Plausible scenario for a generic violation of the weak cosmic censorship conjecture in

asymptotically flat four dimensions

Felicity C. Eperon,1, ∗ Bogdan Ganchev,1, † and Jorge E. Santos1, ‡

1
Department of Applied Mathematics and Theoretical Physics,
University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, UK
We present a plausible counterexample to the weak cosmic censorship conjecture in the four-
dimensional Einstein-Scalar theory with asymptotically flat boundary conditions. Our setup stems
from the analysis of the massive Klein-Gordon equation on a fixed Kerr black hole background.
In particular, we construct the quasinormal spectrum numerically, and analytically in the WKB
approximation, then go on to compute its backreaction on the Kerr geometry. In the regime of
parameters where the analytic and numerical techniques overlap we find perfect agreement. We
give strong evidence for the existence of a nonlinear instability at late times.
arXiv:1906.11257v4 [gr-qc] 28 Feb 2020

Introduction – Quantum gravity remains terra Were they to dissipate energy faster than superradiance
incognita, largely because it is hard to access experi- creates them, the above scenario would not be possible.
mentally. One might wonder why, since singularities are We numerically compute the GW emission for fixed grav-
known to form under a variety of circumstances [1]. How- itational coupling, M µ, and spin parameter, J/M 2 , as a
ever, general relativity stubbornly conceals these regions function of ` = m and find that it leads to energy and
of high curvature, where quantum gravity is likely to play angular momentum dispersion that would not be able to
a leading role, by hiding them behind an absolute event counter the efficiency of superradiance.
horizon. This phenomenon is observed for such large Our paper is organized as follows: first we present our
classes of initial data that it was promoted to a conjec- setup and provide analytic and numerical data for the
ture in [2]. This is the weak cosmic censorship conjecture instability time scales at large `. We then compute, nu-
(WCCC), which forbids the formation of naked singular- merically, the energy radiated towards future null infinity
ities, i.e. singularities in causal contact with future null in this process as well as the backreaction of the scalar on
infinity, starting from generic initial data. one of the components of the Weyl tensor. We see that
Here we propose a four-dimensional asymptotically flat modes with higher ` radiate less, implying that energy
counterexample to the WCCC motivated by the superra- is accumulated at small scales more efficiently for larger
diant instability [3–12], afflicting massive scalar pertur- values of `. Finally we end with discussion of the results.
bations around Kerr black holes (BHs) [13] - rotating, Setup – We work with the Einstein-Hilbert action
with spherical topology and considered the most general minimally coupled to a real massive scalar field ψ
BH solutions of the vacuum Einstein equation [14].
√
 
R
Z
The angular dependence of such perturbations around S= d4 x −g − ∇a ψ∇a ψ − µ2 ψ 2 , (1)
M 16 π G
Kerr is parametrized by two integers {m, `}: m counts
the number of nodes in the azimuthal direction and `−|m| where µ is the scalar field mass, gab the spacetime metric
the number of zeros in the polar direction. We show that and R its Ricci scalar. The equations of motion are
for any scalar field of mass µ and any nonzero value
of the BH spin, for sufficiently large values of ` = m, R
Rab − gab = 8π G Tab (2a)
these perturbations herald instabilities around the BH, 2
extracting energy and angular momentum. Furthermore,
where Rab is the Ricci tensor of gab ,
the time scales associated to these instabilities grow para-
metrically as e4 ` log ` , indicating that each of the ` modes 2ψ = µ2 ψ , (2b)
decouples from the rest, evolving independently.
As time progresses, modes with smaller values of ` sta- and
bilize one by one, forming scalar clouds around the BH,
similar to those in [15, 16]. However, these BHs were Tab = 2∇a ψ∇b ψ − gab ∇c ψ∇c ψ − µ2 ψ 2 gab . (2c)
shown to be unstable to higher m-modes [17], giving rise
to the expectation of a cascade towards larger values of An important solution to these equations is the Kerr BH
`. This corresponds to a transfer of energy from lower [13], with ψ = 0 and
`-modes to higher ones, indicating an evolution towards
dr2
 
smaller scales - a phenomenon akin to turbulence in non- ∆ 2

2
ds2 = − dt − a sin θdφ + Σ 2
dθ 2
+
relativistic 3 + 1 fluids. Σ2 ∆
A possible stabilizing mechanism is the emission of sin2 θ
+ [a dt − (r2 + a2 )dφ]2 , (3)
gravitational waves (GWs) by the scalar clouds [18, 19]. Σ2
 2

where ∆ = r2 + a2 − 2 M r, Σ2 = r2 + a2 cos2 θ, φ ∈ 2, see [21]). For large m, Eq. (5b) can be approximated by
(0, 2π) is a periodic coordinate and θ ∈ (0, π) is a polar the usual equation for spherical harmonics on a 2-sphere
coordinate. The BH event√ horizon is a null hypersurface so that Λ = `(` + 1) + O(`−1 ) and `, m are integers with
with r = r+ ≡ M + M 2 − a2 , angular velocity ΩK = ` ≥ 0, |m| ≤ `. In the following we set m = ` and take
a/(a2 +r+ 2
) and surface gravity κK = (r+2
−a2 )/[2r+ (r+
2
+ the limit `  1.
2
a )]. The constant M is the BH mass and a parametrizes To determine the large ` limit at fixed µ and a, we used
its angular momentum via J = M a. The absence of the same method as in [22], which combines a matched
naked singularities demands |a| ≤ M with the inequality asymptotic and WKB type approach. We quote the final
being saturated at extremality, when the Kerr BH event result and leave the details to the Supplemental Material,
horizon becomes degenerate with κK = 0.
We study Eq. (2b) on a fixed Kerr background (3),
µ̂2
 
which is stationary and axisymmetric with respect to the Re(ωM ) = µ̂ 1 − 2 + O(`−3 ) , (6a)
Killing fields ∂/∂t and ∂/∂φ, respectively. We consider 2`
perturbations of the form
9
ψ(t, r, θ, φ) = e−i ω t+i m φ ψbωm (r, θ) `−4`− 2 +p
 
(4) 4`+5 π (`ΩK − µ)
Im(ωM ) = 2`+1−p √ µ̂ sinh ×
2 πp! κK
and assume that ψbωm (r, θ) is separable, ψbωm (r, θ) = 
2

µ̂
 
ΩK
 
Rω`m (r)Sω`m (θ), with the label ` anticipating that the exp − `ΩK − arctan − 2(1 − ` + p) ×
κK r+ κK
separation constant will be parametrized by an integer `.
1 + O(`−1 ) , (6b)
 
Not all solutions to Eq. (2b) are separable, but we are in-
terested in those composed of the sum (possibly infinite)
of such separable solutions. ω ∈ C is a complex fre- where µ̂ ≡ µM and p ∈ N0 is a radial overtone. First note
quency, determined by imposing appropriate boundary that if we set a = 0, the argument of the sinh becomes
conditions. We are interested in finding unstable mode negative, and the instability disappears. Furthermore, its
solutions for which Im(ω) > 0. onset sits precisely at the onset of superradiance, namely
Inserting the ansatz (4) into Eq. (2b) yields a system of `ΩK = Re(ω). More importantly for our purposes, in the
two second order ordinary differential equations, coupled limit ` → +∞, the growth rate of the instability scales
via the separation constant Λ, as e−4` log ` , and no matter what the value of µ or a, one
can always find a value of ` = `? ≡ dµ/ΩK e above which
the instability sets in. This shows that all Kerr black
 
∆ ∆ Rω`m,r ,r + V (r)Rω`m = 0, (5a)
holes are unstable to massive scalar field perturbations,
irrespective of their initial spin |J| < M 2 and of the mass
1   µ of the scalar perturbation.
sin θ Sω`m,θ ,θ
sin θ One can test the regime of validity of our approxima-
m2 tion, by comparing with data obtained by solving numer-
 
