Delayed Radio Flares from a Tidal Disruption Event
DDelayed Radio Flares from a Tidal DisruptionEvent
A. Horesh , ∗ , S. B. Cenko , , I. Arcavi , Racah Institute of Physics, The Hebrew University of Jerusalem, Jerusalem 91904, Israel Astrophysics Science Division, NASA Goddard Space Flight Center, Mail Code 661, Greenbelt, MD 20771, USA Joint Space-Science Institute, University of Maryland, College Park, MD 20742, USA The School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel CIFAR Azrieli Global Scholars program, CIFAR, Toronto, Canada ∗ [email protected] Radio observations of tidal disruption events (TDEs) - when a staris tidally disrupted by a supermassive black hole (SMBH) - providea unique laboratory for studying outflows in the vicinity of SMBHsand their connection to accretion onto the SMBH. Radio emissionhas been detected in only a handful of TDEs so far. Here, we reportthe detection of delayed radio flares from an optically-discoveredTDE. Our prompt radio observations of the TDE ASASSN-15oishowed no radio emission until the detection of a flare six monthslater, followed by a second and brighter flare, years later. We findthat the standard scenario, in which an outflow is launched brieflyafter the stellar disruption, is unable to explain the combined tem-poral and spectral properties of the delayed flare. We suggest thatthe flare is due to the delayed ejection of an outflow, perhaps fol-lowing a transition in accretion states. Our discovery motivatesobservations of TDEs at various timescales and highlights a needfor new models.
Synoptic time-domain surveys have been increasingly fruitful indiscovering nearby tidal disruption events (TDEs) over the last sev-eral years . These transient events which are interpreted as stars beingtidally disrupted by super-massive black holes (SMBHs), may providea window into many diverse astrophysical questions. Uncovering dor-mant SMBHs is only one of the revelations made through these events.The complex physical process of accretion of matter onto a SMBH isanother. The nature of TDEs and their emission mechanisms are stillpuzzling. For example, what generates the ultraviolet (UV) and opticalemission? Is it a process related to accretion or maybe internal shocksin streams of stellar debris ? Does an accretion disk form aroundthe SMBH, and if so, when and in which geometry ? Panchromaticstudies of a growing number of events hold the key for unlocking manyof the remaining open questions.An example is the progress made in recent years by uncoveringpotentially two distinct sub classes of TDEs: (1) thermal TDEs, dis-covered via their optical/UV emission, and (2) relativistic events suchas Swift
J1644+57 [ref.7–9], which exhibit high-energy non-thermalemission. Until recently, thermal TDEs were not seen to exhibit strongradio emission, but this changed with the discovery of a prompt radiosignal from the nearby event ASASSN-14li [ref.12]. The weak ra-dio emission has been interpreted as originating from a sub-relativisticshockwave launched into the SMBH circumnuclear material (CNM),driven by either outflows from an accretion disk or by the unboundstellar debris traveling away from the central SMBH (another plau-sible explanation is that the outflow is a jet that slowed down ). On theother hand, Swift
J1644+57 exhibited a strong radio afterglow , or-ders of magnitude more luminous than the emission in ASASSN-14li,originating from a relativistic jet that is viewed on-axis.The field of TDE radio observations have seen additional devel-opments in recent years. Very long baseline radio observations of aninfrared (IR) transient that is considered a TDE candidate (Arp 299B-AT1), revealed a radio jet . Moreover, a recent radio transient, discov- ered independently of detection at other wavelengths, is also attributedto a TDE . A recent excitement is the discovery of a coincident neu-trino with an the optically discovered TDE , AT 2019DSG, which alsoexhibits radio emission similar to the one in ASASSN-14li. These pastdiscoveries are just additional pieces of the puzzle, which will hope-fully allow building a coherent picture of the overall physical processesin play.In search for similar radio emission from other nearby TDEs, wecarried out radio ( - GHz) observations, using the Karl Jansky VeryLarge Array (VLA) telescope and recently also using the ArcminuteMicrokelvin Imager (AMI) telescope, of a number of optically discov-ered TDE candidates at various time scales from early to late times (A.Horesh et al., manuscript in preparation). While most of the observa-tions resulted in null-detections, a single event, named ASASSN-15oi,revealed a delayed radio flare, months after its optical discovery, fol-lowed by somewhat even more surprisingly a second flare years later.ASASSN-15oi was discovered in optical wavelengths by the All-Sky Automated Survey for SuperNovae (ASASSN ) on 2015 August14 [ref. 21] at a distance of Mpc. At the time of discovery ithad an optical magnitude of V ∼ . while previously it was notdetected on 2015 July 26 down to V (cid:46) . , suggesting it was discov-ered relatively young. An optical spectrum obtained by the PESSTOcollaboration on 2015 August 20, provided the initial classification ofASASSN-15oi as a TDE . Upon discovery, multiple groups launchedpanchromatic monitoring campaigns. In the optical, additional spec-troscopy of ASASSN-15oi confirmed the initial classification of theevent as a TDE and showed a rapid spectral evolution (compared toother optically discovered TDEs). A search for high-energy emissionby the Neil Gehrels Swift
Observatory, initially showed no significantemission . However, deeper X-ray observations revealed what initiallyseemed as a non varying weak X-ray source, but which later increasedin flux . In the radio, we launched a monitoring campaign using theVLA. Discovery of a Delayed Radio Flare
Our radio campaign began on 2015 August 22, eight days afterthe optical discovery. Our initial observation performed in both GHzand GHz, resulted in null-detections with σ limits of ≈ µ Jy( . × erg s − Hz − ) and ≈ µ Jy ( . × erg s − Hz − ),respectively. Despite the non-detections, we kept observing the source,motivated by some theoretical models that suggest a delay betweenthe TDE optical flare and the formation of the accretion disk .We observed ASASSN-15oi twice more on 2015 September 6 ( ∆ t =23 days, where ∆ t is the time since optical discovery) and November12 ( ∆ t = 90 days), observations which also resulted in null-detections,until the discovery of significant radio emission on 2016 February 12( ∆ t = 182 days; Figure 1 and Extended Data Figure 1), with an ap-proximate flux density of ≈ µ Jy ( . × erg s − Hz − ), at a1 a r X i v : . [ a s t r o - ph . H E ] F e b he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.peak frequency of ∼ . GHz. Once ASASSN-15oi was detected inthe radio, we embarked on a followup observing campaign, carryingout observations in multiple radio frequencies in order to characterizethe properties of the broadband radio spectrum and its evolution.Separately from our followup observing campaign, the position ofASASSN-15oi was observed recently as part of the Very Large ArraySky Survey (VLASS ). Inspection of the quick-look images (producedby the National Radio Astronomy Observatory [NRAO]) reveals a re-brightening of the radio emission at GHz on 01 July 2019 (almost4 years after the initial optical discovery), to a flux density level of ≈ µ Jy ( . × erg s − Hz − ). The Peculiar Evolution of the Delayed Radio Flare
Our followup campaign (which includes six additional observ-ing epochs since radio discovery) reveal an unusually evolving radiospectrum (see the full spectral evolution in Extended Data Figure 2).During the first weeks after the initial detection of the radio emis-sion, the peak frequency slowly evolved, from ν p = 9 . ± . to ν p = 8 . ± . GHz (see section ‘Radio Observations’ in the Methods),while at the same time, the peak flux density dropped by ≈ . Incontrast, at later times, the radio flare evolved quickly. As seen in Fig-ure 2, the peak frequency of the radio emission decreased to < GHzonly months after radio discovery.Another oddity is the shape of the radio spectrum. In general, theradio spectral peak ( ν p ) observed in transient phenomena is usually dueto either the minimum energy of the emitting electrons (in which casethe flux density at ν < ν p is F ν ∝ ν / ) or due to absorption (usuallydue to synchrotron self absorption [SSA] and thus the flux density at ν < ν p is F ν ∝ ν / , as seen for example in most radio supernovae).Neither of these two cases is observed in ASASSN-15oi. Instead, ourdata show that the flux density, F ν ∝ ν α , has a power-law of α ≈ (in contrast to both ASASSN-14li [ref.10] and Swift
