aa r X i v : . [ a s t r o - ph ] M a y Submitted to ApJ Letters
Preprint typeset using L A TEX style emulateapj v. 05/04/06
RECOILING BLACK HOLES IN QUASARS
E. W. Bonning , G. A. Shields, S. Salviander Submitted to ApJ Letters
ABSTRACTRecent simulations of merging black holes with spin give recoil velocities from gravitational radiationup to several thousand km s − . A recoiling supermassive black hole can retain the inner part of itsaccretion disk, providing fuel for a continuing QSO phase lasting millions of years as the hole movesaway from the galactic nucleus. One possible observational manifestation of a recoiling accretion diskis in QSO emission lines shifted in velocity from the host galaxy. We have examined QSOs from theSloan Digital Sky Survey with broad emission lines substantially shifted relative to the narrow lines.We find no convincing evidence for recoiling black holes carrying accretion disks. We place an upperlimit on the incidence of recoiling black holes in QSOs of 4% for kicks greater than 500 km s − and0.35% for kicks greater than 1000 km s − line-of-sight velocity. Subject headings: galaxies: active — quasars: general — black hole physics INTRODUCTION
Recent breakthroughs in numerical relativity haveallowed the possibility of simulating in complete generalrelativity the final orbits of a binary black hole up toand past the merger. Analytical approximations havefor some time indicated that the merger of unequalmass black holes will, through anisotropic emission ofgravitational radiation, impart to the final black hole asubstantial ‘kick’ of up to more than 1000 kilometersper second (Fitchett 1983; Damour & Gopakumar2006; Sopuerta et al. 2006; Lousto & Price 2004,and references therein). Numerical simulations haveonly begun to explore the parameter space of blackhole mass ratio, spin magnitude, and spin-orbitorientations in such mergers(Herrmann et al. 2006;Baker et al. 2006; Gonz´alez et al. 2007; Herrmann et al.2007; Koppitz et al. 2007; Campanelli et al.2007; Gonzalez et al. 2007; Baker et al. 2007;Tichy & Marronetti 2007). Interestingly, they areconsistently showing kicks of order 100 to 1000 km s − depending on black hole spin inclinations, up to amaximum of ∼ − for spins anti-alignedand perpendicular to the orbital angular momentum(Gonzalez et al. 2007; Tichy & Marronetti 2007). Suchlarge recoil velocities have significant astrophysicalimplications for galactic mergers, since even velocitiesof order 1000 km s − can be greater than the escapevelocity of moderate-sized elliptical galaxies and spiralbulges, and much greater than the < ∼
300 km s − escapevelocity for dwarf galaxies (Campanelli et al. 2007;Merritt et al. 2004, and references therein).Super-massive black holes ( ∼ M ⊙ ) will be formedduring the merger of galaxies (Begelman et al. 1980).The binary orbit will decay quickly as a result dynam-ical friction due to the stellar background. The orbitmay stall at a radius ∼ LUTH, Observatoire de Paris, CNRS, Universit´e Paris Diderot;Place Jules Janssen 92190 Meudon, France; [email protected] Department of Astronomy, University of Texas, Austin, TX78712; [email protected],[email protected] merger taking place in an AGN, the accretion disk willremain bound to the recoiling black hole inside the radius R = 1 . × M /v cm where the orbital velocityis equal to the recoil velocity. Here, M = M/ M ⊙ and v = v/ − . The retained disk massin such a case, assuming an α disk (Shakura & Sunyaev1973; Frank et al. 2002) will be: M ( R ) = (10 . M ⊙ ) α − / − M / ˙ M / R / f / (1)or M ( v ) = (10 . M ⊙ ) α − / − M / ˙ M / v − / (2)where ˙ M is the accretion rate in solar masses per year.Stability requires M d isk < M B H (see Loeb (2007)).For a black hole binary contained in an accretiondisk, the two holes will empty out a ‘gap’ with aradius of approximately twice the binary semi-majoraxis (Macfadyen & Milosavljevic 2006). This will refillquickly after the kick (Loeb 2007), and QSO activitywill resume. The disk mass will be sufficient to fuel QSOactivity over a disk consumption time t d ≈ M ( v ) / ˙ M ≈ (10 y r ) α − / − M / ˙ M − / v − / . This ‘wandering QSO’phase could last for a time comparable to the pre-mergerphase, and would result in either a QSO displaced a num-ber of kpc from the galactic nucleus or QSO emissionlines shifted relative to galactic systemic velocity.Observations of nearby AGN do not show displacednuclei (Libeskind et al. 2006). This may be one indica-tion that large kicks rarely occur during an active AGNphase. Alternatively, these kicks may be observed in thevelocity of the AGN emission lines. The broad emission-line region (BLR) of QSOs corresponds to radii withinwhich the disk will remain bound to the post-mergerblack hole. The BLR dynamical timescale is ∼ years,so the BLR should be regenerated quickly from the diskfollowing the recoil and resumption of AGN activity. Thenarrow emission lines, in contrast, arise from gas pre-dominantly orbiting in the potential of the host galaxy The quasar HE0450-2958 (Magain et al. 2005) has beensuggested as such a candidate, but questions remain aboutwhether an ejected black hole is indicated (Hoffman & Loeb 2006;Merritt et al. 2006; Kim et al. 2007).
