Mode stability for the Teukolsky equations

on Kerr-anti-de Sitter spacetimes

Olivier Graf ∗ , Gustav Holzegel ∗, †
∗
Westfälische Wilhelms-Universität Münster, Mathematisches Institut,
Einsteinstrasse 62 48149 Münster, Bundesrepublik Deutschland

†
Imperial College London, Department of Mathematics,
South Kensington Campus, London SW7 2AZ, United Kingdom
arXiv:2205.02801v1 [gr-qc] 5 May 2022

May 6, 2022

Abstract
We prove that there are no non-stationary (with respect to the Hawking vectorfield) real mode solutions
to the Teukolsky equations on all (3+1)-dimensional subextremal Kerr-anti-de Sitter spacetimes. We further
prove that stationary solutions do not exist if the black hole parameters satisfy the Hawking-Reall bound
√  √
and a −Λ < 203 . We conclude with the statement of mode stability which preludes boundedness and
decay estimates for general solutions which will be proven in a separate paper. Our boundary conditions
are the standard ones which follow from fixing the conformal class of the metric at infinity and lead to a
coupling of the two Teukolsky equations. The proof relies on combining the Teukolsky-Starobinsky identities
with the coupled boundary conditions. In the stationary case the proof exploits elliptic estimates which fail
if the Hawking-Reall bound is violated. This is consistent with the superradiant instabilities expected in
that regime.

1 Introduction and main results

The Kerr-anti-de Sitter spacetimes (Kerr-adS) are solutions to the Einstein equations

Ric(g) = Λg, (1.1)

with negative cosmological constant Λ =: −3k 2 < 0, given by

∆
2 Σ Σ ∆ϑ
2
gKadS := − 2
dt − a sin2 ϑdϕ + dr2 + dϑ2 + 2 sin2 ϑ adt − (r2 + a2 )dϕ ,
Ξ Σ ∆ ∆ϑ Ξ Σ
where

Σ := r2 + a2 cos2 ϑ, ∆ := (r2 + a2 ) 1 + k 2 r2 − 2M r,

∆ϑ := 1 − a2 k 2 cos2 ϑ, Ξ := 1 − a2 k 2 .

The following parameters define Kerr-adS spacetimes containing a non-degenerate subextremal black hole region.
Definition 1.1. The parameters (M, a, k) belong to the set of admissible subextremal Kerr-adS black hole
parameters if 1

M, k > 0, a ≥ 0, ak < 1,

and if the set {r ≥ 0 : ∆(r) = 0} has two distinct elements. The maximum element r+ (M, a, k) > 0 is called
the black hole radius.
1 Up to making the change of variable ϕ → −ϕ, we can always assume that a ≥ 0. The constant k can be chosen to be positive

by definition.

1
 We consider the metric gKadS with parameters as above on the manifold MKadS := Rt × (r+ , +∞)r × S2ϑ,ϕ .
It is well-known that (MKadS , gKadS ) can be smoothly extended to a larger manifold containing a black hole
region and with MKadS embedding isometrically as its domain of outer communications. In particular, we can
attach a null boundary r = r+ to MKadS , which is known as the event horizon. See Figure 1 below.

Unlike their Λ > 0 and Λ = 0 counterparts, whose exterior stability properties have been intensely stud-
ied over the past two decades [Vas13, Dya16, HV18, Sch16, Mav21a, Mav21b, Fan21, Fan22] and [DHR19b,
ABBM19, HHV21, DHRT21, KS20, KS21], the stability properties of Kerr-adS spacetimes have remained more
elusive. A first distinct feature of the Λ < 0 case is that stability is to be understood in the context of an initial
boundary value problem for the Einstein equations, for which boundary conditions have to be imposed at the
conformal infinity of the spacetime (see [Fri95, EK19]). It is expected that stability will strongly depend on the
choice of boundary conditions (see [HLSW20]). A second distinct feature is that the nature of the (in)stability
properties of the Kerr-adS black hole will crucially be tied to conditions on its parameters (the mass M , the
specific angular momentum a and the cosmological scale k).

This rich dynamical behaviour can already be illustrated in the context of the so-called “toy-stability prob-
lem”, which is the problem of understanding the asymptotic properties of solutions to the conformal covariant
wave equation on Kerr-adS gKadS ψ +2k 2 ψ = 0. For Kerr-adS spacetimes satisfying the so-called Hawking-Reall
bound [HR99]
2
a ≤ kr+ , (1.2)

all solutions with Dirichlet boundary conditions decay inverse logarithmically in time [HS13] and this decay
rate is sharp for general solutions [HS14] (see also [Gan14]). This weak decay is due to a stable trapping mech-
anism caused by the reflecting Dirichlet boundary conditions. On the other hand, if the Hawking-Reall bound
is violated, [Dol17] established the existence of periodic and exponentially growing solutions. Geometrically,
the Hawking-Reall bound (1.2) guarantees the existence of a globally causal Killing vectorfield, the absence
of which allows for so-called superradiant instabilities.2 Note that if the Dirichlet conditions are replaced by
optimally dissipative boundary conditions, strong inverse polynomial decay is expected [HLSW20]. The expo-
nential growth or the (too) slow decay of solutions with Dirichlet conditions suggested the following instability
conjecture.3

Conjecture ([HS13]). The Kerr-anti-de Sitter spaces are (non-linearly!) unstable solutions to the initial bound-
ary value problem for Einstein equations (1.1) with Dirichlet-type boundary conditions.
The present paper initiates a program to address the above conjecture and extend the results for the toy-
stability problem to the actual Einstein equations near the Kerr-adS family. Our first objective is to establish
the slow logarithmic decay rate for the linear problem to provide further support to the above conjecture:

Prove that Kerr-adS spacetimes satisfying the Hawking-Reall bound (1.2) are linearly stable solutions of the
Einstein equations (1.1) – i.e. perturbations decay inverse logarithmic in time (and generally not better) to a
linearised Kerr-adS spacetime – if Dirichlet-type conditions are imposed on the linear perturbation of the metric.

1.1 The Teukolsky equations

Unlike the toy-stability problem, the problem of linear stability for the Einstein equations immediately forces
one to address the question of gauge, i.e. which coordinate system the Einstein equations should be linearised
in. However, it is also well-known – in complete analogy with the asymptotically flat case – that there is a large
gauge independent aspect of linear stability that can be addressed. From the work of Teukolsky [Teu72] we
2 Superradiant instabilities are absent in the asymptotically flat case but appear for instance for the massive wave equation on

Kerr even with Λ = 0, the mass acting as a confinement mechanism just as the reflecting Dirichlet boundary conditions do in the
Kerr-adS case (see [Shl14]). For the full linearised Einstein equations on Kerr-adS, the existence of superradiant instabilities have
been obtained by a virial-type argument in [GHIW16] using the so-called canonical energy.
3 The (in)stability of the anti-de Sitter limiting space M = a = 0 with Dirichlet-type conditions is still an open question.

See [BR11] for numerical simulations revealing a turbulent instability mechanism in the context of the Einstein-scalar field sys-
tem, [Mos18, Mos20] for recent breakthrough results establishing the instability of anti-de Sitter space in the context of the
spherically symmetric Einstein-Vlasov system, and [CS22] for the construction of periodic solutions to non-linear wave equations
on anti-de Sitter space.

2
know that two of the linearised null curvature components4 α[+2] and α[−2] are gauge invariant to linear order
and satisfy the following decoupled wave equations (see [CT21a] for their expression in the general Λ 6= 0 case)
a2 k 2
 
s d∆ 2s a d∆ cos ϑ
0 = gKadS α[s] + 2 2 ∂r α[s] + 2 Ξ +i 2 +i cos ϑ ∂ϕ α[s]
Σ Ξ dr Σ Ξ 2∆ dr sin ϑ ∆ϑ
 2 2
  2
 (1.3a)
2s r + a d∆ cos ϑ [s] s 2 2 2 2 2 cot ϑ [s]
+ 2 − 2r − ia ∂t α + 2 2 1 + a k + 2(3 + 2s)k r − Ξ s α ,
Σ 2∆ dr ∆ϑ Σ Ξ ∆ϑ

for s = ±2. The Teukolsky quantities α[±2] are spin-weighted complex functions, for which we give the following
compact ad hoc definition (see [DHR19a, Section 2.2.1]). We refer the reader to [DHR19a, Mil21] for more
geometric definitions.
Definition 1.2. Let s ∈ R. A smooth s-spin-weighted complex function on MKadS is a function
α : Rt × (r+ , +∞)r × (0, π)ϑ × (2πT)ϕ → C such that α is smooth on Rt × (r+ , +∞)r × (0, π)ϑ × (2πT)ϕ
and such that eisϕ (Ze1 )k1 (Z e2 )k2 (Ze3 )k3 α and e−isϕ (Ze1 )k1 (Z
e2 )k2 (Ze3 )k3 α extend continuously at ϑ = 0 and ϑ = π
respectively for all k1 , k2 , k3 ∈ N, where
Ze1 := − sin ϕ∂ϑ + cos ϕ (−is csc ϑ − cot ϑ∂ϕ ) , Ze3 := ∂ϕ ,
Ze2 := − cos ϕ∂ϑ − sin ϕ (−is csc ϑ − cot ϑ∂ϕ ) .
Another question to be addressed is what “Dirichlet conditions” should mean for the system of linearised
Einstein equations. The most natural and fully geometric “Dirichlet conditions” for the Einstein equations in
the context of the initial boundary value problem of [Fri95, EK19] is to impose that the conformal class of
the induced metric on the boundary at infinity coincides with that of the anti-de Sitter metric. This directly
translates into the following boundary conditions for the null curvature components α[±2]