J1644+57 [ref. 16]in which α ≈ / ). Although this differs from the spectral shape ofstandard SSA (from a homogeneous CNM shockwave) model, it maybe explained by a more complex model in which the SSA originatesfrom an inhomogeneous source (see section ‘Spectral Modeling of theRadio Emission’ in the Methods). This spectral index is also typical ofradio emission originating from some quiescent GHz-peaked spectrum(GPS) sources which are considered to be young active galactic nu-clei (AGNs), but do not exhibit flares on timescales as observed here(see section ‘Spectral Modeling of the Radio Emission’ in the Methodsand Extended Data Figure 3). The spectral index of the optically thinsynchrotron emission above the peak ( F ν>ν p ∝ ν α syn ), is a functionof the power-law index ( p ) of the energy distribution of the emittingelectrons ( N e ∝ E − p ). For ASASSN-15oi, the spectral index is ini-tially in the range − ≤ α syn ≤ − , which is somewhat steeper thanthe spectral index observed in both Swift
J1644+57 and ASASSN-14li.Later, when the peak flux density decreased below GHz, the opticallythin spectrum became shallower ( α syn (cid:38) − . ; which is also shallowcompared to ASASSN-14li and Swift
J1644+57).
The Nature of the Delayed Radio Flare
The delayed radio emission we observe from the TDE ASASSN-15oi, its properties and evolution raise several key questions. Can thelate-time radio emission and the earlier null-detections be reconciledunder a standard CNM shockwave model? If so, does it originate in arelativistic jet, such as observed in
Swift
J1644+57 or possibly an off-axis jet which becomes visible only at late times? Or does it point to asub-relativistic shockwave (driven by accretion disk outflows or stellardebris) such as the one observed in ASASSN-14li? Does a delayedradio detection require a delayed outflow formation?To address these questions, first we test the standard dynamicalmodels that have been used to explain the radio emission from TDEs so far . In these models, a single shockwave (either relativisticor sub-relativistic) is launched into the CNM around the time of op-tical discovery. Both the optically thick and optically thin emissionhave a power-law temporal evolution with a range of values for thepower-law indexes (depending on the properties of the shockwave; seesection ‘Temporal Evolution of the Radio Emission’ in the Methods).The steep rise of the observed radio emission from non-detection on ∆ t = 90 days to detection on ∆ t = 182 days requires that the tempo-ral evolution of the flux density will be steeper than F ν ∝ t . In theanalytical models that we explore, the fastest rise rate of the emissionoccurs in the relativistic jet case. When an on-axis relativistic jet isinteracting with CNM profiles in the range between a constant densityto ρ CNM ∝ r − . (which is the steepest density profile found so farin TDEs ), the flux density fastest rise is F ν ∝ t , when the emis-sion is optically thick (for comparison, the optically thick radio emis-sion in a sub-relativistic supernova usually results in shallower rise of F ν ∝ t . ). Extrapolating backward in time from our initial radiodetection in C-band on 2016 Feb 12, results in a predicted flux den-sity of . mJy on 2015 Nov 12, well above our detection limit (seesection ‘Temporal Evolution of the Radio Emission’ in the Methods).A steeper rise can be obtained if the relativistic jet is observed off-axis. Exploring numerical models of such a scenario , we find that itcannot account for both the steep rise from non-detection to detection,and for the subsequent spectral and temporal evolution of the detectedradio emission (see section ‘Temporal Evolution of the Radio Emis-sion’ in the Methods). Another possible model, used to explain there-brightening of radio emission at late times (such as the one observedin the relativistic TDE Swift
J1644+57 [ref. 16]) is a structured rela-tivistic jet launched promptly after stellar disruption. In this case,the slower parts of the jet, will provide additional power source for theobserved emission in late times. However, even in this case the steepestrise due to such a jet structure will result at most in a ≈ t increase influx density which is not enough to explain the steep rise in the initialradio flare we observe. The big jump in flux density that we observefrom null-detection to detection thus seems to disfavor the predictionsof the existing models that we discussed above, that invoke the interac-tion of a relativistic (on or off-axis) or sub-relativistic outflow with theCNM promptly after the disruption of the star, and therefore points toa radio emitting process that occurs at late-times.The temporal evolution of the optically-thin emission, followingits temporal peak, is inconsistent with the above models. Whether weconsider an on- or off-axis relativistic jet or a sub-relativistic outflow,the optically thin radio emission is expected to have a power-law tem-poral evolution ( F ν ∝ t β ) with a predicted power-law index of − ≥ β ≥ − . However, here, we see a varying rather steep temporalevolution, where the temporal power-law reaches a value β < − (seeExtended Data Figure 4).A possible solution for explaining the initial radio null-detectionsis if unbound material from the TDE was initially traveling in a cavityaround the SMBH until it suddenly reached an extended high-densityCNM structure. However, once the outflow enters the high-density ex-tended CNM structure, the radio emission from the shockwave shouldfollow a spectral evolution similar to ASASSN-14li (or Swift
J1644+57)in contrast to what we observe (see Figure 2 and Extended Data Fig-ure 4) making this scenario less plausible. A scenario in which an out-flow from the TDE suddenly encounters a spatially thin confined denseCNM shell or filament, producing a brief late-time flare, is also disfa-vored. In this latter case, the radio flux density is expected to decreaseextremely rapidly after peak ( F ν ∝ t − ), once the outflow crosses thefilament , but such a steep decline is not observed here. One may alsoconsider the possibility that an outflow was launched only at late-timesdue to delayed accretion of bound stellar debris (as suggested by somestudies ). But also in this case, once the outflow is launched into thePage 2he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.CNM, the resulting radio emission should have spectral and temporalproperties according to the standard model discussed above, in con-trast to the data presented here. It is possible though that some delayedactivity (accretion) is in play, but one that involves processes that arecurrently not included in standard TDE models.The re-brightening of the radio emission on July 2019 may be themost recent evidence against the scenarios we examine above. A sec-ondary flare with an increase in emission by a factor of > , years af-ter the onset of the TDE, is not expected in any of the above scenarios.Even if considering a radio re-brightening in a structured jet model ,then currently, none of these models predict a re-brightening of the fluxdensity by more than an order of magnitude over such a long time scaleas observed here. Moreover, explaining a re-brightening at late timeswith a structured jet that has been launched early on, requires that theinitially observed delayed late-time emission will be explained by thisoutflow as well. However, as we already showed, it cannot. A sec-ondary flare, years after the onset of the TDE, is therefore not expectedin any of the above scenarios. One possibility is that the re-brighteningis driven by the same process responsible for the initial delayed flarewe detected. One may also consider the possibility of a recurring TDEflare due to repeated partial disruptions of a star . Another possibleexplanation is that the TDE occurred around a binary SMBH system.In such a scenario, the accretion rate may be highly variable with mul-tiple peaks including also several years after the initial disruption .Still, all of these proposed theoretical explanations currently do notoffer clear predictions for late-time radio emission that can be testedagainst our measurements. Furthermore, unfortunately, informationabout the 2019 re-brightening event is also limited (K. Alexander etal., manuscript in preparation).Late-time ultraviolet (UV) and optical observations show no signa-ture of any renewed activity or a secondary flare during the first yearafter the optical discovery of ASASSN-15oi [ref. 26]. UV emissionis still detected from the TDE after the time in which we discoveredthe delayed radio flare (with observations ongoing up to a year afteroptical discovery), and is consistent with a simple power-law declineof the UV emission detected at early times. A series of optical spectrataken starting at ∆ t = 301 days