Bonning et al.that will not follow the recoiling black hole (Merritt et al.2006). The displaced QSO will still ionize the interstellargas, producing narrow emission lines, albeit different indetail from a normal narrow line region (NLR). There-fore, the broad emission lines associated with a recoilingdisk will appear shifted with respect to the galaxy sys-temic redshift as expressed by the narrow emission lines.We have carried out a search for candidate kickedQSOs using spectra from the Sloan Digital Sky Survey Data Release 5 (DR5). We have focused attention onthe broad H β and narrow [O iii] lines, but also considerbroad Mg ii where available. Note that velocity shifts ofbroad emission lines are a well studied phenomenon oftenattributed to BLR physics or even orbiting binary blackholes (e.g. Gaskell 1996; Richards et al. 2002, and refer-ences therein). This complicates the task of identifyingtrue examples of recoil. OBSERVATIONS
The QSOs in our study were spectroscopically iden-tified as QSOs in the SDSS DR5, within the redshiftrange 0 . < z < .
81 such that H β and [O iii] emis-sion were both measurable. The lines were measured bymeans of a least squares fit of a Gauss-Hermite func-tion (Pinkney et al. 2003) to the line profile, togetherwith a linear fit to the continuum in the vicinity of theline. The broad H β line was fit after removing an as-sumed narrow H β line with the profile and 10% of theflux of the [O iii] λ ii line, and the narrow emission lines [O ii] and [S ii] were measured when accessible. We assumeddoublet ratios of unity for Mg ii and 1.2 red-to-blue for [O ii] . Spectral fits were accepted if the line widths andequivalent widths had an accuracy better than 15% andvisual inspections showed a good fit and no artifacts.For more details on the measurement procedure, includ-ing Fe ii subtraction, see Salviander et al. (2007). Ourfinal data set consists of 2598 objects. The redshifts ofthe peaks of the lines were calculated from the fits to theline profiles. The relative displacement of the broad H β line with respect to the peak of [O iii] was calculated as∆ v H β = c ( z H β − z [ O iii ] ) / (1 + z [ O iii ] ). The displacementof Mg ii from [O iii] , ∆ v M g , is defined analogously to∆ v H β .A histogram of ∆ v H β and ∆ v M g is shown in Figure1. A Gaussian fit to the distribution of ∆ v H β gives amean displacement of +100 km s − , and the FWHM ofthe distribution is 500 km s − . In our sample, 40 QSOs(1.5%) have ∆ v H β displacements greater than 600 km s − from the mean.The fact that there is an overall redshift may largelyresult from physical processes in the BLR. However, wefind an average [O iii] blueshift of 30 km s − relativeto [S ii] and 40 km s − relative to [O ii] , similar to theresults of Boroson (2005). The largest blueshifts of [O iii] relative to [O ii] were found for objects with small ∆ v H β .Interestingly, the 36 objects with ∆ v H β more than 600km s − from the mean and measurable [O ii] showedgood agreement between [O iii] and [O ii] . Therefore,use of [O iii] as a reference velocity does not significantlyaffect our calculation of the high H β line displacements. -3000 -2000 -1000 0 1000 2000 3000 ∆ v (km s -1 )050100150200250300 N u m b e r o f Q S O s ∆ v H β ∆ v Mg II
Fig. 1.—
Histogram of ∆ v H β and ∆ v M g along with Gaussian fitsto the data. The distribution is somewhat broader than a Gaussian,with small but significant numbers of outliers with | ∆ v | > − . COMPARISON WITH NUMERICAL SIMULATIONS
Schnittman & Buonanno (2007) compute the blackhole recoil velocities for a range of mass ratios, spinmagnitudes and directions from post-Newtonian equa-tions of motion calibrated to the results of the numer-ical simulations. For mergers of two black holes withequal spin parameter a ∗ = 0 .
