0705.4172/ms.tex
1: \documentclass[12pt]{emulateapj}
2: %\documentclass[12pt,preprint]{aastex}
3: \usepackage{epsfig}
4: \usepackage{graphicx}
5: \usepackage{amssymb}
6: \usepackage{lscape}
7: \usepackage{longtable}
8: \usepackage{natbib}
9: \usepackage{lscape}
10: %\usepackage{txfonts}
11: \bibpunct{(} {)} {;} {a} {} {,}
12: \newcommand*{\DISKBB}{\texttt{DISKBB}}
13: \newcommand*{\DISKPN}{\texttt{DISKPN}}
14: \newcommand*{\EQPAIR}{\texttt{EQPAIR}}
15: \newcommand*{\THCOMP}{\texttt{THCOMP}}
16: \newcommand*{\COMPTT}{\texttt{COMPTT}}
17: \newcommand*{\DBBTT}{\texttt{DBBTT}}
18: \newcommand*{\DBBEQPAIR }{\texttt{DBBEQPAIR}}
19: \newcommand*{\DBBTTPEG }{\texttt{DBBTTPEG}}
20: \newcommand*{\PEGPWRLW}{\texttt{PEGPWRLW}}
21: \def\sco{Scorpius X-1}
22: \def\msun{$M_{\odot}$}
23: \def\mdot{$\dot{M}$}
24: 
25: \shorttitle{Spectral Evolution of \sco}
26: \shortauthors{A. D' A{\'\i} et al.}
27: 
28: \begin{document}
29: 
30: \title{Broad-band Spectral Evolution of Scorpius X-1 along its Color-Color Diagram}
31: \author{A. D'A\'i\altaffilmark{1}, P. $\dot{Z}$ycki\altaffilmark{2}, 
32: T. Di Salvo\altaffilmark{1},R. Iaria\altaffilmark{1}, 
33: G. Lavagetto\altaffilmark{1},  N.R. Robba\altaffilmark{1}}
34: 
35: \altaffiltext{1}{Dipartimento di Scienze Fisiche ed Astronomiche, 
36: Universit\'a di Palermo, Archirafi 36, 90123 Palermo, Italy; email: dai@fisica.unipa.it}
37: \altaffiltext{2}{Nicolaus Copernicus Astronomical Center, Bartycka 18, 00-716 Warsaw, Poland}
38: 
39: \begin{abstract}
40:   We analyze a  large collection of RXTE archive  data from April 1997
41:   to August 2003  of the bright X-ray source Scorpius  X-1 in order to
42:   study the  broadband spectral evolution of the  source for different
43:   values  of  the inferred  mass  accretion  rate  by studying  energy
44:   spectra  from selected  regions in  the Z-track  of  its Color-Color
45:   Diagram.   A  two-component  model,  consisting of  a  soft  thermal
46:   component interpreted as thermal emission from an accretion disk and
47:   a  thermal Comptonization  component,  is unable  to  fit the  whole
48:   3--200 keV energy spectrum  at low accretion rates. Strong residuals
49:   in the highest energy band of the spectrum require the addition of a
50:   third component that can be  fitted with a power-law component, that
51:   could represent  a second thermal Comptonization from  a much hotter
52:   plasma,  or a hybrid  thermal/non-thermal Comptonization,  where the
53:   electrons in  the Comptonizing cloud have  a Maxwellian distribution
54:   of velocities with a power-law hard tail.  The presence of this hard
55:   emission in  \sco~ has been previously reported,  however, without a
56:   clear  relation with  the accretion  rate.  We  show, for  the first
57:   time, that there exists a  common trend in the spectral evolution of
58:   the source, where the spectral parameters change in correlation with
59:   the position of the source in the CD.  In particular, using a hybrid
60:   thermal/non-thermal Comptonization model (EQPAIR code), we show that
61:   the power supplied to the non-thermal distribution can be as high as
62:   half  of the  total  hard  power injected  in  heating the  electron
63:   distribution.  We  also found that  a small sample of  spectra, when
64:   the source resides  at the top of the FB can  also show intense hard
65:   X-ray emission.   We discuss the physical  implications derived from
66:   the  results of  our analysis,  with  a particular  emphasis on  the
67:   hardest part of the X-ray emission and its possible origins.
68: \end{abstract}
69: \keywords{accretion discs -- stars: individual: \sco\ --- stars: neutron
70: stars --- X-ray: stars --- X-ray: general --- X-ray: binaries} 
71: \maketitle
72: \section{Introduction}
73: \subsection{Scorpius X-1}
74: \sco~  is the brightest  persistent X-ray  source in  the sky  and the
75: first  identified X-ray  extra-solar  source \citep{giacconi62}.   The
76: X-ray source  is an old,  low magnetized neutron star  (NS), accreting
77: matter  transferred  through   Roche-lobe  overflow  from  a  low-mass
78: companion \citep[recently identified as an  M class star of $\sim$ 0.4
79: \msun,~][]{steeghs02}.  \sco~  is a  prototype  of  the  class of  the
80: Low-Mass X-ray Binaries  (LMXBs), and assuming a distance  of $2.8 \pm
81: 0.3$ kpc \citep{bradshaw99}, the source emits close to the theoretical
82: Eddington limit for a  1.4 \msun~ NS ($L_{Edd} \sim 2~\times~10^{38}~$
83: erg s$^{-1}$).\\
84: Based on the timing behavior of LMXBs in correlation with the position
85: of   a  given   source  in   the  X-ray   color-color   diagram  (CD),
86: \citet{hasinger89} grouped  these sources  into two categories:  the Z
87: sources  and the Atoll  sources.  The  former are  brighter, radiating
88: close to $L_{Edd}$, the latter  are less bright, emitting at 0.01--0.1
89: $L_{Edd}$.   Z  sources exhibit  the  classical three-shaped  branches
90: describing a  Z pattern in the  CD: the Horizontal Branch  (HB) at the
91: top of the Z track, followed by the Normal Branch (NB) and the Flaring
92: Branch  (FB)  at  the  bottom   of  the  pattern.   There  are  strong
93: indications \citep[e.g.][]{vrtilek91} suggesting  that what drives the
94: changing  in the  spectral and  temporal  properties of  LMXBs is  the
95: instantaneous  accretion rate  ($\dot{M}$), which,  for Z  sources, is
96: believed to increase  monotonically from the HB to  the FB. Similarly,
97: Atoll sources, display two  different spectral/timing states: the soft
98: and luminous  banana state, associated to higher  accretion rates, and
99: the hard  and less luminous  island state, associated to  lower \mdot.
100: This early and straightforward classification  has now moved to a more
101: complex picture, as the patterns  that Atoll and Z sources describe in
102: their CDs,  when displayed on long  timescales, appear to  be more and
103: more similar to each  other \citep{muno02, gierlinski02}, although some
104: observational  facts still  hold: Z sources  move on the  CDs on
105: shorter timescales, have higher luminosities and present a timing
106: phenomenology different from that of the Atoll sources.\\
107: \sco, as  all the sources  of its class, shows  several quasi-periodic
108: oscillations (QPOs) along  all the branches of its  CD: the horizontal
109: branch oscillations (HBOs), the  normal branch oscillations (NBOs) and
110: the  flaring branch  oscillations (FBOs).   In \sco~  NBOs,  with peak
111: frequencies in the  range 4.5--7 Hz, and FBOs,  in the frequency range
112: 6--25 Hz,  seem to be physically  related to each other  since the NBO
113: peak frequency  smoothly joins  the FBO peak  frequency as  the source
114: moves from  the NB to the  FB \citep{casella05}.  Van der  Klis et al.
115: (1996) reported the first observation of  HBOs in \sco~ at 45 Hz, with
116: an inferred  harmonic near 90 Hz;  the power spectrum can  also show a
117: pair of QPOs, whose frequencies  are in the range 800-1100 Hz, denoted
118: kHzQPOs,  that  shift  simultaneously  in frequency,  with  an  almost
119: constant  or  weakly  frequency-dependent peak  separation  \citep[see
120: e.g.][]{zhang06}.   The   physical  interpretation  of   these  timing
121: features is not unique and the related scientific debate is still open
122: \citep[see][~for a review]{klis04}.\\
123: Spectral  studies of  \sco~  have been  so  far not  so extensive  and
124: detailed  as  timing studies.   This  is in  part  due  to the  strong
125: brightness of  the source that actually prevents  its observation with
126: the  most  sophisticated  and  high-resolution X-ray  satellites  like
127: BeppoSAX and  ASCA in  the recent past  and Chandra and  XMM-Newton at
128: present.   Three  articles have  investigated  so  far the  broad-band
129: spectral  behavior of  the source,  using  data from  the Rossi  X-ray
130: Timing  Explorer (RXTE).   \citet{barnard03} used  data from  both the
131: Proportional  Counter Array  (PCA) and  the High  Energy  X-ray Timing
132: Experiment  (HEXTE), showing that  the spectrum,  in the  energy range
133: 2.5--50  keV,  can  be  fitted  by a  simple  two-component  model,  a
134: black-body  soft  component  and  a  cutoff power-law,  plus  a  broad
135: Gaussian line, and  interpreted these results in the  framework of the
136: so  called Birmingham model  \citep{church01}. In  this interpretation
137: the spectral emission consists of black-body emission from the NS plus
138: Comptonized   emission  from  an   extended  accretion   disk  corona.