and up to ∆ t = 455 days shows thatthe broad emission lines, typical of TDEs, which are detected early on,have diminished, and no new emission lines are present .We now compare the evolution of the X-ray emission with thatof the radio emission (Figure 3). The X-ray emission, that is de-tected early on slowly and steadily rose with time, and peaked (afteran increase in flux by a factor of ∼ ) about a year after the opti-cal discovery . A direct comparison with the evolution of the ra-dio emission is somewhat limited by a gap in X-ray data during thetime of the first late-time radio detection and the radio subsequent fol-lowup observations. It is clear that the radio emission does not in-crease in parallel to the X-ray emission (Figure 3), as the radio emis-sion already fades away at (cid:46) days, while the X-ray emission isstill rising. Interestingly enough, the X-ray emission, after the gap inX-ray observations increased beyond of the Eddington luminosity,and its thermal (soft) component became brighter .This behaviour isusually observed in X-ray binaries (XRBs), although while this transi-tion is observed at a level of ∼ in some XRBs, it usually occurson average at higher Eddington luminosities ( ∼ − ) [ ].A possible explanation of this behaviour in XRBs is that this transi-tion occurs when the accretion rate increases and fresh material witha high Lorentz factor is injected into an existing jet . Following thisstage in XRBs, a radio flare is observed . We also note that combinedX-ray observations during June - August 2019, the period in whicha re-brightening of the radio emission is detected in VLASS, showthat the X-ray flux slightly increased, after declining, to a flux levelof . ± . × − erg cm − s − . This translates to . of the Eddington luminosity only (lower than the X-ray luminosity increaseto a level of L Edd around the time of the initial delayed radio flarewe observed). However, these recent X-ray data are limited and av-eraged over several months (see section ‘A Comparison Between theX-ray and Radio Emission Temporal Evolution’ in the Methods), thusmaking their interpretation difficult.Radio emission with a similar spectral shape as the one we observein ASASSN-15oi and similar temporal behaviour (in contrast to the onein GPS sources), has been observed in some AGN or blazar radio flares(but which is not necessarily typical of the whole AGN/blazar flarepopulation). In September 2011, a radio flare from M81 exhibited aninverted radio spectrum with a peak frequency of ∼ GHz. The flareradio flux density slowly decreased on a timescale of weeks with thespectral peak frequency roughly staying the same, until a second flarewas observed . In another case, a year long radio flare from the blazarCTA 102 had a complex temporal and spectral behaviour . The latetime evolution phase of this radio flare also shows similar characteris-tics to those observed in ASASSN-15oi (Figure 2). In general, theseAGN/blazar radio flares have been partially explained by the shock-in-jet model in which a shock propagates in an existing radio jet leadingto what seems as a flare. However, a phase in which the radio peakfrequency does not vary while the peak flux density decreases, as ob-served in both the M81 and the CTA 102 flares, is not captured by thismodel. Nonetheless, it is possible that there is a radio weak quiescentjet associated with the SMBH of ASASSN-15oi, which is shocked bya delayed injection of energy by the TDE. Such preexisting weak qui-escent radio emission, that suggests a non-TDE related activity of theSMBH, has been found in ASASSN-14li , at a level which is too faintfor detection at the distance of ASASSN-15oi.It has been suggested, that emission in both XRBs and AGNs isdominated by a weak non-thermal jet when the accretion rate is consid-erably sub-Eddington (the accretion becomes radiatively inefficient) .A phase transition in accretion occurs when the accretion rate increasesabove a critical threshold, at which point the emission in X-ray be-comes disk dominated and a high velocity outflow is launched (some-times observed as spatially discrete knot ejections ) resulting in aradio flare. It is therefore possible that such a phase transition occurredin ASASSN-15oi resulting in the delayed launch of an outflow whichled to rapidly-rising radio emission at late times. However, what trig-gers this phase transition, how this transition and the outflow launchingcoupled to it depend on the nature and properties of the relevant phe-nomena (e.g., TDE vs XRB), and what the typical signatures of thistransition in TDEs are (e.g will all TDEs exhibit an increase of their X-ray emission above a certain threshold characterized by some percent-age of their Eddington luminosity), are still open questions. The detailsof what follows any transition in the accretion phase are also unclear.Recall, that once detected, we observed an initially slowly evolving in-verted spectrum, but, shortly after, the spectral peak frequency rapidlyevolved. It is possible that the termination of the slow spectral evolu-tion phase marks the point at which the shock reached the edge of apre-existing jet and that the emission that follows is of the slowly cool-ing jet. Testing this scenario and answering any other open questionwill hopefully become possible with the discovery of additional eventssuch as ASASSN-15oi.Yet another open question relates to how common are such late-time flares (due to a delayed outflow ejection) in TDEs. There is oneother case of a possible late-time radio flare in the IR TDE ARP 299B-AT1 [ref. 17]. In that case, a single frequency (8.4 GHz) radio obser-vation, taken 12 days after the first possible indication of an increasein the IR flux, resulted in a null-detection. The next observation, car-ried out just days later, detected increasingly bright radio emission.The full set of radio measurements (spanning thousands of days) ofARP 299B-AT1 is consistent with a relativistic jet launched briefly afterPage 3he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.the stellar disruption. The late-time radio spectrum, which is consistentwith originating from an electron energy distribution of N e ∝ E − ,was also slowly evolving, in agreement with the predictions of knownmodels , but in contrast to the evolution observed in ASASSN-15oi.It is possible that the delayed launch of an outflow, as we observehere, has been missed in other TDEs due to limited observational cov-erage. First, in several past TDEs there is a substantial gap betweenthe time of disruption and the time at which the first radio observationis carried out (e.g. the disruption time in the case of ASASSN-14li ispoorly constrained). Thus even if radio emission is detected initially insuch cases, the exact time at which the outflow was launched with re-spect to disruption is unknown. Moreover, radio observations, in mostcases, whether radio emission is detected or not, are curtailed after sev-eral months, leaving any flaring event occurring later on, undetected.The peculiar delayed flares we discovered in ASASSN-15oi on timescales of months and years thus motivate carefully planned observa-tional campaigns of TDEs from early times until very late times. Summary
Our radio observing campaign of the optically discovered TDEASASSN-15oi since discovery and up to more than a year later, re-vealed a delayed radio flare with odd spectral and temporal proper-ties. A second even more luminous radio flare has been detected inVLASS observations. The various models that we explore here, thathave been suggested so far for explaining the radio emission originat-ing from TDEs, are unable to explain the combined properties of theobserved radio emission. Specifically, it seems that such a delayedbright radio flare, following an extended period of null radio detec-tions, requires some sort of an outflow to be launched at late times (intoa possibly inhomogeneous CNM; see Methods), suggesting a delayedonset of enhanced accretion. Some of the properties of the emissionhave similarities to XRBs and to AGN/blazar radio flares, thus raisingthe possibility that a transition in accretion phase state, that is proposedas an explanation of these latter flares, is also in play in TDEs. Thedetails of this process, that has not been observed in TDEs until now,and what triggers it are yet to be discovered. Understanding this pro-cess, requires that we first better characterize it. Our discovery thusmotivates late-time radio campaigns of TDEs that will hopefully iden-tify additional delayed flares. These could help us study the processresponsible for triggering delayed enhanced accretion, the subsequentoutflow launching, and the emission that accompanies it, thus helpingin unveiling the nature of this new puzzling phenomenon in TDEs.
