Ultimate fate of apparent horizons during a binary black hole merger II: Horizons weaving back and forth in time

Daniel Pook-Kolb,1, 2 Ivan Booth,3 and Robie A. Hennigar3, 4, 5 1Max-Planck-Institut f¨urGravitationsphysik (Albert Einstein Institute), Callinstr. 38, 30167 Hannover, Germany 2Leibniz Universit¨atHannover, 30167 Hannover, Germany 3Department of Mathematics and Statistics, Memorial University of Newfoundland, St. John’s, Newfoundland and Labrador, A1C 5S7, Canada 4Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1 5Department of Physics and Computer Science, Wilfrid Laurier University, Waterloo, Ontario, Canada N2L 3C5 In this second part of a two-part paper, we discuss numerical simulations of a head-on merger of two non-spinning black holes. We resolve the fate of the original two apparent horizons by showing that after intersecting, their world tubes “turn around” and continue backwards in time. Using the method presented in the first paper [1] to locate these surfaces, we resolve several such world tubes evolving and connecting through various bifurcations and annihilations. This also draws a consistent picture of the full merger in terms of apparent horizons, or more generally, marginally outer trapped surfaces (MOTSs). The MOTS stability operator provides a natural mechanism to identify MOTSs which should be thought of as black hole boundaries. These are the two initial ones and the final remnant. All other MOTSs lie in the interior and are neither stable nor inner trapped.

I. INTRODUCTION to a MOTT foliated by apparent horizons as dynamical apparent horizon (DAH). The now numerous detections of gravitational wave Whether we consider dynamical horizons or MOTTs events leave little doubt that black hole coalescences are in general, one seemingly basic question remained open: a regularly occurring phenomenon in our universe [2–8]. What happens to the horizons of two black holes when With the help of numerical relativity simulations, the they merge? It is well known that a common apparent produced gravitational waves travelling to distant ob- horizon forms around the two individual ones when they servers are analysed and modeled with steadily increas- are sufficiently close to each other. This common horizon ing efficiency and accuracy [9, 10]. However, the details immediately splits into an outer and an inner branch. of the merger of the black holes themselves is less well This fact together with the observation that MOTTs may understood. This is partly for conceptual and partly for in principle weave back and forth in time (see e.g. [15, 20]) numerical reasons. sparked speculations that all the horizons in a binary On the conceptual side, one needs to answer the ques- merger might, in fact, be parts of a single world tube tion of how to describe black holes in highly dynamical [21–23]. situations. When a black hole is at rest or only slightly Prior to recent advances in methods for locating perturbed, the event horizon is a suitable description [11]. MOTSs numerically [24], it was not possible to further In non-perturbative cases, however, its teleological na- investigate these ideas. The world tubes that had to be ture makes it unsuitable for gaining an understanding of tracked developed extremely distorted shapes, which the the dynamics [12–15]. A much better alternative is pro- typically used algorithms failed to resolve. The funda- vided by the quasilocal horizon framework [12, 16, 17]. mental assumption often employed to simplify the nu- The central concept in this framework, the dynamical merical task is that the surface we wish to locate is horizon, presents a notion of black holes that is valid star shaped, i.e. it can be represented using an angle- and satisfies physical laws even in the highly nonlinear dependent (coordinate) distance function from some ref- arXiv:2104.11344v1 [gr-qc] 22 Apr 2021 phases of the merger. Dynamical horizons are based on erence point. See Ref. [25] for a review. the numerically accessible marginally outer trapped sur- By removing this limitation, it was shown recently that faces (MOTSs), i.e. surfaces defined as having vanish- the individual apparent horizons connect indirectly to ing outward expansion. FollowingS such a MOTS through the outer common horizon by merging (non-smoothly) the time evolution of a spacetime generates a world tube, with the world tube of its inner branch at the time when called a marginally outer trapped tube (MOTT). In their the individual apparent horizons touch [26–29]. However, original definition [12, 16, 17], dynamical horizons are a they continue to exist afterwards and their later fate was certain subset of MOTTs. In the present work, MOTS not fully resolved in these studies. One reason for this is stability [18, 19] will play a central role and we shall re- that in the utilized methods, one still had to anticipate fer to a stable MOTS as apparent horizon1 (AH) and the possible shapes with appropriate initial guesses. As we shall see, resolving their full fate requires even more exotic guesses, and for this a new method was needed. We have developed and presented such a method in the 1 The compatibility with the traditional terminology of apparent first paper of this two-part series, henceforth denoted as horizons as boundaries of trapped regions will be discussed below. paper I [1]. As we shall see in the remainder of this second 2 paper, having such a method is the key to resolving this 2. Section V connects the various observations made in question. theS previous sections with the MOTS stability properties. One might argue that in classical general relativity, The signature and the expansion of the ingoing null rays, whatever happens in the interior of the event horizon re- together important for understanding the behavior of the mains – by definition – causally disconnected from far area, are presented in Section VI. Finally, Section VII will away observers. MOTTs are always located in this inte- conclude with a discussion of the main results. rior region and, after the outer common apparent horizon has formed, the individual apparent horizons are even further away from it. Nevertheless, the question of their II. BASIC NOTIONS fate seems highly relevant if one aims at using MOTTs to understand the merger. To allow for a physically mean- A. Marginal surfaces and their world tubes ingful interpretation, these should be well-behaved ob- jects in the first place. But how is this compatible with We consider four-dimensional spacetime ( , gαβ, α) the results of paper I? There, we have shown that a two- with Lorentzian four-metric g of signatureM ( + +∇ +). αβ − black-hole configuration may contain a large number of For a smooth spacelike two-surface ( , qAB, A), let `± MOTSs not previously known. Are these merely artifacts be two linearly independent future pointingS nullD normals + of an “unphysical” configuration? Since they were found scaled such that ` `− = 1. In the present paper we in time-symmetric initial data with a common AH al- will only consider closed· surfaces− and we assume it ready present, neither the past nor the future of this is possible to assign an outward directionS on . Then, + S configuration contains two separate black holes. In the ` is taken to be pointing outward and `− inward. The present paper, we therefore look at the merger of initially expansions Θ of a congruence of null rays travelling in ± separate black holes and aim to answer two questions: i) the `± directions is then given by Do such additional MOTSs also form dynamically in a αβ merger of initially separate black holes? ii) What physi- Θ = q α`β± , (1) cal significance do they have and how can we differentiate ± ∇ them from the intuitively more relevant apparent hori- αβ α β AB α where q = eAeBq with eA being the pull-back from zons we associate with the individual black holes ( ) S1,2 to . The expansions can be seen as the trace of the and the final common one ( outer)? M S S extrinsic curvatures kAB± of associated with `±. The To answer these questions, we perform numerical simu- (symmetric) trace-free part isS given by the shear lations of the head-on collision of two non-spinning black holes with no initial momentum. In particular, we show 1 α β 1 σ± = A`± Θ qAB = eAe α`± Θ qAB . (2) explicitly that i) such additional MOTSs do form dynam- AB ∇ B − 2 ± B∇ β − 2 ± ically and ii) the new MOTTs we track do, in fact, bifur- We will call Θ the outgoing and Θ the ingoing expan- cate and annihilate with the other MOTTs. This means + sion. − that they are indeed weaving back and forth in time, but The signs of Θ allow us to classify . In particular, if as we will see later, they do not connect to form just Θ < 0 then is± called a trapped surfaceS . The existence one single smooth surface. A remarkable result is a sur- of± such a surfaceS has been proven to imply that spacetime prisingly clear and predictable behavior in terms of the is causally geodesically incomplete and thus singular [30]. stability of these MOTTs: Whenever a MOTT switches is called a marginally trapped surface if Θ < 0 and direction in time, it gains an additional negative eigen- ΘS = 0 and a marginally outer trapped surface− (MOTS) value of the stability operator, i.e. it effectively becomes + if Θ = 0 with no restriction on Θ . We mention here “more unstable”. Furthermore, only three MOTTs are + that Andersson et al. show in [31]− that existence of a stable (and hence DAHs) in the sense of [18, 19] and are strictly stable MOTS (introduced in Section II B) is suf- thus distinguished from all other MOTTs we find. These ficient for the singularity theorem mentioned above to are the two DAHs traced out by and associated 1 2 hold. Note that we can still scale the null normals by with the individual black holes andS the oneS final com- arbitrary positive functions f > 0 via mon DAH traced out by . Souter The rest of this paper is organized as follows. We start + + 1 ` f` and `− `− . (3) by fixing the notation and introducing the required math- → → f ematical concepts in Section II. Section III gives the numerical details of our setup, the simulations and the Fortunately, the signs of the expansions Θ and conse- method to locate and track the marginal surfaces. The quently the above characterization of is invariant± under new MOTSs and the world tubes they trace out are intro- these transformations. S duced in Section IV. Here, we also describe a mechanism Let the spacetime be foliated by spatial slices M occurring multiple times along otherwise smoothly evolv- (Σt, hij,Di,Kij) with Riemannian three-metric hij and ing MOTTs where a cusp forms followed by a new self- extrinsic curvature Kij. Following a MOTS through intersection. Additionally, we present MOTSs of toroidal slices of the foliation provides the notion ofS a MOTT. topology that exist inside the individual MOTSs and More precisely, a smooth three-manifold is called a S1 H 3 marginally outer trapped tube (MOTT) if it admits a fo- liation of MOTSs. Note that this definition of a MOTT makes no use of the foliation of spacetime by the Σt. We will, however, only consider MOTTs with a foliation of H MOTSs contained in the slices Σt. For a spacelike future (Θ < 0) MOTT , the foliation can be shown to be unique− [32]. H The above objects are closely related to various terms involving the word “horizon”. For instance, a spacelike fu- FIG. 1. Spherically symmetric evolution of a MOTT H (thin ture MOTT is defined as a dynamical horizon in [17, 33] solid line). It is perfectly possible for H to weave back and while in [28, 29], the definition of a dynamical horizon forth through the time foliation. See, for example, [20, 21] for has been generalized to refer to any MOTT. Additional exact solutions exhibiting this behaviour. qualifiers (e.g. future, spacelike) were used to then spe- cialize where needed. One of the reasons for this gen- eralization is that MOTSs and MOTTs were found to then be labeled as outer trapped, outer untrapped or appear in a much wider variety than expected and the marginally outer trapped as shown in FIG. 1. A MOTT original dynamical horizon did not cover all the interest- traces a curve through the (t, r) space and it can weave ing cases. However, the examples in [28, 29] were just the itsH way back and forth through the foliation of spacetime. start and [34] showed that even in a single slice of the Tangents to this curve can be written in the form Schwarzschild spacetime there are an infinite number of MOTSs. This together with the results shown in the first α = N(uα + vRα) , (4) V and this second paper clearly suggest that not all of these α α objects should be thought of as black hole boundaries. where u and R are respectively the unit vectors in the t As mentioned in the introduction, the notion of MOTS and r directions. N can be any function of (t, r), though a convenient choice is the lapse. v is the speed of relative stability turns out to reliably select those MOTSs which H possess reasonable physical properties and can thus be to the foliation. is spacelike if v > 1, timelike if v < 1 and null if v =H 1. If becomes| tangent| to the foliation| | called horizons. We will therefore introduce the MOTS | | H stability operator in the following section. (as at B or C) then at that point v . The speed v may also be calculated| | → from ∞ the fact that Θ+ = 0 on . Then writing the derivative in the direc- tion as δ (theH notation is chosen to be compatibleV with B. MOTS stability V the next section), δ Θ+ = 0. It follows that V