139: \citet{bradshaw03} used  only PCA data  (in the 2.5--18.2  keV range),
140: adopting  a model  consisting of  a black  body emission  plus  a bulk
141: motion Comptonization component and a reflection component in the form
142: of a  broad Gaussian line.  \citet{damico01} studied  HEXTE data above
143: 20 keV, in order to test the presence of high energy X-ray emission in
144: the spectra of  the source.  Data were fitted,  using a bremsstrahlung
145: component,  to mimic  the  effects  of the  thermal  component, and  a
146: power-law component,  whenever the  statistics and clear  residuals in
147: the fit  required it.  In this  way no apparent  correlation was found
148: between the  presence of the hard  tail and position of  the source in
149: the CD, contrarily to what had been reported for other Z sources.\\
150: More recently,  \citet{disalvo06}, using INTEGRAL data  in the 20--200
151: keV energy band, detected a  hard X-ray tail, with photon index values
152: between  2 and  3.4, whose  intensity  decreased as  the source  moved
153: towards  higher accretion  rates.  Data  did  not show  evidence of  a
154: high-energy  cutoff up  to $\sim$  200 keV,  suggesting  a non-thermal
155: origin for the hard tail.\\
156: \sco~   is   also   an   interesting   source   of   radio   emission.
157: \citet{formalont01},  through an  extensive VLBA  monitoring campaign,
158: have shown that  the radio emission is composed  of a point-like radio
159: core emission at the position of  the X-ray source and of two opposite
160: radio  lobes, moving  through the  ISM with  relativistic  speeds $v/c
161: =0.45 \pm 0.03$, and with an angle of 46$^{\circ} \pm$6$^{\circ}$ with
162: respect to our line of sight.  It is also evident a connection between
163: the  phenomenon of  radio flare  at  the core  of the  source and  the
164: following flaring of  the lobes, so that it is  argued that the energy
165: production  of the radio  emission is  confined near  the NS  and only
166: afterward transported to the lobes via the working surface of a jet
167: \citep[see e.g.   the magnetohydrodynamical simulations of][]{kato04}.\\
168: These  results  stimulated  the  search for  the  possible  connection
169: between inflow  and outflow mechanisms working in  the violent regimes
170: of accretion onto a NS or  a black hole (BH). Energetic mass outflows,
171: collimated sometimes  in a  typical jet pattern,  can be  produced not
172: only  in  extragalactic X-ray  sources,  such  as  quasars and  active
173: galactic  nuclei (AGN),  but  also in  galactic  X-ray binary  systems
174: \citep[see~][~for  a  review]{fender02}.  It  has  been observed  that
175: galactic BH  binary systems  are able to  produce strong  jet emission
176: (sources  for which  a jet  has  been already  spatially resolved  are
177: usually   denoted  \textit{microquasars}),   with   radio-loud  states
178: associated  to  hard/low  X-ray   states,  alternated  to  periods  of
179: radio-quenching  during  soft/high   X-ray  states.   For  NS  systems
180: radio-loud episodes are generally  interpreted as a jet signature.  In
181: this context, all the Z sources seem to owe, under certain conditions,
182: a  radio-jet  nature.  Detections  of  radio-loud  states are  usually
183: associated with spectral  states of low \mdot, i.e.\ on  the HB of the
184: Z-track  \citep{fender00}, while  the  few radio  detections in  atoll
185: systems found  the sources residing in  the island state  of their CDs
186: \citep{migliari03}.
187: \subsection{Spectral properties of Z sources class} \label{introduction}
188: We rapidly  review in  this section the  spectral properties of  the Z
189: sources.  The class of the  Z sources display a homogeneous pattern of
190: spectral properties; their energy spectra can be usually decomposed as
191: follows:\\
192: \begin{itemize}
193: \item a  soft thermal  component, with characteristic  temperatures in
194:   the 0.5--1  keV range, interpreted  as thermal emission from  the NS
195:   surface or an optically thick, geometrically thin accretion disk;
196: \item  a  thermal  Comptonized  component, where  the  electron  cloud
197:   distribution has  a temperature  in the 2.5--3  keV range,  the soft
198:   seed photon  temperature is around 1  keV and the  optical depth can
199:   have  rather large  values  ($\tau \geq$  3);  this relatively  cold
200:   thermal  Comptonization is  thought to  take place  close to  the NS
201:   surface;  the soft  seed  photon temperature  generally exceeds  the
202:   highest  temperature reached  in  the accretion  disk,  so that  the
203:   source of  the soft  emission is probably  confined in  the boundary
204:   layer  between the  inner  edge of  the  accretion disk  and the  NS
205:   surface.   The high  optical depth  values associated  to  the cloud
206:   saturates   the  seed   photon  spectrum   close  to   the  electron
207:   temperature, so  that a quasi  Wien spectrum results.  A theoretical
208:   interpretation  of this  spectral  decomposition has  been given  by
209:   \citet{inogamov99}, where the boundary layer emission is expected to
210:   be radiation pressure  supported and its emission is  locally at the
211:   Eddington rate.  Change in luminosity  is attributed to a  change in
212:   the  emitting area rather  than in  its emissivity.   Phase resolved
213:   spectroscopy of  bright LMXBs \citep{revnivtsev05}  further supports
214:   this interpretation,  establishing in all  the analyzed Z  sources a
215:   common  cut-off, due  to  the boundary  layer  thermal emission,  at
216:   $\simeq$ 2.4 keV.
217: \item  a reflection  component,  often simply  modelled  with a  broad
218:   Gaussian line, in  the 6.4-6.9 keV range. In all  the Z sources this
219:   line  has  been  always  observed;  spectra from  BeppoSAX  or  RXTE
220:   observations  were not  able to  constraint the  shape of  the line,
221:   while  the  relative  width  was  in  the  range  0.1-1  keV.   High
222:   resolution spectra  in this energy  range have been obtained  so far
223:   only with  the BBRXT  \citep{smale93} and the  XMM-Newton satellites
224:   \citep{costantini02} for  Cygnus X-2; in both  observations the line
225:   was associated to highly  ionized iron (centroid energy $\simeq$ 6.7
226:   keV), while the line appears  to be intrinsically broad (FWHM $\sim$
227:   1 keV).  \citet{brandt94} pointed  out that the determination of the
228:   line  profile can be  a primary  diagnostic tool,  but, at  the same
229:   time,  kinematic and  relativistic effects  can greatly  distort the
230:   line from  the simple Gaussian  profile. Moreover, if  reflection is
231:   caused  by hard  X-rays  reflected  by a  cold  accretion disk,  the
232:   contribution of the Compton reflected continuum should be taken into
233:   account, when the iron  line emission is a considerable contribution
234:   to the total energy flux;
235: \item a power-law hard tail, which can contribute up to few percent to
236:   the total energy flux,  whose strenght usually varies in correlation
237:   with  the  position  of the  source  in  the  CD, namely  being  the
238:   strongest on  the HB,  gradually decreasing as  the source  moves to
239:   higher accretion rates and totally  fading in the FB \citep[e.g.  GX
240:   17+2,][]{disalvo00}.   The   photon  index  of   the  power-law  was
241:   generally  found in  the range  1.5--3 with  no evident  high energy
242:   cutoff up to energies of about 200 keV.
243: \end{itemize}
244: In this  article we  report a complete  investigation of  the spectral
245: properties  of \sco~  through an  extensive analysis  of  RXTE archive
246: data, indicating a clear connection  between position of the source on
247: the CD and spectral behavior.
248: \section{Data reduction and analysis}
249: The scientific payload of RXTE consists of three instruments, the PCA,
250: the HEXTE,  and the All Sky  Monitor (ASM).  The PCA  consists of five
251: co-aligned  Xenon  proportional  counter  units (PCUs)  with  a  total
252: effective area  of about 6500  cm$^2$. The instrument is  sensitive in
253: the energy range from 2 keV  up to 60 keV \citep{jahoda}, although the
254: response matrix is best calibrated in the energy range 3-22 keV.  Data
255: can be processed using  several different configuration modes; for our
256: analysis  we exclusively use  the \textit{Standard2}  mode, with  16 s
257: time resolution and 128 energy channels in the 2--60 keV energy range.
258: The   HEXTE  consists   of  two   cluster  of   four  NaI/CsI-phoswich
259: scintillation counters  that are sensitive from  15 keV up  to 220 keV
260: \citep{rothschild}.  We use the \textit{Standard Mode}, with 64 energy
261: channels, for the reduction and analysis of the HEXTE data. Background
262: subtraction  is done  by using  the source-background  rocking  of the
263: collimators.  We use HEXTE response matrices of 1999 August.\\
264: We  collected a  large amount  of RXTE  archive data,  discarding only
265: minor shorter  observations, from  April 1997 up  to August  2003.  We
266: present in Table~\ref{tab1} the  datasets we used for our analysis,
267: indicating  the associated  proposal number,  the starting  and ending
268: times of  each observation in Terrestrial Time,  and the corresponding
269: exposure times.\\
270: Data  have  been  processed  using the  standard  selection  criteria,
271: discarding data  taken during Earth occultations  and passages through
272: the South Atlantic  Anomaly.  We only used data from  PCU2 for the PCA
273: and  data from  Cluster A  for  the HEXTE  instrument. We  constructed
274: color-color   diagrams    (CDs)   of   the    source   by   extracting
275: energy-dependent   lightcurves  using   PCA  energy   channels  5--10,
276: 11--16,17--22 and 23--44,  with a 64 s bintime.   These channel ranges
277: correspond  to  the  energy  ranges 1.94--4.05  keV,  4.05--6.18  keV,
278: 6.18--8.32  keV, 8.32--16.26  keV  respectively.  We  define the  Soft
279: Color (hereafter SC) as the ratio of count rates in the 4.05--6.18 keV
280: and 1.94--4.05  energy bands, while  the Hard Color (hereafter  HC) as
281: the ratio  of count  rates in the  8.32--16.26 keV and  6.18--8.32 keV
282: energy ranges.   However, the channel  to energy conversion  depend on
283: the  period  of activity  of  the  satellite  (there are  5  different
284: instrumental Epochs), most of our data belong to Epoch 3 for which the
285: given  energy bands  are appropriate;  for  the Epoch  5 datasets  the
286: energy  ranges  are  slightly  shifted,  so that  this  results  in  a
287: consequent shift  on the HC  and on the  SC axes.  Because we  are not
288: interested in the  secular shifts of the track in the  CD, nor we want
289: to  accumulate spectra  taken at  different Epochs,  but to  perform a
290: statistical study  of the broad-band  spectral behavior of  the source
291: for different CD  positions in different datasets, we  did not perform
292: the energy dependent corrections to the colors.\\
293: We selected regions in the  CDs for each observation dataset, in order
294: to cover  a homogeneous part  of the Z-track  and to have at  the same
295: time a suitable statistics.  From these selections we derived the good
296: time intervals  (GTI) which we  used for extracting  the corresponding
297: spectra for the PCA and HEXTE data.\\
298: Given the high luminosity of the  source all the PCA spectra have been
299: dead-time corrected. Some observations  were performed with the use of
300: an offset pointing; in these cases we extracted the responses matrixes
301: of both the  HEXTE and the PCA instruments,  following the indications
302: given in the on-line  Data Analysis documentation pages.  We processed
303: and analyzed  data using version 6.0  of the FTOOLS  package suite and
304: version    11.3.1     of    Xspec.   
