background files distributed by the HXD teamand corresponding to our observations. These ”tuned”background suffers from systematic uncertainties of about For the rebin procedure, we use the tool PHARBN developedby M. Guainazzi and adapted to the XIS instrument http://heasarc.gsfc.nasa.gov/docs/suzaku/analysis/abc/ We do not include the GSO data in the spectral analysis dueto their very poor statistics.
Inverted spectrumNon-inverted spectrum
Fig. 2. Top:
Bottom:
Light curve of the 5.5-9 GHz spectral index.
A positive spectral index corresponds to an in-verted spectrum generally characteristic of opti-cally thick synchrotron emission from a stratifiedjet (e.g. Blandford & K¨onigl 1979) . The dates of the5 Suzaku observations are indicated by the vertical dottedlines (data from Corbel et al. 2013). that have been added to the PIN data during ourfitting procedure.Following the ABC guide V.4.0 (chapter 5.7.2) , the nor-malization of the HXD/PIN with respect to XIS is fixedto 1.16. The energy ranges for the XIS and HXD/PIN in-struments are restricted to 0.7-10 keV and 20-70 keV re-spectively. In the following, we name the five observationsOBS1, OBS2, OBS3, OBS4 and OBS5 (see Tab. 1).
Simultaneous or quasi simultaneous radio observationswere taken with the Australia Telescope Compact Array(ATCA) at 5.5 and 9 GHz (Corbel et al. 2013, here-after C13). Simultaneous or quasi simultaneous optical andIR observations were taken with the SMARTS telescope(Buxton et al., 2012; Din¸cer et al., 2012). The correspond-ing light curves are reported at the top of Fig. 2 and theradio spectral index (between the 5.5 and 9 GHz band) isplotted at the bottom of the figure.
3. Results
GX 339-4 was at the end of its outburst and, as expected,its X-ray flux decreases from OBS1 to OBS5. The countrates in the 0.7-2, 2-10, 20-70 and 70-200 keV ranges areindicated in Tab. 1 and indeed they all decrease duringthe campaign. Interestingly the decrease is much morepronounced in the softer energy range than in the harderones. The 0.7-2 keV range count rate decreases by morethan a factor 10 while it decreases by less than a factor 2in the 70-200 keV range. This clearly indicates a spectralchange of the X-ray emission.As shown in Fig. 2, the radio emission begins to in-crease between MJD 55590-55600 and reaches a maximumbetween our two first Suzaku observations ( ∼ MJD 55605).The radio emission being generally associated with the pres-ence of a jet, the increase of radio flux is interpreted as thereappearance of the jet while GX 339-4 is turning back toits hard state (see e.g. C13, Kalemci et al. 2013). This iswell supported by the simultaneous increase of the radiospectral index (bottom plot on Fig. 2), which becomes pos-itive after OBS2. Such inverted spectrum is characteristicof optically thick synchrotron emission from a stratified jet(e.g. Blandford & K¨onigl 1979).It is interesting to note that the RXTE/PCA hardnessratio (cf. Fig. 1) begins to increase a few days before the ra-dio peak, the two events being clearly separated in time (seealso Kalemci et al. 2013 for other outbursts). Concerningthe H band emission, it reaches a local maximum about 10days after the radio peak, in between OBS3 and OBS4 (seethe discussion in C13). see http://heasarc.gsfc.nasa.gov/docs/suzaku/analysis/watchout.html, item 16. (cid:239) (cid:239) c t s s - k e V - ! " Fig. 3. Top panel : Folded spectra of the 5 Suzaku obser-vations (OBS1 to OBS5 from top to bottom) fitted by apower law in the 3-10 keV energy range.
For clarity, onlythe XIS0+XIS3 spectra are plotted. The have beenrebinned in order to have 30 σ per bin . The bottompanel shows the contributions to the χ . We begin our spectral analysis on the Suzaku data above3 keV, a spectral domain known to be dominated by theprimary continuum. We first fit the XIS data only between3 and 10 keV with a simple power law model. The bestfit values for the photon index are reported in Tab. 2and the folded spectra are plotted in Fig. 3. The photonindex decreases rapidly between OBS1 and OBS2, fromΓ =1.72 ± ± α line. We adda gaussian component and fit the spectra again. The fitimproves very significantly for OBS1 (∆ χ =143), less forOBS2 and 3 (∆ χ =29 and 30 respectively) and even lessfor OBS4 and 5 (∆ χ =14) but the improvement is alwayssignificant (at more than 99% following the F-test ). Thebest fit parameter values are reported in Tab. 2. Thecontour plots of the line flux vs. line energy and line widthvs. line energy for the 5 observations are plotted in Fig. 4.The line flux is clearly decreasing between OBS1and OBS2 (at more than 3 σ ). Then, between OBS2 andOBS5, it is consistent with a constant at a confidencelevel of ∼ E gauss = 6 . ± .
13. Theothers measurement are consistent with a neutral iron linepeaking at 6.4 keV. The significativity of the variability ofthe line energy during the campaign is however less than40%. The line width is consistent with a constant at aconfidence level of ∼ ∼ ±
70 eV potentially indicating some broadening. See however Protassov et al. (2002) about the usage of theF-test in line-like features.4.-O. Petrucci et al.: The return to the hard state of GX 339-4 as seen by Suzaku
Obs 4 Obs 1Obs 2 Obs 3Obs 5
Suzaku2008
Obs 4 Obs 1Obs 2 Obs 3Obs 5
Suzaku 2008
Fig. 4.