− a2 k 2 cos2 θ + − Λ Sω`m = 0, (5b) ically, without approximations, the full equations. To
sin2 θ
this end, we perform a change of variables so that we
where only solve for smooth functions in their integration do-
main. This necessarily involves a choice of normalization,
V (r) = −k 2 r4 + 2M µ2 r3 − (Λ + a2 k 2 )r2 + which we describe next. For numerical aid we define
(2M Λ − 4amM ω + 2M a2 ω 2 )r − a2 (Λ − m2 ) , ω−mΩ

(5c)
p r
with k ≡ µ2 − ω 2 . Finding unstable modes amounts
to finding the values of ω for which ψ has ingoing bound-
ary conditions at the event horizon (consistent with
the equivalence principle) and finite energy on a partial
Cauchy surface t = const.
This problem can be tackled numerically (for any val-
ues of the parameters) and analytically (in certain regions
of moduli space). We will first compute the modes using
a WKB expansion in m, which we detail next.
WKB Expansion and Numerical Validation–
Our WKB expansion is valid for any spin parameter |a| <
M , and only assumes m to be large. For small values of a
and µ, it reproduces the results in [7][20] (up to a factor of
 r+ −i 2κK K −k r  r+ γ
Rω`m (r) = 1 − e qr (r) , (7a)
r
Sω`m (θ) = sinm θ qθ (θ) , (7b)
with qr (r+ ) = qθ (0) = 1 and γ ≡ 1 + 2M k − M µ2 /k.
We now use the methods of [23] to solve for the eigenpair
(ω, Λ) using a Newton-Raphson routine [24].
As seen in Fig. 1, our numerical data agrees excellently
with (6b). Furthermore, one can measure deviations of
our WKB expression to the exact numerical result, and
it agrees with the error given in Eq. (6b).
Backreaction – We want the GWs emitted by a
scalar cloud around a Kerr BH and its leading order back-
reaction on the geometry. In the vector field case [25, 26]

�� × ��-��
�� × ��-��
�� × ��-���
�� × ��-���
�� × ��-���
� � �� �� ��
FIG. 1. The superradiant modes of a massive scalar around
Kerr with M µ = 0.42 and J/M 2 = 0.99, as a function of m.
The dashed red curve shows the analytic expression (6b) and
the blue disks our exact numerical data.
it has been shown that the system evolves adiabatically;
the emergence of the cloud due to superradiance, and
the consecutive saturation of the vector mode responsi-
ble, due to the spinning down of the BH, proceed on a
much faster time scale than the dispersion of energy and
angular momentum due to GW emission from the cloud.
We proceed using nonlinear perturbation theory and
declare
+∞ +∞
X X
ψ= ψ (2i+1) ε2i+1 , and g = gK + g (2i) ε2i , (8)
i=0 i=1
where gK is given by the Kerr metric (3). We expand
the equations of motion (2) in a power series in ε. To
first order in ε we solve eq. (2b) subject to a choice of
initial data. For the case at hand, we choose ψ to be
given by the real part of one of the unstable modes we
have determined above. These are labelled by a given
value of m. Furthermore, since Im(ωM )  Re(ωM ), we
take ω to be purely real. We then proceed to second
order and attempt to compute the leading order backre-
action on the metric, g (2) , and its associated curvature.
The standard approach to the linearized Einstein equa-
tion presents us with a daunting task; however, Kerr BHs
are algebraically special, allowing us to bypass comput-
ing g (2) , and directly calculate certain gauge invariant
scalars built out of the Weyl tensor. These do not cou-
ple amongst themselves and we focus on the Newman-
Penrose scalar ψ4 since it also allows us to efficiently
compute the GWs emitted by the scalar cloud. ψ4 obeys
the Teukolsky equation [27–29]
h

BH’s loss of energy due to the superradiant modes active
∆ + 3γ − γ̄ + 4µ + µ̄ D+4 − ρ − 3ψ2 −


i
− δ̄ + 3α + β̄ + 4π − τ̄ δ+4β − τ ψ4 = 4π T4 , (9)
whereby all quantities appearing in Eq. (9) are given in
the Supplemental Material. The source term T4 is recon-
3
structed directly from Eq. (2c) and thus depends on ψ (1)
and its gradient only.
The lhs of (9) can be separated into angular and radial
parts as in the vacuum case [27], allowing us to solve for
ψ4 as an infinite sum of separable solutions using Green’s
method.
From ψ4 , Teukolsky [29] showed us how to compute
the rate of gravitational radiation at future null infinity
d2 Es r2 2
= lim |ψ4 | , (10)
�� ��
dtdΩ r→∞ 4 π ω̂ 2
where ω̂ = 2 ω and dΩ is the induced volume on a unit
2-sphere.
We work with the scaled expression
 2
dEs M
PE = , (11)
dt Ms
where
Z +∞ Z π Z 2π
√
Ms = −g T tt dφ dθ dr (12)
r+ 0 0
is the total scalar field energy, - the energy of the per-
turbed initial data. PE is independent of the scalar field
amplitude and measures the energy radiated per m mode
in the initial data. See the Supplemental Material on how
to compute PE .
As a measure of curvature, we monitored the following
time independent quantity as a function of m:
χ ≡ max |ψ4 |2 /M2s .


(13)
r,θ
The radial and azimuthal location of the maximum of
(13), (r, θ) = (r? , θ? ), track the maximum of [ψ (1) ]2 .
Our results for the GW emission are shown in Fig. 2.
The radiated angular momenta in this process is PJ =
m
Re(ωM ) PE , in accordance with [29]. The fact that both
PE and PJ appear to decrease rapidly with increasing m
shows that the evolution occurs, to very good approxi-
mation, at fixed energy and angular momentum. This
is akin to the time evolution of the superradiance insta-
bility with anti-de Sitter asymptotics [30, 31], simulated
recently in [32], and shows hints of turbulent behaviour.
The data in Fig. (2) is for a fixed value of the di-
mensionless spin parameter a/M . However, during the
aforementioned cascade the BH will be gradually spin-
ning down, hence the gravitational radiation for each
value of m should ideally be computed by accounting for
prior to the one under consideration. Nevertheless, us-
ing the superradiant condition Re(ω) > mΩK , one sees
that ∆(a/M ) for successive superradiant modes ∼ `−1
as ` → ∞, implying that in the regime of interest the
dimensionless spin will be approximately constant and
Fig. (2) represents accurately the qualitative behaviour.

�� × ��-��
�� × ��-��
�� × ��-��
�� × ��-��
� � � �
FIG. 2. GW emission of energy and angular momentum, PE
and PJ respectively, for a single ` = m scalar cloud around
Kerr as a function of m. Same parameters as in Fig. 1.
�� × ��-�
�� × ��-��
�� × ��-��
�� × ��-��
�� × ��-��
� � � ��
FIG. 3. χ, a measure of the spacetime curvature, as a function
of ` = m. Same parameters as in Fig. 1. The black dashed
curve shows our leading order approximation, whereas the
dotted red line includes the next to leading order correction
and the orange circles are numerical data.
In Fig. 3 we show the dependence of χ on the initial
data, here labelled by m. A WKB-type analysis reveals
that the leading behaviour for χ is power law in 1/m
6
µ̂10

