A persistent ultraviolet outow from the accretion disc in a transient neutron star binary


 All disc-accreting astrophysical objects also produce powerful disc winds and/or jets. In compact binaries containing neutron stars or black holes, accretion often takes place during violent outbursts. The main disc wind signatures seen during these eruptions are blue-shifted X-ray absorption lines. However, these signatures are only observed during "soft states", when the accretion disc generates most of the luminosity. By contrast, optical wind-formed absorption lines have recently been detected in "hard states", when the luminosity is dominated by a hot corona. The relationship between these disc wind signatures is unknown, and no erupting compact binary has so far been observed to display wind-formed lines between the X-ray and optical bands, despite the many strong resonance transitions in this ultraviolet (UV) region of the spectrum. In turn, the impact of disc winds on the overall mass and energy budget of these systems remains a key open question. Here, we show that the transient neutron star X-ray binary Swift J1858.6-0814 exhibits wind-formed, blue-shifted absorption features associated with C IV, N V and He II in time-resolved, UV spectroscopy obtained with the Cosmic Origins Spectrograph on board the Hubble Space Telescope during a luminous hard state. In simultaneous ground-based observations, the optical H and He I lines also display transient blue-shifted absorption troughs. By decomposing our UV data into constant and flaring components, we demonstrate that the blue-shifted absorption is associated with the former, which implies that the outflow is always present. The joint presence of UV and optical wind features in the hard state reveals a multi-phase and/or spatially stratified evaporative outflow from the outer disc. This type of persistent mass loss across all accretion states has been predicted by radiation-hydrodynamic simulations and is required to account for the shorter-than-expected outburst durations.

All disc-accreting astrophysical objects also produce powerful disc winds and/or jets. In compact binaries containing 41 neutron stars or black holes, accretion often takes place during violent outbursts. The main disc wind signatures seen 42 during these eruptions are blue-shifted X-ray absorption lines. However, these signatures are only observed during 43 "soft states", when the accretion disc generates most of the luminosity 1 . By contrast, optical wind-formed absorption 44 lines have recently been detected in "hard states", when the luminosity is dominated by a hot corona 2 . The relationship 45 between these disc wind signatures is unknown, and no erupting compact binary has so far been observed to display 46 wind-formed lines between the X-ray and optical bands, despite the many strong resonance transitions in this ultraviolet 47 (UV) region of the spectrum 3 . In turn, the impact of disc winds on the overall mass and energy budget of these systems 48 remains a key open question. Here, we show that the transient neutron star X-ray binary Swift J1858.6−0814 exhibits 49 wind-formed, blue-shifted absorption features associated with C IV, N V and He II in time-resolved, UV spectroscopy 50 obtained with the Cosmic Origins Spectrograph on board the Hubble Space Telescope during a luminous hard state. In 51 simultaneous ground-based observations, the optical H and He I lines also display transient blue-shifted absorption 52 troughs. By decomposing our UV data into constant and flaring components, we demonstrate that the blue-shifted 53 absorption is associated with the former, which implies that the outflow is always present. The joint presence of UV 54 and optical wind features in the hard state reveals a multi-phase and/or spatially stratified evaporative outflow from the 55 outer disc. This type of persistent mass loss across all accretion states has been predicted by radiation-hydrodynamic 56 simulations 4 and is required to account for the shorter-than-expected outburst durations 5, 6 .

58
On October 2018, the Neil Gehrels Swift Observatory (Swift 7 ) detected a bright new X-ray binary transient, Swift 59 J1858.6−0814 (hereafter J1858) 8 . Multi-wavelength observations quickly led to the discovery of radio, optical and near-60 ultraviolet (UV) counterparts 9-11 . The detection of thermonuclear runaway explosions in X-rays (Type I X-ray bursts) 61 established that the accreting object is a neutron star located at a distance of about 13 kpc 12 . The system was also found to 62 undergo eclipses, implying a nearly edge-on viewing angle with respect to the disc (i 70 • ) and revealing the orbital period to 63 be P orb ≃ 21.3 h 13 .