References
1. van Velzen, S., Holoien, T. W. S., Onori, F., Hung, T. & Arcavi, I. Optical-Ultraviolet Tidal Disruption Events. arXiv e-prints arXiv:2008.05461(2020). .2. Rees, M. J. Tidal disruption of stars by black holes of 10 -10 solarmasses in nearby galaxies. Nature , 523–528 (1988).3. Piran, T., Svirski, G., Krolik, J., Cheng, R. M. & Shiokawa, H. Disk For-mation Versus Disk Accretion—What Powers Tidal Disruption Events?
Astrophys. J. , 164 (2015). .4. Jiang, Y.-F., Guillochon, J. & Loeb, A. Prompt Radiation and Mass Out-flows from the Stream-Stream Collisions of Tidal Disruption Events.
As-trophys. J. , 125 (2016). .5. Guillochon, J. & Ramirez-Ruiz, E. A Dark Year for Tidal DisruptionEvents.
Astrophys. J. , 166 (2015). .6. Bonnerot, C., Rossi, E. M., Lodato, G. & Price, D. J. Disc formation fromtidal disruptions of stars on eccentric orbits by Schwarzschild black holes.
Mon. Not. R. Astron. Soc. , 2253–2266 (2016). .7. Levan, A. J. et al.
An Extremely Luminous Panchromatic Outburst fromthe Nucleus of a Distant Galaxy.
Science , 199 (2011). .8. Bloom, J. S. et al.
A Possible Relativistic Jetted Outburst from a MassiveBlack Hole Fed by a Tidally Disrupted Star.
Science , 203 (2011). .9. Burrows, D. N. et al.
Relativistic jet activity from the tidal disruption of astar by a massive black hole.
Nature , 421–424 (2011). . 10. Alexander, K. D., Berger, E., Guillochon, J., Zauderer, B. A. & Williams,P. K. G. Discovery of an Outflow from Radio Observations of the TidalDisruption Event ASASSN-14li.
Astrophys. J. , L25 (2016). .11. van Velzen, S. et al.
A radio jet from the optical and x-ray bright stellartidal disruption flare ASASSN-14li.
Science , 62–65 (2016). .12. Holoien, T. W. S. et al.
Six months of multiwavelength follow-up of the tidaldisruption candidate ASASSN-14li and implied TDE rates from ASAS-SN.
Mon. Not. R. Astron. Soc. , 2918–2935 (2016). .13. Krolik, J., Piran, T., Svirski, G. & Cheng, R. M. ASASSN-14li: A ModelTidal Disruption Event.
Astrophys. J. , 127 (2016). .14. Yalinewich, A., Steinberg, E., Piran, T. & Krolik, J. H. Radio emission fromthe unbound debris of tidal disruption events.
Mon. Not. R. Astron. Soc. , 4083–4092 (2019). .15. Zauderer, B. A. et al.
Birth of a relativistic outflow in the unusual γ -raytransient Swift J164449.3+573451. Nature , 425–428 (2011). .16. Berger, E. et al.
Radio Monitoring of the Tidal Disruption Event SwiftJ164449.3+573451. I. Jet Energetics and the Pristine Parsec-scale En-vironment of a Supermassive Black Hole.
Astrophys. J. , 36 (2012). .17. Mattila, S. et al.
A dust-enshrouded tidal disruption event with a resolvedradio jet in a galaxy merger.
Science , 482–485 (2018). .18. Anderson, M. M. et al.
Caltech-NRAO Stripe 82 Survey (CNSS) PaperIII: The First Radio-Discovered Tidal Disruption Event, CNSS J0019+00. arXiv e-prints arXiv:1910.11912 (2019). .19. Stein, R. et al.
A high-energy neutrino coincident with a tidal disruptionevent. arXiv e-prints arXiv:2005.05340 (2020). .20. Kochanek, C. S. et al.
The All-Sky Automated Survey for Supernovae(ASAS-SN) Light Curve Server v1.0.
PASP , 104502 (2017). .21. Holoien, T. W. S. et al.
ASASSN-15oi: a rapidly evolving, luminous tidaldisruption event at 216 Mpc.
Mon. Not. R. Astron. Soc. , 3813–3828(2016). .22. Smartt, S. J. et al.
PESSTO: survey description and products from thefirst data release by the Public ESO Spectroscopic Survey of TransientObjects.
Astron. Astrophys. , A40 (2015). .23. Prentice, S. et al.
PESSTO spectroscopic classification of optical tran-sients.
The Astronomer’s Telegram , 1 (2015).24. Arcavi, I. et al.
Swift Observations of the TDE ASASSN-15oi.
The As-tronomer’s Telegram , 1 (2015).25. Gezari, S., Cenko, S. B. & Arcavi, I. X-Ray Brightening and UV Fadingof Tidal Disruption Event ASASSN-15oi.
Astrophys. J. , L47 (2017). .26. Holoien, T. W. S. et al.
The unusual late-time evolution of the tidal dis-ruption event ASASSN-15oi.
Mon. Not. R. Astron. Soc. , 5689–5703(2018). .27. Shiokawa, H., Krolik, J. H., Cheng, R. M., Piran, T. & Noble, S. C. GeneralRelativistic Hydrodynamic Simulation of Accretion Flow from a StellarTidal Disruption.
Astrophys. J. , 85 (2015). .28. Dai, L., McKinney, J. C. & Miller, M. C. Soft X-Ray Temperature TidalDisruption Events from Stars on Deep Plunging Orbits.