∗ r→+∞
∗ r→+∞
e[+2] − α
α e[−2] −−−−−→ 0, r2 ∂r α
e[+2] + r2 ∂r α
e[−2] −−−−−→ 0, (1.3b)
∗
where e[±2] are the following renormalisations
denotes the complex conjugation, and where α
e[+2] := ∆2 (r2 + a2 )−3/2 α[+2] ,
α e[−2] := (r2 + a2 )−3/2 α[−2] .
α
We call (1.3b) the conformal anti-de Sitter boundary conditions for the Teukolsky quantities, and we emphasise
that they couple the dynamical variables at the boundary. We refer the reader to [HLSW20, Section 6] for the
details on the derivation of (1.3b). Moreover, we point that the boundary conditions (1.3b) are also the most
natural ones in the context of metric perturbation using Hertz maps [DRS09, DS13], which is expected given
the duality between the two approaches [Wal78].

One usually restricts to study metric perturbations which are regular at the event horizon. For the Teukolsky
quantities α[±2] this imposes the following regularity conditions.
Definition 1.3. Let α[±2] be two smooth ±2-spin weighted quantities. Let r∗ be the tortoise coordinate given
by
dr∗ r2 + a2 π
= , r∗ (r = +∞) = ,
dr ∆ 2
and let t∗ be the regular time function given by 5
1
t∗ := t + r∗ − arctan(kr).
k
We say that α[±2] are regular at the event horizon in the future of t∗0 ∈ R if, for all t∗1 ≥ t∗0 , we have
 
p −1

q  [+2]   p −1

q  −2 [−2]  
L ∆ L αe 
 = Or→r+ (1), L ∆ L ∆ α e 
 = Or→r+ (1), (1.3c)
t∗ =t∗
1 t∗ =t∗
1

4 The quantities α[+2] , α[−2] are obtained from null components of the Riemann tensor defined with respect to the so-called alge-

braically special null frame. They correspond of the respective horizontal tensors α, α in the Christodoulou-Klainerman framework
and to the respective quantities Ψ0 and (r − ia cos ϑ)4 Ψ4 in the Newman-Penrose formalism, where Ψ0 , Ψ4 are defined with respect
to the so-called Kinnersley tetrad.
5 Any other regular time function is equally acceptable. See also the presentation in [HS13, Section 2].

3
for all p, q ∈ N, and where the pair of null vectorfields L, L is defined by
aΞ aΞ
L := Ξ∂t + ∂ϕ + ∂r∗ , L := Ξ∂t + ∂ϕ − ∂r∗ .
r2 + a2 r2 + a2
Under these conditions, the Teukolsky equations give rise to a well-posed initial boundary value problem.
[±2] [±2]
Theorem 1.4. Let t∗0 ∈ R and let α0 , α1 be (smooth) initial data quantities on Σt∗0 := {t∗ = t∗0 }, compatible
with the boundary conditions at infinity (1.3b) and with the regularity conditions at the horizon (1.3c). Then,
there exist two (smooth) regular quantities α[±2] defined on the future J + (Σt∗0 ) of Σt∗0 in MKadS such that
[±2] [±2]
α[±2] |Σt∗ = α0 and ∂t α[±2] |Σt∗ = α1 , which solve the Teukolsky equations (1.3a) and which satisfy the
0 0
conformal anti-de Sitter boundary conditions at infinity (1.3b) and the regularity conditions at the horizon in
the future of t∗0 (1.3c).6
Proof. Defining

∗
∗
αD := ∆2 α[+2] − α[−2] , αN := ∆2 α[+2] + α[−2] ,

the system (1.3) falls in the framework of strong hyperbolic operators with homogeneous (respectively Dirichlet
and Neumann) boundary conditions for the unknowns (αD , αN ) as defined in [War15]. The result then follows
from Theorem 2.3 in [War15].

Figure 1: The spacetime MKadS and the initial boundary value problem of Theorem 1.4.

Theorem 1.4 tells us that the Teukolsky system (1.3) can be studied independently of having to address the
problem of gauge. In fact, it is even believed that the Teukolsky quantities control the full solutions to the
Einstein linearised equations up to pure gauge solutions and linearised Kerr(-adS) solutions. See [Wal73] where
it was shown, in the Λ = 0 case, that if α[±2] vanish identically, then the full solution is trivial (in the above
sense), and [DHR19b] where, in the Schwarzschild case, quantitative decay of the full solution (up to pure gauge
solutions and linearised Kerr) is obtained from the decay of α[±2] . With this in mind, in the Kerr-adS case, a
cornerstone of our objective mentioned before Section 1.1 is the following:

Show that the solutions to the Teukolsky system (1.3) decay inverse logarithmically in time provided that the
background Kerr-adS spacetime satisfies the Hawking-Reall bound (1.2).
6 See Figure 1 for a Penrose diagram depicting the setting of Theorem 1.4.

4
1.2 Mode stability on the real axis and the main theorem
In this paper, we prove mode stability for the initial boundary value problem of Theorem 1.4. Mode stability – in
conjunction with the theory of scattering resonances7 – has been a fundamental ingredient to obtain the decay
of solutions to wave-type equations: see for example [HHV21] for the proof of the linear stability of Kerr with
slow rotation |a|  M , or [HV18, Fan21, Fan22] for the proof of the linear and non-linear stability of the slowly
rotating Kerr-de Sitter solutions, which all rely on resonance expansions and the identification of obstructions
to mode stability. In another framework, quantitative versions of mode stability (see [Shl15, Tei20]) were used
in the proof of the decay of solutions to the wave and Teukolsky equations on Kerr in the full subextremal range
|a| < M in [DRS16] and [ST20].

For the Teukolsky equations (1.3), mode solutions take the following form.
Definition 1.5. Let ω ∈ C. We say that α[±2] are (smooth) modes with frequency ω if there exists (smooth)
time-independent functions α̂[±2] such that
?
α[+2] (t, r, ϑ, ϕ) = e−iωt α̂[+2] (r, ϑ, ϕ), α[−2] (t, r, ϑ, ϕ) = e+iω t α̂[−2] (r, ϑ, ϕ), (1.4)

for all (t, r, ϑ, ϕ) ∈ R × (r+ , +∞) × (0, π) × (0, 2π).

Remark 1.6. In the literature, Teukolsky mode quantities α[±2] are usually defined with the convention α[±2] =
e−iωt α̂[±2] . While any convention is equally acceptable when α[+2] and α[−2] are not coupled, the coupling of
these two quantities via the conformal anti-de Sitter boundary condition (1.3b) yields that the only convention
potentially leading to the existence of non-trivial modes and fitting into the framework of [War15] is the one of
Definition 1.5.
Remark 1.7. For solutions to the full linearised Einstein equations, there can be no non-trivial mode solution
with the convention of Definition 1.5 alone, due to the coupling of α[+2] and α[−2] via the Bianchi equations
(see [NP62] and the Teukolsky-Starobinsky constraint relations [SC74, TP74]). In that case the correct definition
of mode solution is sums of complex-conjugated frequencies mode solutions in the sense of Definition 1.5, i.e.
[+2] ∗ [+2] ∗ [−2] [−2]
α[+2] = e−iωt α̂− + e+iω t α̂+ , α[−2] = e+iω t α̂− + e−iωt α̂+ ,

where
[+2] [+2] [−2] ∗ [−2]
α− := e−iωt α̂− , α− := e+iω t α̂− ,

and
[+2] ∗ [+2] [−2] [−2]
α+ := e+iω t α̂+ , α+ := e−iωt α̂+ ,

are respectively mode solutions to the Teukolsky equations in the sense of Definition 1.5.
Mode stability is the statement that there do not exist regular modes which are solutions of the initial
boundary value problem for Teukolsky equations with Im(ω) ≥ 0. This article is dedicated to the proof of the
following theorem which establishes mode stability on the real axis.
Theorem 1.8 (Main theorem). Let (M, a, k) be admissible subextremal Kerr-adS black hole parameters in the
sense of Definition 1.1. There exists no non-zero smooth modes with real frequency ω ∈ R in the sense of
Definition 1.5 which are solutions of the Teukolsky equations (1.3a), which satisfy the conformal anti-de Sitter
boundary conditions at infinity (1.3b) and the regularity conditions at the horizon (1.3c), except for a potential
set of stationary modes with respect to the Hawking vectorfield
a
K(α[±2] ) = 0, K := ∂t + 2 + a2 ∂ϕ .
r+

Moreover, there are no non-trivial stationary solutions which are also time-independent, i.e. ∂t α[±2] = 0.
Finally, if the Hawking-Reall bound (1.2) is satisfied, and if one of the following two ad hoc hypothesis
1
ak ≤ , (1.5a)
20
7 See [DZ19] for a general introduction to this theory.