The 90% contour plots line flux vs. line energy (
Left ) and line width vs. line energy (
Right ) from the bestfits of the five XIS spectra with a power law + gaussian model in the 3-10 keV energy range. The dashed contourcorresponds to the Suzaku observation of 2008 (Tomsick et al., 2009).To test if the line broadening could be due to relativis-tic effects, we replace the gaussian profile by a Laor profilewhich is expected if the line emission is produced by anaccretion disk around a black hole. To limit the number offree parameters, we fix the outer radius of the disk R out to400 R g and the index of the radial power law dependenceof the disk emissivity to 3. Then we fit the five XIS spectrasimultaneously, imposing only the same disk inclination an-gle (which was let free to vary) for the 5 observations. Allthe other parameters (Laor model inner radius and normal-ization, power law photon index and normalization) wereindependent from one observation to the other and free tovary. The fit gives a best fit value of the inclination angle i = 21 ± (cid:104) E Laor (cid:105) = 6 . ± .
07 keV.Concerning the disk inner radius, it is well constrained to afew Schwarzschild radii for OBS1 and OBS2 while we haveonly upper limits for the three last observations. While itstrongly suggests the presence of the accretion disk veryclose to the black hole at the beginning of the campaign,the data prevent any clear conclusion concerning the diskrecession.
We now include the 0.7-3 keV data in the fitting procedure.The ratios data/model of the five XIS spectra obtained afterextrapolation of the best fit power law + laor modeldown to 0.7 keV are plotted in Fig. 6. A photo-electric
Obs 4Obs 1Obs 2Obs 3Obs 5 Suzaku2008
Constraints from LAOR fits of the iron line profile
Fig. 5.
The 90% contour plots of the disk inclination anglevs. disk inner radius obtained by fitting simultaneously thefive XIS spectra with a power law + Laor model above 3keV. We impose the same inclination angle between the5 models. The dashed contour corresponds to the Suzakuobservation of 2008 (Tomsick et al., 2009)absorption (model tbnew in xspec ) was added but theHydrogen column density was fixed to 6 × cm i.e. thetypical galactic N H observed in the direction of GX 339-4(e.g. Zdziarski et al. 2004; Cadolle Bel et al. 2011). The softX-ray part of the spectra varies in time, in excess above thepower law extrapolation in OBS1 and OBS2, but in deficitin the three last observations. Letting the column densityfree to vary improves the fits, but they are still statisticallyunacceptable especially in the case of OBS1 and OBS2.The soft X-ray excess could be the signature of theoptically thick accretion disk component which dominatesthe X-ray emission in the soft state, then smoothly dis-appearing from OBS1 to OBS5 when the source re-enters All modeling was performed using the ”wilm” abundances(Wilms, Allen & McCray 2000) with the vern cross-sections(Verner et al. 1996) 5.-O. Petrucci et al.: The return to the hard state of GX 339-4 as seen by Suzaku power law
Obs. Γ F − keV L − keV /L Edd χ /dof( × − erg cm − s − ) (%)1 1.72 ± ± ± ± ± ± power law+gauss Obs. Γ E gauss σ gauss F gauss EW χ /dof(keV) (keV) ( × − ph cm − s − ) (eV)1 1.81 ± ± ± +1 . − . +40 − ± ± ± +0 . − . +25 − ± ± ± +0 . − . +40 − ± ± ± +0 . − . +35 − ± ± ± +1 . − . +75 − ± ± ± +0 . − . +10 − power law+laor Obs. Γ E laor R in F laor EW χ /dof(keV) ( R g ) ( × − ph cm − s − ) (eV)1 1.80 +0 . − . +0 . − . +0 . − . +1 . − . +40 − +0 . − . +0 . − . +2 . − . +0 . − . +40 − +0 . − . +0 . − . <
35 1.3 +0 . − . +60 − +0 . − . +0 . − . <
180 0.7 +0 . − . +60 − +0 . − . +0 . − . <
25 0.7 +0 . − . +130 − +0 . − . +0 . − . >
100 0.38 +0 . − . +5 − Table 2.
