Multiwavelength observations reveal a faint candidate black hole X-ray binary in IGR J17285 − 2922

IGR J17285 − 2922 is a known X-ray binary with a low peak 2–10 keV X-ray luminosity of ∼ 10 36 erg s − 1 during outburst. IGR J17285 − 2922 exhibited two outbursts in 2003 and 2010 and went into outburst again in 2019. We have monitored this ∼ 4-month long 2019 outburst with Swift in X-ray and the Very Large Array in radio. We have also obtained four optical spectra with the Gran Telescopio Canarias and Southern Astrophysical Research Telescope, three optical photometry measurements with the Las Cumbres Observatory, and one near-infrared spectrum with the Gemini South telescope. The ratio between its X-ray and radio luminosity is consistent with both samples of neutron star and black hole (BH) X-ray binaries, while the ratio between the X-ray and optical luminosity is consistent with BH X-ray binaries. Studying the evolution of its X-ray power-law index throughout the outburst, we ﬁnd additional evidence for a BH as compact object. The four optical spectra show no H α emission and the nIR spectrum shows no Br γ emission, suggesting that the donor star could be hydrogen-poor and hence that IGR J17285 − 2922 might have an ultracompact binary orbit. The shape of the X-ray light curve is well described by an exponential, followed by a linear decay, from which we obtain a relation between the orbital period P orb and the binary mass ratio. We discuss how this relation is consistent with theoretical predictions and known ultracompact X-ray binaries. Lastly, we discuss how the observed properties are reminiscent of short-P orb BH X-ray binaries.

soft state, the X-ray spectrum is dominated by low energy emission, while during the hard state the emission is dominated by high energy emission.In the hard state, the X-ray binary can launch part of the accreted material in the form of a collimated jet (Spencer 1979;Fender 2006).While the accretion flow is most prominently detected in the X-ray band up to optical, the jet is detected in radio and possibly also in (near-)infrared and optical (Russell et al. 2006;Russell, Fender & Jonker 2007).The matter in the jet is thought to arise from, and thus be correlated to, the accretion flow.For example, the study of X-ray binaries in the hard state in radio and X-ray simultaneously has revealed a coupling between this in-and outflow (see e.g.Corbel et al. 2003Corbel et al. , 2013;;Gallo et al. 2014;Gallo, Degenaar & van den Eijnden 2018).
The outburst-quiescence accretion cycles in transient X-ray binaries can be described by a disc instability model.As matter from the donor star builds up in the accretion disc during quiescence, a thermal instability eventually leads to an outburst with an increased mass accretion rate on to the compact object (see e.g.Lasota 2001;Hameury 2019).Transient X-ray binaries can be loosely classified based on the 2-10 keV peak X-ray luminosity during outburst (Wijnands et al. 2006).The 'bright X-ray binaries' reach peak luminosities of 10 38 -10 39 erg s −1 .However, not all X-ray binaries are this bright.Systems that have a peak luminosity of 10 36 -10 37 erg s −1 are typically called 'faint X-ray binaries'.Other X-ray binaries are even fainter, with peak luminosities of 10 34 -10 36 erg s −1 , which are called 'very faint X-ray binaries' (VFXBs).This faintness makes VFXBs hard to detect, with outbursts that may go unnoticed by X-ray all-sky monitors given their typically low resulting fluxes.
Three promising explanations have been proposed to account for the faint nature of these VFXBs.The first hypothesis is that VFXBs harbour NSs that truncate the accretion disc with their relatively strong magnetic field (Illarionov & Sunyaev 1975;Heinke et al. 2015).This prevents efficient accretion on to the NS, making the system faint in the X-ray band.It could also be possible for such systems to display a propeller driven outflow (D'Angelo & Spruit 2010).To identify a truncation of the inner disc, X-ray reflection spectroscopy can be used to measure the inner disc radius (Fabian et al. 1989).Although there are indications of a truncated disc in some VFXBs, distinguishing a magnetically truncated disc from the formation of a radiatively inefficient accretion flow in the disc has proved difficult (Narayan & Yi 1994;Degenaar et al. 2017;van den Eijnden et al. 2018).
The second hypothesis is that of ultracompact X-ray binaries (UCXBs; King & Wijnands 2006;Heinke et al. 2015).The orbital period (P orb ) of such systems is typically defined as ࣠ 90 min (Nelson, Rappaport & Joss 1986), which requires the donor star to be hydrogen poor to still fit within its Roche lobe.Ultracompact X-ray binaries can have significantly reduced mass transfer rates, making the system faint in the X-ray band.A UCXB can be identified by directly measuring P orb (e.g.measuring periodic eclipses or dips in X-ray or optical, or measuring periodic orbital modulations from optical photometry), indirectly using the requirement of a small accretion disc (e.g. using the ratio of optical to X-ray flux), or other diagnostics such as estimating the composition of the donor through spectral data [for a list of methods see e.g.in't Zand, Jonker & Markwardt (2007)].One of these diagnostics involves the absence of H α in optical spectra, as this may indicate a hydrogen-poor disc and therefore a hydrogen-poor donor star (Nelemans et al. 2004;Werner et al. 2006;Hernández Santisteban et al. 2019).Several UCXBs have been confirmed so far (e.g.Cartwright et al. 2013;Koliopanos et al. 2021, and references therein).The known sample of UCXBs consists (mostly) of NS accretors; BH UCXBs may have been identified in the extragalactic globular cluster NGC 4472 (Maccarone et al. 2007;Zepf et al. 2008;Steele et al. 2014) and a Galactic BH UCXB may have been identified in 47 Tuc X9 (Miller-Jones et al. 2015;Bahramian et al. 2017).
The third hypothesis is that of symbiotic X-ray binaries.In this scenario, the compact object accretes matter from the wind of a giant donor star (see e.g.Masetti et al. 2006Masetti et al. , 2007;;Mattana et al. 2006).Due to the low mass transfer rate from the giant donor star to the compact object, the system can be faint in the X-ray band.Recently, it was suggested that symbiotic X-ray binaries may contribute a significant fraction to the total population of VFXBs (Shaw et al. 2020;Bahramian et al. 2021).
Several other explanations have been put forward.VFXBs could be seen edge-on, making them appear fainter at such a high inclination.This can (possibly) be seen in systems such as CXOGC J174540.0-290031(Muno et al. 2005;Porquet et al. 2005) and Swift J1357.2-0933(Corral-Santana et al. 2013;Mata Sánchez et al. 2015).VFXBs could arise from the so-called period gap X-ray binaries, in which the compact object captures the wind from a detached M dwarf donor star (Maccarone & Patruno 2013;Heinke et al. 2015).VFXBs could also simply be intrinsically bright systems at large distances (Wijnands et al. 2006).A single explanation for the nature of all VFXBs is unlikely, the class of VFXBs is likely to be heterogeneous.
VFXBs are interesting for multiple reasons.First, VFXBs allow an in-depth study of low-level accretion and this has revealed a diagnostic that could allow us to distinguish between NS and BH X-ray binaries (Wijnands et al. 2015).Accompanying this with a simultaneous study of the jet can give insight into the coupling between the accretion flow and jet at relatively low accretion rates.Secondly, VFXBs are also important for understanding binary evolution and population synthesis (see e.g.Maccarone et al. 2015).A complete understanding of the evolution of mass-transferring binary systems has proved to be difficult to develop (see e.g.Paczyński 1971;Tauris & van den Heuvel 2006).Measuring properties such as P orb , the masses of the individual binary components and the nature of the accreting compact object for VFXBs can improve our understanding of their binary evolution and how these systems can get so faint in the X-ray band.Finally, VFXBs with an ultracompact binary orbit are thought to produce low-frequency gravitational waves.They are consequently interesting targets to study with future gravitational wave missions, such as the Laser Interferometer Space Antenna (Nelemans, Yungelson & Portegies Zwart 2001;Nelemans & Jonker 2010).

IGR J17285−2922
IGR J17285−2922 is a borderline faint to very-faint X-ray binary first detected in outburst by International Gamma-Ray Astrophysics Laboratory (INTEGRAL) in 2003, with a 20-150 keV X-ray luminosity of L X ∼ 10 36 erg s −1 for an assumed distance of 8 kpc due to the proximity to the Galactic Centre (Walter et al. 2004;Barlow et al. 2005).X-ray activity coinciding with the position of IGR J17285−2922 was detected with the Rossi X-ray Timing Explorer from an unidentified source named XTE J1728−295 in 2010 (Markwardt & Swank 2010).XTE J1728−295 was confirmed to be the same source as IGR J17285−2922 in subsequent observations with the Neil Gehrels Swift Observatory (Swift; Gehrels et al. 2004) and INTEGRAL (Turler et al. 2010;Yang et al. 2010).The search for the optical counterpart during this 2010 outburst resulted in the identification of a variable star with R ∼ 19 mag and I ∼ 18.5 mag (Russell et al. 2010a, b;Torres et al. 2010).An in-depth investigation on this 2010 outburst was done by Sidoli et al. (2011) using a high quality XMM-Newton observation, along with INTEGRAL data.Sidoli et al. (2011) concluded that IGR J17285−2922 is a transient VFXB, although no definitive answer was found for the nature of the compact object.The absence of thermonuclear X-ray bursts and X-ray pulsations allowed for either a NS or BH primary (Sidoli et al. 2011).
More recently, on 2019 April 8-9, INTEGRAL found renewed X-ray activity from IGR J17285−2922 (Ducci et al. 2019).To further investigate its nature, we monitored this 2019 outburst in the X-ray band with Swift and in radio with the Karl G. Jansky Very Large Array (VLA).On top of this, we obtained throughout the outburst four optical spectra covering H α with the Southern Astrophysical Research Telescope (SOAR) and Gran Telescopio Canarias (GTC), one near-infrared (nIR) spectrum with the Gemini South telescope and three optical photometry measurements with the Las Cumbres Observatory (LCO) telescope network.We will use these multiwavelength observations to constrain the nature of the compact object, donor star, as well as the binary orbital properties of the (V)FXB IGR J17285−2922.