The concept of MOTS stability in the sense of Refs. [18, δ Θ δ + Θ v = u + = 1 2 ` + , (5) 19] is helpful to get a deeper insight into the evolution −δ Θ − δ Θ properties of a MOTS. In particular, it will be useful in R + R + assessing at which time t a MOTT becomes tangent where we made a particular choice for the null vectors H to a spatial slice Σt and thus generically “turns around” relative to u and R: in time. These instances correspond to a MOTT appear- + 1 ing and bifurcating into two branches, as well as when ` = (u + R) and `− = u R. (6) two MOTTs merge and annihilate. Furthermore, we ex- 2 − pect the physically relevant horizons to be boundaries for Then we can again see that if δRΘ+ 0, v (as trapped and untrapped surfaces, at least in a neighbour- → | | → ∞ long as δ`+ Θ+ = 0). hood. This property also turns out to be closely related Next, applying6 the spherically symmetric null Ray- to the notion of MOTS stability. chaudhuri equation we find Before delving into the fully generic case, it will be helpful to first consider a very simple situation in which G v = 1 + 2 ++ . (7) the stability operator appears almost naturally. Spheri- δRΘ+ cally symmetric spacetimes provide such an example and so here we restrict our attention to spherically symmetric α β By the null energy condition, G++ = Gαβ`+`+ 0. If it MOTSs. vanishes, then v = 1 and so is outward null and≥ isolated H α β at that point. If matter falls through (Gαβ`+`+ > 0) and the region outside of is outer untrappedH (from A 1. MOTTs in spherical symmetry H to C and then B onwards in FIG. 1) then δRΘ+ > 0 v > 1 and is spacelike outward at that point. However⇒ We will start by establishing some basic properties of if the regionH outside of is outer trapped (from C to H MOTTs in spherical symmetry. In this setting, each point B) then δRΘ+ < 0 and so could be either spacelike, in the two-dimensional (t, r) space represents a sphere timelike or (inward) null. AtHC it transitions from v = and we can calculate its expansion Θ . A point (t, r) can to and at B from to . ∞ + −∞ −∞ ∞ 4

These quantities also determine the expansion of . for a function ψ. Note that the local deformation of is From (4) and (7) H fully determined by ψ as a function on just . ForS the S   spherical deformations of the last section, ψ would be a G++ α i Θ = N (Θu + vΘR) = Θ . (8) constant and V would be the normal R in a t = const V − δRΘ+ − slice. Next, to each of the ν we can construct a (non-unique) Thus if G++ = 0, then Θ is non-expanding. However S ν V pair of null normals `± and calculate the expansions Θ . if G++ = 0 and Θ < 0 (that is, the (t, r) spheres ν 6 − ± get smaller moving inward from the MOTT), then is They are chosen so that `0± = `± (the original null nor- H expanding if the region just outside is outer untrapped mals on ). The MOTS stability operator LV with re- S α 2 α + (from A to C and then B onwards) and shrinking if the spect to the normal V , scaled such that V `α = 1, is ν region outside of is outer trapped (from C to B). To- then defined as the derivative of Θ+ with respect to ν: gether these resultsH mean that if we consider as a con- tinuous curve running in the direction A to BH then it is ∂ ν LV ψ := δψV Θ+ := Θ+ . (10) increasing in area as one would expect for a black hole ∂ν ν=0 horizon. It is shown in [19] that LV does not depend on the choice Note that the exact location of the transition points of `± away from . On the other hand, the definition of is at least partly a function of the foliation. For exam- ν S LV is not invariant under the rescalings (3) of the null ple, if we chose a foliation rotated relative to our original normals `±. However, since we will be interested only in Σt in FIG. 1, then the points of tangency between and H its eigenvalues, we can use the fact that LV is isospectral the foliation would be different. Their number could even under (3) as shown in [35] and we shall work with the increase. However the possible timelike signature of H particular choice (6). means that this wending through time cannot (always) Just as in the spherical case of the previous section, be understood as simply a by-product of the choice of we will discuss the evolution properties of a MOTT spacetime foliation. Timelike sections necessarily inter- H in the context of some fixed foliation Σt and use this sect with many slices of any foliation. foliation to talk about bifurcations and annihilations. We Finally note that a simulation that only tracked outer- will then need to choose the vector V α as the normal of most MOTSs would see an apparent horizon jump from 3 α α S which lies in Σt, i.e. we choose V = 2R . The stability A to B at Σ2. However a more careful tracker would iden- operator with respect to the slice Σt is then defined as tify B as a MOTS pair creation event with one surface LΣ := L2Rα and it takes the form (e.g. [18, 19, 36]): subsequently moving inwards while the other expands outwards. Ultimately that inward moving MOTS would   ˜ 1 2 annihilate with the original apparent horizon at C. Both LΣψ = ψ + 2 σ+ 2G++ G+ ψ , −4 2R − | | − − − pair creation and annihilation events happen at points (11) where δ Θ 0. R + → where factors of 2 differing from [18, 19] result from our + different cross normalization ` `− = 1 and where 2. The MOTS stability operator · − A A ˜ ψ = ( A ωA)( ω )ψ (12) Away from spherical symmetry everything is more 4 D − D − α β complicated, though many of the themes (and conclu- with ω = e `− ` the connection on the nor- A − A β ∇α + sions) that we have just examined remain. In particular α β mal bundle of , its Ricci scalar, G+ = Gαβ`+` , δ Θ continues to play a key role though it is now gen- S R − − R + G = G `α `β and σ 2 = σ+ σAB. This is a second eralized to become the MOTS stability operator [18, 19]. ++ αβ + + + AB + order, linear, elliptic operator| | with discrete spectrum and Leaving rigid spherical symmetry behind, there is no for a non-vanishing connection ω it is not self-adjoint. longer just one way to expand a surface outwards (or in- A However, its principal eigenvalue λ , i.e. the eigenvalue wards). Since deformations tangent to a MOTS leave 0 with the smallest real part, is always real. A MOTS Θ invariant, we will only consider deformationsS along + with λ 0, λ > 0, λ < 0 is called stable, strictlyS some direction V α normal to . In principle, this need 0 0 0 stable, or≥unstable, respectively. The meaning of this ter- not be limited to a particular sliceS Σ containing . Now, t minology will soon become clear. for such a deformation of , consider a family Sof sim- ν Note that for the spherical cases that we considered in ilar surfaces such that S = . Then we canS consider 0 the last section, ω = 0 and σ+ = 0 and we were only the vector field of normalS vectorsS V α to these surfaces. A AB In turn these generate a congruence of curves that map points between the ν and we can write the tangent vec- tor to these curves,S the generator of the deformations, 2 In particular, V cannot be parallel to `+. as 3 The factor 2 results from our different convention (6) for the ∂ scaling of `+ as compared to [18]. However, this does not change = ψV (9) the spectrum of L . ∂ν V 5

for ψ. The inhomogeneity on the righthand side of this is fixed by the foliation. Note that if LΣ is invertible, then there exists a so- lution for (16) for any lapse N. This is guaranteed if, e.g., λ0 > 0. However, if λ0 0, then invertibility can fail, as eigenvalues may vanish.≤ Equivalently, in that case there are homogeneous solutions to the evolution equa- tion (16): that is we can choose a α that is tangent V to Σt and still satisfies (15). The pair creation event B FIG. 2. Definition of the evolution vector V along a MOTT H. in FIG. 1 is such an event. This association of vanish- V can be split into a component ψR orthogonal to S = H∩Σ t1 ing eigenvalues with pair creation/annihilation has also within Σ and one orthogonal to Σ , i.e. Nu, which is also t1 t1 been observed away from spherical symmetry [24]. In the orthogonal to S. present paper, we shall see multiple instances of horizons appearing and vanishing precisely as one of the eigenval- considering constant ψ. With those restrictions ues – not necessarily λ0 – becomes zero. If it is λ0 that vanishes, then Proposition 5.1 of [31] shows that generi-   1 cally is tangent to Σt and, for fixed slicing, unique at LΣψ = 2 δψRΘ+ = 2G++ G+ ψ , (13) H R2 − − − least in a neighbourhood. Note too that λ > 0 implies that δ Θ > 0 and where R is the areal radius of the MOTS. Then the term 0 ψR + again, just as we saw in the spherically symmetric case, in parentheses is the principal4 eigenvalue of L . Hence Σ this is sufficient to imply that the evolving is space- in spherical symmetry we can only have an “ingoing” H H like at that point [18, 19] and so, if Θ < 0, expand- when the sum of the matter terms is comparable in size to − ing (see also [33, 37]). This area expansion theorem has 1/R2. If it is not then, as we saw earlier, is necessarily been generalized to include timelike “backwards in time” spacelike and (if Θ < 0) expanding. HoweverH if it is segments [38]. However this is not the end of the story: equal so that δ Θ − = 0, then we can have horizon pair ψR + certain assumptions made in that generalized proof have formation (annihilation) as at B (C) in FIG. 1. Physically now been shown to not always hold during black hole this can be thought of as a case where the matter outside mergers [28, 29]. We will see further examples of this in is dense enough to cause the formation of a new horizon Section VI. outside the old one [21]. Stability is not only useful for understanding the evolu- Returning to the general stability operator (11), An- tion of , it also tells us something about local properties dersson et al. proved in [18, 19] that λ > 0 implies ex- 0 of withinH the slice Σ . In paper I, it is shown that the istence of a smooth MOTT containing . That is, t stabilityS operator for MOTS can be understood as the evolves smoothly to the futureH and to theS past for atS analogue of the Jacobi operator for geodesics [1]. Thus least a short time interval. This can be understood intu- if one considers an axisymmetric MOTS to be one of a itively in the following way. Consider a three-dimensional congruence of marginally outer trapped (possibly open) MOTT in spacetime, illustrated in FIG. 2. Again, let surfaces, then the number of negative eigenvalues of the α be theH tangent to orthogonal to the MOTSs and stability operator corresponds to the number of inter- scaledV such that t =H 1, where is the Lie-derivativeS V V sections with other nearby members of the congruence. along α. For aL fixed foliation ΣL, we can split α into t Hence a stable MOTS with λ > 0 does not intersect the componentsV orthogonal and tangent to Σ , V 0 t its neighbours while an unstable one certainly does have α = Nuα + ψRα , (14) such intersections. V More generally, it is shown in [18, 19] that a strictly where the foliation fully determines the lapse N and so stable MOTS (λ0 > 0) has the barrier property. In Nuα is fixed. Clearly, the variation of the expansion along S essence, this means that given a close-by surface 0 then the MOTT vanishes, S if 0 has expansion Θ0 0, it cannot extend into the H S + ≤ exterior of . Similarly if Θ0 0, it cannot enter the δ Θ+ = 0 , (15) + V interior. OnS the other hand if ≥has the barrier property, which means that finding the tangent vector α amounts it is at least stable. S V to solving the inhomogeneous partial differential equation Following from these results, we adopt the following convention in the present work. A stable MOTS shall LΣψ = 2δNuΘ+ (16) S − be called an apparent horizon (AH). Note that an appar- ent horizon is usually defined as the outer boundary of the trapped region in a given slice Σt. Due to Theorem 2.1 4 in [31], this boundary is a stable MOTS, so our definition Allowing non-constant ψ would add a term ∆S ψ, where ∆S is the Laplacian on . The spectrum would then− be that of the includes the previous one and extends it to include sur- Laplacian on a roundS sphere (with lowest eigenvalue being zero) faces that can still reasonably be associated with black shifted by the constant term in parenthesis in Eq. (13). hole boundaries. Examples are the previously outermost 6