305: \section{The Data}
306: \input{tab1}
307: Hereafter  we shall refer  to a  collection of  close in  time pointed
308: observations  used to  extract one  CD by  using its  proposal number.
309: From datasets 20035 (and 30035) we have obtained two different CDs; we
310: will label  them, respectively, 20035A and 20035B  (30035A and 30035B,
311: see  the  {\textit Spectrum}  column  in Table~\ref{tab2}).   Datasets
312: 20035A, 20035B,  30036, 30035A,  30035B and 40020  belong to  Epoch 3,
313: while datasets  70014, 70015  and 80021 belong  to Epoch  5.  Datasets
314: 70014 and 70015 have been merged in one CD, because these observations
315: were close in time and the source pointing was the same.\\
316: The eight CDs that we have  extracted present the source mostly on the
317: NB/FB, as  these are the  states where the  source spends most  of the
318: time.   In  \sco~ the  HB  is  not  \textit{horizontal} but  rather  a
319: vertical continuation  of the NB  (a characteristic shared also  by GX
320: 17+2  and  GX 349+2,  the  so-called  Sco-like  sources).  For  visual
321: clarity  we will  refer to  the \sco~  CD pattern,  mainly  making the
322: distinction between the  left (or FB) track and the  right (NB and HB)
323: track,  and between  the top  (harder) parts  and the  bottom (softer)
324: parts of the tracks.\\
325: The  count  rates  associated   with  the  extracted  spectra  show  a
326: significant  trend, generally correlated  with the  HC; HEXTE  and PCA
327: count rates  are the highest  at the top  of the FB  and monotonically
328: decrease  towards the FB/NB  apex; following  the direction  bottom NB
329: $\rightarrow$  HB, the  trend is  inverted.  This  behavior  is mostly
330: stressed in  the HEXTE data (compare  the countrate in  spectrum 07 in
331: the top FB with spectrum 10  in the transition region FB/NB, for which
332: the rate is reduced by two thirds in the HEXTE, but only by one third
333: in the PCA).\\
334: \begin{figure*}
335: \centering
336: \begin{tabular}{l l l l}
337: \includegraphics[height=3.5cm, width=2.5cm ]{f1a.ps} &
338: \includegraphics[height=3.5cm, width=2.5cm ]{f1b.ps} &
339: \includegraphics[height=3.5cm, width=2.5cm ]{f1c.ps} &
340: \includegraphics[height=3.5cm, width=2.5cm ]{f1d.ps} \\
341: \includegraphics[height=3.5cm, width=2.5cm ]{f1e.ps} &
342: \includegraphics[height=3.5cm, width=2.5cm ]{f1f.ps} &
343: \includegraphics[height=3.5cm, width=2.5cm ]{f1g.ps} &
344: \includegraphics[height=3.5cm, width=2.5cm ]{f1h.ps} \\
345: \end{tabular}
346: \caption{  \small \linespread{1}
347: CDs extracted from RXTE datasets used in our spectral analysis. 
348: Different colors indicate different selected regions from which the CD-resolved
349: spectra were produced.
350: We label the spectra through a progressive numbering as shown in the CDs.}
351: \label{fig1}
352: \end{figure*}
353: In  four datasets  (20035A,  20035B, 40020  and  30036) the  satellite
354: pointed the source on axis;  in three datasets (30035A, 30035B, 80021)
355: an offset  pointing has  been used.  This  influenced the  high energy
356: statistics, where in  the same position on the  CD, an offset pointing
357: produced an HEXTE count rate lowered by almost one order of magnitude.
358: Spectra  extracted  from  datasets  70014  and  70015  have  also  low
359: statistics  in the  HEXTE band,  as  the integration  time is  shorter
360: compared with the other spectra.\\
361: We generally used a 3.0--22 keV  energy band for the PCA spectra and a
362: 20--200 keV energy band for the HEXTE spectra.  HEXTE channels from 28
363: to 63 have been rebinned, grouping four channels into one.  In some of
364: the  spectra that  we extracted  we found,  however, model-independent
365: mismatches in the overlapping energy  region between the HEXTE and the
366: PCA;  because   these  mismatches   are  of  instrumental   origin  we
367: occasionally restrict  these energy bands until these  features did no
368: longer impact on  our fit results.  In Table~\ref{tab2}  we report the
369: exact energy  bands used for  the PCA and  the HEXTE datasets  for the
370: selected spectra.\\
371: The choose to assign an appropriated systematic error for the PCA data
372: is an essential step in the analysis of the spectra. This operation is
373: not straightforward,  as the commonly used standard  candle, the Crab,
374: provides  in this  case only  a  partial calibration  tool.  The  Crab
375: countrate for  every energy channel below  $\sim$ 20 keV  is about one
376: order  of magnitude  less than  the \sco~  countrate, thus  making the
377: extrapolation  of  the Crab  based  systematic  error  to the  \sco~
378: spectra riskful.  The  mostly adopted view to assign  a 1\% systematic
379: error for  all the energy channels  has a strong impact  on the softer
380: part of the spectrum, where there are the mostly weighted channels; we
381: found that  in this case the  fit is strongly driven  by residuals in
382: the 3-6 keV range, leading to a modelization of the spectrum that does
383: not seem  physically plausible. The overall  effect is a  shift of any
384: soft  component outside  the RXTE  energy band,  and  unrealistic high
385: values for the extrapolated flux.\\
386: The only way to avoid the  strong driving effect of the first channels
387: is to arbitrarily  raise the systematics in this  range; we found that
388: assigning a  3\% systematic error for  channels between 3  and 3.9 keV
389: (namely the  first two PCA  channels available in our  spectra), while
390: leaving a 1 \% systematic error  for all the other channels, allows to
391: derive a  physically plausible scenario for the  spectral evolution of
392: the source, does not result  in extreme lower/higher values of reduced
393: $\chi^2$  for  every fit  performed,  allows  us  to exploit  all  the
394: available  energy band  and to  generally constrain  all  the spectral
395: components of the adopted models.  We occasionally still found in some
396: fits systematical residuals in the soft part of the spectrum, that are
397: more stressed  in the softer  spectra, whose origin can  be presumedly
398: instrumental and related to the xenon L-edge near 5 keV.  We choose to
399: ignore these  effects, in  order to not  overparameterize the  fit, as
400: they do not influence the determination of the spectral parameters and
401: do not impact the $\chi^2$ values of the fits.\\
402: No  systematic  error  has  been  associated to  the  HEXTE  data.   A
403: normalization constant is left free to vary between the PCU2 and HEXTE
404: spectra to take into account residual flux calibration uncertainties.
405: \section{Spectral  models} \label{spectralmodels}  As  pointed out  in
406: Section 1,  spectra of LMXBs  are usually described  as the sum  of at
407: least  two spectral components.   Because the  PCA energy  band starts
408: from  3  keV, it  is  not  possible to  constrain  the  effect of  the
409: photoelectric interstellar  absorption on the  source flux.  Following
410: \citet{christian97},  we  fixed  its  value to  3  $\times$  10$^{21}$
411: cm$^{-2}$, for each  fit performed.  To fit the  CDs selected spectra,
412: we tried  at first a  series of models  given by the combination  of a
413: soft  thermal  component,  such  as  a blackbody  or  multicolor  disk
414: blackbody  \citep[\DISKBB~  in  Xspec,~][]{mitsuda84}, and  a  thermal
415: Comptonized component.   For the  latter we, alternatively,  tried the
416: \texttt{COMPPS}          \citep{poutanen96},          {\texttt{COMPTT}
417:   \citep{titarchuck94}  and   \THCOMP~ \citep[not  included   in  the
418:   standard XSPEC  package, see~][]{zdziarski96} Comptonization models.