9, and a restricted set ofmass distributions such that m m / ( m m > ∼ . f =0 .
31 of recoils greater than 500 km s − and a frac-tion f = 0 .
079 of recoils greater than 1000 km s − .Convolving the probabilities of Schnittman & Buonanno(2007) with random kick inclinations to the line of sight,we find the predicted observational kick fractions to be f = 0 .
18 and f = 0 . v H β are f = 0 .
04 and f = 0 . DISCUSSION
The observed incidence of H β shifts over 1000 km/s isseveral times less than theoretically expected for rapidlyspinning holes with random orientations and similarmasses. Moreover, closer inspection of the objects withlarge ∆ v H β suggests that these large displacements mostlikely result from BLR physics rather than recoils.1. The largest shifts occur only for objects withlarge H β FWHM. (This differs from the find-ings of Sulentic et al. (2007) and Richards et al.(2002) for C IV λ β lines peaks at about3000 km s − , but for objects with shifts greaterthan 600 km s − from the mean we find an aver-age FWHM of ∼ − , with only one ob-ject (SDSS J141959.21+610143.6) narrower than3000 km s − . We have no reason to expect recoilsto prefer larger FWHM.2. For recoils, all broad lines should have approxi-mately the same velocity shift. Figure 2 shows thatecoiling black holes in quasars 3 Fig. 2.—
A plot of Mg ii – and H β – [O iii] displacements. Thedotted line shows a fit to the data of ∆ v M g = 0.6 × ∆ v H β , andthe solid line shows the relation of equality. Removal of the highredshift outlier does not change the fit. typically ∆ v M g ≈ . v H β for high as well as lowshifts. No subset of objects stands out to the eyeas noteworthy candidates to be recoils.3. We find more redshifts than blueshifts, contrary toexpectation for random kick directions. Relativeto our measured mean, there are 59 redshifts over500 km s − and 47 blueshifts under -500 km s − (out of 2598 objects). We note that this differenceis not highly significant, and is dependent on theaverage blueshift of [O iii] with respect to systemicvelocity.4. We have constructed composite spectra for objectswith ∆ v H β greater than 600 km s − from the mean,less than -600 km s − from the mean, and with ab-solute value of the shift less than 600 km s − ˙Inall three composite spectra, shown in Figure 3, the[Ne v ] FWHM is several hundred km s − greaterthan that of the lower ionization lines. This is typ-ical for QSOs, where high ionization or high crit-ical density lines are thought to originate close tothe black hole. The close in gas will feel the blackhole potential and have a higher velocity compo-nent leading to a broader profile (Laor 2007, andreferences therein). However, in the case of a kickeddisk, one might expect a ‘left behind’ NLR or in-terstellar medium ionized by the off-center QSO,which would show line widths consistent among thenarrow lines. Alternatively, if there were a com-pact NLR bound to the black hole, it would con-tribute a [Ne v ] component centered on the velocityof the black hole. Neither of these cases appearin the composite spectra, suggesting that high-∆ v H β QSOs have normal NLRs. Table 1 showsthe FWHM and equivalent width (EW) of [O ii] ,[Ne iii ], and [Ne v ] in the composite spectra ofFigure 3. It can be seen that the shifted QSOshave similar narrow line intensities as the unshiftedQSOs. ∆ v H β > 700 km s -1 F λ ( a r b it r a r y ) -500 km s -1 < ∆ v H β < 700 km s -1 ∆ v H β < -500 km s -1 Fig. 3.—
Composite spectra for the highest H β redshifts relativeto [O iii] (top panel, 33 objects), highest blueshifts (lower panel,24 objects), and shifts centering around the mean (middle panel,2541 objects). TABLE 1Composite spectra
EW (˚A) redshift blueshift | ∆ v H β | <
600 km s − [O ii] [O iii] iii ] 2.73 2.10 1.90[Ne v ] 1.95 1.58 1.5FWHM (km s − ) redshift blueshift | ∆ v H β | <
600 km s − [O ii]
582 472 506 [O iii]
541 536 420[Ne iii ] 748 595 516[Ne v ] 775 615 650
5. Finally, we visually inspected the high shiftobjects to see if any of them stood out askick candidates. Of particular interest wasSDSS J091833.82+315621.1, shown in Figure 4,which has the the largest H β shift in our sample(∆ v H β = 2667 km s − and ∆ v M g = 1231km s − ).This is very near the maximum recoil velocityrecorded by numerical simulations (Gonzalez et al.2007; Tichy & Marronetti 2007). While the shiftedH β line is striking in appearance and symmetricin shape, there seems to be little else in the spec-trum to distinguish it from a non-kicked BLR. NLRline ratios and intensities are normal, and its ∆ v M g is consistent with 0.6∆ v H β as shown in Figure 2.While we cannot definitely rule out a kick, theshifted H β line in this object may just as well bedue to some other physical process in the BLR.Some highly blueshifted objects, such asSDSS J120354.76+371137.2 (∆ v H β =-513 km s − )or SDSS J135800.40+404358.1 (∆ v H β =-813km s − ) have a sharp cutoff on the red side of theline, often accompanied by a ledge or shoulder. Inthese cases, the asymmetry shifts the peak, butthe wings more nearly center on the [O iii] redshift.Similarly, Richards et al. (2002) found that largeblueshifts (relative to [O iii] ) in the broad C iv and Mg ii lines can be attributed to a diminution Bonning et al. Fig. 4.—
SDSS J091833.82+315621.1, the largest shifted objectin our sample with ∆ v H β = 2667 km s − . of flux in the red wing of the line rather than to atrue shift. Therefore, it seems likely that in theseobjects, the shape of broad H β is altered in such away as to give a blueshifted peak.Possibly more promising are objects which, be-sides having symmetrical broad lines, also havesimilar ∆ v H β and ∆ v M g in addition to a [Ne v ]line with similar broadening as the lower ioniza-tion narrow lines. Two such objects in our sam-ple, albeit with fairly low signal/noise spectra, areSDSS J134812.36+052402.6, with (∆ v H β ,∆ v M g ) =(-706,-769) km s − and FWHM ([Ne v ], [O iii] ) =(380,308) km s − , and SDSS J103144.53+415420.8, with (∆ v H β ,∆ v M g ) = (-518,-462) km s − andFWHM ([Ne v ], [O iii] ) = (561,572) km s − . Suchobjects may merit further investigation.In summary, we find a number of QSOs with displacedbroad line peaks relative to the narrow lines. However,for a variety of reasons, few if any of these are likelycandidates for recoiling black holes. The retained diskmass in a recoil is enough to power a QSO episode last-ing a substantial fraction of the total QSO phase, so thatkicked QSOs stand a fair chance of being observed if theyoccur. It is therefore likely that some mechanism is atwork to prefer small kicks over large ones. For example,Bogdanovic et al. (2007) propose that in gas-rich merg-ers, the spin and orbital angular momenta of the blackholes become aligned with that of the large scale gas flow.If this process can occur before the final merger, the blackholes will be in a configuration leading to small ( < − ) kicks, which would be undetectable through dis-placed broad lines. Alternatively, if black holes do notmerge until after the QSO phase, there would be no diskto fuel a kicked QSO. Such a phenomenon would have tobe detected through other means, such as tidal disrup-tion of stars in the galactic bulge (Gezari et al. 2006) orenlargement of the galactic core (Merritt et al. 2004).The authors thank Richard Matzner for enlighteningdiscussions. EWB is supported by Marie Curie Incom-ing European Fellowship contract MIF1-CT-2005-008762within the 6th European Community Framework Pro-gramme.) kicks, which would be undetectable through dis-placed broad lines. Alternatively, if black holes do notmerge until after the QSO phase, there would be no diskto fuel a kicked QSO. Such a phenomenon would have tobe detected through other means, such as tidal disrup-tion of stars in the galactic bulge (Gezari et al. 2006) orenlargement of the galactic core (Merritt et al. 2004).The authors thank Richard Matzner for enlighteningdiscussions. EWB is supported by Marie Curie Incom-ing European Fellowship contract MIF1-CT-2005-008762within the 6th European Community Framework Pro-gramme.