64
J1858 displayed extreme variability during its outburst in all energy bands, with the X-ray luminosity changing by 1-2 orders 65 of magnitude on time-scales of seconds (see Figure 1-a) [13][14][15][16] . The X-ray spectrum consisted of a heavily absorbed thermal 66 accretion disc component plus a very steep non-thermal power law tail (photon flux N ph (E) ∝ E −Γ , with Γ < 1) 17, 18 . Both 67 the peculiar X-ray spectrum and spectacular variability are reminiscent of those seen during the outbursts of the well-studied 68 black-hole X-ray binaries V404 Cyg and V4641 Sgr, which are thought to be a consequence of accretion at super-Eddington 69 rates 19-22 . 70 In order to shed light on the accretion and outflow processes associated with the outburst, we carried out strictly simultaneous, 71 time-resolved observations across the electromagnetic spectrum on August 6, 2019 around 00 (UTC). One of our primary goals 72 was to search for outflow signatures in the far-ultraviolet (far-UV) band, since this region contains several strong resonance lines 73 that are very sensitive to the presence of warm, moderately-ionized intervening material.Therefore, the timing of this campaign 74 was centered on far-UV spectroscopic observations with the Hubble Space Telescope (HST). Simultaneous optical spectroscopy 75 was obtained at both the Very Large Telescope (VLT) array and the Gran Telescopio de Canarias (GTC). Additional information 76 about the campaign is provided in the Extended Data section.

77
In line with data obtained at other wavelengths 10, 13, 23 , the far-UV light curve exhibits dramatic flaring activity (Figure 1-b).

78
The X-ray, far-UV and optical variability are clearly correlated, with any lags between these time series being 1 s (Vincentelli 79 et al. in prep). This suggests that the multi-wavelength flaring is driven by a variable central X-ray source.

80
The presence of a large, strongly irradiated accretion disc is the key requirement for a thermally-driven outflow 24-26 , while 81 high inclinations tend to strengthen wind-formed absorption features 1, 4, 27 . All of this makes J1858 an ideal candidate for 82 displaying clear observational outflow signatures. In addition, X-ray spectroscopy of the source obtained earlier in the same 83 outburst revealed redshifted N VII emission, tentatively suggesting that the blue wing of this line was absorbed in an outflow, 84 even though the system was still in the hard state during these observations 28 . Time-resolved optical spectroscopy also revealed 85 clear, but highly variable P Cygni wind features in Hα and He I 5876 Å during the bright hard state 2 ( Figure 2). 86 Figure 3 shows the time-averaged far-UV spectrum we obtained with HST in the hard state. The spectrum is rich in both 87 absorption and emission lines that span a wide range of ionization states. Most of the low-ionization absorption lines are 88 centered at or near the rest wavelength of the relevant transition, with most of these lines not being intrinsic to the system 89 but rather due to interstellar absorption along the line of sight. However, at least two emission lines -N V 1240 Å and 90 C IV 1550 Å -show clear evidence for associated blue-shifted absorption. Since these species are associated with temperatures 91 of T ≃ a few × 10 4 K, their presence unambiguously establishes the existence of a warm and moderately ionized outflowing  Moreover, all strong emission lines in the spectrum -which includes the Si IV 1400Å doublet resonance line and the He II 1640 Å 95 2/12 recombination feature -show evidence for a slight red-shift or a red-skew, suggesting that they are also affected by blue-shifted 96 wind absorption.

97
As shown in the insets of Figure 3, the blue edges of the far-UV absorption features extend up to ≃ −2000 km s −1 , similar 98 to the wind speed inferred from the optical data. However, the far-UV absorption troughs are considerably deeper than those in 99 the optical, which rarely fall below 90% − 95% of the continuum. This is likely because most of the strong far-UV lines are 100 associated with strongly scattering resonance transitions, whereas the optical features are associated with recombination lines 101 that connect two excited levels. Very high (column) densities are required in order for such recombination lines to produce 102 absorption. On the other hand, sensitivity of far-UV resonance lines to intervening material makes this waveband particularly 103 valuable for studying outflows 29 .

104
In order to establish if the far-UV wind signatures are always present -or are instead associated with the strong flaring suggests that the outflow is, in fact, always present, but that its signatures may sometimes be swamped by the flaring component 111 (in which these signatures are absent). The same effect may be responsible for the transience of the blue-shifted absorption seen 112 in the optical data, especially considering how weak these features are (c.f. Figure 2).

113
The presence of detectable blue-shifted absorption associated with the UV resonance lines (e.g. N V 1240 Å, C IV 1550 Å) 114 implies that the optical depth in these transitions must be significant. This, in turn, requires minimum column densities 115 for the relevant ions, which can be cast as approximate lower limits on the mass-loss rate carried away by the outflow 116 (see Methods for details). Conservatively assuming ionization fractions of f = 1 for both C 3+ and N 4+ , these limits are 117Ṁ wind 2 × 10 −11 M ⊙ yr −1 for N V 1240 Å andṀ wind 3 × 10 −12 M ⊙ yr −1 for C IV 1550 Å. The actual ionization fractions 118 may be considerably lower, and the mass-loss rate correspondingly higher.