5
or
1 2
a≤ kr , (1.5b)
4 +
holds, there are no non-trivial stationary mode solutions.

Remark 1.9. Projecting the potentially non-trivial stationary mode solutions of Theorem 1.8 onto the Hilbert
basis (eimϕ )m∈Z of 2π-periodic in ϕ functions, we obtain directly that their associated frequencies must be of the
form

ω = mω+ ,
2
with m ∈ Z \ {0}, and where ω+ := a/(r+ + a2 ).
Remark 1.10. In view of the numerical results of [CDH+ 14], we expect that the second part of the theorem
actually holds without the assumptions (1.5). However, in view of the bounds obtained in the present paper, the
numerical results of [CDH+ 14] and the superradiant instabilities revealed in [GHIW16], we expect that our result
is sharp with respect to the Hawking-Reall bound, i.e. that mode stability does not hold if it is violated. Note
that this violation can happen even under the additional assumption (1.5a). See the discussions in Section 1.4
and Remark 5.3.
Theorem 1.8 is the main result of this paper and is proved in Sections 2 to 5. We provide below a brief
sketch of the proof. The first part of the result relies on classical ODE analysis. Under the assumptions of the
theorem one may use the full separability of the equations to decompose the mode solutions α e[±2] as
X X
αe[+2] (t, r, ϑ, ϕ) = e−iωt m`
R[+2],ω ω
(r)Sm` (ϑ)e+imϕ , αe[−2] (t, r, θ, φ) = e+iωt m`
R[−2],ω ω
(r)Sm` (ϑ)e−imϕ .
m` m`

m` m`
The radial functions R[+2],ω and R[−2],ω satisfy the same second order ODE and the regularity conditions (1.3)
impose that either ω = mω+ , or they live on two different branches of the Frobenius expansion at the horizon.
The key step is to prove that the Wronskian
 
 Rm` m`
R[−2],ω 
[+2],ω
W (r) = 
 
m` m`
 ∂r? R[+2],ω ∂r? R[−2],ω 


m` m`
vanishes when r → +∞, hence W (r) = 0 for all r. This implies that R[+2],ω and R[−2],ω are proportional
which is possible only if ω = mω+ . The vanishing of the Wronskian limit is obtained by combining the
m` m`
Teukolsky-Starobinsky identities – which relate R[+2],ω and the complex conjugate of R[−2],ω through a fourth
order differential operator – and the coupled boundary conditions (1.3b). Crucial to this argument is the strict
positivity of the angular Teukolsky-Starobinsky constant, which also requires a delicate analysis (see Lemma 3.1
and Remark 4.2).

The second part of the theorem is proven by angular and radial elliptic estimates (see Section 5). The
cornerstone is the positivity of a potential, which is linked to the Hawking-Reall bound (see Remark 5.3), and
is obtained in this paper using additionaly the hypothesis (1.5).

1.3 From mode stability on the real axis to full mode stability
It is well-known that one can exploit the continuity of the mode frequencies in the black hole parameters to
deduce the full statement of mode stability for (1.3) from mode stability on the real axis together with mode
stability for specific values of the parameters, e.g. the Schwarzschild case a = 0 (see [AMPW17]).8

For a large class of Schwarzschild-adS spacetimes, we can in fact obtain uniform boundedness statements
(and therefore mode stability for Im(ω) > 0) relatively easily by generalising the so-called Chandrasekhar
transformation theory of [DHR19a, DHR19b] in the asymptotically flat case. We recall that the main difficulty
to obtain uniform boundedness statements for (1.3), even in the case a = 0, is that there is no natural conserved
8 The general idea is that modes “have to cross the real axis” to become exponentially growing modes.

6
energy or integrated local energy decay estimate for solutions. Using the Chandrasekhar transformation, one
obtains for a = 0 the following Regge-Wheeler system
 
[±2] ∆ [±2] 6M
0 = LLΨ + 4 L − Ψ[±2] , (1.6a)
r r

and

∗ r→+∞
∗ r→+∞
Ψ[+2] − Ψ[−2] −−−−−→ 0, r2 ∂r Ψ[+2] + r2 ∂r Ψ[−2] −−−−−→ 0, (1.6b)

where
r4 r4 ∆2 [+2] r4 r4
     
[+2] [−2] 1 [−2]
Ψ := L L α , Ψ := L L α , (1.7)
∆ ∆ r3 ∆ ∆ r3

and where L[±2] are angular Laplace-type elliptic operators with eigenvalues `(` + 1) for ` ≥ 2. Let us introduce
the two null coordinates u := t−r∗ , v := t+r∗ , so that in the (u, v, ϑ, ϕ) coordinate system we have ∂v = 21 L and
∂v = 12 L. Standard energy conservation for (1.6) using the boundary conditions (1.3b) yields the conservation
law (see Figure 2)

Ev0 [Ψ](u, v) = Ev0 [Ψ](u, v0 ) (1.8)

for any v0 ∈ R, v ≥ v0 and u ≥ v0 − π, and with the energy

X Z u n ∆
 p
6M
o
[s] 2 [s] [s] 2 [s] 2
Ev0 [Ψ](u, v) := k∂u Ψ kSū,v 2 + 4 k L Ψ kSū,v 2 − kΨ kSū,v 2 dū
r r
s∈{−2,2} u(r=∞,v)
Z vn  p o (1.9)
[s] 2 ∆ [s] [s] 2 6M [s] 2
+ k∂v Ψ kSu,v̄
2 + 4 k L Ψ kSu,v̄ 2 − kΨ kSu,v̄
2 dv̄ ,
v0 r r
2
√
where k ξkS 2 := S2 |ξ(u, v, ϑ, ϕ)|2 sin ϑdϑdϕ and where L[s] are the square root operators of L[s] .
R
u,v

Figure 2: The conservation law (1.8).

[±2]
Remark 1.11. Denoting by Ψm` the projections of Ψ[±2] on the eigenfunctions of L[±2] , the potential appearing
in (1.9) rewrites
X X  
L[±2] Ψ[±2] 2 2 − 6M Ψ[±2] 2 2 = 6M  [±2] 2
p
Su,v̄ Su,v̄
`(` + 1) − Ψm` .
r r
`≥2 |m|≤`

7
The energies (1.9) are therefore coercive provided the parameters are such that r+ ≥ M (and more generally if
Ψ[±2] are supported in a sufficiently large angular regime ` ≥ `c (M, k)). We highlight that for Schwarzschild-adS
spacetimes with fixed mass M , the black hole radius r+ can range from 0 (when k → +∞) to the Schwarzschild
radius 2M (when k → 0). Moreover – using a standard variational argument – it can actually be shown
that there exists non-trivial regular periodic and growing mode solutions to the Regge-Wheeler system (1.6) if
r+ is sufficiently small. However, these solutions cannot arise from regular mode solutions to the Teukolsky
system (1.3), see Theorem 1.16.
It is well-known that the conservation law (1.8) can be combined with a redshift estimate [DR09] to deduce
(for the parameter range r+ ≥ M ) the non-degenerate boundedness estimate
X Z ∞ n r4 ∆ p o
sup k∂u Ψ[s] k2Sū,v
2 + 4 k L[s] Ψ[s] k2Sū,v
2 + kΨ[s] k2Sū,v
2 dū
v≥v0 ∆ r
s∈{−2,2} u(r=∞,v)
X Z ∞ n r4 (1.10)
∆ p o
. k∂u Ψ[s] k2Sū,v
2 + 4 k L[s] Ψ[s] k2Sū,v
2 + kΨ[s] k2Sū,v
2 dū =: Ev0 [Ψ].
u(r=∞,v0 ) ∆ 0 r 0 0
s∈{−2,2}

Finally, using that the operators Z

ei defined in Definition 1.2 commute with (1.6) we conclude pointwise bound-
edness of Ψ[+2] and Ψ[−2] :
X X 2
X X
kΨ[s] kL∞ (J + (v0 )) . sup kΓi Ψ[s] kSu,v
2
v≥v0 ,u≥v0 −π
s∈{−2,2} s∈{−2,2} i=0 Γ∈{Z
e1 ,Z e3 }
e2 ,Z
v
u 2 (1.11)
uX X
.u
t Ev0 [Γi Ψ],
i=0 Γ∈{Z
e1 ,Z e3 }
e2 ,Z

2
where the first step follows from Sobolev embedding on Su,v and the second from one-dimensional Sobolev
embedding on constant v-hypersurfaces and the commuted version of (1.10).