MOTSs and , as we shall see below. Similarly, a mass ratio of q = m /m = 2 (i.e. m = 1/3, m = 2/3). S1 S2 2 1 1 2 three-surface will be called a dynamical apparent hori- We choose a distance parameter of d := x2 x1 2 = 0.9 zon (DAH) ifH it allows a foliation of apparent horizons. resulting in two black holes which are initiallyk − separatek with no common apparent horizon present. We track the various MOTSs in the simulations using 3. Simplification for vacuum, axial symmetry and no spin the method described in [24, 27] and available from [41], which in turn uses software libraries described in [42– As shown in the first paper, for non-spinning axisym- 49]. Two approaches are used to locate the new MOTSs metric MOTSs in vacuum, we can simplify LΣ to in the simulations. One is the shooting method described S in paper I [1], which can, in principle, locate all axisym-   1 2 metric MOTSs of spherical topology by choosing suit- LΣψ = ∆ ψ + 2 σ+ ψ , (17) − S 2R − | | able starting points on the z-axis. This method has been implemented in [41] and can be applied to both, ana- A where ∆ = A is the Laplacian on . In this case LΣ lytically known initial data as well as to slices obtained S D D S is self-adjoint with purely real spectrum. Another simpli- e.g. from numerical simulations. The other method for fication can be made if we introduce coordinates (θ, φ) generating initial guesses is motivated by the assump- on . Let φ be the coordinate along orbits of the axial tion that MOTSs may vanish and appear only in pairs of S A Killing field ϕ preserving the two-metric qAB and van- two. Based on this idea, we try to track each MOTS to ishing at exactly two points, the poles of . We take φ to S the future and to the past. Whenever a MOTS cannot be be in the range [0, 2π). Let further θ be any coordinate tracked further in either direction, we look for a “close orthogonal to φ, e.g. cos θ = ζ, with ζ constructed as in by” one with which it might merge. Such a merger will Ref. [39]. Then we can write any eigenfunction ψ of LΣ be an annihilation if it happens to the future and a bi- as furcation if it happens in the past direction. Appendix A details the method we use to locate such a corresponding X∞ imφ ψ(θ, φ) = ψm(θ)e . (18) MOTT using families of surfaces of constant expansion m= Θ+. −∞ The simulations themselves are carried out using the For each m Z, the eigenvalue problem LΣψ = λψ then ∈ Einstein Toolkit [50, 51]. The Brill-Lindquist initial data reduces to a one-dimensional problem are generated by TwoPunctures [52], while we use an ax- isymmetric version of McLachlan [53] for evolving these Lmψ := (L + m2qφφ)ψ = λψ . (19) Σ m Σ m m data in the BSSN formulation of the Einstein equations. Kranc The eigenvalues of Eq. (19) are labeled as λl,m, where This uses [54, 55] to generate efficient C++ code. l is chosen to run from l = m over the eigenvalues in We always work with the 1 + log slicing and a Γ-driver ascending order. This guarantees| | that for a round sphere, shift condition [56, 57]. An important feature of these for which the spectrum reduces to that of the Laplacian gauge conditions is that they are “singularity avoiding”, 1 2 which results in simulation time effectively slowing down on a sphere of radius R0 and shifted by 2 = 1/R0, the eigenvalues are labeled in the conventionalR way as close to the punctures. It was seen in [58] that the indi- 2 vidual MOTTs 1,2 essentially stop evolving due to this λl,m = (1 + l(l + 1))/R0. For brevity, we will sometimes S write λl := λl,0. effect, with the precise behavior being highly dependent on the choice of initial data: A smaller initial distance allows the to evolve further during the simulation. S1,2 III. NUMERICAL SETUP AND MOTS We repeat our simulations at different resolutions to en- FINDING sure convergence of our results. Most of these results are obtained using a spatial grid resolution of 1/∆x = 720, We use Brill-Lindquist initial data [40] for our simu- with additional simulations carried out with 1/∆x = 240, lations. These describe a Cauchy slice Σ which is time 360 and 480. Shorter simulations to verify certain fea- symmetric, i.e. with vanishing extrinsic curvature. The tures were performed at 1/∆x = 960 and 1920. We do 3 not use mesh refinement and choose our domain large topology of Σ is R x1, x2 , where x1,2 are the coordi- nates of two punctures.\{ The} Riemannian three-metric is enough to ensure that any boundary effects do not reach 4 the MOTSs for as long as we track them. Additional de- conformally flat, hij = ψ δij, where δij is the flat metric. The conformal factor is given by tails and convergence properties of our simulations are described in Ref. [27]. m m ψ = 1 + 1 + 2 , (20) 2r1 2r2 IV. OVERVIEW OF THE VARIOUS MOTTS where m1,2 are the bare masses of the black holes and r1,2 are the (coordinate) distances to the respective puncture. We shall here focus primarily on one particular configu- The general picture one expects to find in head-on ration with total ADM mass M = m1 + m2 = 1 and a mergers of two black holes has previously been analyzed 7 in great detail [26–29, 59]: Initially, only two individual outer 1 2∗ apparent horizons are present, 1 and 2, belonging to S S S S S outer inner 1∗ 2∗∗ the two separate black holes. At a time tbifurcate, com- S S S inner∗ 1∗∗ 0∗ mon MOTSs outer and inner form as one surface and S S S inner∗∗ 2 ∗∗ bifurcate intoS two branches.S While settles to the S S S0 Souter final Schwarzschild horizon, inner travels inwards and 90 becomes increasingly distorted.S At the precise time when touch, denoted as t , forms a cusp and coin- 80 S1,2 touch Sinner cides with 1 2. Immediately afterwards, 1,2 intersect each otherS while∪S forms self-intersections.S However, 70 Sinner the final fate of 1, 2 and inner had not been resolved in those studies.S WeS shall attemptS to resolve that fate 60 here. 2

In the following, we will encounter several new MOTSs /M 50 and we will, as before, differentiate between them us- area ingS different sub- and superscripts. It is understood that 40 1 2 replacing “ ” with “ ” indicates that we refer to the S ∪ S S H MOTT traced out by . 30 S 20

A. Area evolution outer bifurcate touch t t 10 The main results are most easily visualized in terms 0 1 2 3 4 5 6 of the area of the various MOTSs. FIG. 3 shows that t/M we indeed find multiple new MOTSs previously not known, each forming in a bifurcation as a pair with an outer and an inner branch. Furthermore, we find MOTSs FIG. 3. Evolution of the area of the various MOTSs. Lines of which merge and annihilate in pairs of two. Each bifur- the same color and different line styles smoothly connect and correspond to a single world tube continuing back and forth cation and annihilation connects two MOTTs in a lo- ∗ cally smooth world tube. Upon formation, the area of in time (except for the pair (S1, S1 ), which is discussed in Section IV B 3). The sum of the areas of S is shown as the the outer branch increases while it decreases for the in- 1,2 thin dashed line. This coincides with the area of Sinner at ttouch ner branch. However, with the exception of outer, even marking the merger of Sinner with S1 ∪ S2. Note that despite the areas of the outer branches soon start toS decrease. some MOTSs having larger area than Souter, they are, in fact, This non-monotonic behavior of the area along a smooth outer all contained within Souter for t ≥ tbifurcate. For all curves portion of a MOTT has been previously discussed for that end without smoothly connecting to another curve, we in [26, 28] and has been attributed to properties lost track of the corresponding MOTS for numerical reasons Sinner of the expansion Θ of the ingoing null rays `− and (see the end of Section IV A for details). to the signature of− the world tube. We shall here ex- tend the discussion of these properties to the new world tubes in Section VI. One characteristic that all MOTSs entiate between a MOTS vanishing due to annihilation S along such a MOTT have in common is which punctures and one vanishing due to loss of accuracy. they enclose, i.e. no MOTS crosses a puncture in its evo- lution. Further, all additional bifurcations happen after the formation of the outermost common MOTS outer at B. The world tubes outer S tbifurcate and the new MOTSs are solely contained within . Souter In this subsection, we will give an overall description We note here that we lose track of some of the MOTSs, of the individual connected world tubes. such as 2∗∗, during the simulation without having an in- dicationS of an annihilation. The reason we cannot track these MOTSs further is a purely numerical one. The shapes move very close to one of the punctures in our 1. The world tube of Souter numerical coordinates (the proper distance to the punc- ture is, of course, always infinite). This close proximity The most complicated of the four world tubes is the results in loss of numerical accuracy since a large spatial one that asymptotes to the final Schwarzschild horizon. region is covered by a decreasing number of numerical Starting with , this MOTT is composed of the se- Souter grid points as the puncture is approached, i.e. this region quence outer inner inner∗ inner∗∗ . All MOTSs is numerically underresolved. Fortunately, the analysis of along thisS MOTT→ S enclose→ both S punctures.→ S FIG. 4 shows the MOTS stability spectrum enables us to clearly differ- several examples of these MOTSs as we follow this world 8

outer at t = 6 M outer at t = 0.725 M inner at t = 1.5 M 1.0 S 0.75 S S

0.50 0.50 0.5 0.25 0.25

0.0 0.00 0.00 z/M z/M z/M 0.25 0.25 0.5 − − − 0.50 0.50 − − 1.0 − 0.75 0.75 − −

1.0 0.5 0.0 0.5 1.0 0.5 0.0 0.5 0.5 0.0 0.5 − − − − x/M x/M x/M

at t 3.8628 M at t = 5.5 M ∗ at t = 2.5 M Sinner ≈ Sinner Sinner 0.0 0.50 0.2 0.25 0.0 0.2 − 0.00 0.2 − 0.25 z/M z/M 0.4 z/M − 0.4 − − 0.50 − 0.6 − 0.6 0.75 − − 0.8 1.00 − − 0.5 0.0 0.5 0.4 0.2 0.0 0.2 0.4 0.5 0.0 0.5 − − − − x/M x/M x/M

∗ at t 0.8181 M ∗∗ at t = 3 M ∗∗ at t 4.9667 M Sinner ≈ Sinner Sinner ≈ 0.50 0.4 0.0 0.2 0.25 0.0 0.2 0.00 − 0.2

z/M z/M − z/M 0.25 0.4 − 0.4 − − 0.50 − 0.6 − 0.6 0.04 − − 0.75 0.8 0.06 − − − 0.03 0.00 0.03 − 0.5 0.0 0.5 0.5 0.0 0.5 0.4 0.2 0.0 0.2 0.4 − − − − x/M x/M x/M

∗ ∗∗ FIG. 4. Several shapes of the MOTSs along the connected MOTT following the sequence Souter → Sinner → Sinner → Sinner. The panels are to be read row by row from left to right. The respective inset in the bottom-left indicates the location of the shown MOTS on the connected area curve with a red dot (see FIG. 3 for the precise axes and labels). For reference, the shapes of S1 and S2 are drawn as light grey solid lines (except for the top-left and the center panel, where S2 does not exist). We start ∗∗ in the top-left panel with Souter at t = 6.5M and then go backwards along the MOTT until we reach Sinner at the final time t = 4.4625M when it could be located (bottom-right panel). The inset in the bottom-right of this last panel shows the newly ∗∗ formed second self-intersection of Sinner. tube, starting with (top left panel) and mov- MOTT , we find that coincides momentar- Souter Hinner Sinner ing backwards. We find the well-known bifurcation of ily with 1 2 at ttouch 3.86M (middle left panel) ( , ) at touter 0.702M (top center panel). and afterwardsS ∪ S develops self-intersections.≈ It merges and Souter Sinner bifurcate ≈ Going now forward in time along the inner common annihilates smoothly with ∗ , which retains the self- Sinner 9

at t = 0 at t 5.3399 M ∗ at t = 5.3 M S2 S2 ≈ S2 0.6 0.0 0.0 0.4

0.2 0.2 0.2 − − 0.0

z/M z/M z/M 0.4 0.2 0.4 − − − 0.4 0.6 − 0.6 − 0.6 − − 0.8 0.8 0.8 − − 0.5 0.0 0.5 − 0.4 0.2 0.0 0.2 0.4 0.4 0.2 0.0 0.2 0.4 − − − − − x/M x/M x/M