Astrophys. J. , L39 (2015). .29. Lacy, M. et al.
The Karl G. Jansky Very Large Array Sky Survey (VLASS).Science Case and Survey Design.
PASP , 035001 (2020). .30. O’Dea, C. P. et al.
Multifrequency VLA observations of GHz-peaked-spectrum radio cores.
A&AS , 549 (1990).31. Giannios, D. & Metzger, B. D. Radio transients from stellar tidal disrup-tion by massive black holes. Mon. Not. R. Astron. Soc. , 2102–2107(2011). .32. Metzger, B. D., Giannios, D. & Mimica, P. Afterglow model for the ra-dio emission from the jetted tidal disruption candidate Swift J1644+57.
Mon. Not. R. Astron. Soc. , 3528–3537 (2012). .33. Chevalier, R. A. Synchrotron self-absorption in radio supernovae ,810 (1998). URL http://iopscience.iop.org/0004-637X/499/2/810 .34. Generozov, A. et al.
The influence of circumnuclear environment on theradio emission from TDE jets.
Mon. Not. R. Astron. Soc. , 2481–2498(2017). .35. Alexander, K. D., van Velzen, S., Horesh, A. & Zauderer, B. A. RadioProperties of Tidal Disruption Events. arXiv e-prints arXiv:2006.01159(2020). .36. van Eerten, H., van der Horst, A. & MacFadyen, A. Gamma-Ray BurstAfterglow Broadband Fitting Based Directly on Hydrodynamics Simula-tions.
Astrophys. J. , 44 (2012). . Page 4he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
37. Mimica, P., Giannios, D., Metzger, B. D. & Aloy, M. A. The radio afterglowof Swift J1644+57 reveals a powerful jet with fast core and slow sheath.
Mon. Not. R. Astron. Soc. , 2824–2841 (2015). .38. Granot, J. & van der Horst, A. J. Gamma-Ray Burst Jets and their RadioObservations.
PASA , e008 (2014).39. Harris, C. E., Nugent, P. E. & Kasen, D. N. Against the Wind: Radio LightCurves of Type Ia Supernovae Interacting with Low-density CircumstellarShells. Astrophys. J. , 100 (2016). .40. Guillochon, J. & Ramirez-Ruiz, E. Hydrodynamical Simulations to Deter-mine the Feeding Rate of Black Holes by the Tidal Disruption of Stars:The Importance of the Impact Parameter and Stellar Structure.
Astro-phys. J. , 25 (2013). .41. Campana, S. et al.
Multiple tidal disruption flares in the active galaxy IC3599.
Astron. Astrophys. , A17 (2015). .42. Coughlin, E. R., Armitage, P. J., Nixon, C. & Begelman, M. C. Tidal dis-ruption events from supermassive black hole binaries.
Mon. Not. R. As-tron. Soc. , 3840–3864 (2017). .43. Dunn, R. J. H., Fender, R. P., K¨ording, E. G., Belloni, T. & Cabanac, C.A global spectral study of black hole X-ray binaries.
Mon. Not. R. As-tron. Soc. , 61–82 (2010). .44. Maccarone, T. J. Do X-ray binary spectral state transition luminositiesvary?
Astron. Astrophys. , 697–706 (2003). astro-ph/0308036 .45. Tetarenko, B. E., Sivakoff, G. R., Heinke, C. O. & Gladstone, J. C.WATCHDOG: A Comprehensive All-sky Database of Galactic Black HoleX-ray Binaries.
Astrophys. J. Supp. , 15 (2016). .46. Fender, R. P., Belloni, T. M. & Gallo, E. Towards a unified model for blackhole X-ray binary jets.
Mon. Not. R. Astron. Soc. , 1105–1118 (2004). astro-ph/0409360 .47. King, A. L. et al.
Discrete knot ejection from the jet in a nearby low-luminosity active galactic nucleus, M81 ∗ . Nature Physics , 772–777(2016). .48. Fromm, C. M. et al. Catching the radio flare in CTA 102. I. Light curveanalysis.
Astron. Astrophys. , A95 (2011). .49. Marscher, A. P. & Gear, W. K. Models for high-frequency radio outburstsin extragalactic sources, with application to the early 1983 millimeter-to-infrared flare of 3C 273.
Astrophys. J. , 114–127 (1985).50. Falcke, H., K¨ording, E. & Markoff, S. A scheme to unify low-power accret-ing black holes. Jet-dominated accretion flows and the radio/X-ray corre-lation.
Astron. Astrophys. , 895–903 (2004). astro-ph/0305335 .51. Bright, J. S. et al.
An extremely powerful long-lived superluminal ejectionfrom the black hole MAXI J1820+070.
Nature Astronomy , 697–703(2020). .52. Alexander, K. D., Wieringa, M. H., Berger, E., Saxton, R. D. & Komossa,S. Radio Observations of the Tidal Disruption Event XMMSL1 J0740-85. Astrophys. J. , 153 (2017). .53. Irwin, J. A. et al.
CHANG-ES V: Nuclear Outflow in a Virgo Cluster Spiralafter a Tidal Disruption Event.
Astrophys. J. , 172 (2015). .54. McMullin, J. P., Waters, B., Schiebel, D., Young, W. & Golap, K.
CASAArchitecture and Applications , vol. 376 of
Astronomical Society of thePacific Conference Series , 127 (2007).55. Weiler, K. W., Panagia, N., Montes, M. J. & Sramek, R. A. Radio Emis-sion from Supernovae and Gamma-Ray Bursters.
Annu. Rev. Astron.Astrophys , 387–438 (2002).56. Barniol Duran, R., Nakar, E. & Piran, T. Radius Constraints and Mini-mal Equipartition Energy of Relativistically Moving Synchrotron Sources. Astrophys. J. , 78 (2013). .57. Bj¨ornsson, C. I. & Keshavarzi, S. T. Inhomogeneities and the Modelingof Radio Supernovae.
Astrophys. J. , 12 (2017). .58. Chandra, P. et al.
Type Ib Supernova Master OT J120451.50+265946.6:Radio-emitting Shock with Inhomogeneities Crossing through a DenseShell.
Astrophys. J. , 79 (2019). .59. O’Dea, C. P. The Compact Steep-Spectrum and Gigahertz Peaked-Spectrum Radio Sources.
PASP , 493–532 (1998).60. Chevalier, R. A. & Fransson, C. Circumstellar emission from typeib and ic supernovae.
Astrophys. J. , 381 (2006). URL http://iopscience.iop.org/0004-637X/651/1/381 .61. Granot, J., De Colle, F. & Ramirez-Ruiz, E. Off-axis afterglow light curvesand images from 2D hydrodynamic simulations of double-sided GRB jetsin a stratified external medium.