µ
χWKB = . (14)
m π 2 26
16 r+
One can consistently correct this approximation to next-
to-leading order (see the Supplemental Material).
Note that PE is harder to compute numerically than χ,
hence why we have extended results for χ up to m = 5.
Gedanken experiment – We are now ready to
present our possible counterexample to WCCC with
asymptotically flat boundary conditions. Consider
generic initial data for the Einstein-Scalar system. These
data are controlled by a large functional freedom coming
from the fact that we can choose the initial metric on
a constant time slice, as well as the extrinsic curvature
(so long as the Hamiltonian and momentum constraints
are satisfied). In addition, we can also control the initial
profile for the scalar field and its first time derivative on
4
a constant time slice. We are going to choose our initial
data to be close to that of the Kerr BH, so that devi-
ations from the Kerr metric only occur at order O(ψ 2 ).
This condition can be relaxed by considering initial data
for the purely gravitational sector that is small in some
norm. This essentially means that all the dynamics are
being generated by the scalar field.
For generic scalar field initial data, we expect the scalar
field profile to have some support on the unstable modes
of the preceding sections, i.e. to excite unstable modes.
Since all other modes decay with time [33], we expect
the late time evolution to be dominated by the leading
unstable modes and their backreaction. For each value of
m there is an infinite number of such modes labelled by
` ≥ |m|. However, all of these modes stop being unstable
as soon as the condition Re(ω) > mΩK is no longer satis-
fied. The dynamics of this change in angular momentum
and energy is entirely controlled by the ` = m modes, im-
plying that, after some suitably long time, the dynamics
of the Einstein-Scalar system can be well approximated
by restricting our attention to scalar profiles of the form
"+∞ #
X
−iω` +i`φ
ψ(t, r, θ, φ) = Re a` e Rω` `` (r) Sω` `` (θ) ,
`=0
(15)
and determining its leading order backreaction on the
��
spacetime curvature. The coefficients a` are determined
by our choice of initial data: for finite Sobolev norm ini-
tial data we expect a` to exhibit polynomial behaviour
in 1/`, whereas for C ∞ initial data we expect the coef-
ficients a` to decay faster than any polynomial in 1/`.
Note that for real analytic initial data one can show that
a` ≈ e−α` , for α > 0.
Given that each ` mode evolves on an exponentially
different timescale, as shown by Eq. (6b), they effectively
decouple from each other, allowing us to study each term
in Eq. (15) and its backreaction on the metric separately.
Eventually, a given ` = `? mode becomes stable, but the
system remains unstable to higher values of ` > `? . This
cascading happens slowly, since the time scales for this
effect are exponentially large. One might worry that the
energy contained in these high ` modes is radiated away
as time passes by, but we have seen in Fig. (2) that this
is not the case. In fact, the larger the value of `, the
smaller its radiative power is.
Finally, we have seen in Fig. (3) that χ decays as `−16 .
This in turn implies that a mode with weigh a` will de-
pend on ` as a4` `−16 . The reason for this is simple: the
Teukolsky scalar ψ4 is sourced by [ψ (1) ]2 , and χ is related
to |ψ4 |2 , which translates into the overall scaling men-
tioned above. The curvatures are thus suppressed with
time, and the evolution continues until all of the angular
momentum is deposited into the scalar cloud, and the
central black hole becomes Schwarzschild. However, one
can show [34] that massive scalar field perturbations de-
cay extremely slow around Schwarschild black holes, with
 5

a dependence as weak as 1/ log(log t), for large t. This SUPPLEMENTAL MATERIAL

suggests that the hypothetical end point is itself nonlin-
early unstable through a mechanism similar to the one WKB expansion – We present the details of the
reported in [22, 35, 36]. On the other hand, this slow expansion in large ` that we use to determine the scalar
decay warrants the question whether the gravitational instability growth rate. We work with l − m = n, where
radiation might not just disperse the clouds, before any n ∼ O(1), n ∈ N0 . To this end we rewrite the Klein-
nonlinearities become problematic. We cannot answer Gordon equation from the main text in the following way
this in our analysis, as our approximation for radiation
emission breaks down at late times. Therefore the out-
r − r+  0

come of our thought experiment depends on which of the ∆ Rω`m (r) ,r + V (r) Rω`m (r) = 0, (16a)
competing processes - the GW emission or the nonlinear r+
0
 
issues resulting from slower than logarithmic decay - wins sin θ Sω`m (θ) ,θ  m2

2 2
over. If it is the latter, then the lack of a possible sta- − ã k̃ cos θ + − Λ Sω`m (θ) = 0,
sin θ sin2 θ
tionary end point leads us to conjecture that the spirit, (16b)
if not the letter, of weak cosmic censorship is violated.
r r2 r3 a4
Whether the curvature will be infinite in finite time is a V (r) = a1 +a2 + a3 2 − k̃ 3 + ,
question that we cannot settle with our current methods. r+ r+ r+ r/r+ − ã2
a1 = Λ+ã ω̃ 1 + ã2 ã ω̃ 1 + ã2 − 2 m ,



Note also that r? (`) increases with `, posing a problem
from a numerical perspective: 1) the time scales involved a2 = ã2 ω̃ 2 1 + ã2 − Λ, a3 = µ̃2 + ã2 ω̃ 2 ,


are enormous; 2) the cascading towards high ` values
2
a4 =ã2 m − ã ω̃ 1 + ã2

makes this problem dependent on high frequency modes , (16c)
(as the simulation of turbulence in 3 + 1 nonrelativitic
fluids); and 3) the integration domain must extend to where k̃ = µ̃2 − ω̃ 2 , Λ is the angular separation constant
spatial infinity to observe this effect. and we have introduced dimensionsless variables
Conclusions – We have seen (Fig. 2) that the effi- a M
ciency of superradiance cannot be counteracted by GW ã = , M̃ = , ω̃ = ω r+ , µ̃ = µ r+ , (17)
r+ r+
emission, implying that the system will continue advanc-
ing to higher values of m with curvatures decreasing ap- As explained in the main text, finding unstable modes
propriately (as shown by Fig. 3) until a configuration amounts to determining the values of ω̃ for which ψ has
with a central Schwarzschild black hole is reached. How- ingoing boundary conditions at the event horizon (con-
ever, Schwarzschild is likely to be nonlinearly unstable sistent with the equivalence principle), and finite energy
due to the very slow decay of perturbations induced by on a partial Cauchy surface t = const [37].
massive scalar fields. Reaching this troublesome regime, In the eikonal limit `  1 the spheroidal eigenvalue is
given that effects we cannot account for in our analy- Λ = `(` + 1) + O(`−1 ); this can easily be verified by a se-
sis do not prevent this, will involve time scales much ries expansion of the angular equation after prefactoring
longer than the age of our Universe, of course, as one sinm θ from S(θ). In order to determine ω̃ we divide the
will have to go to large values of ` = m . Nevertheless domain in two intersecting regions, solve (16a) in each
our scenario provides the first plausible example of a sys- of them and then match the solutions in the overlap. To
tem with asymptotically flat boundary conditions, where this end we introduce the new variable x = r/r+ −(1+ã2 )
WCCC is violated. ˜ Rω`m
0
(x + ã2 ) ∆
 
(x) ,x + Ṽ (x)Rω`m (x) = 0, (18)
Acknowledgments – We would like to thank
O. J. C. Dias and H. S. Reall for reading an earlier a4
V (x) = b1 + b2 x + b3 x2 − k̃ x3 + ,
version of this manuscript. We also want to thank 1+x

2
b1 = ω̃ 2 1 + ã2 1 + 4 ã2 − 2 m ã ω̃ 1 + ã2



W. E. East for many useful discussions that led to
this paper. F.C.E. and B.G. are supported by STFC.
2 
−ã2 `(` + 1) + 1 + ã2 µ̃2 ,

J.E.S is supported in part by STFC Grants PHY-
b2 = − 1 + ã2 1 + 3 ã2 µ̃2 − 3 1 + 2 ã2 ω̃ 2




1504541 and ST/P000681/1. This work used the DIRAC
Shared Memory Processing system at the University of −`(`+1),
Cambridge, operated by the COSMOS Project at the
b3 = 3 + 4 ã ω̃ 2 − 2 + 3 ã2 µ̃2 ,
2



Department of Applied Mathematics and Theoretical
Physics on behalf of the STFC DiRAC HPC Facility ˜ = (1 + x)(x + ã2 )
with ∆
(www.dirac.ac.uk). This equipment was funded by BIS
Near-horizon region – Using quasimodes [35, 38], one
National E-infrastructure Capital Grant ST/J005673/1,
can show that k̃ = O(`−2 ). We thus take the near-
STFC Capital Grant ST/H008586/1, and STFC DiRAC
horizon region to be defined by x  `2 and see that
Operations Grant ST/K00333X/1. DiRAC is part of the
National e-Infrastructure. k̃ x3  x m2 , b3 x2  x m2 , (19)
 6

implying that we can drop the cubic and quadratic terms resulting in
inside V (x), leaving us with  
00 1 e1 b2
˜ RH
0
h a4 i j (z) + − − p + 2 j(z) = 0. (28)
(x + ã2 ) ∆
 