∗ at t = 1.5 M ∗∗ at t 0.8181 M ∗∗ at t 3.4333 M S2 S2 ≈ S2 ≈ 0.4 0.50 0.50 0.2 0.25 0.25 0.0 0.00 0.00 0.2 − z/M 0.25 z/M 0.25 z/M − − 0.4 0.50 − − 0.50 0.6 − − 0.75 − 0.75 0.8 − − 1.00 − 0.5 0.0 0.5 0.5 0.0 0.5 0.5 0.0 0.5 − − − x/M x/M x/M

∗ ∗∗ FIG. 5. Same as FIG. 4 but showing MOTSs along the connected MOTT consisting of S2 → S2 → S2 .

intersection, and travels from the annihilation backwards 2. The world tube of S2 outer in time until shortly after tbifurcate (bottom left panel). At this point, it smoothly connects to inner∗∗ in a bifurca- S This MOTT consists of the sequence 2 2∗ 2∗∗, tion. inner∗∗ , initially fully outside and enclosing 2, sub- S → S → S S S with MOTSs shown in FIG. 5. Starting with 2 at t = 0 sequently becomes increasingly distorted. Just like inner, S inner∗∗ S (top left panel), we find the annihilation of 2 with 2∗ it forms a cusp at the time tcusp when the lower part of S S at t 5.35M. ∗ is then followed backwards until the it passes from the outside to the inside of 2, after which ≈ S2 it has a second self-intersection (bottom rightS panel). point where it bifurcates with 2∗∗ at t 0.811M (bottom center panel). All MOTSs alongS this MOTT≈ enclose only the puncture in the interior of 2, i.e. that of the larger of the two original black holes.S The bottom right panel of FIG. 5 shows the last time we were able to reliably locate ∗∗. Shortly after this time, it gets too close (in The two instances where a cusp is formed along this S2 MOTT are very similar in nature. An important ingredi- coordinates) to the puncture inside 1, resulting in loss of numerical accuracy. However, justS as for and ent for understanding these is the maximum principle for Sinner ∗∗ , ∗∗ approaches from the outside. We suspect MOTSs (c.f. Section 3.2 in [22]). This implies that two Sinner S2 S2 that ∗∗ will subsequently move to the inside of 2 by smooth MOTSs and 0, one enclosing the other, must S2 S be identical if theyS haveS a common point with normals forming a cusp at the time of transition. At this time, it must momentarily coincide with in its lower portion pointing in the same direction. This is precisely the sit- S2 and another MOTS ( 1a, shown in Section IV C) in its uation of inner as it approaches ttouch and that of inner∗∗ S innerS ∗∗ S upper portion. If true, this scenario would lead to a self- as t tcusp . At the respective times of cusp formation, the two→ portions of the MOTS separated by the cusp are intersection forming in this world tube. individually each a smooth MOTS. In the case of , The annihilation of with ∗ resolves the previously Sinner S2 S2 this is just and , while for ∗∗ it is and another unknown fate of this individual apparent horizon, al- S1 S2 Sinner S2 MOTS, , discussed in Section IV C. though the fate of the full world tube (including ∗ and S1c S2 10

t = 2.5 M t = 2.5 M

1 0.5 ∗ S S0 ∗ ∗∗ S1 S0 0.5 1∗∗ S

0.0 0.0 z/M z/M

0.5 0.5 − −

0.5 0.0 0.5 0.5 0.0 0.5 − − x/M x/M

∗ ∗∗ ∗ ∗∗ FIG. 6. Examples of the MOTSs S1, S1 , and S1 at simulation FIG. 7. Examples of the MOTSs S0 and S0 at simulation time t = 2.5M. The light grey curve shows S2 for reference. time t = 2.5M. The light grey curves show S1 and S2 for reference.

∗ 2∗∗) is not numerically resolved. We shall defer further 4. The world tube of S0 Sdiscussion of possible scenarios to Section VII.

The last world tube is traced out by the pair ( 0∗, 0∗∗), ∗ S S S0 which bifurcates at tbifurcate 0.888M. These do not con- nect smoothly to any of the≈ above MOTSs. FIG. 7 shows ∗ 3. The world tube(s) of S1 and S1 the MOTSs at a time t = 2.5M. None of the MOTSs along this world tube contain any puncture in their inte-

∗ rior. At tS1 0.756M we find the formation of the bifurcate ≈ Based on the cases of cusp formation and self- pair ( ∗, ∗∗). These enclose the puncture inside but S1 S1 S1 intersections mentioned above, we speculate here on the do not contain the puncture of 2. FIG. 6 shows this pair future of ∗∗ beyond the point where we are able to locate along with at a time t = 2.5SM. S0 S1 it reliably: We already mentioned the cusp formations of A merger and annihilation of 1 with ∗ analogous to ∗∗ and ∗∗ in the previous two subsections. In these two S S1 S1 S2 that of 2 with 2∗ could not be seen in this simulation. It cases, either 1 or 2 coincides with 1∗∗ or 2∗∗ on one of is unclearS if thisS lack of an annihilation is a) purely due to their portions,S respectively.S We alsoS notedS that in both the smaller 1 being closer in coordinates to one of the cases the two remainders, 2a and 1a, respectively, are punctures, whichS increases the effect of the slow-down themselves MOTSs discussedS in SectionS IV C. These lat- discussed above and in [58], or if b) has a qualita- ter two MOTSs do not contain any puncture. They may, S1 tively different behavior than 2 and possibly does not however, at some point touch and start to intersect, just annihilate at all. One option toS pursue the first possibil- like and do at t . At the time when they touch, S1 S2 touch ity is to choose initial data for a simulation which reduces we propose that 0∗∗ coincides with their union, has a the aforementioned slow-down effects. Appendix B shows cusp at this time,S and forms a self-intersection immedi- results for such a simulation, which suggests that the lack ately afterwards. We are unfortunately not able to resolve of annihilation could indeed be due to case a) even in the this idea at this point since we lose numerical accuracy present simulation. near the punctures before this can be observed. Lastly, we would like to point out that, in a very similar way to 2∗∗, we find the inner branch 1∗∗ approaching S S 5. Comparison with the extreme mass ratio merger case one of the individual apparent horizons, this time 1, from the outside. In this case, however, we were ableS to resolve the formation of a cusp as 1∗∗ coincides with It is illuminating to compare these sequences of ( is discussed in Section IVS C). As expected, MOTSs to FIG. 16 in [34]. That paper studied marginally S1 ∪ S2a S2a ∗∗ subsequently self-intersects. This supports again the outer trapped open surfaces (MOTOSs) in Schwarzschild S1 above expectation that this also happens for ∗∗. spacetimes with the argument that they could be used S2 11 to model dynamical apparent horizon and other MOTT be found at later times. The same qualitative behavior evolutions during an extreme mass ratio merger. is found inside . Note that the self-intersecting MOTS S2 That paper did not include true dynamics, but it was 1c later merges with inner∗∗ at the time the latter forms argued that it should be possible to assemble the var- theS cusp at t 4.32MS . FIG. 10 shows this formation ious MOTOSs found in Schwarzschild into a sequence of a cusp and subsequent≈ self-intersection. Similarly, S2a describing the full dynamics of the merger. The authors merges with 1∗∗ as it forms its cusp at about t 4.9M. proposed one such evolution, but what they didn’t con- Another observationS is that these interior MOTSs≈ can- 5 sider is that many of the steps could be happening simul- not “escape” their enclosing 1 or 2, respectively, while taneously. Instead of creations and annihilations, they ex- the latter exist. This is againS easilyS explained by the pected a single continuous sequence of MOTOSs evolving maximum principle for MOTSs given in Section 3.2 of in time. If one reinterprets the sequence in their FIG. 16 Ref. [22], since any common point with a common normal as a marginally outer trapped open tube weaving back- direction would imply the MOTSs necessarily coincide. wards and forwards in time, then the proposed picture Interestingly, 1d and 2d are MOTSs with toroidal becomes much closer to that of FIG. 4 in the current topology, i.e. theirS EulerS characteristic is χ = 0 in con- paper, though still with some mistakes. trast to all other MOTSs which have χ = 2. We numeri- + It is intriguing that that simple model of an ex- cally verified that their shear σAB does not vanish, which treme mass ratio merger can reproduce cusp formation immediately implies that any first order spacelike out- which subsequently evolves into self-intersections. How- ward deformation does not lead to an untrapped surface ever since [34] works with an exact solution, we can (see the remark at the end of Section III A of Ref. [17]). push further and see that this process may repeat in- This is compatible with a negative principal eigenvalue of definitely with surfaces involving more and more self- the stability operator, which we find as well. 1d is shown intersections. This then is extra evidence that the dou- in the last panel of FIG. 8 and closely followsS the loop of ble self-intersection of FIG. 4 at t 4.9667M is likely the self-intersecting 1c. Just like 1a,b,c and 2a,b,c, the ≈ S S S just the beginning of a sequence that may continue in- toroidal MOTSs merge with 1 and 2, respectively, as definitely with each n-times self-intersecting MOTOS ul- t 0. S S timately being annihilating with an (n + 1)-times self- →From their behavior for t 0, one can immediately → intersecting MOTS which in turn was pair-created with deduce that the area of 1a,b and 1d becomes twice the a (soon-to-be) (n + 2)-times self-intersecting MOTOS. area of while hasS three timesS this area in the limit S1 S1c t 0. Analogous arguments hold for 2a,b,c,d. FIG. 11 confirms→ this expectation and shows theS evolution of the C. MOTSs inside S1 and S2 various areas as a function of time. All these interior MOTSs get very close in coordinates None of the MOTSs described thus far are located fully to one of the punctures and we hence lose most of them due to numerical problems before the end of our simula- in the interior of either 1 or 2. As discussed in the S S tion. This happens earlier for than for first paper, we were in fact not able to locate such an S1a,b,c,d S2a,b,c,d interior MOTS in the Brill-Lindquist initial data. From as the former are smaller in coordinates than the latter. the discussion in the previous Section IV B, however, one would expect that at any point where a cusp forms in a MOTS and separates it into two smooth parts, these two D. MOTSs seen from the other asymptotic ends parts are individually separate MOTSs with their own time development to the future and past. This holds for Figures 8 and 9 show curves i and i which we did S1 S2 example for the upper portion of ∗∗ at the time it not yet comment on. These curves belong to marginally Sinner forms a cusp shortly before the last panel of FIG. 4. This inner trapped surfaces (MITSs), where Θ = 0 with no − portion is fully contained inside 1, i.e. at some point condition on Θ . Equivalently, these surfaces are MOTSs S + a MOTS must have formed inside 1 evolving into this with the notion of inward and outward reversed. These S self-intersecting shape. A similar argument holds for the MITSs are interesting for two reasons. lower portion of ∗∗. S1 First, the MITSs provide a means for generalizing a A search for MOTSs in the interior regions of 1 and 2 feature visible in the time-symmetric case (see paper I), S S at a time t > 0 has indeed been successful. FIG. 8 shows where the MOTSs with sharp turns seem to transition the four MOTSs we could locate inside 1 at a simulation between portions staying close to one of the other MOTSs S time t = 1.5M. From FIG. 9 it can be seen that these with fewer or no turns. However, at these turns the notion MOTSs pinch off and merge with 1 as t 0, i.e. they do of inside and outside may switch. This is irrelevant in S → not exist separately in the initial data and thus can only time-symmetry as in that case Θ+ = 0 Θ = 0, but it becomes important during the simulation⇐⇒ − where time-symmetry is lost. As can be seen in FIG. 8, the i MOTSs 1a,b,c,d run close to 1 on an extended portion, 5 We are confident about this as two of them are also authors on consistentS with the notions ofS inside and outside there. the current paper! FIG. 12 visualizes this idea by showing the spatial part of 12

0.6 1 1b 1c 1d Si S S S S1 1a 0.5 S

z/M 0.4

0.3 t = 1.5 M

0.1 0.0 0.1 0.1 0.0 0.1 0.1 0.0 0.1 0.1 0.0 0.1 − − − − x/M x/M x/M x/M

FIG. 8. The MOTSs S1a,b,c,d in the interior of S1 at a simulation time t = 1.5M. S1a (first panel) and S1b (second panel) do not enclose the puncture inside S1, while S1c (third panel) is self-intersecting and does enclose this puncture. The last panel shows i a MOTS S1d of toroidal topology. We also show a marginally inner trapped surface S1 as thin solid line, which is discussed in Section IV D.