Best fit of the XIS data between 3 and 10 keV with a simple power law, a power-law + gaussian line and apower-law + Laor profile emission line. In the last case, the 5 XIS spectra have been fitted simultaneously, letting all theparameters free to vary but imposing the Laor profile inclination angle to be the same between the 5 models. The bestfit value for the inclination angle is i=21 +5 − deg. The errors on the inner disk radius R in of the Laor profile correspondto a confidence level of 90% for 2 parameters. We have reported also the best fit parameter values for the 2008 Suzakuobservation, again assuming the same inclination angle than the 5 other Suzaku observations. With the addition of thesedata, the best fit value for the inclination angle then becomes i = 20 +3 − deg. To compute the 3-10 keV flux in Eddingtonunit, we adopt an Eddington luminosity of L Edd (cid:39) . × erg s − (i.e. we assume a 10 solar masses black hole) anda distance to GX 339-4 of 8 kpc.in the hard state. To test this hypothesis, we have addeda multicolor disk component diskbb in our fits. Thecolumn density is still free to vary independently for the5 observations . The best fit parameters are reported inTab. 3. The addition of a multicolor disk component im-proves strongly the fit of OBS1 and OBS2 with ∆ χ =1332and 173 respectively. This component is not statisticallyrequired however in the three last observations potentiallysuggesting the disappearance of the disk component duringthe state transition.We have reported on Fig. 7 the contour plots, forthe 5 observations, of the diskbb normalization vs. R in,laor the best fit value of the inner disk radius ofthe laor profile. Assuming an inclination angle of 21deg., a distance of 8 kpc and a black hole mass of 10solar masses, the diskbb normalizations can also beconverted to an ”apparent” disk inner radii R in,app . Imposing N h to the same value for the 5 observations givea much worse fit with ∆ χ =171 for 4 less dof. Taking into account a color temperature correction factor f col = 1 . R in,diskbb = f col R in,app . The valuesof R in,diskbb have been reported on the right axis of Fig. 7a.Both estimates of the disk inner radius (i.e. from diskbb or laor ) roughly agree one with each other. Notea few differences however: R in,laor are consistent betweenOBS1 and OBS2 while R in,diskbb of OBS1 is smaller andinconsistent with the OBS2 value. In both cases however,the inner radius is found to be lower than or of the orderof 10 R g . In OBS3 R in,laor is constrained to be in between ∼ ∼ R g while R in,diskbb is unconstrained. In OBS5 R in,laor is upper limited to ∼ R g while R in,diskbb isunconstrained. We reach however the same conclusionsfor the evolution of the disk inner radius, R in being smallin OBS1 and OBS2 ( < R g ), but then the contours forOBS3, OBS4 and OBS5 becomes too large to constrain its tbnew*(diskbb+power law+laor) Obs. N h T in N diskbb Γ E laor R in,laor F laor EW χ /dof ∆ χ × (eV) × (keV) ( R g ) × − (eV)1 0.47 +0 . − . +10 − +0 . − . +0 . − . +0 . − . +0 . − . +1 . − . +30 − +0 . − . +10 − +2 . − . +0 . − . +0 . − . +2 . − . +0 . − . +30 − +0 . − . < > +0 . − . +0 . − . +24 . − . +0 . − . +40 − +0 . − . +10 − +2 . − . +0 . − . +0 . − . +2 . − . +0 . − . +30 − +0 . − . < × < +0 . − . +0 . − . <
30 0.7 +0 . − . +120 − +0 . − . +20 − < × +0 . − . +0 . − . >
210 0.4 +0 . − . +10 − Table 3.
Best fit of the XIS data between 0.7 and 10 keV with a diskbb + power-law + Laor line and a photo-electricabsorption (model tbnew , Wilms et al. 2000). The line flux is in units of ph cm − s − . We assume a 10 solar massesblack hole for R g . We report also the ∆ χ fit improvement due to the addition of the multicolor disk component. . Fig. 6.
Ratios data/model of the five XIS0+XIS3 spectrafor OBS1 (top) to OBS5 (bottom). The model is a powerlaw + laor fitted above 3 keV and then extrapolateddown to 0.7 keV. The Hydrogen column density is fixed to6 × cm . We use tbnew to model the X-ray absorption(Wilms et al., 2000).behavior.We have checked that these results do not significantlychange when using a more physical model (via e.g. up-scattering of disk photons) for the high energy continuumlike comptt in xspec . Such models indeed differ to apower-law shape especially in the low energy portion ofthe spectrum where a low energy roll-over is expected. Thecorresponding contour plots obtained when fitting with comptt are reported in Fig. 7b. The main differenceswith Fig. 7a are larger upper limits on R in,laor in OBS4and OBS5.Note that the fits reported in Tab. 3 are relatively badespecially for OBS1 and indeed residuals are visible in thesoft part ( < diskbb+ power law model, as well as neutral absorption (due totheir limited band pass at low energy, they fixed the hydro-gen column density at 5 × cm − ) so that our resultscan be safely compared to theirs. As can be seen on Fig.8, both best fits parameter values agree very well one witheach other. Compared to RXTE however, the lower limitof the energy band of the XIS suzaku instrument allows tofollow the disappearance of the diskbb component down tolower flux and lower inner disk temperature. GX 339-4 was observed by Suzaku in 2008 during a long ex-posure where flux and spectral index values were very closeto the one of OBS5 (Tomsick et al. 2009, T09 hereafter).It is worth noting however that this observation was made1.6 years after the peak of its 2007 outburst and that thesource was in a persistent, but faint, hard state (e.g. Russellet al. 2008; Kong 2008). In comparison, in 2011 GX 339-4turned back in the hard state since ∼ R in > R g at 90%confidence for a disk inclination of 0 deg, and R in > R g They also added a smeared edge at a fixed energy of 10 keVto fit the iron K absorption edge seen around 7.1 keV. 7.-O. Petrucci et al.: The return to the hard state of GX 339-4 as seen by Suzaku
Obs 4Obs 1 Obs 2Obs 3 Obs 5
Suzaku2008
Obs 4Obs 1 Obs 2Obs 3 Obs 5
Suzaku2008
Fig. 7.