From the estimate (1.11) we deduce the following theorem.

Theorem 1.12. For Schwarzschild-adS spacetimes with parameters satisfying r+ ≥ M , all sufficiently regular
solutions to the Teukolsky system (1.3) remain uniformly bounded. More specifically, for characteristic initial
data on v = v0 , we have the uniform boundedness estimate
[+2] −2 [−2]
α ∞ + + w α ∞ +
L (J (v=v0 )) L (J (v=v0 ))
e e
v
u 2
uX X (1.12)
. w−2 αe[−2] L∞ (v=v ) + L(w−2 α
e[−2] ) L∞ (v=v ) + u

t Ev0 [Γi Ψ],
0 0
i=0 Γ∈{Z
e1 ,Z e3 }
e2 ,Z

∆
where w := r4 . In particular, there exist no non-zero regular growing mode solutions Im(ω) > 0.
Remark 1.13. The estimates (1.12) are stated for initial data posed on a characteristic hypersurface v = v0
as it takes a particularly simple form. The analogous estimates for spacelike data are easily obtained from the
proof below.
Remark 1.14. The w-weights in (1.12) ensure that all quantities that appear in the norms are consistent with
the regularity conditions (1.3c) at the horizon and the asymptotics (1.3b) at infinity.
Remark 1.15. The estimate (1.12) loses many derivatives and is far from being optimal. However, it can be
proven in an entirely elementary fashion, which is what we focus on here. Improved boundedness statements not
losing derivatives will be discussed in our subsequent paper in the more general context of Kerr-adS spacetimes.
Proof of Theorem 1.12. By the estimate (1.11) it suffices to prove (1.12) with the square root on the right hand
side replaced by Ψ[+2] L∞ (J + (v=v0 )) + Ψ[−2] L∞ (J + (v=v0 )) and that is what we will do. From the definition
(1.7), we derive the following general transport estimates (see Figure 3):
kw−2 Le
α[−2] kL∞ (J + (v=v0 )) . kw−2 Le
α[−2] kL∞ ((v=v0 )) + kΨ[−2] kL∞ (J + (v=v0 )) ,
(1.13a)
kw−2 α
e[−2] kL∞ (J + (v=v0 )) . kw−2 α
e[−2] kL∞ ((v=v0 )) + kw−2 Le
α[−2] kL∞ (J + (v=v0 )) ,

8
and, integrating outwards from the conformal boundary,

kw−1 Le
α[+2] kL∞ (J + (v=u0 )) . kw−1 Le
α[+2] kL∞ (r=∞,u≥u0 ) + kΨ[+2] kL∞ (J + (v=v0 )) ,
(1.13b)
ke α[+2] kL∞ (r=∞,v≥v0 ) + kw−1 Le
α[+2] kL∞ (J + (v=u0 )) . ke α[+2] kL∞ (J + (v=v0 )) .

Using that w → k 2 when r → ∞, the boundary conditions (1.3b) imply

α[+2] kL∞ (r=∞,v≥v0 ) = kk 4 w−2 α

ke e[−2] kL∞ (r=∞,v≥v0 ) ,
kw−1 Le
α[+2] kL∞ (r=∞,v≥v0 ) = kk 2 w−2 Le
α[−2] kL∞ (r=∞,v≥v0 ) .

Using these identities and coupling the transport estimates (1.13), we deduce the desired estimate (1.12) and
this finishes the proof of the theorem.

Figure 3: The transport estimates (1.13).

Unfortunately, the above argument is not as simple in the case of general Schwarzschild-adS black holes –
in view of the potentially negative potential in (1.6) –, or in the case of non-vanishing angular momentum – in
which case the equations for Ψ[±2] involve a coupling to α[±2] . Nonetheless, the continuity of mode frequencies
together with the real axis mode stability of Theorem 1.8 enables to improve the mode stability result of
Theorem 1.12 to the following full mode stability statement.
Theorem 1.16. Let (M, a, k) be admissible subextremal Kerr-adS black hole parameters such that the Hawking-
Reall bound (1.2) and the additional hypothesis (1.5) hold. Then there exists no non-zero mode solutions to the
Teukolsky system (1.3) with Im(ω) ≥ 0
Proof. The Teukolsky system (1.3) belongs to the class of strongly hyperbolic operator with homogeneous
boundary conditions considered in [War15]. The Laplace-transformed Teukolsky operator and its dual are
therefore locally uniformly Fredholm, see [War15, Theorems 4.3 and 4.6, Equations (4.4) and (4.11)]. Using
the associated estimates one shows directly that the set of frequencies ω ∈ C associated to non-trivial modes
is continuous in the parameters (M, a, k) (see also [HV18, Proposition 5.11] and [Vas13, Paragraph 2.7] where
these arguments are carried out in greater generality).
Assume that there exists a mode solution with Im(ω) > 0 for admissible parameters (M, a, k) satisfying the
Hawking-Reall bound and the additional hypothesis. There exists a smooth path of such parameters joining
(M, a, k) to (1, 0, 1). Since r+ (1, 0, 1) = 1, by Theorem 1.12, there exists no mode with Im(ω) > 0 for these
parameters. Therefore, by continuity of the frequency ω ∈ C, there exists admissible parameters satisfying
the Hawking-Reall bound and the additional hypothesis such that the Teukolsky equations admit a real mode
Im(ω) = 0. This contradicts Theorem 1.8 and finishes the proof of the result.

9
1.4 Related work
In the Λ = 0 case, the history of mode stability took a significant turn with [Whi89] which showed mode
stability for Im(ω) > 0 on all subextremal Λ = 0 Kerr black holes. This has been extended to include mode
stability on the real axis for the wave equation in [Shl15] and for the Teukolsky equations in [AMPW17], and
to include the extremal case in [Tei20]. Recently, the transformation of Whiting [Whi89] has been given a new
profound interpretation in terms of spectral symmetries in [CT21a], which also allowed the authors to provide
partial mode stability results in the Kerr-de Sitter case Λ > 0. See also [Hin21] where mode stability results are
obtained for a specific class of such spaces. The general Λ > 0 case remains an open problem.

In the Λ < 0 case, a mathematical study of modes for the wave equation on Schwarzschild-adS spacetimes
was initiated in [Gan14]. Note also that for a general class of equations, boundary conditions and asymptot-
ically anti-de Sitter black holes, the papers [War15, Gan18] provided general definitions of modes and general
Fredholm alternative type results.

Numerical results for Teukolsky equations on Kerr-anti-de Sitter black holes have been obtained starting
with [CL01, MN02] in the Schwarzschild-adS case. We mention in particular the remarkable [CDH+ 14] where
modes for the full range of admissible Kerr-adS black hole parameters are numerically computed and which
in particular corroborates Theorem 1.8 and Theorem 1.16. In that work, the main observation is that mode
stability holds as long as the Hawking-Reall bound (1.2) is satisfied, while, as soon as the Hawking-Reall bound
is violated, there exists growing modes Im(ω) > 0.9 These growing modes come from exponentially decaying
modes of the Hawking-Reall regime Im(ω) < 0, which have crossed the real axis Im(ω) = 0 exactly at stationary
frequencies ω = mω+ after leaving the Hawking-Reall regime. In particular, this gives evidence for the existence
of periodic stationary solutions when the Hawking-Reall bound is violated (see [DSW15] for further works in
that direction). We mention that the boundary conditions used in [CDH+ 14] are Robin boundary conditions
which were first derived in [DS13] and that these Robin conditions also play an important role in our proof (see
Section 4).

1.5 Final comments

We comment briefly on how the above results will be used for our program (see the discussions before and at
the end of Section 1.1). In a follow-up paper we will prove the following result:
Theorem (in preparation). All solutions to the Teukolsky system (1.3) on Kerr-adS spacetimes satisfying the
Hawking-Reall bound (1.2) and a slow rotation hypothesis (ak  1 is sufficient) are bounded and decay in time
with an inverse logarithmic decay rate.
The proof of the theorem will exploit that for sufficiently large angular parameters ` ≥ `c (M, a, k), the (inho-
mogeneous) Regge-Wheeler equations produced by the physical space Chandrasekhar transformation theory are
perturbations (under the slow rotation hypothesis) of decoupled wave equations with positive conserved energies.
See the Regge-Wheeler equations (1.6), the conserved energies (1.9) and Remark 1.11 in the Schwarzschild-adS
case. The boundedness and decay results of [Hol09, HS13] can be applied to such equations, and these prop-
erties can be integrated back to the original Teukolsky solutions (see the proof of Theorem 1.12). In the finite
angular regime 2 ≤ ` ≤ `c , where the Regge-Wheeler potential is negative, the above analysis fails. However,
here Theorem 1.8 or the full mode stability result of Theorem 1.16 in a resonance expansion of the Teukolsky
quantities can be used to obtain directly the boundedness and decay.

Finally, a third paper will be devoted to obtaining the full linear stability by linearising the Einstein equa-
tions in a suitable gauge and using the above theorem.

Acknowledgements. Both authors thank Sam Collingbourne, Christopher Kauffman, Rita Teixeira da Costa
and Claude Warnick for interesting discussions around this project. Special thanks to Allen Fang for precious in-
sights on resonance theory and pointing out references for Theorem 1.16. O.G. thanks Corentin Cadiou for useful
Mathematica coding suggestions. G.H. acknowledges support by the Alexander von Humboldt Foundation in
the framework of the Alexander von Humboldt Professorship endowed by the Federal Ministry of Education and
9 The closer one is from the Hawking-Reall regime, the larger the azimuthal number m of the growing mode must be. This is
2 provided that |m| → +∞. See Remark 5.3.
consistent with our observations of the degeneracies of the potential right when a > kr+

10
Research as well as ERC Consolidator Grant 772249. Both authors acknowledge funding through Germany’s
Excellence Strategy EXC 2044 390685587, Mathematics Münster: Dynamics-Geometry-Structure.

The rest of this article (Sections 2 to 5) is dedicated to the proof of Theorem 1.8.

2 The radial and angular Teukolsky equations

Following [Kha83], we have preliminary definitions of frequency-dependent radial and angular operators.
Definition 2.1. Let ω ∈ R and m ∈ Z. Let

K := am − ω(r2 + a2 ), H := aω sin ϑ − m csc ϑ.

For n ∈ Z, we define the following radial operators

ΞK ΞK
Dn := ∂r + i + n∂r log ∆, Dn† := ∂r − i + n∂r log ∆,
∆ ∆
and the following angular operators10
ΞH p ΞH p
Ln := ∂ϑ − + n∂ϑ (log( ∆ϑ sin ϑ)), L†n := ∂ϑ + + n∂ϑ (log( ∆ϑ sin ϑ)).
∆ϑ ∆ϑ
Let define the following radial and angular decompositions.
ω
Lemma 2.2. Let ω ∈ R. There exists real functions Sm` (ϑ) and eigenvalues λωm` ∈ R indexed by ` ≥ 2, |m| ≤ `
which are solutions to the eigenvalue equation

0 = ∆ϑ L†−1 ∆ϑ L2 Sm`
p p
ω
+ −6aΞω cos ϑ + 6k 2 a2 cos2 ϑ + λω

ω m,ω ω ω
m` Sm` =: L [λm` ]Sm` , (2.1)

with asymptotics (together with the consistent asymptotics for derivatives)

ω
Sm` (ϑ) = Oϑ→0 ((1 − cos ϑ)(m+2)/2 ), ω
Sm` (ϑ) = Oϑ→π ((cos ϑ + 1)(m−2)/2 ), if m > 2,
ω
Sm` (ϑ) = Oϑ→0 ((1 − cos ϑ)(m+2)/2 ), ω
Sm` (ϑ) = Oϑ→π ((cos ϑ + 1)(2−m)/2 ), if |m| ≤ 2, (2.2)
ω (−2−m)/2 ω (2−m)/2
Sm` (ϑ) = Oϑ→0 ((1 − cos ϑ) ), Sm` (ϑ) = Oϑ→π ((cos ϑ + 1) ), if m < −2.
ω


The functions Sm` (ϑ)e+imϕ are smooth +2-spin weighted complex functions and the family Sm`ω
(ϑ)eimϕ `≥2,|m|≤`
forms an L2 -Hilbert basis of the smooth +2-spin weighted complex functions. Moreover, the eigenfunctions Sm`
ω
ω
and eigenvalues λm` are analytic in the admissible Kerr-adS black hole parameters (a, k) and in the frequency
ω, and in the Schwarzschild-adS case a = 0 we have

λω
m` = `(` + 1) − 2.

Proof. The proof is a consequence of Sturm-Liouville theory, see the proof of [Tei20, Proposition 2.1]. The
asymptotics (2.2) are given by the only regular branches obtained by Frobenius analysis at ϑ = 0 and ϑ = π.
Definition 2.3. Let ω ∈ R and let α[±2] be two smooth ±2-spin weighted complex
 modes  with frequency ω. We
m`
define the associated radial Teukolsky quantities to be the radial functions R[±2],ω (r) projections of
`≥2,|m|≤`
e[±2] on the Hilbert basis of Lemma 2.2. That is,
the renormalised functions α
X X
e[±2] (t, r, ϑ, ϕ) =
α e∓iωt e±imϕ R[±2],ω
m` ω
(r)Sm` (ϑ),
`≥2 |m|≤`

for all (t, r, ϑ, ϕ) ∈ R × (r+ , +∞) × (0, π) × (0, 2π).

Remark 2.4. The complex conjugate of a smooth ±2-spin weighted complex function is a smooth ∓2-spin
weighted complex function. The decomposition of Definition 2.3 in the −2-spin case is obtained using this
conjugacy.
10 The definitions of the two operators L and L† were wrongly interchanged in [Kha83].

11
 The Teukolsky equations satisfy the following separability properties.
Lemma 2.5 (Radial Teukolsky equations). Let ω ∈ R. Let α[±2] be two ±2-spin weighted complex modes with
frequency ω. Then, α[±2] satisfy the Teukolsky equations (1.3a) if and only if the associated radial Teukolsky
m`
quantities R[±2],ω as defined in Definition 2.3 satisfy the following radial Teukolsky equations

∆3
 2
(r + a2 )3/2 m`

Im,ω
0=e ω m`
[+2] [λm` ]R[+2],ω := Im,ω ω
[λ ] R[+2],ω ,
(r2 + a2 )7/2 [+2] m` ∆2
(2.3)
∆  
I−m,−ω
0=e[−2] [λω m,`
m` ]R[−2],ω := 2 I−m,−ω ω
[−2] [λm` ] (r 2
+ a2 3/2 m,`
) R [−2],ω ,
(r + a2 )7/2

where
†
Im,ω ω 2 2 ω
[+2] [λm` ] := ∆D1 D2 + 6iΞωr + 6k r − λm` ,

I−m,−ω
[−2] [λω † 2 2 ω
m` ] := ∆D−1 D0 + 6iΞωr + 6k r − λm` .

Moreover,
e[±2] satisfy the conformal anti-de Sitter boundary conditions at infinity (1.3b) if and only
 the functions α
if
 ∗  ∗
m` m`
R[+2],ω − R[−2],ω → 0, r2 ∂r R[+2],ω
m`
+ r2 ∂r R[−2],ω
m`
→ 0, (2.4)

when r → +∞,
e[±2] at the horizon imply
 if ω − mω+ 6= 0, the regularity conditions (1.3c) for the functions α
∗ ∗
−iΞ(ω−mω+ )r
m`
R[+2],ω ∼ Am`
[+2],ω e , ∆−2 R[−2],ω
m`
∼ Am`
[−2],ω e
+iΞ(ω−mω+ )r
, (2.5a)

as well as
∗
−iΞ(ω−mω+ )r
∂rn∗ R[+2],ω
m`
∼ (−iΞ(ω − mω+ ))n Am`
[+2],ω e ,

m,`
 ∗ (2.5b)
∂rn∗ ∆−2 R[−2],ω ∼ (iΞ(ω − mω+ ))n Am`
[−2],ω e
+iΞ(ω−mω+ )r
,

when r∗ → −∞, where n ∈ N and where Am`

[±2],ω are complex-valued constants,

e[±2] imply
 if ω − mω+ = 0, the regularity conditions (1.3c) at the horizon for the functions α
m`
R[±2],ω = ∆2 Am`
[±2],ω (r), (2.5c)

with Am`
[±2],ω : [r+ , +∞) → C smooth functions of r.

Proof. Equations (2.3) follow from [Kha83, Equations (3.19) and (3.21)].11 See also [CT21a, Equation (3.9)].
The equivalence from the boundary conditions (1.3b) and (2.4) is directly obtained from the angular and radial
decomposition of Definition 2.3. To prove (2.5a), (2.5b) and (2.5c) we first define
1
c2 (M, a, k) := 2 + a2 ∂r ∆(r = r+ ) > 0.
r+

e[+2] and the angular decomposition of Definition 2.3

From the regularity condition at the horizon (1.3c) for α
we have
 
m`
(−iΞ(ω − mω+ ) − ∂r∗ )R[+2],ω = L e−iωt eimϕ R[+2],ω
m`
+ Or→r+ (∆)
= Or→r+ (∆) (2.6)
2 ∗
= Or∗ →−∞ (ec (a,M,k)r
).
11 We think that the definition of Φ4 in [Kha83] should be Φ4 := (r − ia cos ϑ)4 ψ4 instead of Φ4 := (r − ia cos ϑ)3 ψ4 .

12
Integrating (2.6) in r∗ , we obtain that there exists a complex-valued constant Am`
[+2],ω such that
∗
m` −iΞ(ω−mω+ )r
R[+2],ω ∼ Am`
[+2],ω e ,

when r∗ → −∞. The analogous limit for ∆−2 R[−2],ωm`

as well as the higher regularity asymptotics follow similarly
and this finishes the proof of (2.5a) and (2.5b). In the particular case ω − mω+ = 0, we deduce from (2.6) that
m`
∂r R[+2],ω = Or→r+ (1),

and by using the regularity conditions (1.3c) further

∂rn R[+2],ω
m`
= Or→r+ (1), (2.7)

for all n ∈ N. A Frobenius analysis at the regular singular point r = r+ for the radial Teukolsky equation (2.3)
shows that there are two non-trivial solutions Rirreg and Rreg with

Rirreg (r) = f (r) + ∆2 log(∆)g(r), Rreg = ∆2 g(r),

where f, g are smooth functions on [r+ , +∞) and such that f (r+ ) = g(r+ ) = 1. We deduce from (2.7) that
m` m`
R[+2],ω must be proportional to Rreg . Along the same lines, we also have that R[−2],ω is proportional to Rreg
and this finishes the proof of (2.5c).
A direct computation gives the following relations for the radial Teukolsky operators.
Lemma 2.6. We have
 ∗  ∗
−m,−ω ω −m,−ω ω
Im,ω
e [λω
[+2] m` ] = I
e
[−2] [λm` ] = I
em,ω ω
[λ
[−2] m` ] = I
e
[+2] [λ m` ]
  2 2 4

(r + a )
= ∂r2∗ + ∂r∗ log ∂r∗ + V m,ω [λω
m` ],
∆2
with
∆ ∂r ∆ Ξ2 K 2
V m,ω [λω 8iΞωr + 6k 2 r2 − λω


m` ] := m` + 2iΞK + 2
(r2 2
+a ) 2 2
(r + a ) 2 2 (r + a2 )2
∆∂r ∆ ∆2
− 3r 2 2 3
+ 3(2r2 + a2 ) 2 .
(r + a ) (r + a2 )4

3 Teukolsky-Starobinsky transformations
ω
Lemma 3.1 (Angular Teukolsky-Starobinsky transformations). Let Sm` be a solution to (2.1) with λ = λω
m`
ω imϕ
such that Sm` e is a smooth +2-spin weighted function. Let define the associated Teukolsky-Starobinsky
ω
quantity Sm`,c by
p p p p
ω ω
Sm`,c := ∆ϑ L−1 ∆ϑ L0 ∆ϑ L1 ∆ϑ L2 Sm` .

Then, we have

∆ϑ L†2 Sm`,c
p p
0 = L−m,−ω [λω ω ω
+ 6aΞω cos ϑ + 6k 2 a2 cos2 ϑ + λω

ω
m` ]Sm`,c = ∆ϑ L−1 m` Sm`,c .

Moreover, we have the following formula

∆ϑ L†−1 ∆ϑ L†0 ∆ϑ L†1 ∆ϑ L†2 Sm`,c

p p p p
ω
= ℵ(m, `, λω ω
m` )Sm` ,

where ℵ(m, ω, λω
m` ) is the angular Teukolsky-Starobinsky constant defined by

ℵ(m, ω, λω ω 2 ω 3 ω 4 ω ω
m` ) := 4(λm` ) + 4(λm` ) + (λm` ) + 8aλm` (6 + 5λm` )(Ξm)(Ξω)
2 2
− 144a3 (Ξm)(Ξω) k 2 (−2 + λω 2
+ 4a4 k 2 (−6 + λω



m` ) + 2(Ξω) m` ) + 6(Ξω)
+ 4a2 k 2 λω ω ω 2 2
+ 8a2 6λω ω 2 2
(Ξω)2



m` −12 − 4λm` + (λm` ) + 24(Ξm) m` − 5(λm` ) + 18(Ξm)
> 0.

13
Proof. All these identities except the positivity of the angular Teukolsky-Starobinsky constant are direct compu-
tations and are left to the reader.12 To obtain the positivity of ℵ we revisit arguments from the proof of [CT21b,
Lemmas 2.9 and 2.11].13 Note that the a = 0 case is immediate from the expression of ℵ and the expression of
λω
m` (see Lemma 2.2) and we shall therefore assume that a > 0. Note also that the problem is invariant by the
transformation

ϑ → π − ϑ, m → −m, ω → −ω,

and we shall therefore assume m ≥ 0. By integration by parts we have

Z π Z πp
∆ϑ L†−1 ∆ϑ L†0 ∆ϑ L†1 ∆ϑ L†2 ∆ϑ L−1 ∆ϑ L0 ∆ϑ L1 ∆ϑ L2 SS sin ϑdϑ
p p p p p p p
2
ℵ |S| sin ϑdϑ =
0
Z0 π p p p p 2 (3.1)
=  ∆ϑ L−1 ∆ϑ L0 ∆ϑ L1 ∆ϑ L2 S  sin ϑdϑ ≥ 0,
 
0

from which we obtain ℵ ≥ 0. We also infer from (3.1) that ℵ(m, ω, λω

m` ) = 0 iff
p p p p
ω ω
Sm`,c = ∆ϑ L−1 ∆ϑ L0 ∆ϑ L1 ∆ϑ L2 Sm` = 0. (3.2)

In the m ≥ 2 case, a direct computation using the asymptotics (2.2) gives

 
−(m−2)/2 ω ϑ→0 2 −(m+2)/2 ω
(1 − cos ϑ) Sm`,c (ϑ) −−−→ −4Ξ (m − 1)m(m + 1)(m + 2) lim (1 − cos ϑ) Sm` (ϑ) ,
ϑ→0

Thus, (3.2) holds only if limϑ→0 (1 − cos ϑ)−(m+2)/2 Sm`

ω
(ϑ) = 0. From a Frobenius analysis of the ODE (2.1) at
ω
ϑ = 0 this implies that Sm` = 0.

In the m = 0 case, a computation gives

 
ϑ→0
(1 − cos ϑ)−1 S0`,c
ω
(ϑ) −−−→ (κ0 + κ00 ) lim (1 − cos ϑ)−1 S0`
ω
(ϑ) ,
ϑ→0
 
ϑ→π
(1 − cos ϑ)−1 S0`,c
ω
(ϑ) −−−→ (κ0 − κ00 ) lim (1 − cos ϑ)−1 S0`ω
(ϑ) .
ϑ→π

with

κ0 := λω ω 4 2 2 6 4 2 2
k 2 (−6 + λω 2
κ00 := 8aλω


0` (2 + λ0` ) − 24a k ω + 12a k ω + 2a 0` ) + 6ω , 0` Ξω.

ω
From a Frobenius analysis of (2.1) at ϑ = 0, π, identity (3.2) can hold for non-trivial S0`,c only if κ0 = κ00 = 0.
Using that a > 0, this only holds in the following four cases

m = 0, λ = 0, Ξω = ±k, (3.3a)
p
m = 0, λ = −1 − a2 k 2 ± 1 + 14a2 k 2 + a4 k 4 , Ξω = 0. (3.3b)

In the m = 1 case, a computation gives

 
−1/2 ω ϑ→0 −3/2 ω
(1 − cos ϑ) S1`,c (ϑ) −−−→ −κ1 lim (1 − cos ϑ) S1` (ϑ) ,
ϑ→0

with

κ1 := 12Ξ(6a2 k 2 + λ + 6aΞω).

Thus we shall assume that κ1 = 0 (i.e. λ = −6a2 k 2 − 6aΞω) otherwise S1`

ω
= 0. Under this assumption, a direct
computation gives that
ω
S1`,c = −8κ01 a3 sin2 ϑ∆ϑ−2 (f (ϑ)S1`
ω ω
(ϑ) − sin ϑ∆ϑ ∂ϑ S1` (ϑ)) , (3.4)
12 A Mathematica notebook containing the verifications of all the computations of the present article is included as an ancillary

arXiv file.
13 We think that there is small gap in that proof in the |m| ≤ 1 case which can be fixed by setting k = 0 in our argument.

14
with

κ01 := (ak 2 + Ξω)(k − ak 2 − Ξω)(k + ak 2 + Ξω)

f (ϑ) := 2 + cos ϑ + aΞω sin2 ϑ + a2 k 2 (1 − 2 cos ϑ − 3 cos2 ϑ + cos3 ϑ).

Assume that κ01 6= 0. Then, integrating the first order ODE given by combining (3.2) and (3.4), we have
   
ω −3/2 1/2 2 Ξω
S1` (ϑ) = C(1 − cos ϑ) (1 + cos ϑ) ∆ϑ exp − ak + arctanh(ak cos ϑ) ,
k

where C is the integration constant. The above does not satisfy the required asymptotic ∼ (1 − cos ϑ)3/2 when
ϑ → 0 except if C = 0 and S1`ω
= 0. Thus, we must have κ01 = 0. The vanishing of κ1 and κ01 is equivalent to
the following three cases

m = 1, λ = 0, Ξω = −ak 2 , (3.5a)
m = 1, λ = ∓6ak, Ξω = ±k(1 ∓ ak). (3.5b)

We now prove that there are no solutions to equation (2.1) which are regular at ϑ = 0 and ϑ = π for triplets
(m, λ, Ξω) as given in the seven cases (3.3a) (3.3b) (3.5a) (3.5b).14 Using that Sm`
ω
is regular at ϑ = 0, π,
ω
multiplying (2.1) with S = Sm` and integrating by parts gives
Z π
∆ϑ |∂ϑ S|2 + G(ϑ)|S|2 sin ϑdϑ,


0= (3.6)
0

with
1 −1
G(ϑ) := ∆ − 8 + 34X 2 − 17X 4 ± 16X(1 − X 2 ) cos(ϑ) − 10X 2 cos(2ϑ)
4 ϑ
+ X 4 cos(4ϑ) + 16(1 − X 2 )2 csc(ϑ)2


in the (3.3a) cases,
p
G(ϑ) := 3 + X + X 2 ∓ 1 + 14X 2 + X 4 − X 2 cos(2ϑ) − 4∆−1 2 2
ϑ (1 − X ) + 4(1 − X ) csc(ϑ)
2
in the (3.3b) cases,
1 2 − 2X 2
G(ϑ) := 4(3 + X 2 ) + + 4(1 − X 2 )∆−1 −5 + 2X − X 2 + 4(−1 + X)X cos ϑ


ϑ
4 1 + cos ϑ
− 4X 2 cos(2ϑ) + 9(1 − X 2 ) csc(ϑ/2)2


in the (3.5a) case,
2
1 2 − 2X
G(ϑ) := 12 + 8X(±2 + X) + + 4∆−1 2
ϑ (1 − X )(−5 + 4X cos ϑ)
4 1 + cos ϑ
− 4X 2 cos(2ϑ) + 9(1 − X 2 ) csc(ϑ/2)2


in the (3.5b) cases,

where X := ak. The functions G are polynomials in X and x := cos ϑ which can be checked to be positive for
ω
−1 < x < 1. Thus, we deduce from (3.6) that S = Sm` = 0 if m = 0 or m = 1 and this finishes the proof of the
lemma.
m`
Lemma 3.2 (Radial Teukolsky-Starobinsky transformations). Let R[±2],ω be solutions to the radial Teukolsky
m`
equation (2.3). Let define the associated Teukolsky-Starobinsky quantities R[±2],ω,c by

m` ∆2 † 4

2 2 3/2 m`

R[+2],ω,c := (D ) (r + a ) R [−2],ω ,
(r2 + a2 )3/2 0
(3.7)
m` ∆2 † 4

2 2 3/2 m`

R[−2],ω,c := 2 (D ) (r + a ) R [+2],ω .
(r + a2 )3/2 0

Then, the Teukolsky-Starobinsky quantities satisfy the following Teukolsky equations

I−m,−ω
e
[+2] [λω m`
m` ]R[+2],ω,c = 0, Im,ω
e ω m`
[−2] [λm` ]R[−2],ω,c = 0, (3.8)
14 For these values (m, λ, Ξω) a direct computation gives that for any function S solution of Lm,ω [λ]S = 0 its Teukolsky-
√ √ √ √
Starobinsky transformation Sc = ∆ϑ L−1 ∆ϑ L0 ∆ϑ L1 ∆ϑ L2 S is identically zero (i.e. identity (3.2) is satisfied).

15
and we have the following inversion formulas

m` −1 ∆2 4
 
R[+2],ω = ℘(m, ω, λω
m` ) (D 0 ) (r 2
+ a2 3/2 m`
) R [−2],ω,c ,
(r2 + a2 )3/2
(3.9)
m` −1 ∆2 4
 
R[−2],ω = ℘(m, ω, λω
m` ) (D 0 ) (r 2
+ a2 3/2 m`
) R [+2],ω,c ,
(r2 + a2 )3/2

where ℘(m, ω, λω
m` ) is the radial Teukolsky-Starobinsky constant defined by

℘(m, ω, λω 2 2 ω
m` ) := 144M (Ξω) + ℵ(m, ω, λm` ).

Proof. The proof of the lemma are direct computations which are left to the reader.12 15
We have the following lemma providing asymptotics at the horizon for Teukolsky-Starobinksy transforma-
tions.
m`
Lemma 3.3 (Teukolsky-Starobinsky horizon asymptotics). Let R[±2],ω be solutions to the radial Teukolsky equa-
tion (2.3) satisfying the the regularity conditions (2.5a) and (2.5b). The Teukolsky-Starobinsky transformations
m`
R[+2],ω,c satisfy the following asymptotics.

 If ω − mω+ 6= 0, we have
∗
m`
R[+2],ω,c ∼ C(m, ω)Am`
[−2],ω e
+iΞ(ω−mω+ )r
,
−1 ∗ (3.10)
∆−2 R[−2],ω,c
m`
∼ ℘(m, ω, λω ∗
m` ) (C (m, ω)) Am`
[+2],ω e
−iΞ(ω−mω+ )r
,

when r∗ → −∞, where Am`

[±2],ω are the same constants as in (2.5a) and (2.5b), and where

2
+ a2 ) 2∂r ∆(r+ ) + 2iΞ(ω − mω+ )(r+
2
+ a2 )



C(m, ω) := − 2iΞ(ω − mω+ )(r+

2 
× (∂r ∆(r+ ))2 + 2Ξ(ω − mω+ )(r+
2
+ a2 ) ,

=: −iξ (Ξ(ω − mω+ )) C(m,

e ω),
2
with ξ := ∂r ∆(r+ ) + iΞ(ω − mω+ )(r+ + a2 ). Moreover, we have the following asymptotics for higher order
derivatives
∗
∂rn∗ R[+2],ω,c
m`
∼ (iΞ(ω − mω+ ))n C(m, ω)Am`
[−2],ω e
+iΞ(ω−mω+ )r
,
 
−1 m` ∗ (3.11)
∂rn∗ ∆−2 R[−2],ω,c
m`
∼ (−iΞ(ω − mω+ ))n ℘(m, ω, λω ∗
m` ) (C (m, ω)) A[+2],ω e−iΞ(ω−mω+ )r ,

when r∗ → −∞.
 If ω − mω+ = 0, we have
m`
R[±2],ω,c = ∆2 Am`
[±2],ω,c (r), (3.12)

where Am`
[±2],ω,c : [r+ , +∞) → C are smooth functions of r.