X-rays
We monitored the outburst of IGR J17285−2922 with Swift to track the outburst evolution and the accretion state.Between April 10 (2 d after the initial INTEGRAL detection) and 2019 September 20, a total of 45 observations were taken (Target ID 00011287 and 00011303, see Table 1 for an overview) with the Swift X-ray Telescope (XRT; Burrows et al. 2005).These Swift/XRT observations had a typical duration of ∼ 1 ks.Only the first observation was taken in Window Timing (WT) mode, while all other observations were taken in Photon Counting (PC) mode.We extracted the 0.3-10 keV count rates with the Swift/XRT product generator1 (Evans et al. 2007(Evans et al. , 2009)).We calibrated the observations using the XRTPIPELINE (version 0.13.5) and the CALDB in the HEASOFT package (version 6.26.1) provided by HEASARC. 2 The images and spectra were extracted using XSELECT (version v2.4g).For the single WT observation (ObsID 00011287001), we used a circular source extraction region with a radius of 35 arcsec and two circular background extraction regions with radii of 35 arcsec each, placed sufficiently far away from the source.For the PC observations, we first correct for pile-up if needed.For the 2nd up to and including the 8th observations (ObsID 00011303002-00011303010), correction for pile-up was required and we used an annular source extraction region with an inner radius of 10 arcsec and outer radius of 35 arcsec.For the 9th, 10th, 11th, and 13th observation (ObsID 00011303011-00011303013 and 00011303015), correction for pile-up was also required and we used an annular source extraction region with an inner radius of 6 arcsec and outer radius of 35 arcsec.For all other PC observations, we used a circular source extraction region with a radius of 25 arcsec.In all the PC observations, we used three circular background extraction regions with radii of 60 arcsec each.The ancillary response files were created with the observation-specific exposure-maps using XRTMKARF (version 0.6.3).The response matrix files swxwt0to2s6 20131212v015 and swxpc0to12s6 20130101v014, for WT and PC mode, respectively, were obtained from the CALDB (version 20190412).All spectra were grouped to have a minimum of 1 count per bin with GRPPHA.On top of this we also grouped the first five spectra (ObsIDs 00011287001 and 00011303002-00011303005) separately to have a minimum of 20 counts per bin with GRPPHA.We fitted the Swift/XRT spectra using XSPEC (v.12.10.1f;Arnaud  1996).We used an absorbed power-law model (TBABS * POWERLAW) and a combined absorbed power law plus blackbody model (TBABS * [BBODYRAD + POWERLAW]).We performed these fits with the hydrogen column density parameter (N H in TBABS; Wilms et al. 2000) as three different options; N H as a free parameter, fixed at N H = 0.679 × 10 22 cm −2 (see Section 4.5), and fixed at N H = 0.99 × 10 22 cm −2 (determined by the simultaneous fit of all Swift/XRT spectra with N H tied for each spectrum).The impact of N H on the determined spectral parameters is discussed in Section 4.5.We adopted the crosssections by Verner et al. (1996) and abundances by Wilms et al. (2000).For consistency, and due to the low total counts in the last half of the outburst, the model fitting was done using Cash statistics (Cash 1979) in all spectra.In addition, we also used χ 2 statistics for the first five spectra to perform the F-test (see Section 3.3).The results using both Cash and χ 2 statistics were consistent with each other.
We determined the 0.5-10 and 1-10 keV unabsorbed fluxes with CFLUX.For radio epochs 1, 3, and 4 (see Section 2.2), no quasisimultaneous Swift/XRT observations were taken.We determined the X-ray flux (1-10 keV) during these radio epochs using a linear interpolation between the closest Swift/XRT observations before and after each of these radio epochs.The two fluxes used for each interpolation are similar down to 30 per cent.We use the largest positive and negative 1 σ error of the interpolated X-ray fluxes for the 1-10 keV flux during each radio epoch.The 1-10 keV X-ray fluxes adopted during each radio epoch are listed in Table 2.

Radio
We monitored the 2019 outburst of IGR J17285−2922 with the VLA over 7 epochs (project code SF8027, see Table 2 for an overview).In epochs 1 to 4, the VLA observations were taken in the B configuration, in epochs 5 and 6 in a BnA configuration, and in epoch 7 in the A configuration.In all epochs, IGR J17285−2922 was observed at C band in 8-bit mode, with two subbands at central frequencies of 4.5 and 7.5 GHz, with 1 GHz bandwidth each.The primary flux calibrator was 3C 286 = J1331+3030 and the secondary phase calibrator was J1743−3058 (∼ 3.6 • angular distance).
We analysed the observations using the Common Astronomy Software Application3 (CASA version 5.6.1;McMullin et al. 2007).Radio frequency interference and other data artefacts were removed by careful visual inspection, in combination with automated CASA routines.We imaged the calibrated 4-5 and 7-8 GHz Stokes I data separately using TCLEAN, with a Briggs weighting scheme robust parameter of 0, balancing sensitivity and the impact of other nearby sources.We determined the flux density in the image plane by fitting a 2D elliptical Gaussian using IMFIT, with the ellipse parameters fixed to those of the synthesized beam.We determined the 1 σ error on the flux density by measuring the RMS of a nearby area containing no sources in the image plane.When the source was not detected in either the 4-5 or 7-8 GHz subband, we determined a 3 σ upper limit as three times the RMS over the source location in the image plane.The details of the VLA observations are given in Table 2.
Table 1.Overview of the Swift/XRT observations and spectral fits for the 2019 outburst of IGR J17285−2922.We used an absorbed power-law model with the hydrogen column density fixed at N H = 0.99 × 10 22 cm −2 .gives the X-ray power-law index.For the non-detections as IGR J17285−2922 becomes quiescent, we assume that the spectra are equal to the last detected spectrum (ObsID 00011287002) to determine the 3 σ upper limits on the 0.5-10 keV X-ray flux by scaling the count rates.All uncertainties are 1 σ .Upper limits are 3 σ .Fixed parameters are indicated with an asterisk.To determine the spectral index (α) of the radio emission, we performed Monte Carlo (MC) simulations.For each radio epoch in which IGR J17285−2922 was detected in both the 4-5 and 7-8 GHz subband, we drew 10 6 frequencies between 4 and 5 GHz (ν 4-5 GHz ) and between 7 and 8 GHz (ν 7-8 GHz ) assuming a uniform distribution.For each of these individual frequencies, we drew a flux density (S 4-5 GHz and S 7-8 GHz ) assuming a Gaussian distribution with mean and standard deviation equal to the observed flux density and RMS, respectively.We determined α using S 7-8 GHz = S 4-5 GHz × (ν 7-8 GHz / ν 4-5 GHz ) α .We determined α and 1 σ errors, as the 50th, 16th, and 84th percentile, respectively.When IGR J17285−2922 was detected in only one subband, we determined a 3 σ upper limit on α using the procedure described in van den Eijnden et al. (2019).In this procedure, an MC simulation is performed to determine at what α (3 σ upper or lower limit) the non-detected subband would have been detected.

ObsID
Table 2. Overview of the VLA observations for the 2019 outburst of IGR J17285−2922.The spectral index α for each VLA observation is calculated as described in Section 2.2.For each VLA observation, we give the (quasi-)simultaneous 1-10 keV X-ray flux and the Swift/XRT ObsID(s) used for this flux.Swift/XRT ObsID(s) followed by an asterisk indicate that the (quasi-)simultaneous 1-10 keV X-ray flux has been interpolated.Upper and lower limits are 3 σ , while all uncertainties are 1 σ .

Optical photometry
We observed IGR J17285−2922 with the LCO 2-m Faulkes and 1-m network of telescopes during the 2019 outburst, as part of a monitoring campaign of ∼50 LMXBs (Lewis et al. 2008;Russell et al. 2010a, b).Imaging data were taken mostly in the Bessell I filter.
The newly developed 'X-ray Binary New Early Warning System' (XB-NEWS) pipeline (Russell et al. 2019) was used to compute astrometric solutions, perform multi-aperture photometry, and flux calibrate the photometry.The pipeline produces a calibrated light curve for the target (for more details see Russell et al. 2019;Goodwin et al. 2020).For images in which the target is not automatically detected above the detection threshold by the pipeline, XB-NEWS performs forced multi-aperture photometry at the known position of the source.All derived magnitudes with an uncertainty >0.25 mag were considered unreliable, and were rejected.

Optical spectroscopy
We obtained four epochs of optical spectroscopy of IGR J17285−2922 during the outburst with the purpose of investigating the presence of H α. We observed the target on the nights of 2019 April 30 -May 1, May 2-3, and June 29-30 with SOAR, using the Goodman Spectograph (Clemens, Crain & Anderson 2004).The first two runs both consisted of two exposures of 1800 s each, using a 400 l/mm grating with a 0.95 arcsec slit, yielding a full width at half-maximum (FWHM) resolution of ∼5.6 Å over the wavelength range from ∼3800-7800 Å.The final run consisted of two 1500 s exposures, using the same grating on a 1.2 arcsec slit, resulting in a ∼7.6 Å FWHM resolution between ∼4850 and 8850 Å.The spectra were reduced and optimally extracted following standard practices using IRAF.
We also observed IGR J17285−2922 using the OSIRIS instrument (Cepa et al. 2000) mounted on the 10.4-m GTC, on the night of 2019 July 21-22.We obtained two spectra with exposures times of 600 s each, using the grism R2500R (5575-7685 Å) with a 1 arcsec slit, providing a velocity resolution of 160 km s −1 .The data reduction and calibration was performed using IRAF, after which we used MOLLY and PYTHON routines to analyse, normalize, and plot the spectra.

nIR spectroscopy
On 2019 July 8, we obtained long slit spectroscopy of the nIR counterpart to IGR J17285−2922 with Flamingos-2 on the 8.1m Gemini South telescope at Cerro Páchon, Chile as part of program ID GS-2019A-FT-106 (PI: Shaw).We obtained 21 × 120 s exposures with the HK grism and a slit width of 2 pixels (0. 36) which provides a typical FWHM resolution of ∼24 Å at Brackett γ (Brγ ; 2.166 μm).
To reduce the effect of the rapidly changing background at nIR wavelengths we dithered along the spatial directions of the slit in an ABBA pattern.On the same night we also used the same set up to observe the telluric standard star Hip 82714, obtaining 4 × 2 s images.
Data were reduced using the Gemini package in IRAF4 following standard procedures.We normalized the averaged spectrum of the telluric standard star and removed the hydrogen Brackett-series absorption features by subtracting a best-fitting Voigt profile.We then used the task nstelluric to shift and scale the science and telluric spectra in order to optimally divide out telluric features from the science spectra.

X-ray and radio light curves
We show the Swift and VLA light curve of the 2019 outburst of IGR J17285−2922 in Fig. 1.The Swift light curve shows a globally decreasing count rate as the outburst proceeds, with a plateau between 2019 June 6 and July 6 (MJD 58640 and 58670, respectively).IGR J17285−2922 was first observed with Swift on 2019 April 10  Table 1, while the data of the VLA observations are given in Table 2. Bottom: The radio spectral index α for each VLA observation with at least one detection in either the 4-5 or 7-8 GHz subband.Uncertainties on all data are 1 σ .Upper and lower limits are 3 σ .
(MJD 58583), and was last detected on 2019 August 16 (MJD 58711), for a total outburst duration of 128 d as observed with Swift.The 42nd Swift/XRT observation (MJD 58734) is reported as a detection by the Swift/XRT product generator (Evans et al. 2007(Evans et al. , 2009)), but manually analysing this observation with XIMAGE, with both DETECT and SOSTA, shows that IGR J17285−2922 is not detected at even 1 σ confidence.Instead, we adopt a 3 σ upper limit on the count rate for this observation by tripling the background count rate as determined with XSELECT.We show the flux densities in the 4-5 and 7-8 GHz subbands of the VLA observations of IGR J17285−2922 in Fig. 1, which are listed in Table 2. IGR J17285−2922 is detected during radio epochs 1, 2, 4, and 6 in both the 4-5 and 7-8 GHz subband with flux densities of ∼ 50-120μJy and is consistent with a flat spectrum of α ∼ −0.6 to 0.2.However during radio epochs 3 and 5, the source is only detected in the 7-8 GHz subband, yielding 3 σ lower limits on α of 0.9 and −0.2, respectively.During radio epoch 7, IGR J17285−2922 is no longer detected in either frequency subband with 3 σ flux density upper limits of ∼ 30 μJy.This radio non-detection occurs around the time of a sharp decline in Swift/XRT count rate as IGR J17285−2922 becomes quiescent.
Using radio epoch 1 (7-8 GHz subband), with the highest signalto-noise ratio (S/N), to determine the best-fitting (J2000) radio position gives RA = 17 h 28 m 38.9 s ± 0.3 s Dec = −29 • 21 43.2 ± 0.1 , where the uncertainties are estimated from the astrometric accuracy of the VLA.Since the S/N is larger than 10 for epoch 1, we use 10 per cent of the synthesized beam to determine the uncertainties.This position is fully consistent with the position determined with the Chandra X-ray Observatory during the 2010 outburst (Chakrabarty, Jonker & Markwardt 2010).

Radio/X-ray coupling
The coupling between the accretion flow and jet has revealed a difference between NS and BH X-ray binaries.In the radio/X-ray luminosity plane (L R − L X plane), BH X-ray binaries have a (e.g. 5 GHz) radio luminosity that is typically a factor of 10 greater than the radio luminosity of NS X-ray binaries at equal (e.g.1-10 keV) X-ray luminosities (see e.g.Fender & Kuulkers 2001;Gallo et al. 2018).This difference is clearly visible in Fig. 2, where we show the position of NS and BH X-ray binaries in the L R − L X plane for 5 GHz radio luminosities and 1-10 keV X-ray luminosities (data Figure 2. The L R − L X plane for X-ray binaries adopted from Bahramian et al. (2018) including the measurements of IGR J17285−2922.We show the L R − L X data of IGR J17285−2922 with different colours for the adopted distances of 4, 8, 12, and 16 kpc.We show the BH X-ray binary sample with black circles and the NS X-ray binary sample with grey squares.The dotted black line shows the best-fitting relation for BHs as determined by Gallo et al. (2006).We have included two L R − L X points for the strong candidate BH (V)FXB Swift J1357.2−0933 for a distance of 6 kpc with purple stars to provide context (Sivakoff, Miller-Jones & Krimm 2011;Plotkin et al. 2016;Paice et al. 2019).Uncertainties on all data are 1 σ and smaller than their respective marker if they are not shown.Upper and lower limits are 3 σ .base5 was consulted 2020 January; Bahramian et al. 2018).We can therefore use the (quasi-)simultaneous radio/X-ray observations of IGR J17285−2922 to determine whether these are consistent with the NS or BH X-ray binary samples to gain insight into the nature of the compact object.
In order to study the position of IGR J17285−2922 in the L R − L X plane, we determine the 5 GHz radio-and 1-10 keV Xray luminosities.The distance to IGR J17285−2922 has only been constrained to d 4 kpc, based on a non-detection (R > 21 mag) in an archival optical image (Sidoli et al. 2011).We adopt a range of distances of 4, 8, 12 and 16 kpc to investigate the nature of the compact object for each of these distances.The (quasi-) simultaneous 1-10 keV unabsorbed X-ray fluxes are determined with the method described in Section 2.1 and the best-fit model determined in Section 3.3.The 1-10 keV X-ray luminosities are consequently calculated with L 1-10 keV = 4π d 2 F 1-10 keV , where F 1-10 keV are the unabsorbed 1-10 keV X-ray fluxes.The radio luminosities are calculated with L ν = 4π d 2 νS ν .The 4-5 and 7-8 GHz radio flux densities, along with the (quasi-)simultaneous 1-10 keV X-ray fluxes are given in Table 2.
Fig. 2 shows IGR J17285−2922 (for assumed distances of 4, 8, 12, and 16 kpc) in the L R − L X plane along with other NS and BH X-ray binaries.We show the 7-8 GHz instead of the 4-5 GHz radio luminosities assuming a flat spectrum to avoid cluttering, which is valid for all VLA observations except the third VLA observation with a strongly inverted spectrum (α > 0.91).The results we obtain here from using the 7-8 GHz radio luminosities in Fig. 2 are consistent with the results using the 4-5 GHz radio luminosities.For a distance of 4 kpc, the L R − L X location of IGR J17285−2922 is more consistent with NS X-ray binaries.For increasing distances, the L R − L X location of IGR J17285−2922 is consistent with both BH-and NS X-ray binaries.For a distance of 16 kpc and greater, the L R − L X location of IGR J17285−2922 becomes more consistent with that of BH X-ray binaries.Without knowing the distance of IGR J17285−2922, we can thus not determine the nature of the compact object based on its location in the L R − L X plane.
Our Swift/XRT observations sample a L X range of nearly 2 orders of magnitude, allowing us to determine the correlation coefficient between L R and L X .We investigate the L R − L X luminosity correlation using a linear fit in log-log space in the form of log(L R ) (1) Here δ is the offset (scaling factor in linear space), γ is the slope of the line (power-law index in linear space) to be determined, and is a normal random variable representing the intrinsic random scatter about the regression line with mean equal to 0 and standard deviation σ as a free parameter (Kelly 2007).Following Gallo et al. (2014) and Gusinskaia et al. (2020), we use the PYTHON port of LINMIX6 method developed by Kelly (2007).In this method, Markov Chain Monte Carlo (MCMC) simulations are performed to fit the linear model in equation ( 1) taking upper limits on the data into account.The LINMIX performs a fit of the parameters γ , δ, and σ .We estimate γ , δ, and σ by calculating the mean for each parameter from the marginalized posterior distributions (100 000 iterations) and determined the 1 σ uncertainties on these parameters by taking the 16-84th percentile of the marginalized posterior distributions.We have performed this method for the 7-8 GHz radio flux densities and X-ray fluxes given in Table 2 (including the upper limit in radio in epoch 7), converted these to their respective luminosities as described above (for a distance of 8 kpc).Similar to Gallo et al. (2014), we use an uncertainty of 0.05 and 0.10 dex on L R and L X , or keep the statistical uncertainty, depending on which is larger.This is to account for a lack of strict simultaneity and model-dependent count rate to flux conversion.The best-fitting parameters are γ = 0.4 ± 0.2/0.3,δ = 0.4 ± 0.1/0.2, and σ = 0.1 ± 0.7/1, where these uncertainties correspond to the case of 0.05/0.10dex uncertainties on the luminosities.We can compare this result for γ to the most recently inferred slope of the NS and BH X-ray binary sample by Gallo et al. (2018).Our slope determined here for IGR J17285−2922 (γ = 0.4 ± 0.2/0.3) is consistent with the NS X-ray binaries (γ = 0.44 +0.05 −0.04 ) and BH X-ray binaries (γ = 0.59 ± 0.02).We note that our correlation for IGR J17285−2922 is measured over a modest range in X-ray luminosity, spanning only ∼ 2 orders of magnitude, with a radio non-detection in the last epoch.Corbel et al. (2013) show that an X-ray luminosity range extending across > 2 orders of magnitude is needed to accurately measure the L R − L X correlation index γ .This is reflected in the adopted uncertainties on the luminosities, as using uncertainties of ∼ 0.15 dex or higher results in too poor quality data to constrain the L R − L X correlation.

X-ray spectral evolution
Another diagnostic we can use is the X-ray spectral evolution of IGR J17285−2922 at low X-ray luminosities.Wijnands et al. (2015) studied the spectral properties of NS and BH X-ray binaries.They found that when using an absorbed power-law model to fit X-ray spectra, NS X-ray binaries have a significantly softer spectrum than BH X-ray binaries between 0.5 and 10 keV X-ray luminosities of 10 34 -10 36 erg s −1 .
To measure the X-ray fluxes and consequently luminosities, we first determine the best-fitting model to the 0.5-10 keV spectra.We compare the Swift/XRT spectra with and without the addition of a blackbody component (BBODYRAD) using an F-test for the first five Swift/XRT observations described in Section 2.1 with N H as a free parameter (FTEST in XSPEC).We use the first five spectra as these have the highest count rate and we expect the blackbody spectral component to be most prominent during this time.In all five observations, we cannot prove the inclusion of a blackbody component to be significant at 1 σ -confidence.Next, we fixed the temperature of the blackbody component between 0.2 and 2.6 keV with increments of 0.4 keV for all five observations, which shows that the inclusion of a blackbody component is only significant at 1 σconfidence for the third and fifth observation (ObsID 00011303003 and 00011303005).Having no evidence for a blackbody component, we will only use the power-law component from now on in all observations.
Having determined what spectral model to use, we measure N H = (0.99 ± 0.05) × 10 22 cm −2 with a simultaneous fit of all Swift/XRT spectra with N H tied for each spectrum.We also determined N H = (1.02± 0.03) × 10 22 cm −2 by the weighted least squares of the Swift/XRT spectra with N H as a free parameter for each spectrum.We adopt a fixed N H of 0.99 × 10 22 cm −2 in further analysis.We determine 3 σ upper limits on the 0.5-10.0keV unabsorbed flux for the non-detections by scaling the upper limits on the count rate to the count rate of the last detection.We assume that the spectral parameters of the non-detections are equal to that inferred from the last detection.The details of the spectral fits for N H = 0.99 × 10 22 cm −2 are listed in Table 1.
We plot in Fig. 3 the X-ray power-law index as a function of the 0.5-10 keV X-ray luminosity, using the data in fig. 1 of Wijnands et al. (2015), including our results for the 2019 outburst of IGR J17285−2922, assuming a distance of 8 kpc.The evolution of of IGR J17285−2922 seems by eye to be more consistent with the BH X-ray binaries.Wijnands et al. (2015) use a 2D Kolmogorov-Smirnov (KS) to quantify the probability that the two samples are drawn from two different distributions.We find that using a 2D KS test, which uses cumulative distribution functions (CDFs) at its core, introduces observational biases.An increased density of Xray observations at a specific time, causes a significantly increased density in the X-ray luminosity space and photon index space.This can significantly alter the CDFs, where they can significantly impact the conclusions.
Since we have a large, not perfectly spaced, number of X-ray observations, we use MC simulations instead of a 2D KS test.For each of the three − L X data sets (NS and BH X-ray binaries, along with IGR J17285−2922), we draw 10 5 data set samples assuming a Gaussian distribution with mean and standard deviation equal to the determined − L X and 1 σ errors, respectively.Next, we bootstrap these data set samples with replacement obtaining a new data set sample with equal size.We fit to each of these new data set samples the modified linear regression equation to account for a logarithmic X-ray luminosity axis.Here a is the slope of the curve and b is the offset of the curve at log(L X ) = 34, at this value the offset between the NS and BH X-ray binary samples should be maximally visible.For each of the NS and BH X-ray binaries, and IGR J17285−2922 data sets, the distribution of these slopes and offsets are used to quantify the differences.We have performed these MC simulations for distances of 4, 8, 12, and 16 kpc.
We show a visualization and corner plot of the MC simulations for a distance of 8 kpc in Fig. 3.The NS visualization is cut-off at L X = 10 34 because the NS sample itself also has this cut-off in Wijnands et al. (2015).We find that the NS and BH X-ray binary samples are different at 3 σ -confidence.For distances of 4, 8, 12, and 16 kpc, the IGR J17285−2922 samples are found to be different from the NS X-ray binary sample at 2 σ -confidence in all cases, while the IGR J17285−2922 samples are not found to be different from the BH X-ray binary sample at even 1 σ -confidence in all cases.Our X-ray spectral analysis thus favours IGR J17285−2922 having a BH primary rather than a NS primary.

X-ray light-curve fitting
Now that we have investigated the nature of the compact object in IGR J17285−2922 via the L R − L X plane and the X-ray spectral evolution at low X-ray luminosities, we move on to investigate the binary parameters.In particular, we are interested in the P orb of the binary system.Heinke et al. (2015) have used the light curves of two VFXBs to estimate P orb .This method is based on the analytical expressions derived by King & Ritter (1998) that explain the shape of the outburst light curve of a typical transient X-ray binary.The overall shape of the light curve is described by an exponential decay above a 'transition' luminosity L t , below which the shape of the light curve can be described by a linear decay.In physical terms, the exponential decay arises from an entirely ionized disc due to irradiation by the central X-ray source.After irradiation is no longer able to ionize the outer edge of the disc, a linear decay sets in.We can use this transition and the exponential decay to constrain the outer disc radius.
The light-curve shape described in the previous paragraph is visible in Fig. 1, as IGR J17285−2922 during this 2019 outburst starts off with an exponential decay followed by a linear decay, where the plateau could be identified as a transition.We can fit [see Powell, Haswell & Falanga (2007) and Heinke et al. (2015) for a detailed description] the exponential decay part of this light curve with Here, F t is the transition flux at which the light curve changes shape from an exponential to a linear decay.F e is the limit of the exponential decay, τ e is the time-scale of the exponential decay, and t t is the time of the transition.We can fit the linear decay part of the light curve with with τ l the time-scale of the linear decay.The results of the exponential into linear decay fit are given in Table 3, with the 0.5-10 keV fluxes of the 2019 outburst of IGR J17285−2922 given in Table 1.The 1 σ errors on the parameters are determined with an MCMC simulation, for which we give the corner plot in Appendix A. We show the fit to the light curve, along with the residuals in Fig. 4. The disc outer radius R 0 is given by with ν the viscosity (we assume ν = 4 × 10 14 cm 2 s −1 following Powell et al. 2007).The transition radius R disc can be derived from the transition flux using Here, φ H relates to the amount of matter available for accretion in the disc and accounts for how the disc is irradiated, recalibrated by Heinke et al. (2015) to be φ H = 1.6 × 10 −18 cm 12/7 s erg −1 .Finally, we can estimate P orb with where R circ is the circularization radius and q is the mass ratio between the donor star and compact object (Frank, King & Raine 2002;Heinke et al. 2015).We assume that R circ is given by the disc radius calculated with either equation ( 5) (dependent on τ e ) or 6 (dependent on F t and d).
We give P orb determined with τ e for a mass ratio q = 0.1, 0.01, 0.005, and 0.001 in Table 3.We also give P orb determined with F t and d for a mass ratio q = 0.1 and is consistent with P orb Table 3. Results for the exponential, followed by a linear decay fit, to the light curve of the 2019 outburst for IGR J17285−2922.The orbital period P orb is determined as described in Section 3.4.We give, where possible, the results for CXO J174540.0−290005 and XMM J174457−2850.3calculated from Heinke et al. (2015).Uncertainties for IGR J17285−2922 are 1 σ , while uncertainties for CXO J174540.0−290005 and XMM J174457−2850.3are 90 per cent confidence, with * representing hard limits reached due to model constraints.P orb,τe (q = 0.01) (h) 2.17 ± 0.05 --P orb,τe (q = 0.005) (h) 1.44 ± 0.03 --P orb,τe (q = 0.001) (h) 0.61 ± 0.02 -- determined with τ e at a distance of ∼ 15 kpc.For a mass ratio q = 0.001 we obtain P orb = 0.61 ± 0.02 h, which could suggest that IGR J17285−2922 has an ultracompact binary orbit.We will further discuss this possibility in Section 4. For comparison, we show the fit results for the two other VFXBs in Table 3, which were proposed to be UCXBs based on the discussed light-curve fitting method by Heinke et al. (2015).We note that the 1 σ uncertainties quoted for IGR J17285−2922 are statistical errors and systematic uncertainties introduced in our underlying assumptions can contribute significantly (e.g.differences in R disc and R circ as seen in Powell et al. 2007).

Optical photometry
The optical field of IGR J17285−2922 is extremely crowded, and the source was blended with a brighter nearby star on the majority of dates.On three dates under good seeing conditions, the source was automatically detected by the XB-NEWS pipeline and accurate magnitudes could be extracted: I = 18.24 ± 0.02, I = 18.38 ± 0.02, I = 18.78 ± 0.01 on MJD 58621.66,58625.58,and 58659.42,respectively.We use these accurate measurements in the optical-X-ray correlation diagram in Fig. 5. IGR J17285−2922 is shown for three sample distances (4, 8, 16 kpc), alongside a sample of BH  et al. (2007).For all plotted distances, the optical counterpart of IGR J17285−2922 is consistent with hard state BH X-ray binary systems, while it is between a factor ∼5-10 brighter than NS systems.

Optical spectroscopy
Further evidence in favour of IGR J17285−2922 having an ultracompact binary orbit comes from our optical spectroscopy around H α. In Fig. 6, we show the four normalized optical spectra taken by GTC and SOAR, where the former is averaged from two spectra taken on the same night.The red-dashed line indicates the H α rest wavelength, while the blue lines show the rest wavelength of three HeI lines in the bandpass.We have indicated, in grey, telluric, and interstellar absorption features.The leftmost band, close to the 5875.6Å He I line, is the interstellar Na I doublet at 5889 and 5895 Å.
No evidence for H α or He emission is observed in any of the spectra.Therefore, we use the GTC spectrum and MOLLY to measure 3 σ upper limits on the equivalent width of four lines, assuming a 2000 km/s velocity width.This yields an equivalent width upper limit of 0.6 Å for H α at 6562.8 Å.For the HeI lines, this results in equivalent width upper limits of 0.8 Å, 0.6 Å, and 0.5 Å for the lines at 5875.6 Å, 6678.1 Å, and 7065.2Å, respectively.For the He II line at 4686.0 Å, we use the two SOAR spectra taken on April 30th and May 2nd, resulting in equivalent width upper limits of 3.7 Å and 6.5 Å respectively.The H α upper limit lies below typical equivalent widths seen in both BH and NS X-ray binaries at similar luminosities (see e.g.Fender et al. 2009, for a sample study).Combined, given these four observations, we do not detect H α at any point during the outburst.The absence of H α emission is consistent with the scenario of IGR J17285−2922 having an ultracompact binary orbit, which would require an hydrogen-poor donor.

nIR spectroscopy
In Fig. 7 we show the K-band segment of the Gemini/Flamingos-2 spectrum of the nIR counterpart to IGR J17285−2922.Much like the optical spectra, the nIR spectrum is featureless and we see no evidence for hydrogen, which is often present as Brγ emission in the nIR spectra of X-ray binaries with low mass donors (e.g.Bandyopadhyay et al. 1997Bandyopadhyay et al. , 1999;;van den Berg & Homan 2017).The nIR spectrum does not show any evidence for lines that might be associated with a late-type (main sequence) donor (such as the CO bandheads or neutral metal species such as Al or Ti).However, this is perhaps not surprising considering that the source was still in outburst at the time the nIR observations were performed.

D I S C U S S I O N
We have monitored the 2019 outburst of the (V)FXB IGR J17285−2922 with Swift/XRT, VLA, SOAR, GTC, Gemini South telescope and LCO telescope network.The Swift/XRT spectra are well described by an absorbed power law, indicative of a hard state throughout the outburst.The source was detected with Swift/XRT as the count rate decreased over the course of ∼ 128 d, after which the source was no longer detected.IGR J17285−2922 was detected in VLA radio observations as well, both at 4-5 and 7-8 GHz in most observations.Only for the third and fifth VLA epoch, the source was detected in the 7-8 GHz subband, but undetected in the 4-5 GHz subband.The radio spectra are consistent with a flat spectrum, with only the third VLA epoch specifically having a strongly inverted spectrum.
IGR J17285−2922 was no longer detected in radio in the last epoch, coinciding with a steep decrease in Swift/XRT count rate.Our VLA observations thus suggest that a compact steady jet (see e.g.Fender 2006;Russell et al. 2016) was detected from IGR J17285−2922 throughout most of the outburst (characterized by a flat radio spectrum), while during at least one epoch discrete ejecta might have been launched (resulting in a steep radio spectrum; e.g.Blandford & Königl 1979).The four optical spectra taken throughout the outburst (both SOAR and GTC) do not reveal H α emission.Similarly, the nIR spectrum taken with the Gemini South telescope shows no Brγ emission.Three optical photometry measurements show detections with I-magnitudes between 18 and 19 mag.

The nature of the compact object
We can separate the binary properties of IGR J17285−2922 into three different categories; the nature of the compact object, the nature of the optical and nIR emission, and the nature of the orbital properties of the binary system.We will start with the nature of the compact object, which we can constrain using the L R − L X and the L opt − L X planes, and with the photon index in the low X-ray luminosity regime.
Let us first turn to the L R − L X diagram, in which the position of IGR J17285−2922 depends on distance.For distances of 8, 12 and 16 kpc in Fig. 2, IGR J17285−2922 is consistent with both the NS and BH X-ray binaries.Approaching shorter distances, IGR J17285−2922 becomes more consistent with NS X-ray binaries, diverging from the BH X-ray binary track.This is visible in Fig. 2 for a distance of 4 kpc.Utilizing the L R − L X plane to determine the nature of the compact object in an X-ray binary requires careful approach.For a fixed distance, no significance can be given on how much IGR J17285−2922 diverges from either the NS or BH X-ray binaries.Moreover, there are clear outliers such as the radio-bright  NS X-ray binary IGR J17591−2342 and the NS candidate X-ray binary 3FGL J0427.9−6704, for which the L R − L X location are similar to that of BH X-ray binaries (Russell et al. 2018;Li et al. 2020).In addition, the L R − L X correlation for NSs and even BHs are not well established below L X ࣠ 10 36 erg s −1 .The scarcity in the number of different X-ray binaries at such low X-ray luminosities is a significant caveat.So, the L R − L X plane cannot be used as an unambiguous differentiator between NS and BH primaries.
In the L opt − L X plane, BH X-ray binaries show optically brighter counterparts than their NS equivalents (Russell et al. 2006(Russell et al. , 2007, see Fig. 5).For all trial distances discussed above, IGR J17285−2922 is consistent with a BH system.However, as highlighted as well for the L R − L X diagram, at smaller distances than 4 kpc, the system would become more consistent with NS X-ray binaries.While the L opt − L X plane is better populated down to low X-ray luminosity, especially for NS systems, there is a larger variety of processes possibly contributing to the optical emission.Therefore, for such small distances, the L opt − L X diagram would also not be an unambiguous differentiator.
Next, we can use the method developed by Wijnands et al. (2015), which is also dependent on distance but nevertheless yields more conclusive results.The hardness of the X-ray spectra (0.5-10 keV) of IGR J17285−2922 throughout the 2019 outburst in Fig. 3 (showing 8 kpc) indicates that a NS X-ray binary can be excluded at 2 σconfidence for all tried distances.This forms a striking contrast with the BH scenario, that is consistent within 1 σ for all tried distances.Because the fitted spectral shape is distance independent, only the Xray luminosity is distance dependent.Changing the distance therefore only affects the offset, as the slope is invariant to the logarithm of a scaling in L X .It is visible in Fig. 3 that the slope is the least constraining parameter.Putting this together, even though the Xray luminosity is distance dependent, the overall conclusion that the hardness of the X-ray spectra suggest that the compact object in IGR J17285−2922 is a BH remains the same.
Whether the conclusion drawn from Fig. 3 is unambiguous and that the conclusion cannot be interpreted in any other way remains to be proven, as already mentioned in Wijnands et al. (2015).On top of this, not all X-ray binaries are included with a known and Xray luminosity between 10 34 -10 36 erg s −1 .For example, we have not included the NS X-ray binaries studied in Parikh et al. (2017), which show similar behaviour to the NS X-ray binaries in Wijnands et al. (2015).We have not included the − L X data for EXO 1745−248 studied in Rivera Sandoval et al. (2018), which shows hard X-ray spectra ( ∼ 1.4) at quiescent / low X-ray luminosity.Increasing the sample of different NS and BH X-ray binaries in this low X-ray luminosity regime is required to confirm the validity of using this tool.If IGR J17285−2922 is confirmed to harbour a BH, it would add another system to the still scarce sample of BH X-ray binaries studied at and below an X-ray luminosity of 10 34 erg s −1 .
Combining the conclusions of the L R − L X and L opt − L X plane, and the hardness of the X-ray spectra, it is clear that for distances of 8 kpc or greater, the three methods are consistent with the scenario of a BH as compact object.On the other hand, the hardness of the X-ray spectra rules out the NS scenario with 2 σ -confidence.For distances around ∼ 4 kpc or shorter, the conclusions on the nature of the compact object become inconsistent.The L R − L X plane is more consistent with the NS X-ray binaries, while Fig. 3 favours a BH as compact object.In order to prevent this inconsistency, the results suggest that the distance to IGR J17285−2922 favours distances around ∼ 4 kpc or larger.This supports the conclusion of Sidoli et al. (2011), who estimated the distance to be larger than 4 kpc based on the non-detection (R > 21 mag) of IGR J17285−2922 in an archival optical image.
How does this conclusion on the nature of the compact object of IGR J17285−2922 compare to the previous outbursts?IGR J17285−2922 has not been observed in radio prior to our VLA monitoring, so our conclusions regarding the L R − L X plane remain unchanged.IGR J17285−2922 has been studied in X-ray in the two previously detected outbursts (Barlow et al. 2005;Sidoli et al. 2011).Similar to the 2019 outburst, IGR J17285−2922 has always been observed in a hard state in both the 2003 and 2010 outburst, with the photon indices reported consistent with the BH sample in Fig. 3.In fact, Wijnands et al. (2015) has already suggested that IGR J17285−2922 might harbour a BH based on the hard spectra at low X-ray luminosities.The compact object in IGR J17285−2922 is therefore consistent with a BH.

The nature of the optical / nIR emission
Next, we will discuss the nature of the optical / nIR emission in IGR J17285−2922.The averaged spectrum taken with the GTC and SOAR (see Fig. 6) show a lack of H α emission, with a stringent 0.6 Å equivalent width upper limit in the GTC data, indicating that the disc might be hydrogen-poor.We also do not detect any Brγ emission in the Gemini/Flamingos-2 nIR spectrum.This could hint towards an ultracompact binary orbit, requiring a hydrogen-poor donor star to support mass transfer through Roche lobe overflow.A lack of H α emission has been reported previously in several optical spectra of UCXBs (Nelemans et al. 2004;Nelemans, Jonker & Steeghs 2006;Werner et al. 2006;Hernández Santisteban et al. 2019).We obtain equivalent width upper limits of 0.8 Å, 0.6 Å, and 0.5 Å for the He I lines at 5875.6 Å, 6678.1 Å, and 7065.2Å respectively and 3.7 Å for the He II line at 4686.0 Å.A lack of He emission may not be representative of the population of UCXBs.While a lack of He emission has been reported in, for example, the UCXB IGR J17062−6143 (Hernández Santisteban et al. 2019), the UCXB XB 1916−05 showed strong He emission lines indicative of a He-rich donor star (Nelemans et al. 2006).
While this creates a consistent picture, a lack of H α emission does not necessarily confirm that the disc is hydrogen-poor.Several other X-ray binaries (excluding UCXBs) have been observed with and without H α emission (during the same outburst).First, the strong candidate BH X-ray binary Swift J1357.2−0933 has been observed with and without H α emission during its 2011 outburst, with the non-detection and detection of H α separated by only 15 h (Casares et al. 2011;Milisavljevic et al. 2011;Torres et al. 2011).Secondly, the strong candidate BH X-ray binary Swift J1753.5−0127 has also been observed with and without H α emission (Torres et al. 2005;Cadolle Bel et al. 2007;Jonker, Torres & Steeghs 2008).Thirdly, a single optical spectrum featured no H α emission in the candidate BH X-ray binary Swift J1539.2−6227(Torres et al. 2009;Krimm et al. 2011).Finally, an optical spectrum of the NS X-ray binary 1RXS J180408.9−342058, a candidate UCXB, showed no H α emission (Baglio et al. 2016;Degenaar et al. 2016).A common denominator in these four X-ray binaries is the H α non-detection is observed in only one optical spectrum for each X-ray binary.
In the case of IGR J17285−2922, four optical spectra were taken at different times during the 2019 outburst (see Fig. 1), all showing a lack of H α. This provides stronger evidence that the donor star in IGR J17285−2922 is hydrogen poor, compared to the sources mentioned.A detailed comparison of the equivalent width upper limits with other X-ray binaries depends on the source distance, due to a possible anticorrelation between equivalent width and X-ray luminosity (Fender et al. 2009); however, the H α upper limit lies firmly below usual detections, implying that the non-detection does not arise due to signal-to-noise limits.Therefore, an ultracompact binary orbit with a hydrogen-poor donor in IGR J17285−2922 offers a viable explanation for the optical spectra.
While the absence of hydrogen might be expected, other elements such as carbon, oxygen, neon or iron could be expected in overabundance.These overabundances could be identified, for example, as absorption features in X-ray spectra (see e.g.Armas Padilla & López-Navas 2019).While our Swift/XRT X-ray spectra of IGR J17285−2922 are of too poor quality to identify these features, the high quality XMM-Newton spectrum obtained by Sidoli et al. (2011) is more promising.As noted by Sidoli et al. (2011), clear negative residual structures remain (see their fig.4) after fitting with an absorbed power law, which were partly attributed to incorrect instrumental modelling.The residual structure around ∼ 1 keV specifically could also be identified as an absorption feature as a result of an enhanced Ne/O ratio, also seen in other UCXBs (Juett, this work, we report I band detections between 18.24 and 18.78 magnitude.From a comparison with the PanSTARRS Database, we find no quiescent counterpart with at a magnitude limit of r > 23.2,7 implying a significant optical brightening during the outburst.However, it still remains unclear how much the accretion disc contributes compared to a possibly irradiated donor star or optical emission from a jet [see e.g.Russell et al. (2006Russell et al. ( , 2007) ) for a discussion].If IGR J17285−2922 is indeed an ultracompact BH system with a small disc, one might expect the optical emission to lie at lower end of the BH distribution if the optical emission is dominated by the accretion disc.Such an effect is, for instance, observed for the NS UCXB 4U 0614+09, highlighted in pink in Fig. 5, which is located on the optically faint end of the NS sample.IGR J17285−2922 is instead not located at a similarly faint position in the BH sample, except for small (<4 kpc) distances.
Possibly, the jet might contribute up to optical bands, increasing the optical flux, while no or little emission from the disc itself is observed.However, with only r-band detections due to the crowded field of IGR J17285−2922 and therefore no optical spectral constraints, this scenario is difficult to test.A less constraining consistency check can however be obtained from the radio observations: while the optical photometric and radio observations were not taken in coordination, the radio flux densities remained relatively constant between radio epochs 2 and 4 (Fig. 1) around 80-100 μJy.Assuming that the jet break frequency lies above the I band, a jet spectral index of α ∼ 0.2 would be sufficient to explain both the radio and optical fluxes from a single, broad-band jet model.This spectral index is close to the approximately flat radio-to-optical spectra typically observed in BH X-ray binaries (Russell et al. 2006).Hence, the radio and optical data are broadly consistent with a jet contributing significantly to the optical increase, instead of the accretion disc.Considering this, it is not obvious that the lack of lines in the optical spectra of these sources would be a result of the jet dominating the optical emission and washing out any lines.For instance, the strong candidate BH Xray binary Swift J1357.2-0933 had a very weak radio jet (Sivakoff et al. 2011) and 1RXS J180408.9-342058 is a NS X-ray binary that did not reach the bright X-ray regime (Gusinskaia et al. 2017) where one might expect a jet to contribute significantly to the optical emission (Russell et al. 2007).

The orbital period and the donor star
Following the discussion on the nature of the compact object and donor star, we will discuss the results of the orbital properties (Section 3.4) for IGR J17285−2922.The overall shape of the light curve can be described by an exponential decay during the first half (࣠ 58631 MJD) of the 2019 outburst, followed by a linear decay during the latter half ( 58631 MJD) as the source becomes quiescent.We list P orb for IGR J17285−2922 obtained from both τ e and L t in Table 3.The derived P orb for IGR J17285−2922 from these two methods are consistent with each other (assuming equal q) for a distance of ∼ 15 kpc.From now on, we will use P orb derived from τ e as this is distance independent.In addition to this, we give P orb for the two VFXBs CXO J174540.0−290005(unknown compact object; Koch et al. 2014) andXMM J174457−2850.3 (confirmed NS;Degenaar & Wijnands 2010;Degenaar et al. 2014) for comparison (Heinke et al. 2015).CXO J174540.0−290005 and XMM J174457−2850.3have the highest quality light curves of transient VFXBs while also showing a clear exponential and/or linear decay.For IGR J17285−2922, P orb (for equal q) is a factor 6 to 7 higher than CXO J174540.0−290005 and XMM J174457−2850.3,which translates to a significantly larger disc radius and consequently τ e .At first glance this seems to argue against an ultracompact nature for IGR J17285−2922, but we will show that this is strongly dependent on the nature of the compact object and the assumed q.
In order to give an estimate of P orb , we first need to determine an accurate range of q in X-ray binaries.We can obtain such constraints on q from both accretion theory and observations of known UCXBs.The mass ratio q can be as high as ∼ 5/6 for X-ray binaries with a low mass donor star (࣠ 1 M ), while above q = 5/6 mass transfer is expected to occur in a short-lived unstable runaway reaction until the mass ratio is brought down to q = 5/6.To determine a lower limit on q, we can turn to X-ray binaries with extremely low donor star masses.Strohmayer et al. (2018) find observational evidence for an extremely small binary mass function of f x = (9.12± 0.02) × 10 −8 M in the NS UCXB IGR J17062−6143 through the measurement of pulsations.The binary mass function f x is given by Here, M d is the donor star mass, i is the inclination, M c is the compact object mass, a c sin(i) is the projected semimajor axis of the binary orbit and G is the gravitational constant.Aside from IGR J17062−6143, six other NS UCXBs have a c sin(i) and P orb determined through the measurement of pulsations.We determine, if not already done in the literature, f x for these NS UCXBs using equation ( 8).We list the seven NS UCXBs for which f x is determined (or already known) and P orb in Table 4, along with their respective references.For each of these seven NS UCXBs, we can set a reasonable lower and upper limit on the NS mass and inclination i to constrain M d and consequently q using equation ( 8).Here, we use a 1.4 M NS as the lower limit and 2.0 M NS as the upper limit ( Özel & Freire 2016).To set a lower limit on i, a random distribution of inclination angles gives a 95 per cent probability to observe a system at an inclination above i 18.2 • (lower limit).An upper limit on the inclination can be set if the light curve shows a lack of eclipses and dips at i < 85 • (Paczyński 1971).For XTE J1751−305 and IGR J17062−6143, we use the constraints on the inclination known from the literature (Markwardt et al. 2002;Strohmayer et al. 2018).
We give for each of the listed NS UCXB in Table 4 the determined lower and upper limits on both i and q.We show the P orb − q relation of IGR J17285−2922 determined from τ e , along with the seven NS UCXBs discussed in the previous paragraph in Fig. 8. On top of this, we show the P orb − q relation for five BH X-ray binaries, which we discuss in Section 4.4.It is visible that for all seven shown NS UCXBs, mass ratios below q ∼ 0.01 are realistic to discuss.Calculations and simulations by van Haaften et al. (2012) support these observational results, showing that a 1.4 M NS primary could have a white dwarf donor star mass as low as 0.0042 M after 10 Gyr of mass transfer (q ∼ 0.003).For a 10 M BH primary, the white dwarf donor star can reach a mass as low as 0.0026 M (q ∼ 0.0003).We give these values for q as lower limits in Fig. 8 in the UCXB regime (P orb < 90 min).We show in Table 3 that for a mass ratio q = 0.005, P orb for IGR J17285−2922 can reach as low as 1.44 h.Considering that this is a crude estimation of P orb due to possible systematic uncertainties and that even lower mass ratios are in theory possible, we cannot exclude an ultracompact nature (P orb < 90 min) in recurrence time scale, distance, bolometric correction and fluence estimations for each outburst, so we only use this as an order of magnitude estimate.Comparing the inferred mass transfer rate to for example fig. 3  The BH UCXB scenario allows for P orb between ∼ 40 and 60 min., without invoking an extreme BH mass or mass transfer rate.This is harder to reconcile for the NS UCXB scenario, as it requires P orb between ∼ 60 and 90 min., an extremely low mass transfer rate and pushes the theoretical limit for q as we approach lower P orb .

Short-P orb BH X-ray binaries
IGR J17285−2922 has a peak L X ∼ 2 × 10 36 erg s −1 at 8 kpc (L X ∼ 8 × 10 36 erg s −1 at 16 kpc).We have discussed that the nature of the compact object is consistent with a BH.If true, such a low peak L X cannot be explained by a magnetically truncated disc as this scenario requires a NS.The optical spectra give strong indications for a hydrogen-poor donor star and the fit to the X-ray flux light curve shows that it is possible for this system to be a UCXB.It has been predicted and shown that the peak luminosity scales with P orb of X-ray binaries (Lasota 2001;Wu et al. 2010).Because UCXBs have P orb ࣠ 90 min, these systems could have a low peak L X .Currently, all confirmed UCXBs have NS primaries, so while we cannot compare IGR J17285−2922 to BH UCXBs, five (strong candidate) BH X-ray binaries are known with a short-P orb (࣠ 5 h), some also showing a low peak L X .Shahbaz et al. (2013) show the P orb distribution of BH X-ray binaries (their fig.7), highlighting short-P orb BH X-ray binaries.The distribution shows two distinct clusters around P orb of 8 and 50 h, separated by a possible gap.For any mass ratio q < 0.1 which we can expect for a 10 M BH in an X-ray binary, the orbital period of IGR J17285−2922 is consistent with the cluster around 8 hours and possibly indicative of a short-P orb BH X-ray binary.
We can compare the observed X-ray behaviour of IGR J17285−2922 to short-P orb BH X-ray binaries to further investigate this possibility.Among these sources are Swift J1357.2−0933(strong BH candidate; Mata Sánchez et al. 2015), XTE J1118+480 (confirmed BH; Gelino et al. 2006) Gelino & Harrison 2003).First, we can compare the derived P orb − q relation of IGR J17285−2922 to these short-P orb BH X-ray binary systems.We list, if known from the literature, f x , i, P orb and q in Table 4 similar to the NS UCXBs discussed in Section 4.3.We note that some of the constraints on i and q for these short-P orb BH X-ray binary systems are (rough) estimates and are not as well constrained as the values listed for the NS UCXBs.These constraints on i and q should only be indicative and should not be taken at face value.Nevertheless, we can include the estimates on q in Fig. 8 for comparison.We can see that the P orb − q relation for IGR J17285−2922 is consistent with the short-P orb BH X-ray binaries.
Second, the donor stars in the short-P orb BH X-ray binaries could be nuclearly evolved stars, having moved off the main-sequence (Haswell et al. 2002;Kuulkers et al. 2013;Shahbaz et al. 2013).Simulations by Pylyser & Savonije (1988) and Ergma & Fedorova (1998) show that these short-P orb BH X-ray binaries, which initially start with a ∼ 10 h P orb and M d ∼ 1 M can end up with significantly reduced P orb of ࣠ 5 h and M d of ∼ 0.15 M .These systems can reach such low P orb due to systemic angular-momentum loss accompanied by several Gyr of mass transfer.It is possible for these donor stars to have evolved off the main-sequence, with a significantly low H content (∼ 0.1-0.2).The evolutionary calculations by Ergma & Fedorova (1998) also show that it is possible for a BH X-ray binary starting with P orb ∼ 17-22 hr to evolve to a BH UCXB with P orb < 1 hr, starting out with a 1.25 M donor star and either a 4 or 12 M BH primary.
Last, the X-ray spectral behaviour of these sources in the low X-ray luminosity regime already match that of IGR J17285−2922, as the BH sample in Fig. 3 partly consists of short-P orb BH Xray binaries and BH X-ray binaries studied in Plotkin, Gallo & Jonker (2013).The X-ray spectral behaviour of Swift J1357.2−0933specifically stands out as this is identical to that of IGR J17285−2922 when fitting Swift/XRT data, requiring only an absorbed powerlaw component in a hard state throughout all observed outbursts (Sidoli et al. 2011;Armas Padilla et al. 2013;Beri et al. 2019).The position of Swift J1357.2−0933 for a distance of 6 kpc (Sivakoff et al. 2011;Plotkin et al. 2016;Paice et al. 2019) in the L R − L X plane in Fig. 2 is consistent with that of IGR J17285−2922 for a distance of ∼ 8 to 12 kpc.Comparing the light curve shape during outburst shows that while the exponential to linear decay transition is clearly seen in IGR J17285−2922, this is not clearly visible in Swift J1357.2−0933.The exponential decay part in the X-ray light curve of Swift J1357.2−0933(their fig.2; Armas Padilla et al. 2013) could be identified as the monotonically decreasing L X .While the linear decay is not clearly visible, it could be identified after the last four or so detections when L X steeply drops and is only constrained through upper limits.This is further supported by the shown UV/optical light curve, which shows a clear correlation to the X-ray light curve and also shows a steeply decreasing magnitude (their fig.2; Armas Padilla et al. 2013).As a first guess, we can estimate τ e using that L X drops by two orders of magnitude over ∼ 150 days monotonically.This results in τ e ∼ 33 days, which considering that P orb is known (2.8 ± 0.3 h) for Swift J1357.2−0933 would result in q ∼ 0.01 from equations ( 5) and ( 7).This is consistent with the previously estimated q for Swift J1357.2−0933taking into account that this is a first guess, there is an uncertainty in P orb and there are systematic uncertainties.This calculation is done for the 2011 outburst of Swift J1357.2−0933 and is consistent for the 2017 outburst (τ e ∼ 37 days from fig. 4 in Beri et al. 2019).
This considers only τ e , but a similar calculation can be done using L t .While the transition cannot be clearly identified, we can make a reasonable guess at L t ∼ 1.5 × 10 33 (d/1.5kpc) 2 erg s −1 (their fig. 2 and table 1; Armas Padilla et al. 2013).For a distance of 6 kpc (Charles et al. 2019) to Swift J1357.2−0933,we would have L t ∼ 2.4 × 10 34 erg s −1 .However, when we now use equations ( 6) and ( 7), we do not converge to a realistic q.To illustrate this, for an assumed q = 0.1, we obtain P orb = 0.3 h for the given L t , with P orb decreasing as q decreases.In order to obtain q < 0.06 (Corral-Santana et al. 2013) and P orb = 2.8 h, we would need to increase L t by more than a factor of 20 for a distance of 6 kpc.In the case where we assume P orb = 1.9 hr (3 σ lower limit), we would still need to increase L t by more than a factor of 10 to bring this in line with the upper limit on q.
This discrepancy could be explained by an overall higher intrinsic X-ray luminosity, which would make L t more in line with the observed P orb and q and to which τ e is insensitive.This has already been suggested in the literature (see e.g.Corral-Santana et al. 2013;Charles et al. 2019;Jiménez-Ibarra et al. 2019).Although the system was initially thought to be at ∼ 2 kpc, Charles et al. (2019) still require a higher intrinsic X-ray luminosity for a distance of MNRAS 507, 330-349 (2021) Downloaded from https://academic.oup.com/mnras/article/507/1/330/6327563 by guest on 14 July 2022 ∼ 6 kpc.Another plausible explanation for this discrepancy is the possibility that the value for φ H recalibrated in Heinke et al. (2015) is significantly larger for Swift J1357.2−0933.This could be similar to XTE J1751−305 and 4U 1543−475, which required φ H to be a factor of ∼ 10 smaller as noted by Heinke et al. (2015).Fitting the X-ray light curve therefore requires a careful approach and the conclusions drawn therefrom should only be indicative.In the case of IGR J17285−2922, the derived P orb − q relation from the τ e and L t method are consistent with one another at d ∼ 15 kpc.At this distance, the conclusions drawn from the L R − L X and L opt − L X plane, and X-ray spectral evolution at low X-ray luminosity are also consistent with a BH primary.

The hydrogen column density N H
Lastly, we will discuss the inconsistency in the best-fitting N H determined here and during the 2010 outburst.The best-fitting N H = (0.99 ± 0.05) × 10 22 cm −2 is determined here with a simultaneous fit of all Swift/XRT spectra for the 2019 outburst of IGR J17285−2922.The XMM-Newton combined with the INTEGRAL X-ray spectrum in Sidoli et al. (2011) has been re-fit with the same model with Wilms et al. (2000) abundances and Verner et al. (1996) crosssections (Sidoli, private communication).The best-fitting parameters are N H = (0.679 ± 0.008) × 10 22 cm −2 and = 1.59 ± 0.01 (90 per cent confidence).A significant discrepancy in N H is present.Two Swift/XRT spectra taken during the 2010 outburst are consistent within their 1 σ error with the absorption column N H determined by XMM-Newton.Therefore, the N H determined here for the 2019 outburst and the N H determined for the 2010 outburst are not consistent with one another, even accounting for uncertainties.
To check whether this difference in the absorption column N H significantly affects our conclusions, we repeated the Swift/XRT fits by assuming that the N H = 0.679 × 10 22 cm −2 is the correct value, regardless of the goodness of fit.In this case, the Swift/XRT spectra are even harder, with photon indices ranging from ∼ 1.0 to 1.5.Similar photon indices were observed in the very hard state of six NS X-ray binaries in Parikh et al. (2017).While five of the six of these NS X-ray binaries in the very hard state followed the NS track, IGR J17285−2922 lies clearly systematically below both the NS and BH tracks.If N H = 0.679 × 10 22 cm −2 is correct, it would suggest a very hard state not observed in other X-ray binaries regardless of the nature of the compact object.However, a similar discrepancy in N H has already been noted in the X-ray spectral fitting of Swift/XRT observations of Aql X-1, but did not alter any conclusions (López-Navas et al. 2020).Parikh et al. (2017) also noted that discrepancies in N H can arise from differences in assumptions, effects introduced by pile-up and adopted model for Swift/XRT observations.For these reasons, we deem it justified to use our best-fitting N H for the analysis and subsequent conclusions.

C O N C L U S I O N S
We have monitored the 2019 outburst of the (V)FXB IGR J17285−2922 in X-ray with Swift, in radio with the VLA, in optical with the GTC and SOAR (spectra) and LCO telescope network (photometry) and in nIR with the Gemini South telescope (spectrum).The location of IGR J17285−2922 in the L R − L X plane for tried distances of 4, 8, 12, and 16 kpc is consistent with both the NS and BH X-ray binaries.The location of IGR J17285−2922 in the L opt − L X plane for tried distances of 4, 8, and 16 kpc is consistent with BH X-ray binaries.From the X-ray photon index in the low L X regime, we also find evidence that the compact object in IGR J17285−2922 is a BH for distances of 4, 8, 12, and 16 kpc.
With four optical spectra from the GTC and SOAR during this 2019 outburst we find that there is considerable evidence for a hydrogen-poor donor star, as all four optical spectra show no H α emission.The nIR spectrum taken with the Gemini South telescope also reveals no Brγ emission.The low peak L X of IGR J17285−2922 and possible hydrogen-poor donor star gives indications of an ultracompact binary orbit.
The shape of the X-ray light curve can be well described by an exponential, followed by a linear decay.For an assumed q ࣠ 0.1, the orbital period of IGR J17285−2922 is consistent with short-P orb BH X-ray binaries.We compare the P orb − q relation for IGR J17285−2922 with seven NS UCXBs with constrained q and theory by van Haaften et al. (2012).This comparison supports the evidence that an ultracompact binary orbit in IGR J17285−2922 cannot be excluded.A determination of the orbital period in IGR J17285−2922 will not only give valuable information on the nature of this system, but will give an estimate on the mass ratio q as well from τ e .
We have shown here that observing a VFXB outburst with a dedicated multiwavelength campaign provides invaluable information about the nature of the compact object, donor star, and the binary orbit.Reproducing such a multiwavelength campaign for the outbursts of other VFXBs allows us to understand why the population of VFXBs in general is extremely faint.This in turn is highly warranted to improve our understanding of accretion physics (at low X-ray luminosity), binary-evolution, and population synthesis.Corner plot for the Monte Carlo Markov-Chain simulations of the simultaneous exponential and linear decay fit described in Section 3.4.We determine the value for each parameter along, with the negative and positive 1 σ error as the 50th, 16th, and 84th percentile, respectively.F t and F e are given in 10 −11 erg s −1 cm −2 , t t is given in Modified Julian Date, and τ e and τ l are given in days.
This paper has been typeset from a T E X/L A T E X file prepared by the author.

Figure 1 .
Figure 1.Top: X-ray and radio light curve for the 2019 outburst of IGR J17285−2922.The black circles show the Swift/XRT count rate, and the red and blue squares show the VLA 4-5 and 7-8 GHz radio flux density, respectively.The SOAR optical epochs are shown with the brown dashed vertical lines, while the GTC optical epoch is shown with the brown dash-dotted vertical line.The optical photometry epochs are shown with the cyan dotted vertical lines.The nIR spectroscopy epoch is shown with the green dashed vertical line.The data of the Swift/XRT observations are given inTable 1, while the data of the VLA observations are given in Table2.Bottom: The radio spectral index α for each VLA observation with at least one detection in either the 4-5 or 7-8 GHz subband.Uncertainties on all data are 1 σ .Upper and lower limits are 3 σ .

Figure 3 .
Figure3.Left-hand panel: X-ray photon index against the 0.5-10 keV X-ray luminosity for BH X-ray binaries (red), NS X-ray binaries (blue), and IGR J17285−2922 (black).We show for each of the three samples the visual representation (cut-off at L X = 10 34 for NSs because the NS sample also has this cut-off) of the results of the MC simulations described in Section 3.3 (showing 1/100th of the simulations of each sample to avoid cluttering).The data for the BH and NS sample can be found inWijnands et al. (2015) and the data for the IGR J17285−2922 sample can be found in Table1.Right-hand panel: Corner plot of the MC simulations for the slope and offset adopting equal colours as in the left-hand figure.

Figure 4 .
Figure 4. Top: The X-ray flux light curve of the 2019 outburst of IGR J17285−2922 indicated with the red circles.The light curve has been fit with an exponential, followed by a linear decay, as described in Section 3.4, indicated by the blue line and blue shaded region.Bottom: Residuals for the light curve fit shown in the top figure.Uncertainties on all data are 1 σ and smaller than their respective marker if they are not shown.

Figure 5 .
Figure 5.The optical-X-ray luminosity diagram for X-ray binaries with a low-mass donor star, based on Russell et al. (2006, 2007).Hard and soft state BH X-ray binaries are shown as the red crosses and blue stars, respectively, while NS X-ray binaries are shown as the open circles.IGR J17285-2922 is shown with the black circles for three example distances (4, 8, and 16 kpc), while the NS UCXB 4U 0614+09 is shown in pink.andNS X-ray binaries based onRussell et al. (2006) andRussell et al. (2007).For all plotted distances, the optical counterpart of IGR J17285−2922 is consistent with hard state BH X-ray binary systems, while it is between a factor ∼5-10 brighter than NS systems.

Figure 6 .
Figure 6.Top: optical spectra of IGR J17285−2922, taken with GTC-OSIRIS on 2019 July 21-22 and with SOAR-Goodman Spectograph on 2019 April 30-May 1, May 2-3, and June 29-30.The GTC spectrum was averaged from two separate spectra taken on the same night.All spectra were renormalized using MOLLY and are plotted with a vertical shift for clarity.The red line indicates H α, the blue lines show the wavelengths of three He I lines and the green line shows one He II line.The grey bands highlight interstellar Na I (left-most) and telluric absorption bands (others).Bottom: zoom-in of the four H and He line regions in velocity space, showing how none of these lines are detected in any of the four spectra.

Figure 7 .
Figure 7.The K-band portion of the Gemini/Flamingos-2 spectrum of the nIR counterpart to IGR J17285−2922.The upper panel shows the telluric transmission spectrum, derived from the standard star Hip 82714, used to correct the target spectrum.The lower panel shows the continuum normalized, telluric corrected spectrum of the target.A red box highlights the region in which we might expect to see Brγ emission.The apparent emission feature between 2.05-2.1 μm is likely a residual from the telluric correction.
in Lasota et al. (2008), fig. 1 in Heinke et al. (2013)  or fig.2inSengar et al. (2017) shows that IGR J17285−2922 aligns well with the transient UCXB systems at P orb of 50 to 60 minutes.This mass transfer rate also aligns well with the comparison of the light curve fit to the (van Haaften et al. 2012) analytic approximations.

Figure A1.
Figure A1.Corner plot for the Monte Carlo Markov-Chain simulations of the simultaneous exponential and linear decay fit described in Section 3.4.We determine the value for each parameter along, with the negative and positive 1 σ error as the 50th, 16th, and 84th percentile, respectively.F t and F e are given in 10 −11 erg s −1 cm −2 , t t is given in Modified Julian Date, and τ e and τ l are given in days.