119
The apparent time-averaged X-ray luminosity during the flaring hard state in which we observed J1858 was L X ≃ 0.01L Edd , 120 although individual flares appear to have reached super-Eddington levels 12 . Taken at face value, this corresponds to an 121 average accretion rate in this state ofṀ acc ≃ 10 −10 M ⊙ yr −1 . In this case,Ṁ wind /Ṁ acc 0.2, suggesting that the wind is 122 dynamically important and could significantly affect the accretion flow 30, 31 . However, it is also possible that the intrinsic 123 luminosity was much higher throughout this state, with time-variable obscuration being responsible for the reduction in the 124 time-averaged flux (and perhaps also the flaring activity). In the extreme case that L ≃ L Edd , the constraint on the wind 125 efficiency isṀ wind /Ṁ acc 10 −3 .

126
The discovery of optical, UV and (probably) X-ray outflow signatures in the luminous hard state of J1858 suggests that 127 disc winds may always be present in transient X-ray binaries, not just in disc-dominated soft states. Our identification of the 128 constant (non-flaring) spectral component as the carrier of these signatures in the far-UV strongly supports this idea. X-ray and 129 far-UV wind signatures have also been observed in some persistent soft-state X-ray binaries, 32-35 , i.e. systems in which the 130 disc is not subject to the instability that drives the outbursts of transient accretors 36-38 131 The emerging physical picture of disc winds being an integral part of the accretion flows in X-ray binaries is consistent 132 with theoretical modeling of outburst light curves 5, 37 . It is also in line with recent radiation-hydrodynamical modeling of 133 thermally-driven outflows from X-ray binary discs 4 . These simulations confirm that strong mass loss is inevitable in any 134 systems with a sufficiently large disc subject to strong irradiation. Both conditions are met in J1858 (see Methods). A key test 135 of the thermally-driven wind scenario will be to check that wind signatures are absent in systems where these conditions are not 136 met 39 .

137
Regardless of the driving mechanism, two key outstanding questions are where and how these outflows manage to sustain 138 a sufficiently low ionization state to allow the formation of optical and UV lines. The most likely answers are that self-139 shielding, probably coupled with clumping, protects parts of the dense base of the wind above the outermost disc regions from 140 over-ionization. Radiative transfer modeling will be needed to confirm this 3, 4, 40 .

158
A total of 4.9 kilo seconds (ks) exposure was obtained in the far-UV with the Cosmic Origin Spectrograph (COS 41 ) and the 159 G140L grating using the primary science aperture (PSA). This configuration provides a spectral resolution of R = λ /∆λ ∼ 900.

160
All the observations were obtained in TIME-TAG mode, yielding a stream of detected events at a time resolution of 32 ms.

161
Data analysis 162 We reduced the far-UV data using the HST CALCOS pipeline 1 . One-dimensional spectra were extracted using the TWOZONE 163 algorithm, which sums over the cross-dispersion direction such that 99% of the flux is extracted at each wavelength. Errors are 164 estimated from Poisson statistics, and the background is modeled with a smooth polynomial and subtracted from the target 165 spectrum. We extracted light curves from the TIME-TAG event files using the same regions defined by the pipeline, except 166 that empirical background correction was directly applied. Regions affected by geocoronal airglow emission associated with 167 Lyman α (λ λ 1208 − 1225 Å) and O II (λ λ 1298 − 1312 Å) were masked when extracting the light curves.

168
Spectral decomposition 169 The highly variable far-UV luminosity during our observations gives rise to a strongly bimodal logarithmic flux distribution 170 (Extended Data Figure 1). This is in line with the visual impression from the far-UV light curve that the dominant variability is 171 due to "shots" or "flares" superposed on a roughly constant background (Figure 1-b).

172
In order to isolate the spectra associated with these two components, we have carried out a simple linear decomposition (1996) 42 , we assume that the flux density F(λ ,t) at wavelength λ and time t can be written as where C(λ ) and V (λ ) are the spectra of the constant and flaring components, respectively. The function D(t) is the driving 176 light curve of the flaring component.

177
In order to estimate D(t), we constructed a far-UV continuum light curve at 10 s time resolution. We then estimated the 178 underlying constant level in this light curve and created a normalized driving light curve from which this estimate was removed.

179
We finally smooth the resulting time series with a 5-point, second-order Savitzky-Golay filter to obtain our estimate of D(t).