(x) ,x + b1 + b2 x + RH = 0. (20) 4 2 k̃ z z
1+x
The solution [39] is a linear combination of Gauss hyper- The general solution to (28) is then a linear combination
geoemtric functions of the Whittaker functions Wk,ν (z) and W−k,ν (−z), thus
h i
RH (x) = Ain (−u)δ (1 + u)φ 2 F1 c− , c+ ; c; −u (21) R∞ = B1 v(z) W−κ,ν (−z) + B2 v(z) Wκ,ν (z),
h i e1 1p
+ Aout (−1)−2 δ (−u)−δ (1 + u)φ 2 F1 λ− , λ+ ; 2 − c; −u , κ=− p , ν= 1 − 4 b2 , v(z) = h(z)/z, (29)
2 k̃ 2
where u = (x + ã2 )/(1 − ã2 ), Ain/out are constants and with B1/2 constants. Expanding for large z, and using
1
ã4 (1 − ã4 )(2 µ̃2 − 3 ω̃ 2 ) 2

i Ω0 p h r i p r
δ= 1+ , z=2 k̃ − (1 + ã2 ) −−−→ 2 k̃ , (30)
1 − ã2 Ω20 r+ r→∞ r+
m − ã ω̃ (1 + ã2 )
φ=−i , Ω0 = m ã − ω̃ (1 + ã2 ) we see that
1 − ã2
√
1 1 − 4 b2
 
c± = ± + φ + δ, c = 1 + 2 δ, lim B1 v(z) W−κ,ν (−z) + B2 v(z) Wκ,ν (z) =
r→∞
2 2 √ √
λ± =c± − c + 1. (22) k̃ r/r+ −1−κ − k̃ r/r+ −1+κ
B1 e r + B2 e r . (31)
Demanding ingoing waves only at the horizon (x = 0) Imposing that the solution decays at infinity sets B1 = 0.
requires setting Aout = 0. To see this, note that in the Matching – We now perform the matching procedure
limit x → 0 the hypergeometric functions take on a con- in the overlapping region 1  x  `2 . To this end, we
stant value to leading order, hence one just needs to know 3
take x ∼ ` 2 and expand the near-horizon and far-away
which of (−u)±δ gives the correct behaviour there. This solutions for large and small variables respectively.
can most easily be deduced by performing a transforma- Looking at the near-horizon region first, we can
tion from BL to Kerr coordinates straightforwardly use the standard asymptotic series of
r2 + a2 a the Hypergeometric function, as all the parameters grow
dv = dt + dr, dχ = dφ + dr. (23) slower than the variable of the function, leaving us with
∆ ∆
Looking at the near-horizon limit allows us to obtain ν− 21
(−1)δ Γ[c+ − c− ] Γ[c]

x
1+ã2

lim RH = Ain
e −i ω t i m φ
e =e −i ω v i m χ
e (r − r+ )
i ω̃ 1−ã m ã
2 − 1−ã2
, (24)
x→∞ Γ[c+ ] Γ[c − c− ] (1 − ã2 )
−ν− 21
(−1)δ Γ[c− − c+ ] Γ[c]

x
which has opposite sign to what we expect at the horizon, + Ain . (32)
as the transformation eliminates any singular behaviour. Γ[c− ] Γ[c − c+ ] (1 − ã2 )
Far away region – Before zooming into spatial infinity Next we consider the expansion of the far-away solu-
we make the following transformation to (18), to ensure tion for small variable and again we can use the standard
that we will get the correct asymptotic behaviour, series in the literature to do so, as the parameters out-
grow the argument of the Whittaker function, thus
q
4 k̃ 2 x2 + 4 k̃ x d2 + d21
Rω`m (x) =h(x) Q∞ (x), h(x) = , p 1
2 k̃ x + d1 (2 k̃)ν− 2 Γ[−2 ν] ν− 1
lim R∞ = −B2 x 2
d1 = 1 + 2 ã2 2 µ̃2 − 3 ω̃ 2 ,



x→0 Γ[ 12 − ν − κ]
d2 = 1 + 3 ã2 µ̃2 − 2 + 5 ã2 ω̃ 2 .


p
(25) 1
(2 k̃)−ν− 2 Γ[2 ν] −ν− 1
− B2 x 2. (33)
We work with x  1, which leaves us[40] with Γ[ 12 + ν − κ]

x2 Q00∞ (x) + 2 x Q0∞ (x) − (k̃ x2 + e1 x − b2 )Q∞ (x) = 0, Afterwards we equate the coefficients in front of the
2 2
e1 = (1 + ã )(µ̃ − 2 ω̃ ),2
(26) equivalent terms in (32) and (33) in order to derive

to solve. The above can be transformed into Whittaker’s Γ[ 12 − ν − κ]  p 2 ν

1 = 2 (1 − ã2 ) k̃ G(`),
differential equation with the help of Γ[ 2 + ν − κ]
z Q (z) Γ[c− − c+ ] Γ[−2 ν] Γ[c+ ] Γ[c − c− ]
x= p , j(z) = ∞ , (27) G(`) = . (34)
z Γ[2 ν] Γ[c+ − c− ] Γ[c− ] Γ[c − c+ ]
2 k̃
 7

Furthermore, in the limit l → ∞ the RHS above has very 1 ± ν̂ i ã µ̃ (m − ã(1 + ã2 ))
σ± = + δ̂, φ̂± = ± ,
small real and imaginary parts, implying that Γ[ 21 +ν −κ] 2 1 − ã2
in the denominator on the LHS must have a pole, thus iξ h ã4 µ̃2 (1 − ã4 ) i 12
δ̂ = 1 − ,
1 1 − ã2 ξ2
+ ν − κ = −N, (35) ξ = ã m − (1 + ã2 )µ̃, (38)
2
with N ∈ N0 , corresponding to the radial node of p the For the imaginary part of ω̃ we allow (35) to be complex
scalar field. This allows us to deduce the real part of k̃
1
+ ν − κ = −N + , (39)
p
e1 2
Re k̃ = − . (36)
2N + 2 ν + 1
with   1,  ∈ C. We can look at the  → 0 limit of
Next, with |ω̃R |  |ω̃I | [41, 42], implying |k̃R |  |k̃I |, we (39), using (36), to derive
can expand (36) for large ` to obtain the real part of ω̃ p
2 i e1
Im k̃ =  , (40)
2 2
(1 + ã ) µ̃ 3 (1 + 2 N + 2 ν)2
ω̃R = µ̃ − + O(`−3 ), (37)
8 `2 which also enables us to find, in the limit ` → ∞,
confirming our expectations. Moreover, our calculation
(1 + ã2 )2 µ̃3
is accurate to O(1) only, hence we can replace ω̃ with µ̃ ω̃I = i Im() . (41)
everywhere except inside k̃, where the leading order term 4 `3
cancels. This reduces (34) to Next, we look at (38) for  → 0 and obtain
Γ[ 12 − ν − κ] h µ̃2 (1 − ã4 ) iν̂ h µ̃2 (1 − ã4 ) iν̂ Ĝ(`)
= Ĝ(`), (−1)N N !  = . (42)
Γ[ 12 + ν − κ] N + 1+ν̂2 N+ 1+ν̂ Γ[−N − ν̂]
2
Γ[−ν̂]2 Γ[σ+ + φ̂+ ] Γ[σ+ + φ̂− ]
Ĝ(`) = , Taking ` large, we can rearrange for , which allows us
Γ[ν̂]2 Γ[σ− + φ̂+ ] Γ[σ− + φ̂− ] to derive an expression for ω̃I via (41). It is valid for any
p
ν̂ = (1 + 2 `)2 − 4(1 + ã2 )(2 + 3 ã2 )µ̃2 , spin parameter |a| < M and scalar field mass µ,

1 +`
2N −5−6 ` (1 + ã2 )3+2 ` (1 − ã2 )2 `2 + 4 ã2 m2 2 µ̃5+4 `


2 π ã m − (1 + ã2 )µ̃

ω̃I = √ 11 sinh ×
π N ! `−N + 2 +6 ` 1 − ã2
 2 ã m   2 ã m  
2 µ̃ (1 + ã2 )2 arctan `(1−ã 4 ã m arctan `(1−ã