Contour plots of the disk normalization (left scale)and the corresponding disk inner radius R in,diskbb (rightscale) vs. the disk inner radius R in,laor deduced from the laor profile. The five XIS spectra have been fitted with a diskbb + laor model and a power law (top) or a comp-tonization model comptt (bottom) for the continuum. Weassume a disk inclination of 20 deg., a distance of 8 kpcand a black hole mass of 10 solar masses. For OBS5, the90% confidence area is on the left of the yellow line. Thedashed contour corresponds to the Suzaku observation of2008 (Tomsick et al., 2009).for a disk inclination of 20 deg.We have re-analyzed these data following the proceduredetailed in Sect. 2. We have fitted them in the 3-10 keVenergy range with the power law + gau model first.The corresponding contour plots of the line flux and linewidth versus line energy are overplotted in Fig. 4 and thebest fit parameter values are reported in Tab. 2. If thespectral shape and flux are in good agreement with OBS5,the line width is clearly inconsistent between the twoobservations, indicating intrinsic differences (geometry?ionisation state?) of the reflecting material.Then we have fitted these data, still in the 3-10 keVenergy range, but with the power law + laor model,either separately or simultaneously to our 5 Suzakupointings, imposing, in the latter case, the inclinationangle to the same value for each data set. The best fitresults are very similar in both cases and are in agreement I n d e x T i n ( ke V ) P L , D BB Fig. 8.
Evolution of the power law photon index (top), theinner disk temperature T in (middle) and the power-law anddisk flux (bottom, filled and empty circles respectively) inthe 3-25 keV energy band in units of 10 − erg cm − s − .The black symbols are the results of this campaign while thegray ones have been obtained by Din¸cer et al. (2012). Themodel used is tbnew × (diskbb+ power law + gaus-sian) in the 0.7 - 10 keV range.with those of T09 (see last row of Tab. 2). We find for2008 an inner radius > R g and a best fit inclinationangle i = 20 +3 − degrees consistent with our previousvalue obtained without the use of these data (see Sect.3.2.1). We have overploted the 90% contour plot of thedisk inclination angle vs. disk inner radius of the 2008observation in dashed line in Fig. 5. While only at a2 σ level, the fact that it does not overlap with the90% contours obtained for our 2011 observationsstrongly suggests that the constraints on the diskinner radius deduced from the laor model areunconsistent between 2008 and 2011 .Following the fitting procedure of the previous sections,we have also fitted the 2008 data set in the 0.7-10 keV rangeadding now a diskbb component to the model. The best fitresults are reported in the last row of Tab. 3 and the cor-responding contour plots of the diskbb normalization vs.inner disk radius have been overploted in dashed line in Fig.7. The constraints on R in deduced from the diskbb nor-malization give now a lower limit of ∼ R g (to be comparedwith the lower limit of ∼ R g from the Laor fit of theiron line), in agreement with our (unconstrained) contourplots of OBS3 and OBS5. These constraints on R in fromthe diskbb component have to be taken with caution how-ever given that the disk component is poorly constrainedin these observations. tbnew*(diskbb+po+kdblr*reflionx) Obs. N h T diskbb N diskbb F diskbb Γ F po R in ξ ref F ref L bol L Edd χ /dof × eV × × − × − R g × − %1 0.42 +0 . − . +20 − +0 . − . +0 . − . +30 − +200 − +0 . − . +20 − +0 . − . +0 . − . +240 − +30 − +0 . − . +30 − +216 . − . +0 . − . >
70 1100 +800 − +0 . − . +20 − +5 . − . +0 . − . >
10 790 +5700 − +0 . − . < < . < +0 . − . >
30 1400 +3900 − +0 . − . +20 − +960 . − . +0 . − . >
180 20 +30 − tbnew*(diskbb+eqpair+kdblr*reflionx) Obs. N h T diskbb F diskbb l h /l s τ T bb F eqpair R in ξ ref F ref L bol L Edd χ /dof × eV × − eV × − R g × − %1 0.46 +0 . − . +10 − +0 . − . +0 . − . +10 − +1 . − . +100 − +0 . − . +10 − +0 . − . +0 . − . +20 − +33 − +185 − +0 . − . +30 − +1 . − . +1 . − . +50 − >
70 1180 +480 − +0 . − . +100 − +0 . − . +2 . − . +50 − >
10 910 +2100 − +0 . − . +30 − +4 . − . +2 . − . +60 − >
30 1450 +3590 − +0 . − . +10 − +0 . − . +0 . − . +20 − >
180 40 +15 − Table 4.