419:   For any model adopted, large residuals in the 6--10 keV energy range
420:   indicated  the presence of  iron features,  that we  simply modelled
421:   with a broad Gaussian line in the 6--7 keV energy range.  We noted a
422:   general tendency for any adopted model to have a very broad Gaussian
423:   line (with $\sigma \geq 0.8$  keV) and, generally, with centroids at
424:   energies $\leq  6.4$ keV.  In  order to avoid  unrealistically large
425:   Gaussian lines,  absorbing the underneath  continuum, we constrained
426:   the line to  have centroid energies only in  the 6.4--7.0 keV energy
427:   range, with a line width less than 0.8 keV, while the normalization
428:   of the line was a completely free to vary parameter.\\
429:   We found an adequate description  of the data, for energies below 30
430:   keV, using  a two-component  model given by  the combination  of the
431:   \DISKBB~  and  the  \COMPTT  components, a  modelization  that  was
432:   firstly introduced  by \citet{mitsuda84}, and  that is known  as the
433:   {\it   Eastern  model}.   Other   Comptonization  models   can  also
434:   statistically  equally well  represent the  spectral sample,  but we
435:   preferred to illustrate  our results using this Compton  model as it
436:   is one of the mostly used  in the LMXBs spectral analysis and allows
437:   an easy way  to compare our results with  results obtained for other
438:   sources.\\
439: Using   the  other   often  used   modelization,  the   Western  model
440: \citep{white86}  that predicts  a  $\sim$ 1--2  keV blackbody  thermal
441: emission and a Comptonization  component with disk soft seed-photon in
442: the 0.4--0.8  keV energy range \citep[see  e.g][]{barnard03}, we found
443: higher  $\chi^2$   values  for  each  spectrum   and  an  extrapolated
444: super-Eddington luminosity along each  part of the Z-track, that would
445: be dramatic for  some FB spectra which would overcome  by more that an
446: order of  magnitude the Eddington  limit, thus making  this assumption
447: not  realistic.  Moreover,  the power  dissipated in  the Comptonizing
448: disk corona would be constantly much greater than the power dissipated
449: as thermal emission on the NS-surface and this breaks our expectations
450: that the power  dissipated near the compact object  should be at least
451: of the same order of the  power dissipated along the disk, if the disk
452: is  very  close to  the  NS  \citep[see  also][~for a  more  extensive
453: discussion]{done02}.\\
454: The \DISKBB~ has two free parameters, the temperature expressed in keV
455: at the inner  disk radius (kT$_{DB}$) and a  normalization factor that
456: depends on the  inner radius of the accretion  disk, the distance from
457: the source and the inclination angle between the line of sight and the
458: normal to  the disk; the \COMPTT~  has four free  parameters: the soft
459: seed  photon temperature,  kT$_{0}$, the  electron temperature  of the
460: Comptonizing  cloud,   kT$_e$,  the  optical  depth,   $\tau$,  and  a
461: normalization constant.  We assumed  that the Comptonizing geometry is
462: spherical.  We will  refer to this two-component model  as the \DBBTT~
463: model.\\
464: As it is evident from Table ~\ref{tab2} the \DBBTT~ model is, however,
465: gradually unable  to satisfactorily fit  the whole energy  spectrum as
466: the  source resides  in zones  of  low inferred  accretion rate.   The
467: residuals, with respect to this  model, clearly indicated an excess of
468: flux at energies greater than 30 keV.  There is also a second group of
469: spectra  (namely spectra  7,8,12,13 and  26)  that gave  for the  same
470: reason an unsatisfactory high $\chi^2$ value. These spectra lie at the
471: top of the  FB in the datasets that have  the higher HEXTE statistics.
472: We will refer to this particular set as the top-FB spectra.\\
473: Because the mechanisms producing this hard excess in the Z sources are
474: not clear yet, in the following we will adopt two working hypothesises
475: to fit the broadband continuum of the source: a) as in the case of the
476: quasi saturated boundary layer  Comptonization, this component is also
477: the result of a thermal Comptonization process, but from a much hotter
478: plasma; b)  a broad band spectrum  is the result of  a hybrid electron
479: velocity  distribution, i.e.  a  thermal electron  distribution, which
480: produces  the   optically  thick  Comptonization,   with  a  power-law
481: non-thermal  tail which  is  responsible for  the high-energy  excess.
482: Because, for case a), the hot  plasma can only reside in regions close
483: to  the  central  compact  object, where  magnetic  and  gravitational
484: effects are the strongest, we assume that the boundary layer radiation
485: field is the primary source  of soft-seed photons also for this second
486: thermal Comptonization.   We note that this assumption  is also needed
487: in order to avoid an  overestimation of the contribution of steep hard
488: tail  at low  energies.  The  use of  an unbroken  power-law  at steep
489: photon indeces can give a major contribution in the lowest energy band
490: of the  spectrum so  that the fit  is driven by  low-energy residuals,
491: possibly   resulting   in   an   erroneous   continuum   modelization.
492: Thereafter, a low-energy cutoff at the seed-photon temperature must be
493: taken into account as the expected photon field spectrum is within the
494: instrumental range.\\
495: We added, thereafter, to our basic two-component model, a power-law in
496: order to mimic this  second Comptonization, multiplied by a low-energy
497: cut-off.  We set  the value of the low-energy  cut-off anchored to the
498: kT$_0$ value of the \COMPTT~  component, while we choose the power-law
499: to have a pegged normalization  expressed as flux of this component in
500: the 20.0--200.0  keV energy range  (\PEGPWRLW, in XSPEC).   We checked
501: for each  spectrum the  $\chi^2$ improvement by  adding a  high energy
502: cutoff, noting  that we could obtain  only lower limits  on the cutoff
503: energy,  generally in the  30--100 keV  range without  any significant
504: improvement in the  $\chi^2$. Thereafter we did not  use a high energy
505: cutoff.   We  will  refer  at  this  model  simply  as  \DBBTTPEG~  or
506: \textit{two Comptonizations} model.\\
507: To model the  possibility given in case b) we  also fitted the spectra
508: showing a  strong hard  excess (namely spectra  for which  the \DBBTT~
509: model gave a  null hypotesis probability less than  0.05 (it generally
510: corresponded to a reduced $\chi^2$ value of 1.4) with a model given by
511: the  sum of  a  thermal  disk component  plus  the recently  developed
512: thermal/nonthermal   hybrid   Comptonization   model  named   \EQPAIR~
513: \citep[see][~for  a full  description of  the model]{coppi99,coppi00}.
514: It  embodies   Compton  scattering,  $e^{\pm}$   pair  production  and
515: annihilation, $pe^{\pm}$ and  $e^{\pm}e^{\pm}$ thermal and non-thermal
516: bremsstrahlung,  and energy exchange  between thermal  and non-thermal
517: parts of the $e^{\pm}$  distribution via Coulomb scattering.  Selected
518: electrons  are accelerated  to suprathermal  energies and  the thermal
519: part of  the $e^{\pm}$ distribution can be  additionally heated.  This
520: model assumes a spherical  plasma cloud with isotropic and homogeneous
521: distribution of photons and  $e^{\pm}$, and soft seed photons produced
522: uniformly within the plasma,  with a thermal temperature of $kT_{BB}$.
523: The  properties of  the plasma  depend  on its  compactness, ${\it  l}
524: \equiv   \mathcal{L}   \sigma_T  /   (\mathcal{R}   m_e  c^2)$   where
525: $\mathcal{L}$ is the  total power in the source,  $\mathcal{R}$ is the
526: radius of the sphere and $\sigma_T$ is the Thomson cross section.  The
527: compactness  is divided  in  a hard  compactness,  ${\it l_h}$,  which
528: corresponds  to  the power  supplied  to  the  electrons, and  a  soft
529: compactness, ${\it  l_s}$, corresponding to the power  supplied in the
530: form of soft photons.  The compactnesses corresponding to the electron
531: acceleration and to the additional  heating of the thermal part of the
532: $e^{\pm}$  distribution  are  denoted  as ${\it  l_{nth}}$  and  ${\it
533:   l_{th}}$, respectively,  and ${\it l_h} \equiv {\it  l_{nth}} + {\it
534:   l_{th}}$.  The  non-thermal energy distribution of  the electrons in
535: the  plasma is  assumed to  be a  power law  replacing  the Maxwellian
536: exponential tail from  a gamma Lorentz factor $\gamma_{min}  = 1.3$ up
537: to a $\gamma_{max} = 1000$  with an energy index $G_{inj}$.  The total
538: Thomson optical depth  and the electron temperature of  the plasma are
539: computed self-consistently from the  assumed optical depth $\tau_p$ of
540: the background electron-proton plasma.\\
541: We  choose to  model the  thermal disk  emission using  as in  the hot
542: Comptonization  model the  \DISKBB~  component for  spectra in  HB/NB,
543: while    we     adopted    the    post-Newtonian     model    \DISKPN~
544: \citep{gierlinski99} in  the case  of the topFB  spectra, as,  in this
545: accretion state,  we expect the disk  to reach its  minimum inner disk
546: radius, close  to the  NS; the \DISKPN~  model takes into  account the
547: effective gravitational  potential in the neighborhood  of the compact
548: object  by   computing  the  emergent  disk  spectrum   using  a  more
549: appropriate  post-Newtonian  potential.   We  assumed  that  the  disk
550: reaches the last stable orbit at  6 R$_g$ and fixed this parameter for
551: the  fits performed,  while  we left  free  to vary  the maximum  disk
552: temperature (T$_{max}$)  and a normalization factor (K  = $M^2 cos(i)/
553: D^2 \beta^4 $, where $M$ is the central mass expressed in solar units,
554: $i$ is the inclination of the  disk, $D$ is the distance to the source
555: in kpc,  $\beta$ the color/effective temperature).  We  have left free
556: to  vary   for  the  \EQPAIR~  component   the  following  parameters:
557: $kT_{BB}$, $l_h/l_s$, $l_{nth}/l_{h}$,  $\tau_p$ and the normalization
558: factor.   The $G_{inj}$  was not  very sensible  to variations  in the
559: range between 1.5 and 3 for  spectra in the HB/NB, while for the topFB
560: spectra had  to be left as  a free to  vary parameter, as this  gave a
561: significant  decrease in  the $\chi^2$  when left  as a  free  to vary
562: parameter.   The  value  of  the  soft  compactness  ${\it  l_s}$  was
563: unconstrained by the  fit (it is mainly driven  by the pair production
564: rate, which  manifests itself with  the annihilation line  at energies
565: $\simeq$ 500 keV, obviously far beyond our energy limits), and we kept
566: it frozen  at the value  of 10 \citep[see  also][]{gierlinski99}.  All
567: the other parameters of the model were frozen at the default values.
568: We will refer to this modelization as the \DBBEQPAIR~ model, or simply,
569: the \textit{hybrid Comptonization} model. 