2) 2)
exp 2 ` − 2(N + 1) + 2
− 2
. (43)
1 − ã 1 − ã

Setting ` = m (n = 0) we get the growth rate of the dominant modes in the spectrum

2N −5−6 ` (1 + ã2 )4+4 ` µ̃5+4 ` 2 π(ã ` − µ̃(1 + ã2 ))

ω̃I,`=m = √ 9 sinh[ ]×
π N ! `−N + 2 +4 ` 1 − ã2
−2 ã 
   −2 ã  
µ̃ (1 + ã2 )2 arctan 1−ã 4 ã arctan 1−ã
  
2 2
exp − 2 N + 1 + +` 2+ , (44)
1 − ã2 1 − ã2

which is rewritten in the main text for compactness. produce the correct behaviour for Schwarzschild [34]
Moreover, for ã = 0 the sinh changes sign and we re-
2N −5−6 ` e−2(N +1)+2 ` µ̃5+4 `
ω̃I,ã=0 = − √ 9 sinh[2 π µ̃]. (45)
π N ! `−N + 2 +4 `
 8
 
KG Numerically– We apply spectral collocation
 m 
Sω`m,θ − a ω sin θ − Sω`m Rω`m Sω`m ,
methods on a discretised Chebyshev grid with sin θ
1 + cos θ r+ e−2 i ω t e2 i m φ 2
z= , x=1− , (46) Tm̄m̄ = 2 Rω`m ×
2 r 4 (r − i a cos θ)
 2
where x = 0 and x = 1 correspond to the event horizon
 m 
Sω`m,θ − a ω sin θ − Sω`m . (49)
and spatial infinity, and z = 0 and z = 1 to the north and sin θ
south poles, respectively. We use dimensionless variables Furthermore, in the main text it is stated that the GW
(17), factor out the singular behaviour at the boundaries frequency and mode number are related to the scalar field
α
Rω`m (x) = (1 − x)β e 1−x xγ Rω`m (x), ones by ω̂ = 2 ω, m̂ = 2 m. To see this we first remind
ourselves of our ansatz for the scalar field
Sω`m (z) = z m/2 (1 − z)m/2 Sω`m (z),
p (1 + ã2 )(µ̃2 − 2ω̃ 2 ) ψ(t, r, θ, φ) = e−i ω t+i m φ Rω`m (r)Sω`m (θ) . (50)
α = − µ̃2 − ω̃ 2 , β = 1 + p ,
2 µ̃2 − ω̃ 2 Next, we define the stress tensor of a scalar as
 1 + ã2 m ã 
Tµν Υ, Ψ = 2∇µ Υ∇ν Ψ−gµν ∇σ Υ∇σ Ψ+µ2 ΥΨ , (51)



γ = −i ω̃ − , (47)
1 − ã2 1 − ã2
and finally apply Newton’s method. We then solve for and plug in ψR = 1/2(ψ + ψ̄) for Υ and Ψ, leading us to
Rω`m (x) and Sω`m (z).
1h


i
No special treatment at the boundaries is needed as Tµν ψR , ψR = Tµν ψ, ψ + Tµν ψ̄, ψ̄ + 2Tµν ψ, ψ̄ ,
4
(47) force the system to pick the right solution. An initial
guess is (43), Λ = `(` + 1), Rω`m (x) = 1, Sω`m (z) = 1. with bar indicating complex conjugation. The ansatz
Computation of ψ4 – We want to compute GWs (50) implies that the terms above will contain exponen-
emitted by a real scalar cloud around a Kerr BH. A com- tial prefactors, which reveal that the first and second
plex scalar is not realistic and produces negligible grav- one source outgoing waves with m̂ = 2 m, ω̂ = 2 ω and
itational radiation, since its stress-tensor is stationary - m̂ = −2 m, ω̂ = −2 ω, respectively, whereas the third
its use is a computational convenience for the determina- one corresponds to the energy of the cloud, given by the
tion of the field’s QNMs, hence to find GWs we consider gravitational mode with m̂ = ω̂ = 0.
the real part of the superradiant modes - ψR = < ψ. The LHS of (48a) can be separated into angular and
As explained in the main text, we treat the cloud as a radial parts as in the vacuum case [27]. Thus take
perturbing source and use Teukolsky’s equation [27–29]
ψ4 = e−i ω̂ t ei m̂ φ ρ4 R(r) S(θ), (52)
h


∆ + 3γ − γ̄ + 4µ + µ̄ D + 4 − ρ − 3ψ2 − with spin coefficient ρ = −1/(r − i a cos θ), and multiply


i both sides by 2 Σ(r, θ)/ρ4 . S(θ) can be identified with a
δ̄ + 3α + β̄ + 4π − τ̄ δ + 4β − τ ψ4 = 4π T4 , (48a) spin-weighted spheroidal harmonic, satisfying


T4 = ∆ + 3γ − γ̄ + 4µ + µ̄ × 1 h i 
2
h

i sin θ s S`ˆm̂,θ + (c cos θ) − 2 c s cos θ + s
δ̄ − 2τ̄ + 2α Tnm̄ − ∆ + 2γ − 2γ̄ + µ̄ Tm̄m̄ sin θ ,θ
2
+ δ̄ − τ̄ + β̄ + 3α + 4π ×

(m̂ + s cos θ)
+ sA`ˆm̂ω̂ − S = 0, (53)
s `ˆm̂
h

i sin θ2
∆ + 2γ + 2µ̄ Tnm̄ − δ̄ − τ̄ + 2β̄ + 2α Tnn . (48b)
with c = a ω̂, sA`ˆm̂ω̂ the separation constant and s = −2.
to extract the NP scalar, ψ4 , encoding the informa- Multiplying by s S̄`ˆm̂ and integrating over θ results in the
tion about the outgoing GWs. Our sign convention is following equation for the radial function
(−, +, +, +), opposite to Teukolsky’s. Nevertheless, the    2
equation in terms of NP variables is unchanged. For their 2 −1 K + 4 i (r − M ) K
∆k ∆k R`ˆm̂,r + − 8 i ω̂ r
definitions, in our convention, we use [43]. ,r ∆k
The tetrad projections we need in T4 are Tnn , Tnm̄ and 
Tm̄m̄ . Moreover using a null tetrad implies we can ignore −a2 ω̂ 2 + 2 a m̂ ω̂ − sA`ˆm̂ω̂ R`ˆm̂ = T`ˆm̂ω̂ , (54)
the gµν term. In a Kinnersley tetrad [44] we have

where K = r2 + a2 ω̂ − a m̂ and the source term is the
−2 i ω t 2 i m φ 
e e 2 2 integrated over angles stress-energy tensor
Tnn = i K Rω`m + ∆k Rω`m,r Sω`m ,
8 Σ2
2 Σ(r, θ)
Z
e−2 i ω t e2 i m φ 4π
Tnm̄ = √
 
i K Rω`m + ∆k Rω`m,r × T`ˆm̂ω̂ = S̄ T4 sin θ dθ, (55)
4 2 Σ (r − i a cos θ) η`ˆm̂ ρ4 s `ˆm̂
 9

with the normalisation condition
Z π
|s S`ˆm̂ (θ)|2 sin θ dθ = η`ˆm̂ .
0 Knowing ψ4 , we can compute the gravitational radia-
Equation (54) should be evaluated for ω̂ =
±2 ω, m̂ =
±2 m, s = −2

on the LHS with Tµν ψ, ψ , for the +,
and Tµν ψ̄, ψ̄ , for the −, inside T4 on
the RHS, and for
m = ω̂ = 0 on the LHS with Tµν ψ, ψ̄ in T4 on the RHS.
However, the Teukolsky equation is invariant under
complex conjugation followed by m → −m and ω → −ω,
hence for m̂ 6= 0 6= ω̂ we

only need to determine the
contribution of Tµν ψ, ψ and double the result.
Moreover, by looking at the asymptotic behaviour of
the homogeneous radial equation ((54) with T`ˆm̂ω̂ = 0)
we can show that the m̂ = ω̂ = 0 mode is subleading to
all the rest at spatial infinity. To this end, transform the
homogeneous (54) with the help of
−s
∆K 2 dr∗ r2 + a2
R`ˆm̂ (r) = √ Y ˆ (r), = ,
r2 + a2 `m̂ dr ∆K
and then take the limit of r → ∞, leaving us with
h 2 i s ω̂ i
Y`ˆ00m̂ (r∗ ) + ω̂ 2 + Y`ˆm̂ (r∗ ) = 0, (58)
r
for finite[45] frequency ω̂ and azimuthal number m̂. The
outgoing part of Y`ˆm̂ behaves as ei r∗ ω̂ r−s , leading to an
ei r∗ ω̂ r−2s−1 asymptotic behaviour for R`ˆm̂ , which to-
gether with the definition (52) allows us to deduce that
the outgoing contribution of ψ4 near spatial infinity, for
finite ω̂ and m̂, behaves as
latter as a source on the RHS of (54), which sets the val-
ues of the GW frequency ω̂ and mode-number m̂, letting
(56) us solve (54) for any allowed value of `.ˆ
tion at infinity and the maximum of the curvature com-
ponent represented by the NP scalar ψ4 with the formulae
given in the main text. The argument for ignoring the
m̂ = ω̂ = 0 mode in the computation of the gravitational
radiation at infinity does not hold for the maximum of
ψ4 , as the latter is expected to be located at a finite dis-
tance from the black hole, near the peak of the scalar
cloud, as we find in our numerics. Nevertheless, we have
checked numerically that its contribution to the maxi-
mum is negligible compared to the rest.
Numerical integration – We use the same numerical
method for (53) and (54) as for (16a) and (16b).
The angular equation – (53) is an eigenvalue problem
which we integrate using Newton’s method. We write
= z ι− (1 − z)ι+ sS`m ,
(57) s S`m (62)
where ι± = | m±s2 | and solve for sS`m . We drop the hats
on m and ω, as we solve the equation generally. The
eigenvalue sA`mω is treated as an unknown and we nor-
malise sS`m to 1 using (56). Appropriate boundary con-
ditions are imposed at the edges of the domain. These
can be obtained by expanding the equation for sS`m in
power series around the poles. At z = 0 we require
− 4 + (m − s)2 − 8 ι− − (2 ι− )2 ∂z sS`m (z)|z=0 =
 
− m2 − 4 s + 2 m s − 4 sA`m − 8 ã s ω − 4 ã2 ω 2 +

(2 ι− )2 + 4 ι+ + 4 ι− (1 + 2 ι+ ) sS`m (z)|z=1 , (63)


lim ψ4 ∼ ei ω̂ r∗ /r (outgoing mode). (59)

r→∞
However, were we to set ω̂ = m̂ = 0 before we take the
limit of r → ∞, instead of (58), we are left with
00 `(` + 1)
Y00 (r∗ ) − Y00 (r∗ ) = 0, (60)
r2
where we have used sA`00 ˆ = `(` + 1) − s(s + 1). The solu-
tion with finite energy at infinity corresponds to asymp-
totic behaviour of the form r−`−3 for ψ4 , which is clearly
subleading to non-zero ω̂ and m̂ modes.
Therefore, for the computation of the gravitational ra-
diation from the scalar clouds, which is performed at spa-
tial infinity, we only need to look at the ω̂ = 2 ω, m̂ = 2 m

in (47) from the scalar field inside T4 , so that we can
case for (54), with Tµν ψ, ψ inside T4 on the RHS.
In this way, ψ4 has been projected onto a basis of spin-
weighted spheroidal harmonics
X analysis, and transfer them to the RHS. Specifically
ψ4 = e−i ω̂ t ei m̂ φ ρ4 R`ˆm̂ (r) s S`ˆm̂ (θ), (61)
`ˆ
where the definition of s S`ˆm̂ (θ) requires `ˆ ≥ m̂.
Thus, given a Kerr BH, for fixed M µ and J/M 2 , and
its ` = m scalar superradiant spectrum, we can use the
and similarly at z = 1

− 4 + (m + s)2 − 8 ι+ − (2 ι+ )2 ∂z sS`m (z)|z=1 =

 2
m − 4 s − 2 m s − 4 sA`m + 8 ã s ω − 4 ã2 ω 2 +
(2 ι+ )2 + 4 ι+ + 4 ι− (1 + 2 ι+ ) sS`m (z)|z=1 . (64)

A starting point to our iteration is the series solution,
provided in [46], up to twelfth order.
The radial equation – (54) is a sourced ODE, which we
can invert once the RHS is known. As explained above,
we need the superradiant modes of the scalar field on a
fixed Kerr background, which we have numerically, so we
ˆ We take out the prefactors
can integrate (54) for any `.
directly substitute our numerical solutions Rω`m (x) and
Sω`m (z). We then factor out the singular behaviour of
R`ˆm̂ at infinity and the horizon, derived via Frobenius
α̂
2
R`ˆm̂ (x) = r+ (1 − x)β̂ e 1−x x2+2 γ̂ R`ˆm̂ (x), (65)
with R`ˆm̂ (x) dimensionless and
α̂ = 2 i ω̃, β̂ = −3 − 2 i(1 + ã2 ) ω̃,
 10

1 + ã2
 
m̂ ã 100
γ̂ = −i ω̃ − . (66) M µ “ 0.42, J {M 2 “ 0.99
1 − ã2 1 − ã2 10´2

10´4
Moreover, x2 γ̂ gets cancelled by the xγ from the radial
scalar functions (47) inside T4 . We then solve for R`ˆm̂ (x). 10´6

ˇ
P max ˇ
`1
ˇ
We need to impose boundary conditions at spatial infin-

ˇ1 ´ PE,`E,`max
10´8
ity, since [27] reveals that the ingoing behaviour of the
10´10
radial function (the one we do not want) is subleading
10´12

ˇ
ˇ
2 i ω̃
(in) − 21−x
i ω̃ (out) e 1−x
10´14
R`ˆm̂ (x) ∼ Z`ˆm̂ e (1 − x) + Z`ˆm̂ , (67)
(1 − x)3
10´16

The boundary condition that selects the outgoing waves 10´18

8 9 10 11 12 13 14 15
only can be deduced by expanding the homogeneous `max
Teukolsky equation for large radial variable , which shows
(in)
that setting Z`ˆm̂ = 0 is equivalent to demanding
FIG. 4. Convergence of the radiated GW energy at infinity
PE for the m = 4 scalar cloud as a function of the number of
1 h
included ` modes in the projection of ψ4 onto the basis of spin-
∂xx R`ˆm̂ (x) + (1 − ã2 )2 sA`ˆm̂$̂ 2
x=1 4(1 − ã2 )2 $̂2 weighted spheroidal harmonics at the highest grid resolution
+ 2(1 − ã2 ) sA`ˆm̂$̂ ]2 ã m̂ $̂ + ã4 $̂2 + 2 ã6 $̂2


− (i + $̂)2 + ã2 (−1 + 4 i $̂ − 3 $̂2 )

104

11 M µ “ 0.42, J {M 2 “ 0.99
+ 2 $̂ − 3 i − 12 $̂ + 20 i $̂2 + ã8 (8 i − $̂)$̂2 103
2
102
+ 8 $̂3 + 2 ã12 $̂3 + 2 ã10 $̂2 (i + $̂ − 2 ã7 m̂ $̂(i − 2 $̂)
ˇ
P x ˇ
ˇ

i 1 3i 3
x`1000

101
+ 4 ã5 m̂(1 + $̂ + $̂2 ) − 8 ã3 m̂(1 − $̂ + $̂2 )
ˇ1 ´ PE,NE,N

2 2 2 4
100
1
+ 4 ã m̂(1 − 3 i $̂ − 2 $̂2 ) + ã4 (3 i − 2 $̂ − 20 i $̂2 + $̂3 )
2 10´1
ˇ
ˇ