Results of the fits of the XIS and HXD data with a power law (top table) or eqpair (bottom table) for thecontinuum and the ionized reflection component reflionx . In the case of the power law fit, the photon indexused in reflionx is fixed to the one of the power law continuum. In the case of eqpair , we froze thephoton index of reflionx to the one obtained in the power law fit. . The inclination was fixed to 20 deg. Weuses the kernel from the laor line profile to account for the gravitational effects close to the black hole (model kdblur )of xspec ). All errors are 90 per cent confidence for one parameter. The fluxes are in unit of ergs.cm − .s − and arecomputed in the 0.7-70 keV energy range. The bolometric luminosity L bol is assumed to be the sum of the luminosities ofthe three spectral component diskbb, eqpair and reflionx . We adopt an Eddington luminosity of L Edd (cid:39) . × erg s − (i.e. we assume a 10 solar masses black hole) and a distance to GX 339-4 of 8 kpc. From the previous sections, fits of the soft X-ray excessand iron line profiles, independently one with each other,both suggest an increase of the disk inner radius fromOBS1 to OBS2. Its evolution is however uncertain forthe three last observations. We also confirm the largevalue of R in deduced from the iron line fit in the 2008Suzaku observation. To go a bit further, we use, in thissection, more realistic models for the continuum and thereflection component. We take also advantage of the HXDinstrument of Suzaku and include the HXD/PIN (20-70keV) data in our fits.We have reported In Fig. 9 the χ contribution whenwe extend the best fit model tbnew × (diskbb+ powerlaw + laor) of the XIS data (the best fit parameters arereported in Tab. 3) above 10 keV in the HXD/PIN energyrange. A clear excess between 20 and 40 keV is visible es-pecially in OBS1 and OBS2 and suggests the presence of areflection bump.To provide a more physical description of the reflectioncomponent giving birth to the iron line, we use the combi-nation of the reflionx code of Ross & Fabian (2005), con-volved with the relativistic kernel kdblur (Laor, 1991). Inagreement with the results of Sect. 3.2.1, we fix in kdblur the outer disk radius R out to 400 R g , the index of the radial power law dependence of the emissivity to 3 and the diskinclination angle to 20 degrees.In reflionx the illumination has a power law shape.So we first fit the different observations with a power-lawfor the primary continuum, fixing the photon indexin reflionx to that of the power law continuum .However, as already discussed in Sect. 3.2.2, comptoniza-tion of the soft disk photons in a hot corona is widelyaccepted as the mechanism at the origin of the X-raycontinuum of X-ray binaries. Thus, instead of a simplepower law, we also use the Comptonization model eqpair (Coppi, 1999) even if it is not perfectly consistent withthe use of reflionx . In this case, we fix the photonindex in reflionx to the precedent values obtainedwhen fitting with the power law continuum Wewill see that our results are qualitatively similar betweenthe two models, the advantage of eqpair being that itsmain parameters (i.e. the ratio between the compactness ofseed photons, l s , and hot electrons, l h , the corona opticaldepth τ and the temperature of the soft disk photons T bb )have a direct physical meaning.The best fit parameters obtained with these modelsare reported in Tab. 4. Note that the fits are alwaysbetter with eqpair than with a power law. The lack ofa low energy roll-over in this latter case could explain The slope of eqpair can be estimated from the l h /l s ratio (e.g. Malzac et al. 2001) and appears to be veryclose to the photon index of the power law fits. Fig. 9. χ contribution in the 0.7-70 keV range when ex-tending the best fit model tbnew × (diskbb+ power law+ laor) (see Tab. 3) of the XIS data above 10 keV inthe PIN energy range. The laor component has been sup-pressed in these plots. Strong residuals are visible near 6.4keV and above 10 keV especially for OBS1 and OBS2.the larger residuals observed in the soft band ( < R in is still well constrained to a small value ∼ R g in this observation. Note also that the ionizationparameter is not larger in OBS3 compared to the otherobservations and then does not support the presenceof a more ionized iron line as suggested by the fits be-tween 3 and 10 keV with a simple gaussian (see Sect. 3.2.1).Since both models (with a power law or eqpair for thecontinuum) give similar parameter constraints and sincethe fits with eqpair gives always a better χ , from now onwe will only discuss the results obtained with this model.We find good fits in all cases but OBS1 for which somefeatures are still present in the soft energy range ( < diskbb+ power law + gaussian fits below 10 keV discussed in Sect. 3.2.2 (and reported in Tab. 3), we have computedthe corresponding χ /dof of these new fits but limited tothe 0.7-10 keV energy range. We find the following values:530/382, 401/382, 392/379, 417/372 and 355/341 forthe five (from OBS1 to OBS5) observations respectively.The improvement of the fits in the 0.7-10 keV energyrange with the blurred ionized reflection model are reallysignificant for OBS1 and OBS2 with ∆ χ = 49 and 27 fortwo less degrees of freedom. No significant improvement isobtained for the other observations, potentially because oftheir lower statistics.The hard to soft compactnesses ratio l h /l s increasessmoothly all along the monitoring in agreement with theobserved spectral hardening (see Tab. 2). It is of the orderof 5-8 suggesting a photon starved geometry for the hotcorona. On the other hand, the hot corona optical depth, τ , increases significantly between OBS1 and OBS2 by afactor ∼ ∼ ξ ref ∼ F diskbb , F eqpair and F ref ) are alsoreported in Tab. 4. They show a clear decrease along themonitoring in agreement with the fact that the sourcereturns back to its quiescent state. However, the relativeratios F diskbb /F eqpair and F ref /F eqpair are not constant,as we could expect if all the spectral components fade inthe same way. These ratios decrease from ∼
14 to 1% and ∼
25 to 10% respectively.The diskbb temperature, T diskbb , as well as the eqpair soft photon temperature, T bb , show also clear decrease fromOBS1 to OBS5. Interestingly, while the two temperatureswere let free to vary during the fitting procedure, T diskbb is about half T bb (see Fig. 11a). Such correlation may havea physical origin. Indeed, in eqpair the soft temperature T bb corresponds to the disk temperature actually ”see” bythe hot corona. On the other hand, T diskbb is the effectivedisk temperature. The ratio between the two, the so-calledspectral hardening factor, is generally estimated to be ofthe order of 2-3 (e.g. Shimura & Takahara 1995; Sobczaket al. 1999; Merloni et al. 2000; Davis et al. 2006) inagreement with what we find.For comparison we have also fitted the 2008 Suzakuobservation with the same model. The best fit parametersare reported in the last row of Tab. 4. Interestingly, apartfrom the reflection ionization parameter which is of theorder of ∼
40 i.e. well below our best fit values observed in201 and a larger disk inner radius, the other parametersagree well with the spectral evolution from OBS1 to OBS5.More precisely, and even if there is a 3 year gap betweenthem as well as a poorer statistics in 2011, the 2008observation seems to be very similar to OBS5 but with adifferent accretion disk state.We have checked if the uncertainties on R in in OBS5could be due to the lower statistics compared to 2008 bysimulating a set of data from the best fit model obtained for2008 but with a combined XIS0+XIS3 exposure time of 40 Fig. 10.
Unfolded best-fit model of the five Suzaku observations (XIS: black crosses; HXD/PIN: red crosses) and corre-sponding data/model ratios using eqppair for the continuum, a blurred reflection kdblur ⊗ reflionx for the reflectioncomponent and a multicolor disk diskbb for the soft X-rays.ks (i.e. of the order of the exposure time of OBS5) instead of210 ks. A fit of this simulated spectrum gives R in > R g and ξ ref = 70 +60 − . So if the OBS5 spectrum and line shapewere exactly the same as in 2008, we should have obtainedvalues of R in and ξ ref in agreement with those observedin 2008, even if the statistics of OBS5 is low. This is incontradiction with our results and this suggests that the accretion disk is in an intrinsically different state between2008 and 2011. As said previously, the best fit of OBS1 is not verysatisfactory and features are visible, especially in the softX-ray range, in the data/model ratio reported in Fig. 10.
The presence of absorption lines in the soft X-ray spectrumof GX 339-4 have already been reported in the literaturewhen the source was in an intermediate state transitingto the low/hard state (e.g. M04, Juett et al. 2006). Ifmost of these lines may be produced by the InterstellarMedium (Juett et al., 2006), a sizable contribution ofa few of them (like e.g. the Ne ix line at 0.922 keV)could be due to a local contribution from an intrinsicAGN-like warm-absorber perhaps produced by a disk wind.We have tested the presence of such absorber byadding an ionized absorber component (model absori of xspec ) in our fits. The fit of OBS1 improves significantly(∆ χ =102!) thanks to the addition of this spectral com-ponent. The corresponding parameters of the absorberare reported in Tab. 5 i.e. an hydrogen column density of N h,abs ∼ cm − and an ionization parameter ξ abs ∼ absori componentto the other observations (including 2008). We do not findany significant improvement. The best fit are also reportedin Tab. 5. The decrease of the signal-to-noise ratio mayhowever limit the correct detection of absorbing material .Then, we can have a rough estimate of the maximaldistance d between the absorbing material and the X-raysource by assuming that d is necessarily larger than theradial extension ∆ r of the absorber. Since N h,abs = n ∆ r (nbeing the density of the warm absorber) and ξ abs = L bol d n then d > ∆ r implies d < L bol ξ abs N h,abs . Using the best fitparameter values obtained for OBS1 i.e. L bol (cid:39) ergs − , ξ abs (cid:39)
300 erg cm s − and N h,abs (cid:39) cm − wefind that the absorber should be at a maximal distance of ∼ R g (assuming a 10 black hole solar masses) from thecentral X-ray source. This is of the order of the binaryseparation in GX 339-4 as estimated by Zdziarskiet al. (2004), thus agreeing with the fact that theabsorber may be produced in the inner parts ofthe binary system.
To have a qualitative idea of the ions potentiallyresponsible for the observed absorption features in OBS1,instead of absori we simply add gaussian absorptionlines, with width fixed to 0 eV (a more detailed work onthese absorption line is dedicated to a future work). Wefocus only on OBS1 since the other observations do notneed apparently the addition of absorption components.As previously said, a few absorption lines were alreadyobserved in GX 339-4 in past Chandra observations (M04),the most intense one being Ne ix at 0.922 keV. We thusadd an absorption gaussian line with energy in the range0.9-0.95 keV. The fit improves strongly with ∆ χ =61 for3 less degrees of freedom. The gaussian best fit energy Note that, despite the difference in the the soft banddue to the lack of a low energy roll-over when fittingthe continuum with a power law, we obtain consistentvalues of the absorber parameters when we use thismodel instead of eqpair
Ionized absorber
Obs. N h ξ abs ∆ χ χ /dof × +0 . − . +90 −
107 597/5132 0.4 +0 . − . +230 − < >
90 -1 520/5104 < − +3 . − . > < − Table 5.