570: \section{Results and discussion}
571: 
572: \input{tab2}
573: We have examined 43 energy spectra of \sco~ in the 3.0--200 keV energy
574: band using RXTE  pointed observations from April 1997  to August 2003;
575: the spectra  have been extracted  from selected regions chosen  in the
576: X-ray CD. We produced a CD for  each close in time dataset in order to
577: avoid  shifts of instrumental  origin, and  repeated our  analysis for
578: each  CD  obtained  in  this   way,  thus  having  at  hand  a  robust
579: representation  of the  spectral evolution  of the  source in  all its
580: accretion states.\\
581: To fit the  spectra we firstly adopted a  two-component model given by
582: the  sum  of a  soft  thermal disk  emission  and  a hard  Comptonized
583: emission.  We checked  for the  presence of  harder X-ray  emission by
584: adding a high-energy power-law to  our basic model, while for a subset
585: of spectra  we adopted a hybrid  thermal/non-thermal Comptonization as
586: described in the previous section.\\
587: We find that  the basic two-component model \DBBTT~  can fit the 3-200
588: keV energy band  spectrum (fits for which we  obtained a $\chi^2_{red}
589: \leq 1.4  $) in  25 out of  a total  43 spectra; these  spectra mostly
590: belong to the FB and to the  NB. Hard tails, in this group of spectra,
591: are not significantly detected in  14 spectra (these are spectra 1, 9,
592: 14, 15, 17, 18, 22, 23,  24, 29, 34, 35, 39, 40, see Table~\ref{tab2})
593: when the  source lies in the  FB/NB apex of  its CD. For the  other 11
594: spectra, although  the addition  of a new  component is  not strictly
595: required by  the fit, we found  a general improvement  in the $\chi^2$
596: value.\\
597: On the contrary, we find that this model fails to describe the spectra
598: in  18 of  the  43 selected  spectra  (fits for  which  we obtained  a
599: $\chi^2_{red} \geq 1.4 $); the  fits generally getting worse and worse
600: from the  bottom of the FB to  the upper NB.  From  the residuals with
601: respect to this  model, it is clearly seen an  excess of hard emission
602: at  energies  above  30  keV.   We report  in  Figure~\ref{fig9}  four
603: representative deconvolved  spectra for the case in  which the \DBBTT~
604: model does not give an adequate description to the broad-band spectral
605: behavior, choosing three  spectra from the HB/NB zone  and one for the
606: topFB  zone: spectrum  06 (at  the top  of the  left track  of dataset
607: 20035A), spectrum 16 (at the top  of the left track of dataset 30036),
608: spectrum  31 (at  the top  of  the left  track of  dataset 40020)  and
609: spectrum 07 (at the top of the right track of dataset 20035B).\\
610: In the following  we will group the spectra  according to the relative
611: position in  their CDs, in: topFB  spectra (spectra 7, 8,  12, 13, 17,
612: 22, 26, 32, 37), FB spectra (1, 9, 14, 18, 27, 28, 33, 38), NB spectra
613: (2, 3, 10, 15, 19, 20, 23, 24,  29, 30, 34, 35, 39, 40) and HB spectra
614: (4, 5, 6, 11, 16, 21, 25,  31, 36, 41, 42, 43).  We caution the reader
615: that this  distinction is  based only on  the results of  our spectral
616: analysis for spectra taken in particular zones of the Z-track, without
617: any reference to  the temporal behavior of the  source. Although it is
618: clear,  looking just  at the  lightcurves, to  distinguish  within one
619: dataset the FB from the NB,  the distinction between the NB and the HB
620: is rather subtle, as the two  track mostly overlap.  For this reason a
621: HB  spectrum,  for this  analysis,  is to  be  considered  only as  a
622: simple label  for a  spectrum extracted from  the top part  of left
623: track of each CD.\\
624: 
625: \section{The hot Comptonization model}  
626: \subsection{Energetics}
627: In the following we will discuss  the fit results based on the adopted
628: model \DBBTTPEG. Although in some spectra this model over-parameterizes
629: the fit, the spectral behavior of the source and our conclusions would
630: not  change,  if  we  used  the best-fitting  values  of  the  \DBBTT~
631: modelization (we show in Table~\ref{tab3a} and~\ref{tab3b} the best-fit values of the
632: spectral  parameter, together  with associated  errors quoted  at 90\%
633: confidence level).  In some cases  we had to freeze some parameters in
634: order  to  make the  fit  physically  consistent  and avoid  solutions
635: without physical  meaning. In  spectra 7,  8, 12, 13,  17, 37,  38 the
636: electron temperature  had to  be fixed to  10 keV (a  value consistent
637: with what we found in other CDs  for the same CD zone; if free the fit
638: would give  a best-fit with kT$_0  >$ kT$_e$ which  is clearly wrong),
639: the  photon index  of the  \PEGPWRLW~ component  in spectra  where this
640: component  is  not  significantly  detected,  or  when  its  value  is
641: unconstrained by  the fit,  is frozen  to a reference  value of  2; the
642: \DISKBB~  parameters of  spectra 42  and  43 had  to be  fixed to  the
643: best-values as obtained  by the best-fit model \DBBTT~  (if free, this
644: component vanishes, while the  \COMPTT~ and \PEGPWRLW~ have comparable
645: flux values).\\
646: In this spectral decomposition we calculated
647: the total luminosity as follows:
648: \begin{equation}
649: L_x = 4 \pi D^2 (F_{TT} + F_{PEG} + a \xi N_{DB} T_{DB}^4)
650: \end{equation}
651: where $F_{TT}$  is the \COMPTT~  flux and $F_{PEG}$ is  the \PEGPWRLW~
652: flux  extrapolated in  the 0.1--200  keV energy  range; $\xi  = 1/(f^4
653: \kappa^2)$,   where    $f$   is   the    spectral   hardening   factor
654: \citep{shimura95},  is assumed  equal to  1.7, and  $\kappa  \sim 0.4$
655: corrects for the fact that the  radius at which is reached the maximum
656: temperature    is    greater   than    the    inner   disk    distance
657: \citep{vierdayanti06}.  $a$ is a constant dependent on the distance of
658: the source, which we assumed to be 2.8 kpc, and the inclination of the
659: disk, which we assumed  45$^{\circ}$. In Figure~\ref{fig1} we show the
660: contribution of the disk and of the Comptonized component to the total
661: flux.  The luminosities on both axis  are in Eddington units for a 1.4
662: \msun.\\
663: The  source span  a luminosity  range  between 0.9  L$_{Edd}$ and  2.4
664: L$_{Edd}$, thus being in  a super-Eddington state almost independently
665: of the position of the source on the CD.  We note that the only way to
666: drastically reduce  this range  is to suppose  a much  massive compact
667: object (the possibility of a massive  NS hosted in the \sco~ system is
668: also   discussed  in   \citet{stella99}  in   the  framework   of  the
669: relativistic  precession  model used  to  explain  the frequencies  of
670: quasi-periodic oscillations in NSs and BH systems) given that both the
671: distance and the inclination angle  are known with good accuracy (10\%
672: and 13\% relative error, respectively).\\
673: The disk flux  and the hard flux correlate as  expected with the total
674: flux,  although the disk  emission presents  a more  pronounced linear
675: relation,  while the hard  flux seems  to saturate  as the  total flux
676: reaches the  highest values. If we  assume a linear  dependency of the
677: two contributions  versus the total  emission we derive a  constant of
678: proportionality  equals to 0.56  for the  disk flux  and 0.44  for the
679: Comptonized flux.  When we look  at the ratio of the two contributions
680: along the Z-track  (see Figure~\ref{fig2}}, a 5\% error  on this ratio
681: is assumed, after  having tested the corresponding hard  and soft flux
682: values  for  different parameter  values  of  the spectral  components
683: inside the uncertainties errorbars as derived from the fits), there is
684: a general trend  to have a higher hard/soft ratio  as the source moves
685: from higher to  lower accretion states; this is  in agreement with our
686: expectation that, at lower accretion  rate the disk could be truncated
687: and the fraction of the  power dissipated in the boundary layer should
688: correspondently increase.  Spectra 7, 12, 13 and 32 do not follow this
689: trend; these spectra belong to  the topFB spectra, and in these cases,
690: we   are   possibly,  underestimating   the   disk  contribution,   or
691: overestimating the hard flux.
692: 
693: \begin{figure*}
694: \centering
695: \begin{tabular}{l l}
696: \includegraphics[height=6cm, width=4cm,  angle=-90]{f2a.eps} &
697: \includegraphics[height=6cm, width=4cm,  angle=-90]{f2b.eps} \\
698: \end{tabular}
699: \caption{ \small \linespread{1} Left panel: soft (disk) luminosity for
700:   the extracted  spectra vs.  total (0.1-200 keV)  luminosity.  Right
701:   panel: hard (Comptonized) luminosity  for the extracted spectra vs.
702:   total (0.1-200  keV) luminosity.  All luminosities are  in Eddington
703:   units for a 1.4 \msun~ NS. }
704: \label{fig2}
705: \end{figure*}
706: 
707: \begin{figure*}
708: \centering
709: \begin{tabular}{l l l l }
710: \includegraphics[height=4cm, width=2.8cm]{f3a.eps} &
711: \includegraphics[height=4cm, width=2.8cm]{f3b.eps} &
712: \includegraphics[height=4cm, width=2.8cm]{f3c.eps} &
713: \includegraphics[height=4cm, width=2.8cm]{f3d.eps} \\
714: \includegraphics[height=4cm, width=2.8cm]{f3e.eps} &
715: \includegraphics[height=4cm, width=2.8cm]{f3f.eps} &
716: \includegraphics[height=4cm, width=2.8cm]{f3g.eps} &
717: \includegraphics[height=4cm, width=2.8cm]{f3h.eps} \\
718: \end{tabular}
719: \caption{  \small \linespread{1}
720: Ratio of the hard/soft luminosities for all the CD extracted spectra.
721: In the x-axis the number refers to the spectrum as indicted in Table ~\ref{tab2}.