1 1 0.5 S S S1a 0.5 S1a S1b S1b S1c S1c S1d S1d S2 S2 S2a S2a 2b 2b 0.0 S 0.0 S S2c S2c 2d 2d

z/M Si z/M Si 1 1 Si Si S2 S2

0.5 0.5 − −

t = 0.2 M t = 1 M

0.5 0.0 0.5 0.5 0.0 0.5 − − x/M x/M

FIG. 9. MOTSs in the interior of the individual MOTSs S1 and S2 at the two different times t = 0.2M (left panel) and t = 1M (right panel). Any MOTS outside or partially outside the individual ones is not shown here. The MOTSs S1a,b,c,d are shown i i individually in FIG. 8. We furthermore show the MITSs S1, S2, which are discussed in Section IV D. the respective null normals with vanishing expansion, i.e. observers far away from any black hole. An observer near + of ` for the MOTSs and `− for the MITS. We find that x1 or x2 will see a quite different picture of what “we” (i.e. all MOTSs in the interior of 1 fully lie in the annular observers for whom x is considered outside) see as S i → ∞ region between 1 and the MITS 1. The MOTSs in the a binary black hole merger. In particular, at t = 0 where interior of behaveS analogously.S are not only MOTSs but also MITSs, an observer S2 S1,2 near, say, x1 interprets 1 as common apparent horizon The second reason to consider MITSs is that the two enclosing both punctures.S FIG. 13 shows that the area of punctures in the Brill-Lindquist initial data belong to two these MITSs is monotonically increasing. Furthermore, different asymptotically flat ends of the slice Σ. In fact, considered as MOTSs seen from the respective asymp- these data contain three ends with x representing i i → ∞ totic region, 1 and 2 are strictly stable and, in fact, the end we commonly choose to be in the “outward” di- the outermostScommonS MOTSs representing a perturbed rection. The ends x x1 and x x2, where x1,2 are the Schwarzschild black hole. While many more MITSs are coordinates of the two→ punctures,→ contain equally valid 13

∗∗ ∗∗ Sinner Sinner S1 S1 S1c S1c 0.0 S2 0.0 S2 z/M z/M

0.5 0.005 0.5 0.01 − − − − 0.02 0.010 − − 0.03 t 4.25052 M t = 4.45 M − 0.01 0.00 0.01 0.02 0.00 0.02 ≈ − − 0.5 0.0 0.5 0.5 0.0 0.5 − − x/M x/M

∗∗ FIG. 10. Cusp formation with subsequent self-intersection in the evolution of Sinner. The left panel shows a time before cusp formation and the right panel a time after. Note that the cusp forms at the time when S1c and S2 touch. The interior self- intersecting MOTSs are very similar to those see in pure Schwarzschild in [34].

30 S1 S2 S1a 80 S2a S1b S2b 1c 2c

2 20 S 2 60 S S1d S2d /M /M 40 area 10 area 20 outer bifurcate outer bifurcate 0 t 0 t 0 2 4 6 0 2 4 6 t/M t/M

FIG. 11. Evolution of the area of the MOTSs S1a,b,c inside S1 (top panel) and of S2a,b,c inside S2 (bottom panel). The areas of S1 and S2 are shown for reference. Note that the smooth annihilation of S2 is not visible since we do not include the area of ∗ the connecting MOTS S2 . likely present in our simulations, we shall leave a more balances several effects resulting from the interplay of exhaustive study of these surfaces for future work. the employed slicing and the numerical setup: A larger value for d makes the individual apparent horizons 1,2 slow down in evolution before they start to intersectS [58] E. Remarks on initial distance and mass ratio and thus prevents us from observing the annihilations of ( , ∗) and ( , ∗ ). A smaller value leads to S2 S2 Sinner Sinner common MOTSs outer and inner already being present Most of the discussed numerical results are obtained in the initial data.S As shownS in paper I, many more com- from simulations starting with a single physical configu- mon MOTSs exist in these kinds of setups and hence at ration, namely Brill-Lindquist initial data consisting of a least some of the various bifurcations do not exist in the conformally flat three-metric with conformal factor (20) simulation. where the bare masses are m1 = 1/3 and m2 = 2/3 and the distance parameter is chosen as d = 0.9. This choice Similar considerations led to the chosen mass ratio of 14

find that this requires a very small initial distance d. 0.65 Thus, while we are indeed able to resolve some of the S1 1a previously presented features for many choices of initial 0.60 Si S1 data, the particular choice we made combines most of 0.55 these in one simulation. Furthermore, the reasons we can- not resolve the full set of these features for other mass 0.50 ratios or initial distances are well understood and due to numerical issues. We therefore have no clear indica- 0.45

z/M tion that the observed behavior should indeed be specific to our choice of these parameters. The exception is a 0.40 possible qualitative change in case of a large mass ra- 2 0.35 tio. For large q, we find that the time tannihilateS where 2 annihilates with 2∗ occurs earlier and much closer to theS time t whenS and start to intersect. Recall 0.30 touch S1 S2 t = 1 M that t is precisely the time when the union touch S1 ∪ S2 0.25 coincides with inner and that this provides the connec- 0.2 0.1 0.0 0.1 0.2 tion between S and [26, 27]. We have verified for − − 1,2 outer x/M S S 2 d = 0.9 that ttouch < tannihilateS up to q = 14. However, if one finds for even larger q that 2 annihilates with before it intersects with , oneS may still be able to FIG. 12. Spatial outward normal directions on S1 and one of 2∗ 1 findS a sequence of MOTSs connectingS with . In the interior MOTSs S1a as well as the inward normal direction S1,2 Souter on the MITS Si . this case, the connection may occur after has turned 1 S2 around in time, i.e. one may see a merger ∗ with S1 ∪ S2 ∗ . Sinner 4 10− 2 V. STABILITY 5

/M 10− ) 0

A 6 10− The various bifurcations and annihilations described − i thus far can also be understood with the help of the A

( 7 1 10− Si MOTS stability operator. As we shall see, this not only S2 gives strong numerical support for our claims of smooth 8 10− bifurcations and annihilations, it also provides a useful 0 1 2 3 characterization of the respective two branches in terms t/M of the eigenvalues of this operator. Let LΣ be the stability operator (11). As shown in i i FIG. 13. Evolution of the area of the MITSs S1 and S2. We Proposition 5.1 of [31], the vanishing of the principal plot the difference A−A0 on a logarithmic scale to emphasize eigenvalue of LΣ is closely related to bifurcations and the area change. Here A0 is the area of the MITS at t = 0 annihilations of a MOTS. The intuitive picture is that where it coincides with the area of S1 and S2, respectively. of a MOTT which is tangent to one of the slices Σ , H t∗ where t∗ is the time of bifurcation or annihilation. At this time, the cross section of is a MOTS with vanishing St∗ H q = 2. In particular, it is true that more unequal masses principal eigenvalue λ0 = 0. Essentially, the proposition keep the larger MOTS further away (in coordinates) proves, under suitable genericity conditions satisfied in all S2 from the puncture in its interior. As a consequence, it suf- our cases, that the existence of such a MOTS t∗ with fers less from the slow-down described in [58] and we can λ = 0 implies existence of a unique MOTT tangentS to 0 H observe that it annihilates with ∗ “earlier” with respect Σt and containing t . Hence, if we do find two MOTTs S2 ∗ S ∗ to simulation time. However, this comes at the cost of 2∗ connecting smoothly at t∗ with λ0 0 as t t∗, then we now getting too close (again, in coordinates) to the punc-S have a clear numerical indication→ for such→ a bifurcation ture inside the smaller MOTS and thus we were not or annihilation. This is precisely the case for the bifur- S1 able to resolve the full world tube of ∗ and show that it cation of the pair ( outer, inner) and the annihilation of S2 S S turns around in time at both of its ends. This also means the pair ( , ∗) shown in Figures 14 and 15. However, S2 S2 that the prospect of resolving the fate of 1 crucially for the other pairs of MOTSs, we in fact find that instead depends on keeping it as far away from itsS puncture as of the principal eigenvalue, it is one of the higher eigen- possible. Without modifying the used gauge conditions, values which tends to zero at bifurcation or annihilation which we do not try to do in the present work, this can time. only be achieved with more equal masses. Our attempts An interesting observation we can make here is that to do so are summarized in Appendix B. However, we the number of negative eigenvalues necessarily changes 15

λ0 for outer λ2 for inner∗∗ TABLE I. Number of negative eigenvalues of L for the S S Σ λ0 for inner λ for ∗ S 1 S2 MOTSs along the different MOTTs. The arrows indicate λ1 for 1∗ λ for ∗∗ S 1 S2 when we have found a smooth connection between the re- λ1 for 1∗∗ λ for ∗ S 2 S0 spective world tubes, while the arrow in parentheses indi- λ2 for ∗ λ for ∗∗ Sinner 2 S0 cates a suspected smooth transition which we could not re- 0.2 0 solve numerically. Shown is N−, i.e. the number of eigenvalues λ < 0 as well as N for the number of all negative eigen- 0.1 l,m=0 − values. This latter value changes for some of the MOTTs, in which case we list all the occurring cases (not in order of

2 0.0 appearance). M l

λ 0.1 ∗ ∗∗ − MOTT H1 (→) H1 → H1 N 0 0 1 2 0.2 −

− outer bifurcate N 0 1 2,4

t − ∗ ∗∗ 0.3 MOTT H2 → H2 → H2 − 0.70 0.75 0.80 0.85 0.90 0.95 0 N− 0 1 2 t/M N− 0 1 2,4 ∗ ∗∗ MOTT Houter → Hinner → Hinner → Hinner FIG. 14. Eigenvalues of LΣ for the ten MOTSs participating in 0 N− 0 1 2 3 the five bifurcations. For each MOTS, we show the respective N− 0 1,3 2,4 3,5,7,9 eigenvalue which tends to zero as t → tbifurcate. ∗ ∗∗ MOTT H0 → H0 0 N− 2 3 λ0 for 2 λ1 for inner N 2 3,5,7 S S − λ0 for 2∗ λ1 for inner∗ 0.3 S S MOTT H1a H1b H1c H1d N 0 1 1 2 2 0.2 − N− 3 1,3 4,6 4,6