Best fit parameters of the ionized absorber com-ponent. The ∆ χ is the χ variation compared to the bestfits reported in Tab. 4 . E abs =0.92 ± ix . The line has anequivalent width (EW) of ∼ × and 10 cm − (see M04 for the computation of the Neon column density).This agrees completely with the estimates measured byM04 and Juett et al. (2006) and this is still higher thanthe expected value from the ISM, suggesting also an originin the local environment of the X-ray binary. Luo &Fang (2014) reach the same conclusion for GX339-4 as well as for a sample of 11 other X-raybinaries . We have tested that the addition of such gaus-sian absorption line is not needed in the other observations.We have also tested the presence of (weak) blue-shiftedionized absorption lines from Fe
XXV and Fe
XXVI sincethey have been observed in a few microquasars and inter-preted as signature of fast ionized outflows. Their expectedEW are generally of the order of 10-20 eV (e.g. Ponti et al.2012). We do not detect such components is our data andfind only upper limit for their equivalent width of 5-10 eV.
4. Hints of disk recession
To be correct, when we talk about disk recession, we aretalking about the recession of the optically thick part ofthe accretion flow. It is possible that the accretion diskextends down to the inner more stable orbit but that,for different reasons (e.g. most of its accretion power isadvected or ejected), it does not radiate any more.While we do not find clear indications of disk recessionduring our monitoring, a few arguments are consistentwith this interpretation. First, and contrary to fits below10 keV, fits of the broad band spectra with blurred ionizedreflection give good constrains on R in for OBS1 and OBS2,of the order of 10 R g and ∼ R g respectively, but it alsoputs lower limits for OBS3 ( > R g ), OBS4 ( > R g ) andOBS5 ( > R g ). For the 2008 data set, our fit gives a lowerlimit R in > R g , in agreement with T09 and clearlysuggesting a disk recession (but see Fabian et al. (2014)for the limitations of the use of X-ray reflection toestimates the disk inner radius) . The fact that this re-cession was apparently stronger in 2008 could be due to the Note that this absorption line energy is well below the Siedge which is known to contaminate the XIS response around1.8 keV12.-O. Petrucci et al.: The return to the hard state of GX 339-4 as seen by Suzaku b) Fig. 11. a)
Best fit diskbb temperature vs eqpair softphoton temperature. The solid line is the linear best fit. Itas a slope of 1.9. The dot-dashed lines corresponds to the1 σ error on the slope i.e. 1.9 ± b) Flux of the diskbb component versus diskbb temperature. The black solid linecorresponds to the log-log linear best fit F diskbb ∝ T . diskbb and the dot-dashed lines to the 1 σ . The red solid line cor-respond to the best fit assuming a T diskbb law. c) Plotsof R proxy (see Sect. 4) versus the disk inner radius R in deduced from our fits (reported in Tab. 4) and assumingthe X-ray luminosity ∝ ˙ M α with α =2.5 (the results arevery similar for α =2 or 3). The dashed line corresponds to R proxy = R in . The blue points in each figure correspond tothe 2008 Suzaku observation.fact that the source was in a persistently faint hard statesince months. This could have let enough time for the in-ner accretion flow to evolve into the observed configuration.Our estimates of l h /l s from our fits also agree with thework of Sobolewska et al. (2011). Note however that weare not using exactly the same model as these authors.No reflection component was taken into account in theirwork but an iron line. We do not expect however thatthis would have any strong effect on our estimate of l h /l s compared to their methods given the low flux we foundin the reflection component (i.e. a factor 10 below thecontinuum flux, see Tab. 4). Thus we believe that ourresults can be safely compared to theirs. These authorsstudied the changes of the l h /l s ratio with the bolometricluminosity L bol in the hard states of GX 339-4 and GROJ1655-40. At luminosities of the order of ∼ l h /l s with L bol . At luminosities lower than ∼ L bol . According to these authors, this behavioris consistent with a scenario where seed photons changefrom cyclo-synchrotron, at the lowest luminosities, tothose from a (truncated) disk, at higher luminosities. Ourobserved increase of l h /l s from OBS1 to OBS5, i.e. witha decrease of L bol , suggests then that the emission of theaccretion disk is still dominating the cooling process of the hot corona.But then, the changes of l h /l s implies a variationof the disk-corona geometry. The increase of l h /l s fromOBS1 to OBS5 indicates a faster decrease of the softphoton flux compared to the corona heating power. Thissituation is naturally expected if R in increases from OBS1to OBS5. Note that the increase of R in would also implya diminution of the reflecting area and consequently afaster decrease of the reflecting component compared toprimary one. This is also in agreement with the decreaseof F ref /F eqpair as discussed in Sect. 3.3.1.We have reported in Fig. 11b the flux F diskbb of the diskbb component versus the diskbb temperature T diskbb .The log-log linear best fit (black solid line in Fig. 11b)gives F diskbb ∝ T . ± . diskbb i.e. a bit smoother than the Stefan-Boltzmann law in T expected in the case of an opticallythick accretion disk with fixed inner radius. This discrep-ancy could be explained by a variation of the hardeningfactor with the disk temperature (e.g. Salvesen et al. 2013).However, we can again interpret this result in term of vari-ation of the disk inner radius. Indeed, in an standard ac-cretion disk of inner radius R in , inner temperature T diskbb and accretion rate ˙ M we expect: T diskbb R in ∝ ˙ M . (1)
On the other hand, at the end of the outbursts, X-ray bina-ries are known to be in a radiatively inefficient state wherethe X-ray luminosity is proportional to ˙ M α with α ∼ − F /αeqpair can be used as a proxy for the accretion rate of our system.In consequence, the ratio R proxy = [ F /αeqpair /T diskbb ] should be a proxy of the disk inner radius R in . We haveplotted R proxy normalized to its value in OBS1 (andassuming α = 2 .