722: We assumed a 5\% error for ratio of the two fluxes.}
723: \label{fig3}
724: \end{figure*}
725: 
726: 
727: \subsection{Disk emission and thermal Comptonization}
728: We  present in  Figure~\ref{fig3}  the dependency  of  the inner  disk
729: radius  and  of  the  inner  disk  temperature,  plotted  against  the
730: hard/soft   luminosity   ratio.   It   is   shown,   both   in   Table
731: ~\ref{tab3a},~\ref{tab3b}  and  in  Figure ~\ref{fig4},  the  apparent
732: radius, derived without  considering important correction factors, due
733: to  the color/effective temperature  ratio, relativistic  effects, and
734: the non-zero torque boundary  conditions; moreover, it should be noted
735: that  these  corrective  factors  can  also  vary,  depending  on  the
736: accretion state.   The sum of all  these effects could  be a drastical
737: rescale of the measured inner radii up to almost an order of magnitude
738: higher \citep{merloni00},  but the general trend  that correlates this
739: parameter with the accretion state of the source should be
740: qualitatively preserved.\\
741: The  apparent inner  disk radius,  for spectra  in the  topFB  has the
742: lowest values,  $\leq$ 10 km, while  the average value  for NB spectra
743: and HB spectra is considerably higher (average value: $\sim$ 18 km for
744: the HB and  NB, $\sim$ 13.5 km for  the FB and $\sim$ 10.2  km for the
745: topFB).  The inner  disk temperature, on the other  hand, presents the
746: highest values, as one could  expect, on the FB, while the temperature
747: diminishes with the accretion rate. A small subset of FB spectra falls
748: out of the trend, but as  pointed out earlier, in this case, the exact
749: determination  of the  relative  contributions of  the two  components
750: could suffer  of systematic  uncertainties linked to  the particularly
751: luminous state of  the source.  On the FB the  disk temperature has an
752: average value of 2.12 keV, while on  the lower part of the FB sinks to
753: 1.73 keV. There  is not any appreciable difference  on temperature for
754: spectra taken on the NB and on the HB with an average value of 1.43
755: keV.\\
756: \begin{figure*}
757: \centering
758: \begin{tabular}{l l}
759: \includegraphics[height=6cm, width=4cm,  angle=-90]{f4a.eps} &
760: \includegraphics[height=6cm, width=4cm,  angle=-90]{f4b.eps} \\
761: \end{tabular}
762: \caption{ \small \linespread{1} Left  panel: inner disk radius vs. the
763:   hard/soft ratio.  Right panel:  temperature at the inner disk radius
764:   vs.  the  hard/soft  ratio.}
765: \label{fig4}
766: \end{figure*}
767: 
768: For   the   hard,   thermal   Comptonized,  component   the   accurate
769: determination of all the  spectral parameters was not always possible.
770: The soft-seed photon temperature  presents, for each spectra analyzed,
771: a substantially higher  value with respect to the  disk temperature at
772: the  inner  radius.   We  hence  propose that  the  soft  photons  are
773: originated from  the boundary layer/NS  surface, partially thermalized
774: with the softer photons of inner part of the accretion disk around the
775: NS. Moreover, calculating the radius of the soft-seed photons emitting
776: region $R_w$ from:
777: \begin{equation} 
778: R_w = 3 \times 10^4 d \sqrt{\frac{f_{bol}}{1+y}}/(kT_0)^2,
779: \end{equation}
780: where  $d$ is  the distance  to the  source in  kpc, $f_{bol}$  is the
781: \COMPTT~  bolometric flux and  $y$ (see  Eq. \ref{ypar}),  the Compton
782: parameter, we derived for our sample values in the 3--6 km range, that
783: clearly indicate  a rather small emitting region,  thus supporting our
784: identification.\\
785: \begin{figure*}
786: \centering
787: \begin{tabular}{l l}
788: \includegraphics[height=6cm, width=4cm,  angle=-90]{f5a.eps} &
789: \includegraphics[height=6cm, width=4cm,  angle=-90]{f5b.eps} \\
790: \end{tabular}
791: \caption{   \small  \linespread{1}   Left   panel:  soft   seed-photon
792:   temperature  kT$_{0}$  vs.  the  inner  disk temperature  kT$_{DB}$.
793:   Right  panel:  soft   seed-photon  temperature  kT$_{0}$  vs.   hard
794:   (\COMPTT+\PEGPWRLW)  luminosity.}
795: \label{fig5}
796: \end{figure*}
797: The CD  correlated changing of the thermal  temperatures is compatible
798: with  a scenario in  which the  position of  the source  on the  CD is
799: determined  by the  instantaneous  \mdot, and  higher accretion  rates
800: correspond to higher seed-photons  temperatures, resulting in a hotter
801: radiation field.  The soft seed-photon temperature has  a 2.2 keV,
802: 2.4 keV, 2.9 keV average values for spectra on the HB/NB, FB and topFB
803: respectfully.  When  plotted against each other,  the disk temperature
804: and the soft seed-photon temperature (see Figure~\ref{fig5}, left panel)
805: follow each other  quite closely, being these two  parameters the main
806: driving  physical quantities  related to  the accretion  state  of the
807: source.  The correlation  between the CD resolved spectra  and the two
808: temperatures shows, although in a qualitative representation, that the
809: link between  the spectral evolution  of the source and  its accretion
810: state is well motivated.\\
811: \begin{figure*}
812: \centering
813: \begin{tabular}{l l}
814: \includegraphics[height=6cm, width=4cm,  angle=-90]{f6a.eps} &
815: \includegraphics[height=6cm, width=4cm,  angle=-90]{f6b.eps} \\
816: \end{tabular}
817: \caption{ \small \linespread{1} Left panel: optical depth vs. electron
818:   temperature (all spectra).  Right  panel: optical depth vs. electron
819:   temperature (subset  of spectra on  the HB/NB).  Errorbars  show 1.0
820:   $\sigma$  errors.}
821: \label{fig6}
822: \end{figure*}
823: The  other  two spectral  parameters,  that  describe the  high-energy
824: curvature of  the Comptonized  component, the electron  temperature of
825: the Comptonizing  plasma (kT$_e$) and  the optical depth of  the cloud
826: $\tau$, are not always well constrained  by the fit.  As it is evident
827: from Figure~\ref{fig6}, for spectra at high accretion rate we obtained
828: high values for the electron  temperature (kT$_e \geq$ 10 keV) and low
829: values for the optical depth  ($\tau \leq$ 1).  As the sources resides
830: on  zones  of  lower   \mdot,  the  Comptonizing  cloud  substantially
831: thickens, while  the electron temperature  correspondently diminishes.
832: Plotting the  subset of the spectra in  the NB and HB  helps to better
833: visualize this  tendency (Figure ~\ref{fig6}, right panel).   It is to
834: be noted that  the determination of these two  parameters is partially
835: related with the hard  power-law component; although there are spectra
836: where the  \PEGPWRLW~ component is  not strictly necessary for  a good
837: fit,  its introduction  slightly  shifts the  the thermal  Comptonized
838: component to lower energies, thus giving a noticeable reduction in the
839: uncertainties of the Compton curvature.\\
840: In all the  spectra it is required a Gaussian  line at energies $\sim$
841: 6.4 keV. It  mostly appear rather broad, with a  line width that often
842: reaches our  constraint of 0.8 keV.  The  equivalent widths associated
843: with the line  are in 80-200 eV  range, and are to be  taken as rather
844: indicative given the low energy resolution of the PCA and the presence
845: of systematic residuals in this energy range.
846: 
847: \subsection{The hard tail behavior}
848: The presence of  a hard X-ray excess in the  extracted spectra of \sco~
849: is mostly evident  in 18  spectra from  a total of  43 CD selected spectra.
850: There is  little evidence  of the  presence of   hard X-ray  excess for
851: spectra lying in the  bottom part of the FB and of  the NB, namely for
852: spectra taken  near the  apex that connects  the left and  right track
853: that compose the V pattern.\\
854: In  all the  examined CD  patterns  we always  detect a  power-law
855: high-energy  excess,  dominant above  $\sim$  40  keV,  as the  source
856: resides at the top of the  left track, i.e.~ a portion of the diagram
857: that  we  tentatively  identified   with  the  HB,  and  which  should
858: correspond to the  lowest mass accretion rate.  This  group of spectra
859: are characterized by  the lowest values of inner  disk temperature and
860: soft seed-photon temperature; the  derived inner disk radii correspond
861: to  a disk  truncated at  about  10-15 R$_g$,  while the  Comptonizing
862: optical  cloud is  substantially  thicker with  respect  to any  other
863: spectrum in  other zones  of the  CD; the fraction  of the  total power
864: dissipated in the  Comptonizing corona has the highest  values, with a
865: ratio hard/soft luminosity well above unity.\\
866: \begin{figure*}
867: \centering
868: \begin{tabular}{l l}
869: \includegraphics[height=6cm, width=4cm,  angle=-90]{f7a.eps} &
870: \includegraphics[height=6cm, width=4cm,  angle=-90]{f7b.eps} \\
871: \end{tabular}
872: \caption{ \small  \linespread{1} Left panel: hard  tail luminosity (in
873:   the 20-200 keV energy range) vs.  the hard/soft ratio.  Right panel:
874:   the energy  index of the  \PEGPWRLW~ component vs.  its  20--200 keV
875:   luminosity (the subset of spectra for which the photon index was not
876:   costrained by the fit is excluded from this figure).}
877: \label{fig7}
878: \end{figure*}
879: Apart from  this consistent  group of HB/NB  spectra, we also  found a
880: minor group  of spectra (namely 5 spectra:  7, 8, 12, 13  and 26), for
881: which the  fit required a third high-energy  component.  These spectra
882: belong to 3 CDs (20035B, 30036 and 40020) and are spectra taken at the
883: top of  the FB.   Although these  spectra are located  in the  same CD
884: region, not all the CD that  display a topFB clearly show this excess.