180
The result is shown as the red curve in Extended Data Figure 2.   The reference values adopted for f osc , λ and A in Equation 2 are representative of the C IV resonance line (treated as a 211 singlet). The reference velocity, v ≃ 1500 km s −1 , is chosen based on the location of the blue-shifted absorption trough in the 212 far-UV line profiles (cf Figure 4). Our adopted value of R(v) ≃ 10 10 cm corresponds to the radius in the disc beyond which a 213 thermally driven outflow is expected to be launched (see below); it is also roughly the radius where v esc ≃ 1500 km s −1 . Finally, 214 by taking f ion = 1, we ensure that our estimate ofṀ w is a lower limit (modulo uncertainties in the other parameters).

215
Based on the depth of the absorption features in the far-UV line profiles, we expect that τ 1 for both N V and C IV. The 216 estimated lower limits on the mass-loss rates are thenṀ w 2 × 10 −11 from N V andṀ w 3 × 10 −12 from C IV.

217
A thermally driven disc wind in Swift J1858.6−0814 ? 218 The accretion discs in luminous X-ray binaries are subject to strong irradiation. As a result, the upper layers of the atmosphere 219 can be heated to the inverse Compton temperature, which depends only on the spectral energy distribution of the radiation 220 field. The X-ray spectrum of Swift J1858.6−0814 in the hard state can be approximated as a power law with photon index 221 Γ = 1.5 and an exponential cut off at E max ≃ 30 keV 28 . For such a spectrum, the Compton temperature is approximately 222 kT IC ≃ E max /12 47 , which gives T IC ≃ 3 × 10 7 K for Swift J1858.6−0814 .

223
Mass loss from these heated layers is inevitable at radii where the characteristic thermal speed of the ions, v th ≃ 3kT IC /m p 224 exceeds the local escape velocity, v esc = 2GM/R. Discs larger than the so-called Compton radius, R IC = (2GMm p )/(3kT IC ), 225 are therefore expected to produce thermally driven outflows. For Swift J1858.6−0814 , we obtain R IC ≃ 5 × 10 10 cm. In reality, 226 the radius at which this mechanism turns on is typically R min ≃ 0.1 R IC 48 . In our mass-loss rate calculation above, we have 227 adopted a characteristic radius R ≃ 0.3R IC for the line-forming region in the outflow.

228
The disc in Swift J1858.6−0814 is certainly large enough to drive such an outflow. The orbital period of the system is 229 P orb ≃ 21.3 h 13 . From Kepler's third law, and assuming that q = M 2 /M 1 1, the binary separation is a bin ≃ 3 × 10 11 cm. If the 230 disc is tidally limited, its outer radius will be roughly R disc ≃ 0.9R 1 , where R 1 is the Roche-lobe radius of the neutron star 49 .

231
The outer disc radius is therefore expected to be R disc ≃ 1 − 2 × 10 11 cm -much larger than R IC , let alone R min ≃ 0.1 R IC .

232
The final condition for strong thermally driven mass loss is that the irradiating luminosity should be sufficiently strong, 233 L L crit = 0.05L Edd 24 . This is comparable to the time-averaged luminosity in the flaring hard state of Swift J1858.6−0814 22 .

234
It is therefore likely that the system was luminous enough to drive a powerful thermal disc wind.  wanted to know about accretion but were afraid to ask.    Figure 3. Average far-UV spectrum of Swift J1858.6−0814 during the luminous hard state. Numerous emission and absorption lines are present; the dominant transitions have been labeled with their corresponding rest position indicated with a green tick. All the emission components are skewed toward shorter wavelengths with blue absorption troughs, which are the characteristic footprint of disc outflows. Insets show a zoom-in to the N V (λ λ 1284 − 1437 Å) and C IV (λ λ 1513 − 1668 Å) profiles with the blue-shifted absorption signatures highlighted in blue, in the latter nearby Si II interstellar absorption is indicated with connected green ticks. These signatures indicate the presence of a warm, moderately ionized accretion disc wind with characteristic velocities similar to those observed in the optical.  Figure 4. Spectral decomposition into a constant (blue) and flaring component (red). Being the latter driven by the observed continuum variability in the far-UV. The average normalized spectrum is displayed with a thick black line for reference, and all are normalized to the continuum level. An offset has been added to the spectra for clarity. The regions of geocoronal emission like Lyman α and Si II were removed to avoid artifacts in the spectral decomposition. Rest positions of the dominant ions are marked with a green tick and labeled in the top spectrum. Insets are zooms into the two transitions in which the presence of the outflow is more prominent. Specifically, regions covered in the insets are λ λ 1284 − 1474 Å for N V and λ λ 1525 − 1717 Å for C IV.