5) in Fig. 11c versus the best fit innerradius values R in obtained from our broad band fits.While the values of R proxy for the last 4 observations arealways larger than the one computed for OBS1, and thensupporting the scenario of the recession of the opticallythick part of the accretion disk, they are smaller than R in in most cases. This discrepancy may be due to the badestimate of the disk inner temperature T diskbb from ourfits due to the limited energy range of Suzaku in the soft X.
5. Disk-wind to disk-jet transitions?
In the recent context of outflow/wind signatures detectionin black hole X-ray binaries (e.g. Miller et al. 2006c; Pontiet al. 2012; D´ıaz Trigo et al. 2012) and their potential linkto the jet evolution during outburst (Miller et al., 2006d;Neilsen & Lee, 2009), the absorption features significantlydetected in OBS1 (and only in OBS1) is an interesting re-sults. Let’s recall that the re-ignition of the radio emissionwas detected just before the beginning of our campaign,peaking between OBS1 and OBS2, and interpreted as thebeginning of the jet-structure re-building while the sourceturned back to the hard state (e.g. C13). By analogy withthe observation of the evolution between a jet-dominated toa wind-dominated accretion flow during a hard-to-soft tran-sition in GRS 1915+105 (Neilsen & Lee, 2009), our resultscould correspond to the reverse situation i.e. the evolutionfrom a wind-dominated to a jet-dominated accretion flowduring a soft-to-hard transition.The absorption features in OBS1 agree with the pres-ence of Neon absorption lines, especially from Ne ix andits column density suggests that part of this line could beproduced locally. Interestingly, this line was detected in aChandra observation of GX 339-4 when the source was inan intermediate state transiting to the low/hard state whileOBS1 was observed just before the complete state transi-tion back to the hard state. The disappearance of the Ne ix in the other observations of the monitoring could suggestthat it is linked to a disk wind only present before the tran-sition to the hard state. The absence of detection of Fe xxv and Fe xxvi however seems to indicate that this wind, ifreally present in OBS1, is not highly ionized.
6. Concluding remarks
Our Suzaku monitoring of GX 339-4 at the end of its 2010-2011 outburst caught the source during its soft-to-hardstate transition. Simultaneous radio-OIR observationsshowed the recovery of the radio emission and were inter-preted as signature of the re-ignition of the compact jets(C13). The onset of the radio emission occured betweenOBS1 and OBS2 while the re-flare observed in OIR reachedits maximum close to OBS3 and OBS4. The Suzaku observations show a global fading of theX-ray flux during the monitoring. For the spectral fits, wefirst use phenomenological models with a simple powerlaw for the continuum. The addition of an iron line isstatistically needed in all the observations. Fits with a laor profile are statistically equivalent to fits with agaussian. Simultaneous fits of the 5 observations with a laor profile give a disk inclination angle of ∼
20 deg. Theconstraints on the inner radius agree with R in < R g inOBS1 and OBS2 at 90% confidence level. Due to the lowerstatistics, these constraints becomes < R g , for the 3last observations.The presence of a soft X-ray excess, above the 3-10keV power law extrapolation, is clearly visible in the twofirst observations and the addition of a multicolor diskcomponent improves statistically the fits. This is not thecase for the three last observations. The evolution of thedisk component (decrease of its inner temperature andtotal flux) agrees with a disk recession from OBS1 ( R in ∼ R g ) to OBS2 ( R in ∼ R g ). However, again, due tothe low statistics, we cannot confirm this recession for thelast observations. A comparison with the very long Suzakuobservation of GX 339-4 during a long faint hard state ofits 2007-2008 outburst (T09), in a flux state similar to ourOBS5, shows a relatively good agreement of the spectralshape between the two observations. The line shape ishowever inconsistent between the two pointings. In 2008the source was observed during a persistent, but faint,hard state while in 2011 it was clearly in the decreasingphase of the outburst. It is then possible that the accretionflow had more time to evolve into a truncated accretiondisk in 2008 compared to 2011.The use of a model including blurred ionized reflectionand thermal comptonisation continuum gives better fitsthan the simple power law + diskbb model usedpreviously, at least for OBS1 and OBS2. For the otherobservations both models give similar results. This modelgive constraints on R in in marginal agreement with adisk recession. Hints of such recession come also from theincrease of the l h /l s compacity ratio of the hot coronaall along the monitoring, and from the deviation of thedisk flux to the Stefan law in T . These hints have tobe taken with care however given the low statistics andthe complexity of the model that may lead to strongdegeneracy between parameters.Finally, signatures of ionized absorption seem tobe present at least in OBS1 but absent in the otherobservations. The radio re-ignition occurring in betweenOBS1 and OBS2 (see C13), we suggest that, duringthese two observations, the accretion flow may havetransited from a disk wind, an ubiquitous characteristicof soft states, and a jet, signature of the hard states.These absorption features may be the last signature ofthe disk wind before the transition to a jet-dominated state. Acknowledgments
The authors thanks the referee for a careful reading of themanuscript and for his/her comments that well improve itsquality. POP acknowledges financial support from CNES.
References