885: This could be due to a statistical reason, as in all the other CDs the
886: HEXTE countrate  (see Table ~\ref{tab2})  drops of about one  order of
887: magnitude  with respect  to  these  3 CDs;  because  the countrate  is
888: essentially concentrated  in the 20--35  keV energy band, that  is the
889: range  that  constrains  the  curvature  of  the  thermal  Comptonized
890: component,  the  energy channels  in  this  energy  band with  a  high
891: statistic  will  drive  the  fit,  and  any  small  excess  above  the
892: exponential tail  of the  Comptonized component will  statistically be
893: more  accentuated.   However,  we  cannot  exclude  that  an  opposite
894: reasoning  is true,  i.e.   that we  are underestimating  instrumental
895: effects  that   appear  only  for  higher   count-rates  as  dead-time
896: corrections or poor background estimate, so that these particular hard
897: tails  are  artifacts of  an  incorrect  modelization  of the  thermal
898: Comptonization  curvature.   As we  have  no  apparent  way to  better
899: calibrate  our  spectra,  we  discuss  also  the  appearance  of  this
900: component for  this group of spectra.  Future  observations with other
901: high-energy satellites, such as INTEGRAL  or SUZAKU, will test if this
902: particular topFB  state of  the source is  also accompanied by  a high
903: energy excess or it is an HEXTE faked detection.\\
904: The hard excess  has no evident relation with  the total luminosity of
905: the source,  but this could be  expected since the  different zones of
906: the CD  track are not related  with the total  X-ray luminosity. Using
907: the pegged power-law, we directly derived from the normalization value
908: of this  component, the  flux in the  20.0-200.0 keV energy  band.  We
909: show in Figure~\ref{fig6}, left panel, the hard tails luminosity in the
910: 20--200 keV range  vs.  the hard/soft ratio.  From the  plot it can be
911: clearly seen  that the hardening of  the spectrum is  reflected in the
912: increasing  luminosity of  the  hard-tail component;  the CD  resolved
913: spectra are,  correspondently, disposed on  this plot: the  FB spectra
914: reside on  the bottom  left of  the diagram with  hard fluxes  $\leq 2
915: \times 10^{-10}$ erg s$^{-1}$, NB  spectra are harder and half of them
916: show significant fluxes ($\sim 4 \times 10^{-10}$ erg s$^{-1}$), while
917: the HB spectra, in the right part of the plot, show the strongest hard
918: X-ray emission (in the range 6--12 $\times 10^{-10}$ erg s$^{-1}$).\\
919: As pointed out in the previous section, we found for the topFB sample,
920: spectra with no  detectable hard X-ray emission, that  occupy the left
921: bottom  part of  this figure  and that  smoothly join  the correlation
922: between hard/soft ratio  and hard flux and spectra  that do not follow
923: this trend  and for  which there is  both in  the \COMPTT~ and  in the
924: \PEGPWRLW~ component  a hard  flux higher than  we expected.   From the
925: figure is  also evident  that one  of the most  luminous hard  tail is
926: found in a particular spectrum on  the topFB (namely spectrum 26 of CD
927: 40020);  the luminosity  of this  component  is related  to the  total
928: broad-band high  luminous state  of the source,  that reaches  in this
929: case the highest value of total luminosity (2.6 L$_{Edd}$) observed in
930: our sample.\\
931: On  the other  hand, the  values of  the photon  index present  a well
932: defined bimodal  distribution according to the position  of the source
933: on the CD. We find quite  flat power-laws, with index values less that
934: unity for all  the FB and topFB spectra, while for  the other group of
935: spectra, values  generally range between  1.5 and 2  (Figure~\ref{fig6},
936: right panel).
937: \subsection{The hybrid Comptonization model}  
938: Models of  hybrid Comptonization  have so far  been mostly  adopted to  
939: explain state transitions both  in black-hole candidate systems (as in  
940: Cyg  X-1, \citealt{gierlinski99}; GRS  1915+105, \citealt{zdziarski01};  
941: or  GX  339+4,  \citealt{wardzinski02})  and in  NS  system  \citep[GX  
942: 17+2,~][]{farinelli05}.\\
943: We  tried to  model the  spectra for  which we  detected a  hard tail,
944: making  the  hypothesis  that  both  the hard  tail  and  the  thermal
945: Comptonized component  were related to each other. In this way,  all the
946: broad-band evolution of the source  is covered in a self-consistent way
947: by only  two spectral components:  the thermal soft disk  emission and
948: the hard hybrid Comptonized component.  In table \ref{tab4} we present
949: the results  of our fits for  the group of  18 spectra, distinguishing
950: between the HB/NB group (top part of the table), for which we used the
951: \DISKBB~ component as in the  case of \DBBTTPEG~ modelization, and the
952: topFB spectra,  for which  we used the  \DISKPN~ component,  under the
953: assumption that for this state  the disk is not truncated, and reaches
954: the surface of the NS (we assumed that this happens at a distance of 6
955: R$_g$).
956: 
957: \begin{figure*}
958: \centering
959: \begin{tabular}{l l}
960: \includegraphics[height=6cm, width=4cm,  angle=-90]{f8a.eps} &
961: \includegraphics[height=6cm, width=4cm,  angle=-90]{f8b.eps} \\
962: \end{tabular}
963: \caption{   \small  \linespread{1}   Left   panel:  soft   seed-photon
964:   temperature kT$_{BB}$ vs.  inner disk temperature kT$_{DISK}$ (which
965:   corresponds to the kT$_{DB}$ value of the \DISKBB~ component for the
966:   HB/NB spectra and to the  kT$_{max}$ value of the \DISKPN~ component
967:   for  the topFB  spectra) for  spectra fitted  using  the \DBBEQPAIR~
968:   modelization.   Right  panel: the  soft  seed-photon temperature  is
969:   plotted against the hard compactness parameter.}
970: \label{fig8}
971: \end{figure*}
972: 
973: 
974: 
975: 
976: As in  the case  of the previously  adopted modelization,  the thermal
977: disk  temperature and the  soft-seed photon  temperature are  the main
978: driving  physical  parameters  that  determine  the  changing  of  the
979: spectral state of the source. We show in Figure~\ref{fig8} (left panel),
980: the correlation between the two temperatures: the topFB spectra occupy
981: the top  right part of  the plot, with  an average $kT_{BB}$  value of
982: 2.73 keV and  inner disk temperature $kT_{DB}$ of  1.97 keV, while the
983: set of  the HB-NB  spectra are  disposed along a  linear trend  in the
984: 1.2-1.8  keV energy  range for  the disk  temperature and  1.9-2.5 keV
985: range for  the soft seed-photon temperature. The  plot closely follows
986: the plot of Figure~\ref{fig5}, right panel.\\
987: The hard compactness ${\it l_h}$  results for all the examined spectra
988: only a small fraction of the soft compactness ${\it l_s}$, with values
989: of hard/soft ratio  ${\it l_h/l_s}$ below 0.2. The  HB/NB spectra span
990: almost an order of magnitude in the ${\it l_h}$ range, while the topFB
991: spectra are all grouped in a more narrow range of values, around ${\it
992:   l_h}$ = 0.6--0.8 (see Figure~\ref{fig8}, right panel).\\
993: The two groups of spectra  significantly differ both in the $\tau$ and
994: ${\it  l_{nth}/l_h}$ values  (it is  to  be noted,  however, that  the
995: optical depth reported from the fits has not the same physical meaning
996: of the classical  Thomson optical depth of the  Comptonizing cloud, as
997: it  more properly  refers to  the  optical depth  associated with  the
998: background  photon  radiation  field):  HB spectra  present  generally
999: higher values of optical depth (with an average value of $\tau_p \sim$
1000: 2.2)   and  a  hybrid   electron  distribution   (with  0.3   $<  {\it
1001:   l_{nth}/l_{h}} <  $ 0.8,  and an average  value of 0.55),  while the
1002: topFB spectra  have considerably lower  values of $\tau_p \leq  1$ and
1003: essentially non-thermal  spectra, with ${\it l_{nth}}  \simeq 1$.  The
1004: slope   of  the   high-energy   power-law  of   the  hybrid   electron
1005: distribution, $\Gamma_{inj}$ is consistent with  a value of 2, for the
1006: HB-NB spectra, while  for the topFB spectra its  value is considerably
1007: lower (between 0 and 1).\\
1008: 
1009: \begin{figure*}
1010: \centering
1011: \begin{tabular}{l l l l}
1012: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9a.eps} &
1013: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9b.eps} &
1014: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9c.eps} &
1015: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9d.eps} \\
1016: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9e.eps} &
1017: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9f.eps} &
1018: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9g.eps} &
1019: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9h.eps} \\
1020: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9j.eps} &
1021: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9k.eps} &
1022: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9l.eps} &
1023: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9m.eps} \\
1024: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9n.eps} &
1025: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9o.eps} &
1026: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9p.eps} &
1027: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9q.eps} \\
1028: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9r.eps} &
1029: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9s.eps} &
1030: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9t.eps} &
1031: \includegraphics[height=4cm, width=4.8cm,angle=-90 ]{f9u.eps} \\
1032: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9v.eps} &
1033: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9w.eps} &
1034: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9x.eps} &
1035: \includegraphics[height=4cm, width=1.8cm,angle=-90 ]{f9y.eps} \\
1036: \end{tabular}
1037: \caption{  \scriptsize \linespread{1}  Deconvolved spectra,  in  the E*f(E)
1038:   representation, and  contributions per  channel to the  $\chi^2$ for
1039:   four  spectra  of our  sample;  first  column:  spectrum 06,  second
1040:   column:  spectrum  16, third  column:  spectrum  31, fourth  column:
1041:   spectrum07.  Deconvolved spectra (upper panels) and contributions to
1042:   the  $\chi^2$ (lower  panels) from  the best-fit  using  the \DBBTT,
1043:   \DBBTTPEG~ and  \DBBEQPAIR~ models for  the first, second  and third
1044:   row  respectively.  In the  spectral  decomposition  the green  line