Mon. Not. R. Astron. Soc. , 2711–2720 (2018). . Acknowledgements
We thank the anonymous referees for improving thismanuscript, and T. Piran and E. Nakar for useful discussions. A.H. is grate-ful for the support by grants from the I-CORE Program of the Planning andBudgeting Committee and the Israel Science Foundation (ISF), and from theUS-Israel Binational Science Foundation (BSF). I.A. is a CIFAR Azrieli GlobalScholar in the Gravity and the Extreme Universe Program and acknowledges support from that program, from the European Research Council (ERC) underthe European Union’s Horizon 2020 research and innovation program (grantagreement number 852097), from the Israel Science Foundation (grant num-ber 2752/19), from the United States - Israel Binational Science Foundation(BSF), and from the Israeli Council for Higher Education Alon Fellowship. Theauthors thank the NRAO staff for approving and scheduling the VLA obser-vations. The National Radio Astronomy Observatory is a facility of the Na-tional Science Foundation operated under cooperative agreement by Associ-ated Universities, Inc. The authors thank the
Swift
TOO team. This researchhas made use of data and/or software provided by the High Energy Astro-physics Science Archive Research Center (HEASARC), which is a service ofthe Astrophysics Science Division at NASA/GSFC. This research has madeuse of the CIRADA cutout service at URL cutouts.cirada.ca, operated by theCanadian Initiative for Radio Astronomy Data Analysis (CIRADA). CIRADAis funded by a grant from the Canada Foundation for Innovation 2017 Inno-vation Fund (Project 35999), as well as by the Provinces of Ontario, BritishColumbia, Alberta, Manitoba and Quebec, in collaboration with the NationalResearch Council of Canada, the US National Radio Astronomy Observatoryand Australia’s Commonwealth Scientific and Industrial Research Organisa-tion.
Author contributions
A.H. has led the radio observing campaign, the dataanalysis and modeling, the interpretation, and the manuscript preparation.S.B.C and I.A. contributed to the interpretation of the results and to themanuscript preparation.
Correspondence and requests for materials should be addressed toA.H. (email: [email protected]).
Additional information
Extended Data Figures and Supplementary infor-mation are available for this paper.
Competing interests
The authors declare that they have no competing fi-nancial interests.
Data availability
The ASASSN-15oi radio data, presented in several plots,can be found in Supplementary Table 1. The raw VLA data is avail-able on the NRAO archive at https://archive.nrao.edu/archive/advquery.jsp . Tools to analyze the VLA data can be found atthe NROA website at https://science.nrao.edu/facilities/vla/data-processing/analysis-packages/analysis-packages . Thecollection of radio data of other TDEs can be found in [ref. 35]. The ASASSN-15oi x-ray emission measurements can be found in [ref. 25]. Any additionaldata that support the findings of this study are available from the correspond-ing author upon reasonable request.
Page 5he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al. Time Since Discovery [days] L [ e r g s - ] ARP299IGR J12580+0134 CNSSJ0019AT 2019DSG
ASASSN-15oi
ASASSN-14li
Figure 1 | The radio luminosity of a handful of TDEs (including the delayed radio flare from ASASSN-15oi) as a function of time.
Note, that the time hereis after optical discovery (in some cases the actual disruption time, such as in ASASSN-14li, is poorly constrained, thus the light curves of these events may shift intime compared to others). The radio measurements (luminosity densities L ν ) are at a frequency of ∼ GHz (data for AT 2019dsg is at . GHz). Empty markersrepresent a phase where the emission is optically thick while filled markers represent optically thin emission. The radio emission of ASASSN-14li (purple circles )and AT 2019DSG (light blue side triangles ) shows a typical evolution of a shockwave in a CNM as the emission peaks when the optical depth is about unity atthe relevant frequency and thereafter the emission declines (similar to what is also seen in radio supernovae). The radio emission from the radio discovered TDE,CNSS J0019 (ref. 18), exhibits similar behaviour as in the two latter TDEs (the radio peak is not shown here as it is at lower GHz frequencies). The radio observationsof XMMSL1 J0740-85 (green diamonds ) and IGR J12580+0134 (blue stars ) took place only at later times, when the radio emission is already optically thin andfading. The radio available for ARP 299B-AT1 (orange circles ) suggests the radio emission originated from a relativistic jet launched promptly after disruption.In contrast, our radio observations, early after discovery and up to days, resulted in null-detections ( σ limits are represented by red line arrows). The delayedradio flare we detect later (red squares) has a peculiar evolution. The light red square at ∆ t ∼ days is the recent detection of a rebrightening emission fromASASSN-15oi at GHz (lacking spectral information). We also present past events with no detected radio emission (detection limits shown as gray triangles). Theemission from
Swift
J1644+57 (not shown) is a few orders of magnitudes brighter than the emission detected from the TDEs presented in the figure, and slowly risingin the first days after discovery.
Page 6he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
Peak Frequency [GHz] P ea k F [ Jy ] ASASSN-14li
Swift
J1644+57ASASSN-15oi15 20 25 30
Peak Frequency (GHz) P ea k F [ m Jy ] CTA 102
Figure 2 | The evolution of the peak flux density and frequency of the delayed radio emission from ASASSN-15oi.
During the first two weeks of the radiodiscovery the peak frequency evolves slowly, while in the six weeks that followed, the peak frequency evolves rapidly to an unknown frequency below GHz (blueline and markers; peak limit marked by blue arrow; σ uncertainties from the peak best fit analysis are presented). In comparison, we show the evolution of the peakin both ASASSN-14li [ref. 10] (dotted line) and Swift
J1644+57 [ref.15] (dashed line). This type of diagnostic plot of the radio peak is used sometimes to follow theevolution of radio flares from AGNs/Blazars. The peak behaviour of ASASSN-15oi is similar to the observed late-stage evolution of two AGN/blazar flares (see maintext). The inset shows a late-time phase in the evolution of the radio flare from the blazar CTA 102 [ref.48]. We emphasize, however, that this behaviour is not observedin all AGN/blazar flares.
Page 7he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
100 200 300 400 500 600
Time Since Discovery [days] X -r a y Lu m i no s i t y ( L E dd ) R ad i o Lu m i no s i t y [ e r g s - ] Figure 3 | Comparison of the temporal evolution of the X-ray luminosity with the optically thin radio luminosity in ASASSN-15oi.
The X-ray emission isdetected soon after the occurrence of the TDE, while the radio emission begins later. Unfortunately, no X-ray data are available when the radio emission is initiallydetected. The x-ray luminosity, however, increases to a level of ≈ the Eddington luminosity when radio emission is detected. The X-ray data are presented as bluecircles ( σ errors are presented). The radio data is at a frequency of GHz (red squares; uncertainties are the image noise and flux calibration uncertainty added inquadrature as defined in Supplementary Table 1). Averaged early radio 3 σ non-detections up to days after optical discovery are represented with red arrows. Page 8he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
MethodsRadio Observations
We observed the field of ASASSN-15oi with the VLA on 2015 Au-gust 22, September 06, November 12, and on 2016 February 12, undera
Swift -VLA joint program (SB 4220). Later observations were per-formed under a director discretionary time (DDT) program (16A-422).The initial four observations were performed in X- and K-bands ( GHzand GHz, respectively) only, as a detection experiment in search forradio emission. Once radio emission was detected in the fourth ob-servation (Supplementary Figure S1), the followup observations wereconducted in a varying wide range of bands from S-band ( GHz) toK-band, as needed, in order to characterize and capture the evolutionof the broadband radio spectrum (see Supplementary Figure S2).We calibrated the radio data using the automated VLA calibra-tion pipeline available in the Common Astronomy Software Applica-tions (CASA) package . Flux density calibration was conducted using3C48, while J2040-2507 was used as a gain calibrator. Images of theASASSN-15oi field were produced using the CASA task CLEAN. Inimages where ASASSN-15oi was detected, the source flux density wasmeasured using the CASA task IMFIT, and the image rms was calcu-lated using the CASA task IMSTAT. In addition, we add a flux densitycalibration error at a conservative level of and to frequenciesbelow and at (or above) Ku-band, respectively. The log of the observa-tions and the resulting measurements are listed in Supplementary Ta-ble 1.As mentioned in the main text, a re-brightening of the delayed flareis detected in VLASS data. VLASS data has been obtained over a longperiod of time and will continue to be collected in the following years.The observation in which ASASSN-15oi was detected was obtained on2019 July. The data were reduced and imaged by NRAO using a soft-ware pipeline designed for VLASS. The flux density from ASASSN-15oi in the VLASS quick-look image is at a level of ≈ mJy (with anassumed general uncertainty level of for all quick look images a ). Spectral Modeling of the Radio Emission
Below we attempt to model the observed broadband radio spec-tra of ASASSN-15oi at the individual observing epochs where apeak in the flux density is observed, but without modeling the tem-poral evolution. Theoretical models predict that radio emis-sion in TDEs originates from a forward shockwave (either relativis-tic or sub-relativistic) traveling in the surrounding environment. Thisshockwave accelerates free electrons that gyrate in the shockwave en-hanced magnetic field and thus emit synchrotron radiation. Therefore,we first model the individual single-epoch broadband radio spectra,we observe, according to a synchrotron self absorption (SSA) spec-tral emission model . This model successfully accounts for theradio emission observed in both
Swift
J1644+57 and ASASSN-14li[ref.10, 13, 15, 16].In the SSA emission model, the radio spectrum exhibits a peakbelow which the emission is self absorbed and thus optically thick andabove which the emission is optically thin. The optically thick emissioncan be described as F ν ∝ πR D B − / ν / , (1)while the optically thin emission is described by F ν ∝ πfR D N B ( p +1) / ν − ( p − / , (2)where R is the radius of the radio emitting shell, D is the distance tothe TDE, f is the emission filling factor, and B is the magnetic field a https://science.nrao.edu/vlass/users-guide-to-quick-look-images strength. The energy density of the magnetic field is a fraction (cid:15) B of theenergy density of the shocked CNM. Thus, the magnetic field strengthdepends also on the square-root of the CNM density (and its profile).Our SSA best fit models of each of the ASASSN-15oi radio spec-tra, separately, in which a radio peak is evident, are presented in Sup-plementary Figure S3. As shown in the figure, the SSA models poorlyaccount for the observed radio spectra (with reduced χ values of χ r > in times ∆ t = 190 , days). This is no surprise, as theoptically thick spectral index of the SSA model is α = 5 / , while ex-amination of the data suggests a spectral index of α ≈ . A shallowerspectral index of the optically thick emission is expected if only inter-nal free-free absorption (FFA) is the dominant absorption mechanisminstead of SSA, although still steeper than the observed α ≈ spectralindex. In this internal FFA model the flux density is: F ν ∝ F ν,syn (cid:18) − e − τ ff τ ff (cid:19) , (3)where F ν,syn is the un-absorbed synchrotron emission, and τ ff is theFFA optical depth. The results of the FFA modeling are presented inSupplementary Figure S3. While the internal FFA models are some-what better in accounting for the observed radio spectra than the SSAmodel ( χ r ≈ . ), they still significantly deviate from the observa-tions. A solution may be found by reverting to the SSA model, but thistime instead of assuming a homogeneous CNM environment, we willassume an inhomogeneous one. Inhomogeneities in the CNM can bemodeled as inhomogeneities of the magnetic field and will result in aSSA spectrum with a broader peak and a shallower spectral index .This explanation was also used recently to describe a shallow opticallythick radio emission in a stripped envelope supernova . We follow thismodel (which is an extension of the SSA model but one that is parame-terized with a distribution of magnetic fields, P ( B ) ∝ B − a , instead ofa single magnetic field). This model thus adds two additional degreesof freedom: the range of magnetic field strengths and the power-lawindex ( a ) of the magnetic field distribution. The best fit results of thismodel are presented in Supplementary Figure S3. We find that theinhomogeneous SSA model provide a better spectral fit ( χ r ≈ . )compared to the previous models. It is important to note that findinga separate good spectral fit to each of the individual observed spectra,does not mean that we have found a dynamical single scenario that canexplain the combined full observed dataset, as we explain below whenattempting a temporal modeling.At this point, it is worth mentioning that the observed radio spec-trum of ASASSN-15oi is reminiscent of GPS sources. These radiosources have a spectrum with peak frequency in the low GHz range,as their name suggests. They are powerful compact ( (cid:46) kpc) ra-dio sources some of which exhibit a morphology of two sided sym-metric sources, when resolved with high angular resolution observa-tions, and are believed to be young AGN radio jets . On a long timescale, GPS sources can exhibit strong variability up to an order of amagnitude in the radio. However, the complex spectral variations ob-served here, over a short timescale of two months, are atypical of GPSsources. Still, considering that the radio spectral shape of GPS sourcesis attributed to inhomogeneities and the resemblance of their spectra tothat of ASASSN-15oi strengthen the conclusion that the radio emissionfrom ASASSN-15oi may originate from a complex CNM environment.Despite the poor fit by a simple SSA model, assuming that the peakflux density is the result of SSA, we use the peak flux density and fre-quency to roughly estimate the shockwave radius by R p = 4 . × (cid:18) (cid:15) e (cid:15) B (cid:19) − / (cid:18) f . (cid:19) − / (cid:18) F p mJy (cid:19) / (cid:18) D Mpc (cid:19) / × (cid:16) ν p (cid:17) − cm , (4)Page 9he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.where (cid:15) e and (cid:15) B are the fractions of shockwave energy deposited intoaccelerating free electrons and enhancing the magnetic field, respec-tively. We adopt standard equipartition value of (cid:15) e = (cid:15) B = 0 . and f = 0 . .The radius of the radio emitting region, when we first detect it(2016 February 12, ∆ t = 182 days), is estimated at R ≈ × cm,which implies (assuming an outflow was launched at optical discovery)a shockwave velocity of , km/s (only slightly higher than the ve-locity expected for the unbound stellar debris ). Note, however, thatif a jet (or an outflow) is launched later, then the velocity estimate willbe higher. We also find that the radio-emitting region’s radius remainsroughly the same, over a two weeks period (when evaluated on 2016February 20 and 27; ∆ t = 190 and 197 days, respectively, followingthe initial radius evaluation on ∆ t = 182 days). Following this pe-riod, the estimated radius makes a big jump to R (cid:38) . × cm,in only six weeks. The above radius estimates (as well as the velocityestimates), however, become lower limits if the emission is originatingfrom an inhomogeneous source (as discussed above). Temporal Evolution of the Radio Emission