+ ã2 (3 i + 19 $̂ + 2 m̂2 $̂ − 4 i $̂2 + 12 $̂3 ) 10´2

i
− ã6 (3 i + 5 $̂ + 6 i $̂2 + 11 $̂3 ) R`ˆm̂ (x = 1) = 0,
where $̂ = 2 ω̃.
In terms of the functions for which we solve numeri-
cally, the formula for the gravitational radiation given in
the main text takes the following form
dEs 1 X 2
= η`ˆm̂ R̂`ˆm̂ (x)
dt (2 ω̃)2 x→1
`ˆ
where we have used (56), the orthogonality of the spin-
weighted spheroidal harmonics, and have multiplied by 2
to accound for the GWs with m̂ = −2 m, ω̂ = −2 ω.
Numerical convergence – As we cannot integrate
ˆ we truncuate the sum until (69)
(54) for infinitely many `,
converges. We will look at the m = 4 case here (m̂ = 8
for the GWs), as this was the hardest one to tackle nu-
merically. We include 9 GW modes, `ˆ = 8 to `ˆ = 16,
to get good convergence for the radiated energy. More-
over, very high grid resolution was needed in the radial
direction, in order to resolve the oscillating behaviour of
the solution far away from the BH (exactly at spatial in-
finity the oscillating part is discarded by the boundary
conditions). This is summarised in Figs. (4) and (5)
10´3
(68)
10´4
12000 13000 14000 15000 16000 17000
Nx
FIG. 5. Convergence of the radiated GW energy at infinity
PE for the m = 4 scalar cloud as a function of the number of
points in the radial direction N x
, (69)
Analytic approximation to χ – In this section we
outline the steps
for the WKB approximation of χ ≡
maxr,θ |ψ4 |2 /Ms2 . We work with ` = m and `  1.
First, we need a WKB expansion of the scalar field
near its maximum - i.e. Rω`m (r), Sω`m (θ), Λ and ω̃, as
well as the location of the minimum of the potential of
the radial KG equation expanded for large ` = m.
We start with the angular equation (16b), which
rewriten in our numerical coordinates (46), together with
the redefinition (47), looks like
(1 − z) z S00ω`m (z) + (m + 1)(2z − 1)S0ω`m (z)
− Λ − m(m + 1) − ã2 k̃ (2z − 1)2 Sω`m (z) = 0. (70)
 
 11

Expand the variables in series in m series expansion in ` = m for T`ˆm̂ω̂ (55). The normalisa-
tion η`ˆm̂ has to be obtained by plugging (75) in (56) and
S1 (z) S2 (z) integrating. The final result is
Sω`m (z) = 1 + + + O(m−3 ),
m m2 √ "
Λ = λ1 m2 + λ2 m + λ3 + O(m−1 ), 2 m(1−u+log u) 16 2 π r+ u
2 4
9 4
T`ˆm̂ω̂ = e m 1−
(1 + ã2 )2 µ̃3 ω3 (1 + ã2 )4 µ̃6 m
ω̃ = µ̃ − + 3 + O(m−4 ), (71)
8 m2 m
 
µ̃2 3 (1 + ã4 )(u − 1)2 + ã2 (u (6 u − 13) + 6)
−
plug them in and solve order by order, requiring regu- 6um #
larity of the solution. The known behaviour of ω̃ from 2 i µ̃(1 + ã ) + 3 µ̃ (1 + ã2 )2 log(u)
2 2 h 1 i
before is utilised. In this way one can obtain Si (z) and − + O 2 . (76)
2m m
λi to any order as a function of the ωi .
Afterwards we transform the radial part (16a) using With all ingredients present, we change variables in (54)
with (72) and WKB expand R`ˆm̂ for large ` = m. Match-
2 r+ m(m + 1) ing order by order resutls in
r= u. (72)
(1 + ã2 )µ̃2 √ "
2 2
2 m(1−u+log u) 2 π r+ u 5 2
The equation will not be given here - the above transfor- R`ˆm̂ = −e m 1−
(1 + ã2 )2 µ̃4 m
mation correponds to the first two terms in the expan-  
sion for large ` = m of the position of the minimum of µ̃2 3 (1 + ã4 )(u − 1)2 + ã2 (u (6 u − 13) + 6)
−
the radial potential. This is straightforward to get af- 6um
ter deriving the leading order behaviour of the angular
#
2 i µ̃(1 + ã ) + 3 µ̃ (1 + ã2 )2 log(u)
2 2 h 1 i
eigenvalue Λ = m(m + 1) + O(1), for which no knowledge − + O 2 , (77)
2m m
of ω3 or higher is required. Afterwards, we take
h R1 (u) i allowing us to to rebuild ψ4 for ` = m via (61).
Rω`m (u) = em f (u) 1 + + O(m−2 ) , (73) We also need an expression for the scalar cloud energy
m
Z +∞ Z π Z 2π
and solve order by order for f (u), Ri (u) and ωi , again √
Ms = −g T tt dφ dθ dr. (78)
demanding regularity of the solution. This can be done r+ 0 0
to any order, but we only give the first few here
Working with coordinates z and u, we substitute our ex-
S1 (z) = c2 , S2 (z) = c3 , pansions for the scalar field in the definition of Ms and
after simplifications the angular integrals can be done
f (u) = c0 − u + log u,
analytically - with Mathematica or with a book on inte-
µ̃2 1 grals. The resulting expression is expanded in ` = m and
(1 + ã2 )2 u + + 3 log u ,
 
R1 (u) = c1 − 2 2
4 u integrated over u from 2µ̃m(1+ã )
(m+1) to infinity, producing
(1 + ã2 )2 µ̃3
λ1 = λ2 = 1, λ3 = 0, ω3 = , (74) h 2
i
4 3 )µ̃2
e2 m π 2 Γ 3 + 2m, (1+ã
m+1
h i
whereby ci are constants that can for example be chosen Ms = 2 m− 52
1 + O(m−1 ) , (79)
24 m−1 m (1 + ã2 )3 µ̃4
so that the scalar field is unity at the maximum (u = 1),
which is what we do to compare with the numerics (where where we also have higher order terms, but the expres-
we can just divide by the numerical maximum). siosn are too cumbersome.
Next, we move onto the Teukolsky equation (54). The Having ψ4 and Ms in the large ` = m limit allows one
expressions are very lengthy and not illuminating at all, to derive the expression for χW KB in the main text by
so we will give only the most important ones. Treating dividing them and expanding in series in ` = m.
the angular equation (53) as the scalar one produces
h
| 2m−s | | 2m+s |
S
s `ˆm̂
= z 2 (1 − z) 2 1−
∗
2 ã µ̃ z(s + ã µ̃ (1 − z)) + d1 i fce21@cam.ac.uk
− + O(m−2 ) , †
bvg25@cam.ac.uk
m ‡
jss55@cam.ac.uk
A
s `ˆm̂ω̂
= 2 m (2 m + 1) − s(s + 1) + O(m−1 ), (75) [1] José M. M. Senovilla. Singularity Theorems and Their
Consequences. 2018. [Gen. Rel. Grav.30,701(1998)].
whereby we have accounted for m̂ = 2 m and d1 is a con- [2] R. Penrose. Gravitational collapse: The role of general
stant that can be chosen to make the normalisation of relativity. Riv. Nuovo Cim., 1:252–276, 1969. [Gen. Rel.
the function easier[47]. Afterwards we need to obtain a Grav.34,1141(2002)].