0.1 MOTT H2a H2b H2c H2d 2 0

M 0.0 N− 1 1 2 2 l

λ N− 1,3 1 4 4 0.1 −

0.2 2 S annihilate inner annihilate

− t t along each MOTT during the evolution and FIG. 16 0.3 0 − 5.0 5.2 5.4 5.6 shows all MOTSs at two different times with N indi- − t/M cated by line thickness and color. In each instance where a MOTS transitions through a bifurcation or annihilation along the indicated sequences of MOTTs, one additional FIG. 15. Eigenvalues of LΣ for the two pairs of MOTSs ∗ ∗ negative eigenvalue of the m = 0 mode appears. (S2, S2 ) and (Sinner, Sinner) close to the time when they anni- Another observation is related to the non-smooth hilate. MOTS mergers where two MOTSs touch at one point and coincide at this time with a MOTS having a cusp. We were able to explicitly resolve three of these mergers as we follow a smooth MOTT across such a bifurcation numerically, namely = , = or annihilation. Three of the MOTSs, namely , , 1 2 inner 1c 2 inner∗∗ outer 1 and = .S Based∪ S on ourS results,S ∪ we S expectS at and , possess a positive principle eigenvalue,S i.e. theyS 1 2a 1∗∗ 2 leastS two∪ Smore suchS mergers, which we could not resolve are strictlyS stable and thus act as barrier for trapped and for numerical reasons. These are = and untrapped surfaces in a neighbourhood. By our terminol- 1a 2 2∗∗ = . For all these casesS where∪ S S = , ogy, they are apparent horizons and the world tubes they 1a 2a 0∗∗ 0 00 weS find,∪ S withS obvious notation, that S ∪ S S trace out are dynamical apparent horizons. The above properties are what one would usually expect from hori- N 0 + N 0 0 + 1 = N 0 00 . (21) zons associated with black holes. All other MOTSs we − − − found possess one or more negative eigenvalues. At any Note that N 0 is constant along each individual MOTT, given time, all MOTSs with a negative eigenvalue are con- even when cusps− and self-intersections form, as they do tained in the interior of one of the strictly stable ones, i.e. for inner, inner∗∗ and 1∗∗. However, in several instances, , , or . To present our results more systemati- weS find thatS eigenvaluesS of the higher angular modes S1 S2 Souter cally, let N be the number of eigenvalues λl,m < 0 and (m = 0) do cross zero on perfectly smooth portions of 0 − 6 N the number of eigenvalues λl,m=0 < 0. Table I lists the MOTT. Due to the axisymmetry and absence of spin the− various values of N and N 0 we find for the MOTSs in our simulation, we have a m degeneracy in the spec- − − ± 16

0.5 0.5

0.0 0.0 z/M z/M

0.5 0.5 − −

1.0 1.0 − t = 1 M − t = 3 M

0.5 0.0 0.5 0.5 0.0 0.5 − − x/M x/M

0 FIG. 16. MOTSs at two different times of the simulation. The line thickness and color reflects N−, i.e. the number of negative stability eigenvalues of the m = 0 mode. The three dark lines in both panels are Souter (common), S1 (upper) and S2 (lower). 0 Lighter colors show MOTSs with larger N−. Note that none of the MOTSs extends beyond Souter.

trum, whence all zero crossings of eigenvalues λl,m=0 hap- 6 λ1, 1 for 2∗ λ1, 1 for 1∗ pen in multiples of 2. Two examples of such cases are ± S ± S λ1, 1 for 2∗∗ λ1, 1 for 1∗∗ depicted in FIG. 17, which shows that the two degener- ± S ± S ate eigenvalues λ of and do cross zero during 1, 1 2∗∗ 1∗∗ 0.4 their evolution. This± crossingS happensS twice for the lat- ter case. Taking invertibility of LΣ as indicator for the 2 existence of a smooth evolution of a MOTS , we here 0.2 M have explicit counterexamples showing that theS converse l,m of this statement is not true. In other words, invertibility λ 0.0 of LΣ is only a sufficient but not a necessary condition for a smooth evolution. 0.2 − 1 2 3 4 5 6 VI. SIGNATURE AND INGOING EXPANSION t/M

As discussed in Section II B 2, a strictly stable MOTS, FIG. 17. Examples of eigenvalues with m 6= 0 crossing zero on ∗∗ λ > 0, belongs to a dynamical apparent horizon that has the smoothly evolving portions of S2 (thick dotted line) and 0 ∗∗ S1 (thin dotted line). The smoothly connecting respective spacelike signature at that point (cf. [18, 19]). Together ∗ ∗ curves for S2 and S1 are added here for reference. Note that with Θ 0, the area will be non-decreasing. Since we ∗∗ ∗∗ − both S and S have a principal eigenvalue λ < 0 and see in FIG.≤ 3 that many of the MOTSs have a decreasing 2 1 0 λ < 0, which are not shown. or non-monotonic area evolution, we expect that those 1,0 with λ0 < 0 cannot have both non-positive ingoing ex- pansion Θ 0 and evolve along a spacelike MOTT. Examples− are≤ shown in Figures 18 and 19 where we see complicated signature changes and indefiniteness of the sign of Θ along the world tubes inner and inner∗ . In these figures,− time increases upwardsH and the signatureH or sign of Θ is shown as color on the world tubes. FIG. 20 and inner∗ connect smoothly). A qualitatively very sim- shows a close-up− of the sign of Θ at the top end where ilar behaviorH of these quantities is found for all MOTTs − and ∗ annihilate (or equivalently where except , and . These are purely spacelike and Sinner Sinner Hinner Houter H1 H2 17

∗ Hinner Hinner

∗ FIG. 18. Signature of Hinner (left panel) and Hinner (right panel). In these tube plots, time goes upwards and z-values increase to the left. These two MOTTs smoothly connect at their top ends but not at the bottom as they connect with different world tubes there. The bottom end of the right panel corresponds to the bottom-left panel of FIG. 4.

∗ Hinner Hinner

∗ FIG. 19. Sign of the ingoing expansion Θ− of Sinner (left panel) and Sinner (right panel) plotted on their respective MOTTs ∗ Hinner and Hinner in the same perspective as in FIG. 18. The portion with Θ− > 0 of Sinner (left panel) smoothly connects to a corresponding portion on Souter, which quickly vanishes after ∆t ≈ 0.04M (not shown here). have Θ 0 at most times.6 even when changes direction in time and has non- A result− ≤ of Bousso and Engelhardt [38, 60] shows that spacelike segments,H it will have a monotonic area evolu- tion provided several conditions hold on . One of these is that Θ 0, which we have alreadyH seen to not be satisfied for− ≤ most MOTTs we found in this simulation. 6 We do find a very short duration of ∆t 0.04M after touter At this point, one could immediately conclude that this ≈ bifurcate where outer has a small portion with Θ− > 0 close to its equa- S result is not applicable to most of our cases and hence tor. This was also found in [29]. A similar portion with Θ− > 0 finding non-monotonic area evolutions is not in tension is found on 2 shortly (∆t 0.02M) before it annihilates with ∗. Both of theseS portions smoothly≈ connect with corresponding with any theoretical expectation. While certainly true, S2 portions on the MOTSs they connect to ( inner for outer and we still think it is worthwhile to show which of the other ∗ S S for 2). assumptions made in the proof are violated, not least S2 S 18

∗ Hinner Hinner

FIG. 20. Close-up of the top ends of the MOTTs shown in FIG. 19. The two world tubes smoothly join at this top end and the portions with Θ− < 0 connect across this merger. because they were believed to be unrestrictive. VII. CONCLUSIONS To state the relevant ones here, we go back to the evo- α lution vector defined as tangent to and orthogonal In the present second paper of this two-part study, to each MOTSV foliating . As before,H we fix its scaling S H the new generalized shooting method introduced in the by requiring t = 1. Since the null normals `± span the first paper was used successfully to uncover new MOTSs two-dimensionalLV space of normals to , we can write S forming during the head-on merger of two non-spinning α α α black holes, including MOTSs of toroidal topology. This = b`+ + c` . (22) V − has vastly increased the number and variety of known MOTSs and also shows that they can have a much richer As = 2bc, the coefficients b and c are related to V·V − range of geometrical properties than had been previously the signature of , i.e. is spacelike, timelike, or null expected. when bc < 0, bc >H0, or bcH= 0, respectively. The proof in However this increase has also highlighted the rarity [60] now requires in addition to genericity assumptions and significance of stable MOTSs. Only three out of all satisfied in all our cases, the following: the multitude that we have observed are stable – even (i) Every inextendible portion of definite sign of c is strictly stable except for the points of annihilation or bi- entirely timelike or contains at least one full MOTS. furcation – and trace out spacelike world tubes. These are exactly the MOTSs that one would naturally asso- (ii) Every MOTS in splits a Cauchy slice Σ that it ciate with black hole boundaries: the two individual black H holes ( and ) and the final remnant ( ). These is contained in into two disjoint portions. S1 S2 Souter world tubes 1, 2 and outer are the dynamical appar- Then, without restrictions on Θ , it is proven that c ent horizons.H OneH mayH ask whether additional strictly cannot change sign on . On all− but the three strictly stable MOTSs may exist if we do not restrict ourselves stable MOTTs, however,H we find that at least one of the to only axisymmetric surfaces. Fortunately, this has been above conditions is violated and that, in fact, both b and ruled out by Theorem 8.1 in Ref. [19]. This unambiguous c do change sign. natural choice of physically relevant horizons provides In cases of self-intersections, it is clearly condition (ii) an additional numerical indication that dynamical ap- that does not hold. But even for MOTSs that do not self- parent horizons are well-behaved objects suitable to de- intersect (or on portions of their world tubes on which scribe the highly dynamical and non-perturbative regime they do not self-intersect), we find that condition (i) is during such a merger. violated. For the case of inner, this was discussed in That said, the apparent horizons cannot forever remain great detail in [28]. With theS annihilation of with aloof from the common herd. appears out of a bi- Sinner Souter ∗ , we are now able to extend these results to later furcation with the unstable while and (likely) Sinner Sinner S2 times. In particular, shortly before inner vanishes, its 1 are ultimately annihilated in mergers with other un- world tube becomes purely spacelikeS with c < 0 stableS MOTSs. The additional MOTTs then significantly Hinner on full MOTSs. At these times, however, inner has self- increase our understanding of the interior structure form- intersections, i.e. this presents an explicitS example that ing shortly after the common apparent horizon appears. (i) is not a sufficient condition. This structure shares certain features with previous spec- 19 ulations that the merger might, in fact, be described by a single smooth MOTT weaving back and forth in time [15, 21–23]. What we find is significantly more compli- 60 2∗ cated, but we do see this back and forth in time. Since S 2 we lose the MOTSs only for (well understood) numerical 50 /M reasons, it seems plausible that they continue to weave ∗∗ S2 back and forth, possibly forming more and more self- area 40 intersections. We also find that all world tubes seem to be 2 connected7 However, since no MOTS crosses a puncture, S some connections are not smooth and instead happen via 30 cusp-formation with subsequent self-intersections. 0.1 0.0 0.1 0.2 0.3 − Together, these observations motivate the following Θ+ suggestion. If one assumes that (i) MOTSs cannot cross punctures, (ii) MOTSs appear and disappear only as FIG. 21. A family of surfaces of constant expansion shown pairs, (iii) all MOTTs connect in some form and (iv) the in the plane of area and expansion. Each point on the solid individual apparent horizons vanish at some point during line corresponds to one surface of this family. Whenever the a merger, then many of the observed behaviors seem to be curve crosses Θ+ = 0, the surface is a MOTS. This plot shows inevitable. It seems conceivable that (i) holds and the re- a family constructed from S2 at simulation time t = 2.5M, ∗ ∗∗ sults of Andersson et al. [31] points toward (ii). While our which connects S2 with S2 and S2 . results certainly do not imply (iii) and (iv), they might still provide an incentive for further investigation in this direction. Appendix A: Families of surfaces of constant expansion The present results for the axisymmetric head-on col- lision of two black holes also have implications for the A MOTS is defined as a closed surface with zero generic case where inspiraling black holes coalesce with- S out any symmetry. We now know which kinds of surfaces outward expansion, Θ+ = 0. However, one may also try to look for surfaces with Θ = c, where c = const, a generalized MOTS finder must be able to resolve and Sc + which surfaces to look for. One might explore whether in a neighbourhood of any given MOTS. These surfaces a generalization of the shooting method could be used can be used to help locating common MOTSs as early as to approximate near-axisymmetric MOTSs to be used as possible during the simulation by tracking common sur- initial guesses for such a finder. An important question faces c>0, which are seen to exist prior to the formation of a commonS MOTS = [61, 62]. We will show will be whether it is still only three MOTTs which are S Sc=0 stable and hence dynamical apparent horizons. here another application of such surfaces, which turned out to be helpful in resolving the various bifurcations and annihilations. This is related to and motivated by the ob- servation made in [24] that a family of such surfaces may connect one MOTS with another. See Figure 13 in [24] ACKNOWLEDGMENTS and its discussion for details. We start with a MOTS found in some particular S Cauchy slice Σt at simulation time t = t1. This surface is We would like to express our gratitude to Graham Cox, then tracked through the simulation forwards and back- Jose Luis Jaramillo, Badri Krishnan, Hari Kunduri and wards in time. If in either direction, the MOTS is lost the members of the Memorial University Gravity Journal and cannot be located anymore, say at t > t , then we Club for valuable discussions and suggestions. 2 1 choose a time t . t2 and construct a family of surfaces IB was supported by the Natural Science and Engi- starting with = . Note that just as there may Sc S0 S neering Research Council of Canada Discovery Grant be multiple MOTSs 0 in any given Cauchy slice, the 2018-0473. The work of RAH was supported by the Natu- surfaces for c = 0S will also not be unique in general. Sc 6 ral Science and Engineering Research Council of Canada However, when varying c in small steps c c0 = c + ε, → through the Banting Postdoctoral Fellowship program one can look for c0 in the vicinity of c by taking c and also by AOARD Grant FA2386-19-1-4077. as initial guess. AsS an example, FIG. 21S shows suchS a family in terms of the expansion and area of the c. In this case, we start from at a time t = 2.5M andS we S2 are able to reliably locate 2∗ and 2∗∗. FIG. 22 shows the shapes of these surfaces ofS constantS expansion and how 7 ∗ they connect 2 with 2∗ (left panel) and 2∗ with 2∗∗ The likely annihilation of 1 with 1 is discussed in Sec- S S S S tion IV B 3. Similarly, basedS on our resultsS we expect mergers (right panel). ∗∗ ∗∗ of with 2 1a and with 1a 2a to happen via A slight complication is encountered whenever Θ = c S2 S ∪ S S0 S ∪ S + cusp-formation. has a local extremum. This happens twice in FIG. 21. In 20

0.50 0.00 0.5 0.30 0.02 0.25 − 2∗ S 0.25 2∗ 0.04 0.0 S − 2 0.00 S 2∗∗ 0.06 0.20 S − 0.5 0.25 0.08 − −

z/M − 0.15 z/M 0.10 0.50 − − 1.0 0.10 0.12 − − 0.75 − 0.14 0.05 − 1.5 t = 2.5 M MOTS 1.00 t = 2.5 M MOTS 0.16 − 0.00 − − 1.0 0.5 0.0 0.5 1.0 Θ+ 0.5 0.0 0.5 Θ+ − − − x/M x/M

FIG. 22. Shapes of the family of constant expansion surfaces of FIG. 21. Shown is the subset of surfaces connecting S2 with ∗ ∗ ∗∗ S2 (left panel) and the subset connecting S2 with S2 (right panel). The color indicates the value of Θ+ = c while the MOTSs are shown as solid red lines. these cases we cannot vary c but instead we can take a individual horizons annihilate. We take this as suggest- small step in area by prescribing the area instead of the ing that the lack of annihilation in the main configuration expansion for this step (see e.g. [61]). Alternatively, one is purely due to the numerical setup and caused by the can anticipate the shape change by extrapolating from MOTS moving too close to the puncture (in the numeri- the previous steps to construct an initial guess surface cal coordinates) as also analyzed in [58]. which overcomes the extremum of c. By using more equal masses, the shape of the smaller Once another MOTS is found this way, it can itself be apparent horizon remains larger in coordinates for a tracked forwards and backwards in time to resolve the longer time, and a smaller initial distance parameter re- possible annihilation or bifurcation. duces the simulation time when the annihilation takes place. FIG. 23 shows the area and principal stability eigenvalue for a simulation with a mass ratio of q = 1.05 and distance parameter d = 0.4. We see that both in-

∗ dividual apparent horizons seem to annihilate, first the Appendix B: Annihilation of S1 with S1 larger one with ∗ and shortly after the smaller one S2 S2 1 with 1∗. Despite the above choice of parameters for The annihilation of the larger individual MOTS thisS simulation,S the MOTSs still approach the punctures S2 with 2∗ was found with high accuracy and is discussed in (in coordinates) before they vanish. We found the anni- the mainS text. Here, the goal is to show that it is plausible hilations of both apparent horizons in simulations with that the smaller individual MOTS also annihilates, in spatial resolutions of 1/∆x = 480, 720, and 960. How- S1 this case with ∗, and that it does not have a qualita- ever, the precise behavior of these curves in the final time S1 tively different behavior than 2 in this regard. To this span after t 5M varies between the resolutions. While end, we perform a simulationS with different initial con- not as convincing∼ as the remaining results we present, a ditions than those in the main text and show that both merger of and ∗ seems at least plausible. S1 S1

[1] Ivan Booth, Robie A. Hennigar, and Daniel Pook-Kolb, ing Run,” Phys. Rev. X6, 041015 (2016), [erratum: Phys. “Ultimate fate of apparent horizons during a binary black Rev.X8,no.3,039903(2018)], arXiv:1606.04856 [gr-qc]. hole merger I: locating and understanding axisymmetric [4] Alexander H. Nitz, Collin Capano, Alex B. Nielsen, marginally outer trapped surfaces,” (2021). Steven Reyes, Rebecca White, Duncan A. Brown, and [2] B.P. Abbott et al. (LIGO Scientific, Virgo), “GWTC- Badri Krishnan, “1-OGC: The first open gravitational- 1: A Gravitational-Wave Transient Catalog of Compact wave catalog of binary mergers from analysis of public Binary Mergers Observed by LIGO and Virgo during Advanced LIGO data,” Astrophys. J. 872, 195 (2019), the First and Second Observing Runs,” Phys. Rev. X arXiv:1811.01921 [gr-qc]. 9, 031040 (2019), arXiv:1811.12907 [astro-ph.HE]. [5] Alexander H. Nitz, Thomas Dent, Gareth S. Davies, [3] B. P. Abbott et al. (LIGO Scientific, Virgo), “Binary Sumit Kumar, Collin D. Capano, Ian Harry, Simone Black Hole Mergers in the first Advanced LIGO Observ- Mozzon, Laura Nuttall, Andrew Lundgren, and M´arton 21

0.5 80 S1 1∗ S 2 0.0 S 60 2∗

2 S 2 2∗ /M M 0.5 S 40 2 0 − S λ 1∗ area S S1 1.0 20 −

1.5 0 − 0 1 2 3 4 5 6 2 3 4 5 6 t/M t/M

FIG. 23. Evolution of the area (left panel) and principal stability eigenvalue (right panel) for a simulation with initial conditions q = m2/m1 = 1.05 and d = 0.4. We focus here on the individual apparent horizons of S1,2 and the MOTSs with which they ∗ ∗ annihilate, S1 and S2 , respectively. These plots show a simulation with resolution 1/∆x = 960. Note that the final part for t & 5M still suffers from the MOTSs being too close to the numerically underresolved puncture regions.

T´apai,“2-OGC: Open Gravitational-wave Catalog of bi- ativity, gravitation and relativistic field theories. Proceed- nary mergers from analysis of public Advanced LIGO ings, 9th Marcel Grossmann Meeting, MG’9, Rome, Italy, and Virgo data,” Astrophys. J. 891, 123 (2019), July 2-8, 2000. Pts. A-C (2000) pp. 568–580, arXiv:gr- arXiv:1910.05331 [astro-ph.HE]. qc/0008071 [gr-qc]. [6] Alexander H. Nitz, Thomas Dent, Gareth S. Davies, and [16] Badri Krishnan, “Fundamental properties and applica- Ian Harry, “A Search for Gravitational Waves from Bi- tions of quasi-local black hole horizons,” Class. Quant. nary Mergers with a Single Observatory,” Astrophys. J. Grav. 25, 114005 (2008), arXiv:0712.1575 [gr-qc]. 897, 2 (2020), arXiv:2004.10015 [astro-ph.HE]. [17] Abhay Ashtekar and Badri Krishnan, “Dynamical hori- [7] Tejaswi Venumadhav, Barak Zackay, Javier Roulet, zons and their properties,” Phys. Rev. D68, 104030 Liang Dai, and Matias Zaldarriaga, “New Binary Black (2003), arXiv:gr-qc/0308033. Hole Mergers in the Second Observing Run of Advanced [18] Lars Andersson, Marc Mars, and Walter Simon, “Lo- LIGO and Advanced Virgo,” (2019), arXiv:1904.07214 cal existence of dynamical and trapping horizons,” [astro-ph.HE]. Phys.Rev.Lett. 95, 111102 (2005), arXiv:gr-qc/0506013 [8] Barak Zackay, Tejaswi Venumadhav, Liang Dai, Javier [gr-qc]. Roulet, and Matias Zaldarriaga, “Highly spinning and [19] Lars Andersson, Marc Mars, and Walter Simon, “Stabil- aligned binary black hole merger in the Advanced LIGO ity of marginally outer trapped surfaces and existence of first observing run,” Phys. Rev. D 100, 023007 (2019), marginally outer trapped tubes,” Adv.Theor.Math.Phys. arXiv:1902.10331 [astro-ph.HE]. 12 (2008), arXiv:0704.2889 [gr-qc]. [9] Geraint Pratten et al., “Let’s twist again: computa- [20] Ishai Ben-Dov, “Penrose inequality and apparent hori- tionally efficient models for the dominant and sub- zons,” Phys. Rev. D 70, 124031 (2004). dominant harmonic modes of precessing binary black [21] Ivan Booth, Lionel Brits, Jose A. Gonzalez, and Chris holes,” (2020), arXiv:2004.06503 [gr-qc]. Van Den Broeck, “Marginally trapped tubes and dynam- [10] Vijay Varma, Scott E. Field, Mark A. Scheel, Jonathan ical horizons,” Class. Quant. Grav. 23, 413–440 (2006), Blackman, Davide Gerosa, Leo C. Stein, Lawrence E. arXiv:gr-qc/0506119. Kidder, and Harald P. Pfeiffer, “Surrogate models [22] P. M¨osta, L. Andersson, J. Metzger, B. Szil´agyi, for precessing binary black hole simulations with un- and J. Winicour, “The Merger of Small and Large equal masses,” Phys. Rev. Research. 1, 033015 (2019), Black Holes,” Class. Quant. Grav. 32, 235003 (2015), arXiv:1905.09300 [gr-qc]. arXiv:1501.05358 [gr-qc]. [11] S. W. Hawking and J. B. Hartle, “Energy and angu- [23] Anshu Gupta, Badri Krishnan, Alex Nielsen, and lar momentum flow into a black hole,” Commun. Math. Erik Schnetter, “Dynamics of marginally trapped sur- Phys. 27, 283–290 (1972). faces in a binary black hole merger: Growth and ap- [12] Abhay Ashtekar and Badri Krishnan, “Isolated and dy- proach to equilibrium,” Phys. Rev. D97, 084028 (2018), namical horizons and their applications,” Living Rev. arXiv:1801.07048 [gr-qc]. Rel. 7, 10 (2004), arXiv:gr-qc/0407042. [24] Daniel Pook-Kolb, Ofek Birnholtz, Badri Krishnan, and [13] Valerio Faraoni and Angus Prain, “Understanding dy- Erik Schnetter, “Existence and stability of marginally namical black hole apparent horizons,” Lecture Notes in trapped surfaces in black-hole spacetimes,” Phys. Rev. Physics 907, 1–199 (2015), arXiv:1511.07775 [gr-qc]. D 99, 064005 (2019). [14] Ivan Booth, “Black hole boundaries,” Can. J. Phys. 83, [25] Jonathan Thornburg, “Event and apparent horizon find- 1073–1099 (2005), arXiv:gr-qc/0508107. ers for 3+1 numerical relativity,” Living Rev. Rel. 10, 3 [15] Sean A. Hayward, “Black holes: New horizons,” in Recent (2007), arXiv:gr-qc/0512169. developments in theoretical and experimental general rel- [26] Daniel Pook-Kolb, Ofek Birnholtz, Badri Krishnan, and 22

Erik Schnetter, “Interior of a binary black hole merger,” CJ Carey, Ilhan˙ Polat, Yu Feng, Eric W. Moore, Jake Phys. Rev. Lett. 123, 171102 (2019). Vand erPlas, Denis Laxalde, Josef Perktold, Robert Cim- [27] Daniel Pook-Kolb, Ofek Birnholtz, Badri Krishnan, rman, Ian Henriksen, E. A. Quintero, Charles R Har- and Erik Schnetter, “Self-intersecting marginally outer ris, Anne M. Archibald, AntˆonioH. Ribeiro, Fabian Pe- trapped surfaces,” Phys. Rev. D 100, 084044 (2019). dregosa, Paul van Mulbregt, and SciPy 1.0 Contributors, [28] Daniel Pook-Kolb, Ofek Birnholtz, Jos´eLuis Jaramillo, “SciPy 1.0: Fundamental Algorithms for Scientific Com- Badri Krishnan, and Erik Schnetter, “Horizons in a bi- puting in Python,” Nature Methods 17, 261–272 (2020). nary black hole merger I: Geometry and area increase,” [45] S. van der Walt, S. C. Colbert, and G. Varoquaux, “The (2020), arXiv:2006.03939 [gr-qc]. NumPy array: A structure for efficient numerical com- [29] Daniel Pook-Kolb, Ofek Birnholtz, Jos´eLuis Jaramillo, putation,” Computing in Science Engineering 13, 22–30 Badri Krishnan, and Erik Schnetter, “Horizons in a bi- (2011). nary black hole merger II: Fluxes, multipole moments [46] Fredrik Johansson et al., mpmath: a Python library and stability,” (2020), arXiv:2006.03940 [gr-qc]. for arbitrary-precision floating-point arithmetic (version [30] Roger Penrose, “Gravitational collapse and space-time 1.1.0) (2018), http://mpmath.org/. singularities,” Phys. Rev. Lett. 14, 57–59 (1965). [47] Aaron Meurer, Christopher P Smith, Mateusz Paprocki, [31] Lars Andersson, Marc Mars, Jan Metzger, and Wal- Ondˇrej Cert´ık,ˇ Sergey B Kirpichev, Matthew Rocklin, ter Simon, “The Time evolution of marginally trapped AMiT Kumar, Sergiu Ivanov, Jason K Moore, Sartaj surfaces,” Class. Quant. Grav. 26, 085018 (2009), Singh, et al., “SymPy: symbolic computing in Python,” arXiv:0811.4721 [gr-qc]. PeerJ Computer Science 3, e103 (2017). [32] Abhay Ashtekar and Gregory J. Galloway, “Some unique- [48] J. D. Hunter, “Matplotlib: A 2d graphics environment,” ness results for dynamical horizons,” Adv. Theor. Math. Computing in Science & Engineering 9, 90–95 (2007). Phys. 9, 1–30 (2005), arXiv:gr-qc/0503109. [49] Michael Droettboom, Thomas A Caswell, John Hunter, [33] Abhay Ashtekar and Badri Krishnan, “Dynamical hori- Eric Firing, Jens Hedegaard Nielsen, Antony Lee, El- zons: Energy, angular momentum, fluxes and balance liott Sales de Andrade, Nelle Varoquaux, David Stansby, laws,” Phys. Rev. Lett. 89, 261101 (2002), arXiv:gr- Benjamin Root, Phil Elson, Darren Dale, Jae-Joon Lee, qc/0207080. Ryan May, Jouni K. Sepp¨anen,Jody Klymak, Damon [34] Ivan Booth, Robie A. Hennigar, and Saikat Mondal, McDougall, Andrew Straw, Paul Hobson, cgohlke, Tony S “Marginally outer trapped surfaces in the Schwarzschild Yu, Eric Ma, Adrien F. Vincent, Steven Silvester, Charlie spacetime: Multiple self-intersections and extreme mass Moad, Jan Katins, Nikita Kniazev, Tim Hoffmann, Fed- ratio mergers,” Phys. Rev. D 102, 044031 (2020), erico Ariza, and Peter W¨urtz,“matplotlib/matplotlib arXiv:2005.05350 [gr-qc]. v2.2.2,” (2018). [35] Jos´eLuis Jaramillo, “Black hole horizons and quantum [50] Frank L¨offler,Joshua Faber, Eloisa Bentivegna, Tanja charged particles,” Classical and Quantum Gravity 32, Bode, Peter Diener, Roland Haas, Ian Hinder, Bruno C. 132001 (2015). Mundim, Christian D. Ott, Erik Schnetter, Gabrielle [36] R P A C Newman, “Topology and stability of marginal Allen, Manuela Campanelli, and Pablo Laguna, “The 2-surfaces,” Classical and Quantum Gravity 4, 277–290 Einstein Toolkit: A Community Computational Infras- (1987). tructure for Relativistic Astrophysics,” Class. Quantum [37] S.A. Hayward, “General laws of black hole dynamics,” Grav. 29, 115001 (2012), arXiv:1111.3344 [gr-qc]. Phys. Rev. D 49, 6467–6474 (1994). [51] EinsteinToolkit, “Einstein Toolkit: Open software for rel- [38] Raphael Bousso and Netta Engelhardt, “New Area Law ativistic astrophysics,” http://einsteintoolkit.org/. in General Relativity,” Phys. Rev. Lett. 115, 081301 [52] Marcus Ansorg, Bernd Br¨ugmann, and Wolfgang Tichy, (2015), arXiv:1504.07627 [hep-th]. “A single-domain spectral method for black hole punc- [39] Abhay Ashtekar, Jonathan Engle, Tomasz Paw lowski, ture data,” Phys. Rev. D 70, 064011 (2004), arXiv:gr- and Chris Van Den Broeck, “Multipole moments of qc/0404056. isolated horizons,” Class. Quant. Grav. 21, 2549–2570 [53] J. David Brown, Peter Diener, Olivier Sarbach, Erik (2004), arXiv:gr-qc/0401114. Schnetter, and Manuel Tiglio, “Turduckening black [40] Dieter R. Brill and Richard W. Lindquist, “Interaction holes: an analytical and computational study,” Phys. energy in geometrostatics,” Phys. Rev. 131, 471–476 Rev. D 79, 044023 (2009), arXiv:0809.3533 [gr-qc]. (1963). [54] Sascha Husa, Ian Hinder, and Christiane Lechner, [41] Daniel Pook-Kolb, Ofek Birnholtz, Ivan Booth, Robie A. “Kranc: a Mathematica application to generate nu- Hennigar, Jos´e Luis Jaramillo, Badri Krishnan, Erik merical codes for tensorial evolution equations,” Com- Schnetter, and Victor Zhang, “MOTS Finder version put. Phys. Commun. 174, 983–1004 (2006), arXiv:gr- 1.5,” (2021). qc/0404023. [42] Erik Schnetter and Jonah Miller, “es- [55] Kranc, “Kranc: Kranc assembles numerical code,” . chnett/SimulationIO: New release to trigger Zenodo,” [56] Miguel Alcubierre, Gabrielle Allen, Bernd Br¨ugmann, (2019). Thomas Dramlitsch, Jose A. Font, Philippos Papadopou- [43] Mike Boyle, “spinsfast: Fast and exact spin-s spherical- los, Edward Seidel, Nikolaos Stergioulas, Wai-Mo Suen, harmonic transforms,” (2018). and Ryoji Takahashi, “Towards a stable numerical evo- [44] Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt lution of strongly gravitating systems in general relativ- Haberland, Tyler Reddy, David Cournapeau, Evgeni ity: The Conformal treatments,” Phys. Rev. D62, 044034 Burovski, Pearu Peterson, Warren Weckesser, Jonathan (2000), arXiv:gr-qc/0003071 [gr-qc]. Bright, St´efanJ. van der Walt, Matthew Brett, Joshua [57] Miguel Alcubierre, Bernd Br¨ugmann, Peter Diener, Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew Michael Koppitz, Denis Pollney, Edward Seidel, and Ry- R. J. Nelson, Eric Jones, Robert Kern, Eric Larson, oji Takahashi, “Gauge conditions for long term numerical 23

black hole evolutions without excision,” Phys. Rev. D67, 044054 (2021), 2010.15186. 084023 (2003), arXiv:gr-qc/0206072 [gr-qc]. [60] Raphael Bousso and Netta Engelhardt, “Proof of a [58] Christopher Evans, Deborah Ferguson, Bhavesh New Area Law in General Relativity,” Phys. Rev. D92, Khamesra, Pablo Laguna, and Deirdre Shoemaker, 044031 (2015), arXiv:1504.07660 [gr-qc]. “Inside the Final Black Hole: Puncture and Trapped [61] Erik Schnetter, “Finding apparent horizons and other Surface Dynamics,” (2020), arXiv:2004.11979 [gr-qc]. two surfaces of constant expansion,” Class. Quant. Grav. [59] Pierre Mourier, Xisco Jimenez-Forteza, Daniel Pook- 20, 4719–4737 (2003), arXiv:gr-qc/0306006. Kolb, Badri Krishnan, and Erik Schnetter, “Quasinor- [62] Erik Schnetter, Frank Herrmann, and Denis Pollney, mal modes and their overtones at the common horizon “Horizon pretracking,” Phys. Rev. D 71, 044033 (2005), in a binary black hole merger,” Physical Review D 103, arXiv:gr-qc/0410081.