The temporal evolution of the observed radio emission can be ex-amined against main theoretical predictions. We first address the an-alytical predictions for an on-axis relativistic jet and a sub-relativisticoutflow. In the case of an on-axis relativistic jet the optically thickemission is expected to rise as t ( k +2) / (4 − k ) , where k is the power-lawindex of the CNM density profile ( ρ CNM ∝ r − k ). Adopting the steep-est density profiles found in some TDEs of ρ CNM ∝ r − . (whichis steeper than the usual wind-like CNM profile used in most numeri-cal simulations), we obtain F ν ∝ t . In the sub relativistic case , theoptically thick emission evolves as F ν,thick ∝ t (28+3 . k ) / , whichis shallower than the relativistic case. We also found that attemptingto model the full observed data, including the observed non-detections,with a sub-relativistic spherical outflow model, fails. Adopting, there-fore, the steepest relation F ν ∝ t (of the relativistic case), the op-tically thick emission detected on 2016 Feb 12 can be evolved backin time to 2015 Nov 12, resulting in a predicted flux density level of . mJy in C-band, well above the detection threshold of our obser-vation at that time (a σ limit of µ Jy). As seen in SupplementaryFigure S4, the jump in flux density from non-detection to detection re-quires a temporal power law slightly steeper than t , which requires aCNM density profile with a power-law steeper than k = 2 . . How-ever, note that while this model predicts that the optically thick emis-sion is rising, in fact, upon detection it is declining. The optically thinemission, on the other hand, is expected to be declining according to F ν ∝ t − p +2) / (2(5 − k )) or F ν ∝ t (42 − k ) / in either the relativisticjet or the sub relativistic outflow scenario, respectively. The steepestdecline in this case for a k = 2 . is t − , while the observed emissiondecline rate becomes steeper than this.We next turn to examine numerical emission models for off-axisrelativistic jets. For that purpose, we use the publicly available b BoxFitcode . In this scenario, a steep rise in the radio emission (steeper thanthe t above) can occur in large off-axis angles when the relativistic jettravels in a CNM with a constant density (a wind-like density results ina shallower rise ). We therefore explore the constant ISM density sce-nario next. We find that the numerical model can provide a reasonablefit to individual epochs of the broadband spectra only. However, thebest-fit parameters vary substantially between the fitted model of eachepoch. For example, the optically thin spectrum observed on day ,requires an off-axis angle of ∼ . rad , while the radio emission ob-served on day requires an off-axis angle of . rad and and ISMdensity an order of magnitude higher than the one in the model for day b https://cosmo.nyu.edu/afterglowlibrary/boxfit2011.html . Moreover, no numerical solution that can explain both the initialsteep rise in flux density, and the spectral and temporal evolution, canbe found.In Supplementary Figure S4, we show several temporal power-lawsfor the rise and the decline of the emission, including the above steepestrise and decline rates. As seen in the figure, assuming that a shockwavewas launched at the time of optical discovery, leading to radio emissionthat peaks at the time we discovered the radio flare, then the radio emis-sion should have been detectable at the time of the third null-detectionobservation. The observed decline rate of the radio emission after thediscovery is also steeper than the one expected from standard theoreti-cal models. A Comparison Between the X-ray and Radio EmissionTemporal Evolution
The X-ray luminosity of ASASSN-15oi [ref.25] is shown in Fig-ure 3 in units of the Eddington luminosity. We estimate the Eddingtonluminosity using the black hole mass estimate of ∼ M (cid:12) [ref.25,26]corresponding to L Edd ∼ . × erg s − . As seen in Figure 3,there is a gap in the X-ray data, when we first detect the radio emis-sion. Despite this, it seems that there is no apparent correlation be-tween the X-ray and radio emissions. The X-ray emission is detectedalready shortly after the optical discovery of ASASSN-15oi, and slowlyincreased by a factor of ∼ up to about a year later and crosses the Eddington luminosity level.In addition to previously published X-ray data , short
Swift snapshot observations were undertaken on 2019 and 2020 and are pub-licly available. The data were analyzed following the same proce-dures used to analyze past
Swift observations of ASASSN-15oi [ref.25, 26]. Due to limited sensitivity, we combined the data sets obtainedover several months in 2019, resulting in a X-ray flux measurementof . ± . × − erg cm − s − . The combined observationsin 2020 show that the X-ray emission then faded to a flux level of . ± . × − erg cm − s − . Page 10he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al. Extended Data Figure 1 | VLA K-band images of the position of the optical TDE candidate ASASSN-15oi, before and after radio detection . The left panel (a)presents the third VLA image we obtained of this field months after optical discovery on 2015 Nov 12, still showing a null-detection. The right panel (b) presents theimage from our forth VLA observation on 2016 Feb 12 which reveals a delayed radio flare, months after optical discovery. The synthesised beam size is shown as awhite ellipse at the bottom left corner of the images. The flux density scale is identical in both images. Page 11he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
Frequency [GHz] F [ Jy ]
182 days190 days197 days233 days283 days369 days576 days
Extended Data Figure 2 | The full observed broadband spectral evolution of the delayed radio flare from ASASSN-15oi.
Each of the radio broadband spectra isfrom a different observing epoch, starting from the initial detection of the delayed flare on days and up to days after optical discovery. Data from each epochis represented by a different marker shape and color as noted in the legend (a dashed line connecting the data has been added for convenience). The error bars representthe image noise and flux calibration error added in quadrature (see Supplementary Table 1).
Page 12he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
Extended Data Figure 3 | Best fit single-epoch spectral models of the radio flare spectra.
Observing epochs at ∆ t = 182 , , days are represented inpurple, yellow and red, respectively. The broadband spectrum in each single epoch was fitted independently, thus not including any modeling of the temporal evolution.The errors of the data modeled here include the flux density calibration error and image noise added in quadrature. The left panel (a) presents the best fit homogeneousSSA model . The middle panel (b) shows the best fit models of the radio flare spectra using the internal free-free absorption model . The right panel (c) is the best fitmodels using the inhomogeneous SSA model . Out of the three models that we try here, the latter model is the best match to the spectral data presented in this figure(see details in Methods). Page 13he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
100 200 300 400 500
Time Since Discovery [days] F [ Jy ] t t t t -1 t -2 t -3 Extended Data Figure 4 | Comparison of the the temporal evolution of the observed optically thin radio emission with different rising and declining power-lawfunctions.
The presented radio emission is at a frequency of GHz (black solid line and markers). The various power-law functions for both the rise of the emission(since the last non detection) and its decline are presented as dashed curves (representing various predictions, see details in Methods). The black triangle represents a3 σ non-detection limit (based on the average between the GHz and GHz limits).
Page 14he Radio Awakens - Delayed Radio Flares from a TDE Horesh et al.
Supplementary Table 1 | ASASSN-15oi VLA observation log
Date Frequency Flux Density Flux Density Error[GHz] [ µ Jy] [ µ Jy]2015 Aug 22 6.1 ≤ ≤ ≤ ≤ ≤ ≤ σ limits and the following broadband observations in various bands. The flux density error quotedis the image rms noise. A flux density calibration error of and (below and above KU-band) should be added in quadrature.(below and above KU-band) should be added in quadrature.