[3] James M. Bardeen, William H. Press, and Saul A Teukol-
sky. Rotating black holes: Locally nonrotating frames,
energy extraction, and scalar synchrotron radiation. As-
trophys. J., 178:347, 1972.
[4] A. A. Starobinsky. Amplification of waves reflected from
a rotating ”black hole”. Sov. Phys. JETP, 37(1):28–32,
1973. [Zh. Eksp. Teor. Fiz.64,48(1973)].
[5] T. Damour, N. Deruelle, and R. Ruffini. On Quantum
Resonances in Stationary Geometries. Lett. Nuovo Cim.,
15:257–262, 1976.
[6] T. J. M. Zouros and D. M. Eardley. Instabilities of mas-
sive scalar perturbations of a rotating black hole. Annals
Phys., 118:139–155, 1979.
[7] Steven L. Detweiler. Klein-Gordon equation and rotating
black holes. Phys. Rev., D22:2323–2326, 1980.
[8] Sam R. Dolan. Instability of the massive Klein-Gordon
field on the Kerr spacetime. Phys. Rev., D76:084001,
2007.
[9] Hirotaka Yoshino and Hideo Kodama. Gravitational ra-
diation from an axion cloud around a black hole: Super-
radiant phase. PTEP, 2014:043E02, 2014.
[10] Richard Brito, Vitor Cardoso, and Paolo Pani. Black
holes as particle detectors: evolution of superradiant in-
stabilities. Class. Quant. Grav., 32(13):134001, 2015.
[11] Richard Brito, Vitor Cardoso, and Paolo Pani. Superra-
diance. Lect. Notes Phys., 906:pp.1–237, 2015.
[12] Yakov Shlapentokh-Rothman. Exponentially growing fi-
nite energy solutions for the Klein-Gordon equation on
sub-extremal Kerr spacetimes. Commun. Math. Phys.,
329:859–891, 2014.
[13] Roy P. Kerr. Gravitational field of a spinning mass as
an example of algebraically special metrics. Phys. Rev.
Lett., 11:237–238, 1963.
[14] David Robinson. Four decades of black holes uniqueness
theorems. In Kerr Fest: Black Holes in Astrophysics,
General Relativity and Quantum Gravity Christchurch,
New Zealand, August 26-28, 2004, 2004.
[15] Otis Chodosh and Yakov Shlapentokh-Rothman. Time-
Periodic Einstein–Klein–Gordon Bifurcations of Kerr.
Commun. Math. Phys., 356(3):1155–1250, 2017.
[16] Carlos A. R. Herdeiro and Eugen Radu. Kerr black holes
with scalar hair. Phys. Rev. Lett., 112:221101, 2014.
[17] Bogdan Ganchev and Jorge E. Santos. Scalar Hairy Black
Holes in Four Dimensions are Unstable. Phys. Rev. Lett.,
120(17):171101, 2018.
[18] Misao Sasaki and Hideyuki Tagoshi. Analytic black hole
perturbation approach to gravitational radiation. Living
Rev. Rel., 6:6, 2003.
[19] Richard Brito, Shrobana Ghosh, Enrico Barausse,
Emanuele Berti, Vitor Cardoso, Irina Dvorkin, Antoine
Klein, and Paolo Pani. Gravitational wave searches for
ultralight bosons with LIGO and LISA. Phys. Rev.,
D96(6):064050, 2017.
[20] The large m limit does not commute with the extremal
limit. It can be shown that at any finite value of m,
Im(ω) = 0 at extremality. However, the region of the
black hole moduli space where this behaviour occurs
scales inversely with m, and is thus absent in the strict
m → +∞ limit. The behaviour in this region could be
investigated by switching the order of limits.
[21] Hironobu Furuhashi and Yasusada Nambu. Instability of
massive scalar fields in Kerr-Newman space-time. Prog.
Theor. Phys., 112:983–995, 2004.
[22] Felicity C. Eperon, Harvey S. Reall, and Jorge E. San-
12
tos. Instability of supersymmetric microstate geometries.
JHEP, 10:031, 2016.
[23] Óscar J. C. Dias, Jorge E. Santos, and Benson Way.
Numerical Methods for Finding Stationary Gravitational
Solutions. Class. Quant. Grav., 33(13):133001, 2016.
[24] The only additional complication is that we are searching
for extremely small growth rates when m increases, so
using extended precision is mandatory.
[25] William E. East and Frans Pretorius. Superradiant Insta-
bility and Backreaction of Massive Vector Fields around
Kerr Black Holes. Phys. Rev. Lett., 119(4):041101, 2017.
[26] William E. East. Superradiant instability of massive vec-
tor fields around spinning black holes in the relativistic
regime. Phys. Rev., D96(2):024004, 2017.
[27] Saul A. Teukolsky. Perturbations of a rotating black hole.
1. Fundamental equations for gravitational electromag-
netic and neutrino field perturbations. Astrophys. J.,
185:635–647, 1973.
[28] William H. Press and Saul A. Teukolsky. Perturbations
of a Rotating Black Hole. II. Dynamical Stability of the
Kerr Metric. Astrophys. J., 185:649–674, 1973.
[29] S. A. Teukolsky and W. H. Press. Perturbations of a ro-
tating black hole. III - Interaction of the hole with grav-
itational and electromagnet ic radiation. Astrophys. J.,
193:443–461, 1974.
[30] Oscar J. C. Dias, Gary T. Horowitz, and Jorge E. Santos.
Black holes with only one Killing field. JHEP, 07:115,
2011.
[31] Benjamin E. Niehoff, Jorge E. Santos, and Benson Way.
Towards a violation of cosmic censorship. Class. Quant.
Grav., 33(18):185012, 2016.
[32] Paul M. Chesler and David A. Lowe. Nonlinear Evolution
of the AdS4 Superradiant Instability. Phys. Rev. Lett.,
122(18):181101, 2019.
[33] This decay can be very complicated to determine and is
not exponential with time.
[34] Óscar J. C. Dias, Jorge E. Santos, Harvey S. Reall, and
Bogdan Ganchev. In preparation.
[35] Joe Keir. Slowly decaying waves on spherically symmet-
ric spacetimes and ultracompact neutron stars. Class.
Quant. Grav., 33(13):135009, 2016.
[36] Joseph Keir. Wave propagation on microstate geome-
tries. Annales Henri Poincare, 21(3):705–760, 2019.
[37] For unstable modes, this condition is equivalent to re-
quiring the modes to be outgoing at spatial infinity.
[38] Gustav Holzegel and Jacques Smulevici. Quasimodes and
a Lower Bound on the Uniform Energy Decay Rate for
Kerr-AdS Spacetimes. Anal. PDE, 7(5):1057–1090, 2013.
[39] Multiply the equation by (1 + x), then change variables
to u = (x + ã2 )/(1 − ã2 ).
[40] First multiply (18) by 1 + x, then perform transfor-
mation (25), followed by dividing the resulting equa-
2 2 3
tion byh (1 + x)(xp+ ã )(t0 + t1 ix + t2 x + t3 x +
4 2 3
2 x )/ 2 x (x + i0 ) x(x + i1 ) + i20 , with t0 = i0 (1 +
ã2 ) − 2 ã2 i20 + ã2 i0 i1 , t1 = i0 (2 i1 (1 + ã2 ) + 2 ã2 ) −
i20 (1 + ã2 ) + 2 i30 − ã2 i1 , t2 = 3 i0 (1 + ã2 + i1 ), t3 =
1+ã2 +4 i0 +i1 , i0 = d1 /(2 k̃), i1 = d2 /k̃, and expand
near spatial infinity (x = 1).
[41] Huan Yang, David A. Nichols, Fan Zhang, Aaron
Zimmerman, Zhongyang Zhang, and Yanbei Chen.
Quasinormal-mode spectrum of Kerr black holes and its
geometric interpretation. Phys. Rev., D86:104006, 2012.
[42] Sam R. Dolan. The Quasinormal Mode Spectrum of

a Kerr Black Hole in the Eikonal Limit. Phys. Rev.,
D82:104003, 2010.
[43] V. P. Frolov and I. D. Novikov, editors. Black hole
physics: Basic concepts and new developments, vol-
ume 96. 1998.
[44] William Kinnersley. Type D Vacuum Metrics. J. Math.
Phys., 10:1195–1203, 1969.
[45] Due to the aforementioned symmetry of the Teukolsky
13
equation, we can work, without loss of generality, with
the case of ω̂ = 2 ω, m̂ = 2 m only.
[46] Emanuele Berti, Vitor Cardoso, and Marc Casals.
Eigenvalues and eigenfunctions of spin-weighted
spheroidal harmonics in four and higher dimensions.
Phys. Rev., D73:024013, 2006. [Erratum: Phys.
Rev.D73,109902(2006)].
[47] Depending on how one does the numerics, s S̄`ˆm̂ inside
T4 might also need to be normalised.