1045:   component  shows  the  disk   emission,  the  blue  line  shows  the
1046:   Comptonized emission, red line the Gaussian line, the thin blue line
1047:   the \PEGPWRLW~ component.}
1048: \label{fig9}
1049: \end{figure*}
1050: 
1051: \section{Conclusions}  
1052: We outline  our main conclusions  from an extensive analysis  of \sco~
1053: RXTE  observations  in the  3--200  keV  energy  range: we  observe  a
1054: spectral  evolution of  the source  that is  clearly dependent  on the
1055: position of the source along  its CD track; the spectral decomposition
1056: consists of four different spectral components that we interpret, from
1057: the softest  to the  highest X-rays respectively,  as follows:  a soft
1058: thermal component  from an optically  thick accretion disk,  a thermal
1059: Comptonization from  the boundary layer, a  reflection component which
1060: mainly manifests  in a broad Gaussian  line at 6.4 keV  and a variable
1061: hard excess above 30 keV, mostly present at low \mdot.\\
1062: We have  shown that  the CD correlated  variations of the  two thermal
1063: temperatures (disk temperature  and soft seed-photon temperature), the
1064: trend  in   the  hard/soft  ratio,  the  Compton   thickening  of  the
1065: Comptonizing  cloud  and  the  apparency  of  significant  hard  X-ray
1066: emission on  the HB/NB  are the main  spectral characteristics  of the
1067: source.\\
1068: We paid particular attention to  the modelization of the hardest X-ray
1069: component, which is of primary importance for the understanding of all
1070: the continuum emission, as it  has a major impact on the determination
1071: of  all the other  spectral parameters.   To fit  this hard  excess we
1072: modelled  our  spectra using  a  phenomenological  component, i.e.   a
1073: power-law  with a low  energy exponential  cut-off at  the seed-photon
1074: temperature   and   a  self-consistent   physical   model,  a   hybrid
1075: thermal/non-thermal Comptonization  code.  In  both cases we  found an
1076: adequate description of the spectra for each source state, and the two
1077: modelizations are statistically equivalent.\\
1078: The use  of simple  power-laws to  fit the hard  tails in  the Z-class
1079: sources, but also in  some atolls \citep{tarana07,fiocchi06}, has been
1080: largely used in the past, given  the lack of broad-band coverage up to
1081: the MeV energies, and a  poor understanding of the physical origins of
1082: this component. Contrary to what previously reported \citep{damico01},
1083: the presence of the hard  tail $is$ related to the broad-band spectral
1084: evolution of the source as inferred  from the source position on the Z
1085: track of its CD: the presence  and the values of the photon indexes in
1086: \sco~ are consistent with the results from fits to spectra on the same
1087: zone  of the  CD for  the  other Z  sources \citep{asai94,  disalvo00,
1088:   disalvo01, disalvo06};  at the same  time, the flux  contribution to
1089: the total  energy output of  the sources is  of the same order  and is
1090: anticorrelated to  the inferred mass accretion rate.   This is clearly
1091: shown  in  the  case of  the  CD  20035A,  where  the hard  tail  flux
1092: monotonically increases as the source moves from the bottom to the top
1093: of its NB/HB.\\
1094: The only difference,  between \sco~ and the other  similar sources, is
1095: constituted by rather flat hard tails that we found when the source is
1096: at the top  of the FB and  which has never been observed  in other NSs
1097: sources;   however,  while  for   every  CD   that  we   obtained,  we
1098: systematically found a hard tail as  the source was at the top of left
1099: V track, the detection on the  FB is presumedly dependent on the HEXTE
1100: statistics.   Past  surveys with  the  use  of scintillation  counters
1101: on-board on  balloons reported the unusual flattening  of the spectrum
1102: of  \sco~ above  40  keV  \citep{agrawal71}, as  well  as more  recent
1103: surveys \citep{manchanda06};  it would be  interesting to definitively
1104: assess the existence of such component; from our analysis, we can only
1105: conjecture that  this component, could be  a signature, on  the FB, of
1106: the trespassing  of the Eddington  limit on the interface  between the
1107: accretion  disk and the  board of  the boundary  layer, followed  by a
1108: violent expulsion of part of the accreting matter.\\
1109: Hard tails on  the HB could have  a thermal origin, as in  the case of
1110: the thermal Comptonized component that dominates the spectrum at lower
1111: energies (less  than 20 keV); in  this case the power-law,  as long as
1112: the optical depth is low and  the electron temperature is not too high,
1113: is  a  good  approximation  of  the Comptonized  spectrum,  whose  $y$
1114: parameter
1115: \begin{equation} \label{ypar}
1116: y = \frac{4 k T_e}{m_e c^2} \times (\tau + \tau^2)
1117: \end{equation}
1118: is related to the photon index of the power-law by the following relation:
1119: \begin{equation}
1120: \alpha = - \frac{1}{2} + \sqrt{\frac{9}{4}+\frac{\pi^2}{3y}}.
1121: \end{equation}
1122: In the case  of \sco~ spectra on the HB/NB,  we derived photon indexes
1123: in  the 1.8--2.2  range, that  imply a  $y$ Compton  parameter  in the
1124: 0.6--1.1 range; a $\tau = 1$  thick plasma would then require a $\sim$
1125: 50 keV thermal  plasma, while a $\tau = 2$, on  the contrary, a $\sim$
1126: 15  keV  electron  temperature.   We  did not  find  any  evidence  of
1127: high-energy cut-off,  obtaining only lower  limits in the  50--100 keV
1128: energy range,  and this constraints the  optical depth to  be less than
1129: one while the electron temperature  would be greater than 50 keV.  The
1130: formation of hot zones, or blobs, of very hot plasma, possibly powered
1131: by  episodes of magnetic  reconnection above  the disk,  constitutes a
1132: plausible scenario \citep{haardt94}; this model has been also proposed
1133: in the case of the  hard/low states of BH candidates \citep{malzac04},
1134: which share with  the bright Z-sources systems, from  a spectral point
1135: of  view,  similar values  of  the hard  photon  indexes  and the  low
1136: accretion state, while they  naturally differ from the strong boundary
1137: layer emission present in the NSs sources and absent in the BH case.\\
1138: A thermal origin, on the other hand, for the topFB hard tails seems to
1139: be  excluded, as  it would  require  unrealistic large  values of  the
1140: electron temperature.\\
1141: Our analysis has shown  that this scenario is statistically equivalent
1142: to a  hybrid Comptonization model,  where the non-thermal  fraction of
1143: the power  injected in the electron  heating counts up to   half of
1144: the total  injected power.  The non-thermal fraction  is significantly
1145: higher with respect  to the value found by  \citet{farinelli05} in the
1146: case of GX 17+2.  The  non-thermal fraction for the topFB spectra must
1147: be distinguished from the non-thermal spectra of the BH sources, as in
1148: the case of \sco, the hard  luminosity is a small fraction of the soft
1149: luminosity, that dominates the 10--30 keV energy spectrum. Most of the
1150: soft photons do not Compton  interact with the electron cloud, because
1151: of the low optical depth values.\\
1152: Another  possible  physical  mechanism   to  explain  the  hard  X-ray
1153: variability      is      the      bulk      motion      Comptonization
1154: \citep[e.g.][]{psaltis01},  present in  systems with  a  high velocity
1155: radial accretion flows.  This radiation mechanism, used to explain the
1156: hard  power-law component  in BH  systems spectra  in  their soft/high
1157: states, is,  however, not able to  account for the the  absence of any
1158: cut-off up  to energies  of 0.5  MeV and for  the hard  photon indexes
1159: observed ($\Gamma  \leq$ 2.0)  \citep{niedzwiecki06}.  In the  case of
1160: bright NS systems,  the high radiation pressure of  the inner boundary
1161: layer constitutes  a strong barrier  for any incoming  convergent bulk
1162: motion flow.   On the other side,  because jets have  been observed in
1163: the radio from \sco, bulk  motion Comptonization inside the jet can be
1164: an  important  production mechanisms  also  of  hard X-ray  radiation.
1165: Theoretical  spectra  and  energetic  contribution  according  to  the
1166: relevant physical parameters involved in the case of a strict coupling
1167: with  inflow  (i.e.   accretion  to  the compact  object  through  the
1168: formation  of  an accretion  disk)  and  outflow  (i.e. jets),  are  a
1169: promising  way to cover  in a  self-consistent way  all the  phases of
1170: accretion.  Details about the geometry of the scattering media, amount
1171: of reflection on the cold disk and dependence from the accretion state
1172: of the  source are yet to  be fully explored.  Moreover,  in this case
1173: synchrotron  emission  of  soft  photons  by a  beamed  population  of
1174: relativistic electrons  becomes a  competitive source of  photons with
1175: respect to the thermal soft  photon emission of the boundary layer and
1176: accretion disk.  An attempt to explicitly compute a jet spectrum, from
1177: radio to hard X-rays, has been recently proposed by \citet{markoff05},
1178: where  the jet  base  subsumes  the role  of  the static  Comptonizing
1179: corona; spectral fits in the case of BH systems (namely Cyg X-1 and GX
1180: 339-4) in hard  states are consistent with this  scenario, but BH soft
1181: states  and NS  systems spectra  need  yet to  be tested  in order  to
1182: understand the limits of validity of the jet model.\\
1183: Questions to be further addressed in future observations are: the exact  
1184: shape and  contribution of the soft  component below the  3 keV range,  
1185: the investigation with good spectral resolution in the 6--10 keV range  
1186: of  the  reflection features,  the  extension up  to  0.5  MeV of  the  
1187: spectral  coverage  in  order  to constrain  the  physical  parameters  
1188: characterizing   the   non-thermal   electron  distribution   of   the  
1189: Comptonizing plasma. The latter  point is well within the capabilities  
1190: of the recently launched SUZAKU satellite, so that future observations  
1191: in this direction can be a stringent test to our conclusions.  
1192: \bibliographystyle{apj}
1193: \bibliography{references}
1194: \clearpage
1195: \input{tab3}
1196: \input{tab4}
1197: \end{document}
1198: