A Wildly Flickering Jet in the Black Hole X-Ray Binary MAXI J1535–571

J1535 was extensively monitored during its 2017 outburst using the 2 m Faulkes Telescope South (FTS) located at Siding Spring (Australia), the Las Cumbres Observatory (LCO) robotic network of 1 m telescopes at Cerro Tololo Inter-American Observatory in Chile (LSC), the South African Astronomical Observatory at Sutherland in South Africa (CPT) and Siding Spring (COJ), and the 1 m ESO Rapid Eye Mount telescope (REM) in La Silla (Chile).

3.1.1. Faulkes/LCO Observations

We observed J1535 with FTS and the 1 m LCO network on 20 nights in September and October 2017, from the first optical detection until the source was no longer visible from the ground. These observations are part of an ongoing monitoring campaign of ∼40 LMXBs (Lewis et al. 2008). Imaging data were taken in the SDSS g'r'i' and Pan-STARRS y-band filters (4770–10040 Å). We detected the source (with errors of ≤0.4 mag) in zero, one, 12, and 11 images in the g', r', i', and y bands, respectively. As for the REM observations, the nondetections at shorter wavelengths are very likely due to the high foreground extinction and led us to schedule only i'- and y-band observations from September 12. A detailed observation log can be found in Table 1.

Table 1.  Detailed Log of the Faulkes/LCO Observations Performed between 2017 September 4 and October 13

Epoch	MJD	Telescope	Filters	Exposures
(UT)				(s)
2017 Sep 04	58000.37035	FTS	r'	3 × 300
2017 Sep 06	58002.42514	FTS	i'	200
2017 Sep 06	58002.78674	1m-S	g'i'	300,200
2017 Sep 07	58003.42129	1m-A	g'i'	300,200
2017 Sep 07	58003.78054	1m-S	g'i'	300,200
2017 Sep 07	58003.78771	1m-S	r'y	200,200
2017 Sep 08	58004.38974	FTS	r'	300
2017 Sep 08	58004.98666	1m-C	y	200
2017 Sep 09	58005.44592	FTS	i'	200
2017 Sep 09	58005.76490	1m-S	y	200
2017 Sep 10	58006.43538	FTS	i'	200
2017 Sep 10	58006.77602	1m-S	g'i'y	300,200,200
2017 Sep 11	58007.42086	FTS	r'i'	2 × 300,200
2017 Sep 12	58008.42079	FTS	i'	200
2017 Sep 13	58009.36657	FTS	i'	200
2017 Sep 13	58009.43914	1m-A	y	200
2017 Sep 14	58010.79202	1m-S	y	200
2017 Sep 21	58017.40310	1m-A	i'y	300,300
2017 Sep 21	58017.43811	FTS	i'y	300,300
2017 Sep 22	58018.40711	1m-A	i'y	300,300
2017 Sep 23	58019.01422	1m-C	i'y	300,300
2017 Sep 25	58021.42373	FTS	i'y	300,300
2017 Sep 25	58021.98683	1m-C	i'y	300,300
2017 Oct 02	58028.99327	1m-C	i'y	200,200
2017 Oct 03	58029.37504	1m-A	i'y	200,200
2017 Oct 04	58030.98582	1m-C	i'y	200,200
2017 Oct 06	58032.37651	1m-A	i'y	200,200
2017 Oct 07	58033.41709	FTS	i'y	300,300
2017 Oct 13	58039.39960	FTS	i'y	300,300

Note. FTS is the 2 m Faulkes Telescope South in Australia, 1m-A is a 1-m Node in Australia, 1m-C is a 1 m node in Chile, and 1m S is a 1 m node in South Africa.

Download table as:  ASCIITypeset image

The Faulkes/LCO data were reduced (debiased and flat-fielded) using the LCO automatic pipeline. Photometry was performed using PHOT in IRAF.³³ For photometric calibration of the r' and i' bands, we used four close-by stars listed in the VST Photometric Hα Survey of the Southern Galactic Plane and Bulge (VPHAS+) Data Release 2 (Drew et al. 2014, 2016). The r, i Vega catalog magnitudes of these stars were converted to r', i' AB magnitudes, which is standard for SDSS filters. We found no catalogs with y-band magnitudes of any field stars (Pan-STARRS does not cover this field). However, we performed log-log polynomial fits to the g', r', i', J, H, and K spectral energy distributions (SEDs) of these four field stars using the VPHAS+ and 2MASS catalogs, and we interpolated the fluxes of each star to the y-band frequency in order to estimate the y-band magnitudes of the four stars. The polynomial curves provided good approximations to the reddened, blackbody SEDs of the stars (Figure 1). In all filters, the errors are dominated by the signal-to-noise ratio (S/N) of the target (not the systematic calibration uncertainty). The LCO light curves are shown in the upper panel of Figure 2.

Zoom In Zoom Out Reset image size

3.1.2. REM Observations

MIR observations of the field of J1535 were made with the Very Large Telescope (VLT) on seven nights from 2017 September 12 to 23, under the program 099.D-0884 (PI: D. Russell). The VLT Imager and Spectrometer for the mid-Infrared (VISIR; Lagage et al. 2004) instrument on the VLT's UT3 (Melipal) was used in small-field imaging mode (the pixel scale was 45 mas pixel⁻¹). Three filters (M, J8.9, PAH2₂) were used on different dates spanning central wavelengths 4.85–12.13 μm. The VISIR observing log, including photometric results, is reported in Table 3. For each observation, the integration time on source was 1000 s, composed of 22 nodding cycles. With chopping and nodding between source and sky, the total observing time was 1800–1900 s per observation.

The data were reduced using the VISIR pipeline in the gasgano environment. Raw images from the chop/nod cycle were recombined and sensitivities were estimated based on standard star observations taken on the same night in the same filters. Aperture photometry was performed using an aperture large enough to ensure small seeing variations did not affect the fraction of flux in the aperture. The standard stars used are listed in the final column of Table 3. The flux densities of J1535 in the table were measured from the resulting combined frames.

We found that J1535 was bright enough on the first four epochs to measure the flux density of the source in individual nodding cycles. We produced short-timescale light curves on these four dates; these are presented in Figure 3. Clear MIR variations are observed on the time resolution of the observations, which is 80–90 s. In order to test whether these variations are intrinsic to the source, we investigate if sky variations, changing conditions, or low S/N could account for the apparent variability. The overall long-term and night-to-night stability of the photometric calibration of VISIR is described in Dobrzycka et al. (2012), and the short-term stability was investigated during commissioning. From these tests, under photometric conditions, the sky transparency variations were less than a few percent, and under clear conditions larger variability (up to ∼10%) was measured, although it is unlikely that the timescale of MIR variability would be as short as minutes, as appears to be the case for J1535. Moreover, the flux of J1535 varies by a factor of ∼2–3 on timescales of minutes, which is in clear excess of (≳20–30 times greater than) the variations expected from different conditions. VISIR has been used to study night-to-night variability in some objects (e.g., van Boekel et al. 2010), but we are not aware of any other published fast photometric variability studies using VISIR.

To further investigate if changing sky conditions could account for the apparent variability in J1535 at the time of the observations, we inspected the variations of the standard stars over the same dates. We found that the ADU/flux conversion factor measured from each standard star observation (in order to flux calibrate the data) varied by 7%–8% over different dates (from September 11 to 23) under different weather conditions (clear or photometric) at different air masses. The variations in conversion factor on the level of 7%–8% in the standards from night to night are likely due to differences in the sky conditions and air mass, as well as intrinsic differences between the conversion factors derived for different standard stars. J1535 clearly has a lower S/N than the standard star observations, but the lower S/N cannot account for the observed variations in J1535 since the error bars on each flux measurement (which take into account variations in the background) are much smaller than the difference between fluxes in the same 30-minute light curves (Figure 3). Possible background variations due to the water vapor content of the atmosphere above Paranal have moreover been monitored during our observations, revealing no clear variability of the background, which remained essentially constant during the observations. We therefore conclude that the MIR variations on minute timescales are almost certainly intrinsic to the source. In Table 4 we quantify the properties of the MIR variability of J1535 on the three dates in which the source was significantly detected in all 22 frames (these three dates were all under photometric conditions). We measured the intrinsic fractional rms variability (which subtracts the contribution to the variations from Poisson noise) by adopting the method of Vaughan et al. (2003) and Gandhi et al. (2010).

J1535 was observed using the imaging camera SALTICAM (O'Donoghue et al. 2006) on the Southern African Large Telescope (SALT; Buckley et al. 2006) near the beginning of the outburst, on 2017 September 8 at 17:37–18:34 UT (MJD 58004.7). In the clear (white light) filter, 110 consecutive images of the field were made, each of exposure 30 s. These observations were done in frame transfer mode, with no dead time between exposures. Automatic image reduction, star identification, and relative magnitude measurements were run under the PySALT pipeline (Crawford et al. 2010). J1535 was clearly detected in all frames. In Figure 4 we present the light curve, which shows the intrinsic variability of J1535. The light curve of a nearby, slightly fainter field star is also shown and is clearly much less variable. The variability properties are included in Table 4, adopting the same method as above to calculate the fractional rms variability. From differential measurements with USNO-B1.0 cataloged field stars, we estimated the brightness of MAXI at the time of the SALT observation to be ∼19.

The optical and NIR light curves obtained with the REM and LCO telescopes are shown in Figure 2 (upper panel), in comparison with the MIR light curves obtained with VLT–VISIR (Figure 2, middle panel) and the Swift BAT 15–50 keV and MAXI 2–10 keV (Figure 2, bottom panel) light curves during the same period of time.

Zoom In Zoom Out Reset image size

The transition from a hard to a softer state reported in Shidatsu et al. (2017b) on MJD 58015 is clearly visible in the Swift light curve as a significant decrease of the hard X-ray flux. Unfortunately, we do not possess an optical/NIR observation on that day, which falls in a gap of our data set; however, a large drop in the flux (more prominent going toward lower frequencies, reaching ∼2 mag in the K band; see Figure 5) is clearly noticeable between MJD 58012 and 58017, that is, around the same time of the X-ray spectrum softening. This frequency dependence of the fade suggests that, at least at NIR frequencies, the component that dominates the emission is likely to be the jet, which is quenched once the X-ray spectrum of the source softens, as expected in the case of BHBs (e.g., Homan et al. 2005; Coriat et al. 2009; Cadolle Bel et al. 2011; Russell et al. 2013b, 2014). Despite the flux drop, the light curves display large variability on daily timescales, with both intraobservation flares and dips during the whole monitoring. In the NIR in particular, variations up to 1.2 mag in the K band between two near epochs (i.e., on a timescale of ∼1 day) are observed. These features are uncommon for BH candidate LMXBs, which normally show a single NIR drop when the source transitions to the soft state when the jet is quenched, and then again a single NIR rise at the end of the outburst, when the source goes back to the hard state (see, e.g., the case of GX 339–4, Homan et al. 2005; Coriat et al. 2009; Cadolle Bel et al. 2011; Buxton et al. 2012).

Zoom In Zoom Out Reset image size

Also in the MIR, a dramatic fading of the flux density corresponding to the softening of the X-ray spectrum around MJD 58015 is observed. In particular, a flux decrease of a factor of 7 in the J8.9 (8.72 μm) band between MJD 58008 and 58018 and of a factor of 6 in the PAH2_2 (12.13 μm) band between MJD 58011 and 58019 is detected (see Figure 2, middle panel).

Moreover, a large amount of intrinsic variability is observed in the MIR (Figure 3). The source being very bright in several epochs (see Table 3), the S/N in the observations was high enough to search for short-term (minutes) variations in the light curves obtained with the VISIR instrument. The brightest observation in particular (MJD 58011 in the PAH2_2 band) can be split into 22 images of ∼80 s time resolution each, and the flux appears to vary significantly, by a factor of ∼2 in less than 15 minutes (see Figure 3). The measured fractional rms is ∼15%–22% (see Table 4), which is comparable to the fractional rms of GX 339–4 in the optical at a similar time resolution (Gandhi 2009; Gandhi et al. 2010) and in the K band (13.20 ± 0.05%; Vincentelli et al. 2018).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To investigate this variability further, we also searched for short-timescale variability in the NIR data set, by splitting the observations at higher S/N into five different images of 30 s integration each. The NIR flux densities varied significantly on some of the dates, with a maximum variation of a factor of ∼2.5 in less than 1 minute in the H band on MJD 58008. Interestingly, MJD 58008 corresponds to the epoch before the softening of the X-ray spectrum, in which the lowest flux was reported in our NIR campaign (see Figure 2), and shows light curves that are variable at a 3σ confidence level. These NIR observations are simultaneous with the last part of our MIR J8.9 band observations, which showed a sudden decrease of the flux density to undetectable levels after the first ∼10 minutes of a high detected flux (∼40 mJy). Interestingly, moreover, this temporary fading of the infrared emission corresponds to a dip in the hard X-ray light curve, that is, to a short-lived softening of the X-ray spectrum.

To obtain an estimate of the absorption column density and associated uncertainties, which is essential in order to deredden our infrared and optical fluxes and to build broadband spectra of the target, we first started by fitting the Swift/XRT spectrum of J1535 acquired during a 1.1 ks observation carried out on 2017 September 11 (obs ID: 00010264004) with an absorbed power-law model. Pile-up effects were mitigated by excising the inner portion of the source PSF (within a radius of 12 arcsec). Absorption by the interstellar medium along the line of sight toward the source was modeled via the TBabs model (Wilms et al. 2000) with abundances from Anders & Grevesse (1989). The fit was statistically acceptable, yielding a reduced ${\chi }_{\nu }^{2}$ = 1.08 for 715 degrees of freedom. The inferred value for the column density and power-law photon index were N_H = (2.62 ± 0.02) × 10²² cm⁻² and Γ = 1.96 ± 0.01. This N_H leads to a V-band absorption coefficient A_V of 9.13 ± 0.50 (Foight et al. 2016).

If we consider the abundances from Wilms et al. (2000) instead of those of Anders & Grevesse (1989), we obtain a higher value of the N_H (3.84 ± 0.03 × 10²² cm⁻²), which is consistent with what is reported in, for example, Kennea et al. (2017). With this N_H, A_V = 13.38 ± 0.73 is inferred. The principal difference between the two abundances lies in the fact that Anders & Grevesse (1989) used the solar abundances as the reference ones for the interstellar medium (ISM), while Wilms et al. (2000) used a more accurate estimate of the ISM abundances, even if still very uncertain (see Wilms et al. 2000 for further details). Anders & Grevesse (1989) abundances typically tend to favor a lower value of N_H, as we found.

The lowest value of the column density estimate for this source is instead given by the Galactic N_H, which is 1.43 × 10²² cm⁻² in the direction of the source (Kalberla et al. 2005), from which we estimate A_V of 4.98 ± 0.27 and ${E}_{B-V}$ of 1.61 ± 0.09. This suggests that the N_H measured from the modeling of the X-ray spectrum may have a strong component that is intrinsic to the source, and therefore may not be a good tracer of the dust absorption along the line of sight.

Moreover, Tao et al. (2018) reported on evidence of variable N_H during the days covered by our optical and infrared observations. However, if the local N_H is not due to the presence of dust, the N_H/A_V relation of Foight et al. (2016) is not applicable, meaning that a changing N_H does not necessarily affect the A_V absorption. To verify if this is the case for MAXI J1535, we first considered a hypothetical source with constant infrared and optical fluxes, and we considered a possible changing A_V that is due to the changing N_H (as reported in Tao et al. 2018). We then dereddened the constant fluxes of the source using these values of A_V, and we built new light curves. This test resulted in an optical i'-band light curve that varied by 2.5 mag (a factor of ∼10 in flux), which is a higher amplitude than observed. This test also predicts the K-band light curve to be approximately constant but slightly brighter after the transition, whereas the opposite is observed (Figure 2).

We then searched for possible correlations between the infrared–optical colors measured for MAXI J1535 and the varying N_H measured from the modeling of the Swift spectra during the outburst (Tao et al. 2018), but no correlation was found. Therefore, we conclude that while the intrinsic N_H is changing wildly (as measured from X-ray spectra), there is no corresponding expected change in the infrared–optical flux or colors, so intrinsic A_V is negligible.

Further arguments that could explain a lack of dust intrinsic to the source might be that (1) the absorber resides between the optical–IR and X-ray emission regions, or (2) the absorber is not large enough to cover the entire optical–IR emitting region, or (3) the absorber may be dust-free or have a very low dust-to-gas ratio. Oates et al. (2018) recently discovered that in V404 Cyg (which had such a high variable N_H that it sometimes was Compton thick), the dust-to-gas ratio must be on the order of 10⁴ smaller than that typically observed in the Milky Way, in order to not show significant change in color as the absorption increased by a factor of 1000. The absorber has to be within or at least close to the dust sublimation radius in order to be dust-free. Oates et al. (2018) found that for V404 Cyg, the X-ray luminosity of V404 in quiescence resulted in a dust sublimation radius comparable to the size of the accretion disk. V404 Cyg is a system with a long orbital period, so it is extremely likely that for MAXI J1535 in outburst, with orders-of-magnitude higher X-ray luminosity than V404 Cyg in quiescence, the dust sublimation radius will be orders of magnitude outside the orbital separation of the X-ray binary. On the other hand, the intrinsic N_H is likely to originate from an absorber located very close to the black hole (as was the case in V404 Cyg; see, e.g., Motta et al. 2017), since it is variable on short timescales.

The difficulty in determining the dust absorption coefficients in the direction of the source introduces large uncertainties when constructing the dereddened optical–IR (OIR) SEDs for J1535. Due to the high variability of the source, moreover, building an average spectral energy distribution of the target on a particular date is challenging. We therefore searched for observations that were as simultaneous as possible, in order to overcome at least this last issue.

Thanks to the REM telescope, we possessed simultaneous data sets in the optical, where unfortunately the target is barely detected only in the reddest bands due to the high extinction of the field. The NIR REM images are almost simultaneous with our optical observations (see Table 2), and in three epochs (MJD 58008, −11, −12), we also have MIR observations that are very near in time to the REM ones (see Table 3; in particular, we notice that the last part of the MJD 58008 MIR observation overlaps with the NIR one). Therefore, we can construct the OIR SEDs at least on these three epochs, bearing in mind that the contribution of the jet might produce significant changes in the SED over timescales of minutes, as observed in the MIR light curves. The three SEDs on MJD 58008, 58011, and 58012 (2017 September 12, 15, and 16, respectively), dereddened using three different values of N_H (1.43 × 10²² cm⁻²; (2.62 ± 0.02) × 10²² cm⁻²; (3.84 ± 0.03) × 10²² cm⁻²; see Section 4.2), are shown in Figure 6.

Zoom In Zoom Out Reset image size

As is clearly visible in Figure 6, considerable changes in the shape of the SEDs are obtained by modifying the value of N_H used to deredden the fluxes. Using the lowest estimate of N_H, the NIR–optical spectrum is described by a power law with a very steep slope (α ∼ −3), which is difficult to interpret. In fact, if we link the OIR emission of the system to the presence of a jet, we would expect to observe in the NIR–optical an optically thin synchrotron spectrum, which is normally described by a less-steep power law than what we observe (typically α ∼ −0.8; see, e.g., Gandhi et al. 2011). Instead, if the accretion disk is contributing to the NIR–optical emission, we would expect to observe a positive slope, due to the blackbody tail of the disk, which is clearly not our case. This is therefore an indication that the galactic N_H is probably an underestimate of the real absorption of light on the target line of sight.

The second estimate of N_H (2.62 ± 0.02  × 10²² cm⁻²; Figure 6, middle panel) gives instead SEDs that are not extremely steep (except for MJD 58008, where the i-band flux is really low with respect to the lower frequency points), with a slope of ∼−1.5 and −0.8 on MJD 58011 and 58012, respectively. However, the contribution of the accretion disk is still missing at the highest frequencies, which is uncommon for an X-ray binary in outburst.

Instead, using N_H = 3.84 ± 0.03  × 10²² cm⁻², the contribution of the disk starts to be evident, and a clear infrared excess due to the emission of a jet is visible (see Figure 6, right panels). In particular, the slope of the power law that we used to fit the lower frequency points of the MJD 58011–12 SEDs are ∼−0.8 to −0.6, respectively, which are values that are typically measured for the optically thin synchrotron spectrum of a jet.

For this reason, from now on, the value of N_H of 3.84 ± 0.03  × 10²² cm⁻² will be used to deredden the fluxes of J1535. The corresponding absorption coefficients evaluated for all wavelengths that are relevant in this work are reported in Table 5. However, we caution the reader that the dereddened spectral slopes depend critically on the uncertain extinction.

Focusing on the right panel of Figure 6, we notice that on all dates the negative-index power law extends up to the MIR. If we interpret it as optically thin synchrotron emission from the jet, this suggests that the jet break frequency might fall at lower frequencies than the MIR, that is, in the radio to far-IR bands (with upper limits to the jet break frequency of 6.2 × 10¹³ Hz and 2.5 × 10¹³ Hz on MJD 58011 and 58012, respectively). On MJD 58008, the slope of the SED at lower frequencies is lower with respect to MJD 58011 and 58012, which suggests a lower contribution of the jet at those frequencies on that day. However, we caution that MJD 58008 is the epoch in which the MIR observation shows the sudden drop in flux after the first ∼15 minutes of observation, until it becomes undetectable. This strong, short-timescale MIR variability might therefore distort the low-frequency shape of the SED at this epoch. Moreover, Tetarenko et al. (2017) reported on radio detections of J1535 with ALMA on the same day, with a preliminary flux measurement of ∼220 mJy at 7 and 140 GHz, which is much higher than what we measure in our J8.9-band VISIR observation (∼63 mJy after dereddening).

In Figure 7, all of the nearly simultaneous SEDs collected in our campaign are shown, from red to purple following the rainbow colors (as explained above, N_H = (3.84 ± 0.03) × 10²² cm⁻² has been used to deredden the fluxes). The drop in the NIR flux that is observed in the light curves (see Figure 2) after MJD 58012 is clearly noticeable in the SEDs, with a large modification of the NIR SED shape between MJD 58012 and MJD 58019. In particular, if we fit the NIR–optical SEDs with a power law, the slope evolves from almost flat to positive (see Figure 7), with a residual infrared flux in the MIR, which might show that the contribution of the jet is lower but still detectable at the lowest frequencies, even if the X-ray spectrum softened (as reported in Shidatsu et al. 2017b). Also in the optical, the small drop in flux due to the decrease of the jet contribution is visible in the SEDs. A small drop is expected if the accretion disk has a stronger contribution to the optical than to the NIR.

Zoom In Zoom Out Reset image size

If we then look at the MIR part of the SEDs, we can observe the fluxes changing considerably between the different epochs. The highest average flux is reached in the PAH2_2 band (12.13 μm), on MJD 58012, with a dereddened value of 157 ± 4 mJy; after that, the average MIR flux decreases, reaching its lowest value on MJD 58020 (27 ± 4 mJy). The first measurement after the softening of the X-ray spectrum is a dereddened flux density in the J8.9 band of 20 ± 3 mJy (MJD 58019), which is a factor of 7 lower than the flux density measured just before the softening in the same band (i.e., 141 ± 12 mJy on MJD 58013). It is therefore evident that we are observing in the MIR the emission of a transient component, which is suppressed as the X-ray spectrum softens, as usually happens in the case of jets quenching over the hard-to-soft transition in BHBs.

As shown in Figure 2, the NIR and optical fluxes stay at a high level during the first phases, which correspond to the rise of the X-ray outburst; at the same time, the X-ray spectrum was reported to be hard by several authors (see Section 2), as typically happens during the rise of a BHB X-ray outburst. After MJD 58015, however, that is, when Swift detected a first softening of the spectrum (Shidatsu et al. 2017b), the infrared and optical fluxes underwent a prominent drop, which indicates that one of the players in the infrared–optical emission was experiencing a suppression in correspondence to the softening of the X-ray spectrum.

From studies of the coupling between accretion and ejection in BHBs, it is clear that a compact jet is produced during the hard state of BHB outbursts, which is then quenched as the system passes to the soft state, due to the observation of a strong suppression of the radio emission after the transition (see Tananbaum et al. 1972; Fender et al. 1999, 2004; Corbel et al. 2004). Less clear is whether the same effect should also be visible at higher frequencies, where the contribution of the jet should not always dominate the emission. However, several cases in which this happens have been observed so far (see, e.g., Homan et al. 2005; Russell et al. 2006, 2013a; Coriat et al. 2009). In particular, if the jet largely contributes to the IR–optical emission, it is likely that we will observe a weakening of the IR–optical flux during a hard-to-soft state transition, which should be more prominent toward lower frequencies, where the contributions of the outer accretion disk and of the X-ray reprocessing are typically lower. This is exactly what we observe: the flux drop corresponding to the softening of the X-ray spectrum is in fact higher in the MIR and in the NIR, where we have a decrease of a factor of 6 at 12.13 μm on an eight-day timescale, than in the optical, where the measured drop corresponds to 0.91 mag in the i' band (i.e., a factor of ∼2 in flux; see Figure 5).

If we follow the jet suppression interpretation to explain the flux drop that we observe, this implies that the jet was contributing at least 83% ± 4% of the flux in the PAH2_2 band (12.13 μm), 86% ± 9% in the K band, and 57 ± 21% in the i' band. This is consistent with what was found in Russell et al. (2006) for a sample of BHBs, for which the jets have been found to contribute ∼90% of the NIR emission and below 76% of the optical emission in the hard state.

Inspecting the SEDs (Figure 7), we see that before the softening occurred on MJD 58015 in the IR regime, the spectrum is described by a power law with a negative index, with slopes ranging from −1 to −0.6, which are typical of optically thin synchrotron emission from a jet (adopting N_H = (3.84 ± 0.03) × 10²² cm⁻²). In the optical, the slope of the power law is always positive, but it becomes steeper after the drop, which shows that the accretion disk contribution becomes more important than the emission of the jet at those frequencies after the drop. This is also further evidence of the fact that the jet is contributing not only in the radio band, as reported by Russell et al. (2017b) and Tetarenko et al. (2017), but also at higher frequencies, up to the optical band, with the jet break frequency probably lying at lower frequencies than the MIR. We therefore interpret the observed drop in the IR–optical flux as the weakening of the emission of the jet while the X-ray spectrum softens.

The fade of the IR–optical flux near the start of the transition toward the soft state has already been observed in other BHBs. After the drop, the IR flux usually does not increase again, until the source transitions back to the hard state in the last phases of the outburst. Indeed, this was the case for XTE J1550–564 (Jain et al. 2001; Russell et al. 2010), MAXI J1836–194 (Russell et al. 2013a; Russell et al. 2014), and from repeated outbursts of GX 339–4 (Coriat et al. 2009; Buxton et al. 2012; Dinçer et al. 2012; see also Kalemci et al. 2013; Corbel et al. 2013). However, here we find that after the softening in J1535, the optically thin component seems to be still present in the MIR, even if strongly suppressed. In addition, inspecting the infrared light curves (Figure 2), we see the NIR flux experiences a sudden increase on MJD 58019–21, which does not have any correspondence with the optical light curves. This excludes that this flaring NIR activity might originate in the reprocessing of the accretion disk emission. It is therefore likely that the origin of these NIR flares is in the jet, which might experience a reflaring after the target started to soften. This hypothesis is supported by the SEDs, where it is clear that after the softening the jet seems to be still present at lower frequencies (MIR). Nearly simultaneous observations in the radio band might confirm or refute this interpretation.

As described in the semiquantitative model of Fender et al. (2004) and Fender et al. (2009; see also Vadawale et al. 2003), when a BHB is in a transitional phase between the hard and the soft state, it is likely to produce several synchrotron spectra from discrete ejections, which evolve toward lower frequencies with time, as the compact jet becomes more intermittent. This is seen most prominently at radio frequencies, later during the transition, close to the "jet line" when bright, optically thin radio flares are commonly observed. Multiple radio flares have been observed when BHBs perform hard–soft excursions in the HID during intermediate states. For example, GRS 1915+105 undergoes small loops in its HID that are accompanied by many radio flares from multiple ejections (Fender et al. 2004). Here, we instead detect IR flaring at earlier (harder) stages of the spectral transition.

In addition to the flaring after the main IR fading, we also notice that on MJD 58008 (hours before our infrared observations), ALMA millimeter and ATCA radio observations have been reported (see Tetarenko et al. 2017; Section 2), revealing that the flux in the millimeter band is found at a much higher value than what we measure in the MIR (>200 mJy at 97 and 140 GHz, versus ∼63 mJy in the J8.9 band of VISIR after dereddening). As described in Section 4.1, the MIR light curve that we extracted for the MJD 58008 epoch is highly variable, the source becoming undetectable after a certain time. Interestingly, the same decrease in the flux is also observed in the NIR on the same date, which on that day is strictly contemporaneous with the last part of the MIR observations (i.e., when the flux has already decreased). This behavior could again be interpreted as an intermittent jet, as well as flares within the jets, and appears to be associated with a small excursion in the HID, before the main softening.

J1535 is the first BHB for which mid-infrared short time variability has been observed so far (with rms ∼15%–22%; Table 4). It is suggestive of the presence of a flickering jet, similar to what was found from MIR observations of GX 339–4 on longer (hour) timescales (Gandhi et al. 2011). Similar levels of fractional rms were observed in other BHBs at higher frequencies, like in the case of GX 339–4 in the optical (Gandhi et al. 2010). In that case, a study of the correlation between optical and X-ray (fast and slow) variability allowed the authors to exclude the accretion disk radiation reprocessing as the cause of the observed ∼15% optical rms, which was moreover higher for redder frequencies. Similar to what was suggested by Gandhi et al. (2010), a model that relates possible perturbations in the accretion flow to variability in the jet might explain the observed variability of our light curves. Such a model has been developed in Malzac (2013) and Malzac (2014; see also Drappeau et al. 2017). In this scenario, a strongly variable accretion flow could inject nonnegligible velocity fluctuations at the base of the jet, which would drive internal shocks at large distances from the BH, where leptons are accelerated, giving rise to strongly variable synchrotron emission, which would in turn affect the radio–infrared emission of the system. This model has been found to be compatible with the MIR variability and SED observed by Gandhi et al. (2011) in GX 339–4 (Drappeau et al. 2017). We have found that the strong observed infrared variability is correlated with changes in the X-ray emission of the system on long timescales (days).

A detailed quantitative comparison of the consistency of the internal shocks model (Malzac 2013, 2014) with our observations goes beyond the scope of this paper. However, a comparison between what we observe and the expectations from the model can be easily performed. Using Figure 6 of Malzac (2014), we can infer the value of the fractional rms that is expected for optical and infrared light curves in a BHB with typical jet parameters (average Lorentz factor γ = 2 for a kinetic power P = 1.3 × 10³⁷ erg s⁻¹, jet half-opening angle Φ = 1°, mass of the BH M = 10 M_⊙, and so on; see Section 4 of Malzac 2014 for more details). In particular, it is clear from the figure that the expected optical rms should be higher than the mid-infrared one in the Fourier frequency range that spans from the frequency corresponding to the total duration of each of our MIR light curves (∼30 minutes, out of scale in Figure 6 of Malzac 2014) to that associated with the time resolution (∼85 s; see Table 4). In the MIR, the fractional rms predicted by the model is ∼17% at a Fourier frequency of 1.2 × 10⁻² Hz, which corresponds to the time resolution of our MIR observation for which the most significant rms has been detected (81.2 s with the PAH2_2 band on September 15; see Table 4). Interestingly, this value is consistent with our findings (rms ∼ 15%; Table 4). In the optical, instead we obtain from the SALT light curve a much lower rms than expected (only 3%–4% rms; see Table 4). However, since J1535 is a BHB in outburst, it is likely that the accretion disk is playing a major role in the optical, thus diluting the jet emission at those frequencies, and therefore decreasing the detected fractional rms due to the internal shocks in the jet. A more detailed calculation of the expected rms should be performed by tailoring a simulation for J1535 so that it can reproduce the SED of the target. However, we do not expect the predicted rms to vary dramatically with the model parameters.