Proof. The asymptotics (3.10) and (3.11) in the R[+2],c case follow from (2.5b), the definition of the Teukolsky-
Starobinsky transformation and the following computations

∆2 † 4

2 2 3/2 2 +iΞ(ω−mω+ )r ∗
 ∗
r →−∞ 2 † 4

2 +iΞ(ω−mω+ )r ∗

(D0 ) (r + a ) ∆ e ∼ ∆ (D0 ) ∆ e
(r2 + a2 )3/2
r ∗ →−∞ ∗
 ∗

∼ ∆2 e−iΞ(ω−mω+ )r ∂r4 ∆2 e+2iΞ(ω−mω+ )r
15 As pointed out using Chamber-Moss coordinates in [DS13] (see also [CM94]), the angular and radial Teukolsky equations share

a similar structure and computations carry over from one equation to the other.

16
and
  ∗
 
∗ r →−∞ ∗
∂r4 ∆2 e+2iΞ(ω−mω+ )r ∼ 2∂r ∆ + 2iΞ(ω − mω+ )(r+ 2
+ a2 ) ∂r3 ∆e+2iΞ(ω−mω+ )r
r ∗ →−∞ 2
+ a2 ) ∂r ∆ + 2iΞ(ω − mω+ )(r+ 2
+ a2 )



∼ 2∂r ∆ + 2iΞ(ω − mω+ )(r+
 ∗

× ∂r2 e+2iΞ(ω−mω+ )r
r ∗ →−∞ 2
+ a2 ) ∂r ∆ + 2iΞ(ω − mω+ )(r+ 2
+ a2 )



∼ 2∂r ∆ + 2iΞ(ω − mω+ )(r+

 ∗

× 2iΞ(ω − mω+ )(r+ 2
+ a2 ) ∂r ∆−1 e+2iΞ(ω−mω+ )r
r ∗ →−∞ 2
+ a2 ) (∂r ∆)2 + (2Ξ(ω − mω+ )(r+ 2
+ a2 ))2



∼ − 2∂r ∆ + 2iΞ(ω − mω+ )(r+
∗
2
+ a2 ) ∆−2 e+2iΞ(ω−mω+ )r .


× 2iΞ(ω − mω+ )(r+
In the R[−2],c case the asymptotics are obtained from the R[+2],c case and the inversion formula (3.9). The
regularity (3.12) are direct consequences of the smoothness of ∆−2 R[±2] from (2.5c). This finishes the proof of
the lemma.
We have the following limits at infinity for the Teukolsky-Starobinsky transformations. The proof of the
following lemma is a direct calculation which is left to the reader.12
Lemma 3.4 (Teukolsky-Starobinsky transmission coefficients). Let m ∈ Z, ω ∈ R and ` ≥ |m|. Let define the
following three real coefficients
−2 −4
℘0 (m, ω, λω ω ω 2
m` ) := −2λm` − (λm` ) + 20a(Ξm)(Ξω) + 8k (1 + λω 2
m` )(Ξω) − 8k (Ξω)4
− 2a2 k 2 (−6 + λω 2


m` ) + 6(Ξω) ,

−4
℘1 (m, ω, λω 2ak 4 (Ξm) + k 2 (2 + λω 3


m` ) := 4k m` )(Ξω) − 2(Ξω) ,

−4
℘2 (m, ω, λω
m` ) := −4k − k 4 λω ω 2 ω 3
m` (1 + λm` )(Ξω) + k (2 + 3λm` )(Ξω) − 2(Ξω)
5

+ a(−3k 6 λω 4 2 2 6 ω 4 3


m` (Ξm) + 8k (Ξm)(Ξω) ) + 2a (k (3 + λm` )(Ξω) − 3k (Ξω) ) .
m`
For all functions R[±2],ω solutions to the radial Teukolsky equations (2.3) we have
m` m` m`
lim R[±2],ω,c = − (℘0 + 12iM Ξω) lim R[∓2],ω + i℘1 lim ∂r∗ R[∓2],ω ,
r→+∞ r→+∞ r→+∞
m` m` m`
(3.13)
lim ∂r∗ R[±2],ω,c = i℘2 lim R[∓2],ω − (℘0 − 12iM Ξω) lim ∂r∗ R[∓2],ω .
r→+∞ r→+∞ r→+∞

4 Robin boundary conditions and the ω − mω+ 6= 0 case

The following proposition shows that the conformal anti-de Sitter boundary conditions (1.3b) are equivalent
to Robin boundary conditions together with the requirement that the coupled radial Teukolsky quantity is
the (renormalised, complex conjugate of the) Teukolsky-Starobinsky transformation. These Robin boundary
conditions were first derived in [DS13].
m`
Proposition 4.1 (Robin boundary conditions). Let m ∈ Z, ω ∈ R and ` ≥ |m| and R[±2],ω : (r+ , +∞)r → C
be two smooth functions. The following three items are equivalent.
m`
1. R[±2],ω satisfy the Teukolsky equations (2.3), the regularity conditions (2.5) at the horizon and the con-
formal anti-de Sitter boundary conditions (2.4).
m`
2. R[+2],ω satisfies the Teukolsky equation (2.3), the regularity conditions (2.5) at the horizon and the fol-
lowing two (linearly dependent) Robin boundary conditions
m` m`
(℘0 + Z + 12iM Ξω)R[+2],ω − i℘1 ∂r∗ R[+2],ω → 0,
(4.1)
m` m`
i℘2 R[+2],ω + (−℘0 + Z + 12iM Ξω)∂r∗ R[+2],ω → 0,

17
 when r → +∞, and we have
 ∗
m`
R[−2],ω = Z −1 R[−2],ω,c
m`
, (4.2)

where
q
Z = Z± (m, ω, λω
m` ) := ± ℵ(m, ω, λω
m` ) − 12iM Ξω. (4.3)

3. Item 2 holds with R[±2] replaced by R[∓2] .

Proof. Assume that Item 1 holds. From the asymptotics of Lemma 3.3 for the Teukolsky-Starobinsky transfor-
mations at the horizon and a Frobenius argument, there exists Z ∈ C∗ such that

∗
R[−2] = Z −1 R[−2],c . (4.4)

Using the conformal anti-de Sitter boundary conditions (2.4) and relation (4.4) we infer
∗
R[+2] − R[−2] = R[+2] − Z −1 R[−2],c → 0,
∗
(4.5)
∂r∗ R[+2] + ∂r∗ R[−2] = ∂r∗ R[+2] + Z −1 ∂r∗ R[−2],c → 0,

when r → +∞. The Robin boundary conditions (4.1) follow from (4.5) and the limits (3.13) for the Teukolsky-
Starobinsky transformations. The Robin boundary conditions lead to non-trivial solutions only if they are
linearly dependent. A direct computation shows that this only occurs provided that Z takes one of two val-
ues (4.3). This finishes the proof of Item 2. That Item 2 implies Item 1 is obtained by rewinding the above
argument and using the results of Lemmas 3.2 and 3.3 for Teukolsky-Starobinsky transformations. The equiv-
alence between Item 1 and Item 3 is obtained along the exact same lines replacing R[±2] by R[∓2] .
We can now prove the main theorem in the non-stationary case.
Proof of Theorem 1.8 in the ω − mω+ 6= 0 case. Define the Wronskian W = W (R[+2] , R[−2] ). The functions
R[+2] and R[−2] satisfy the same second order ODE (see Equations (2.3) and Lemma 2.6) and we have

∆2
 
W (r) = 4 2 lim W (r) . (4.6)
k (r + a2 )4 r→+∞

Using the conformally anti-de Sitter conditions (2.4), we have

lim W (r) = lim R[+2] ∂r∗ R[−2] − R[−2] ∂r∗ R[+2]
r→+∞ r→+∞
  (4.7)
∗ ∗
= − lim R[+2] ∂r∗ R[+2] + R[+2] ∂r∗ R[+2] .
r→+∞

Using that ℵ > 0 by Lemma 3.1, we have

(℘0 , Z± + 12iM Ξω, ℘1 , ℘2 ) 6= (0, 0, 0, 0). (4.8)

Thus, at least one of the Robin boundary conditions (4.1) for R[+2] is non-trivial. This implies that either

R[+2] → 0, or ∂r∗ R[+2] + iσR[+2] → 0,

when r → +∞ and with σ ∈ R. Plugging these in (4.6), (4.7), we obtain that W = 0 and that R[+2] and
R[−2] are proportional. If ω − mω+ 6= 0, from the asymptotics conditions at the horizon (2.5), this implies that
R[+2] = R[−2] = 0 which proves Theorem 1.8 in the non-stationary case.
Remark 4.2. We emphasise that the conditions (4.8) appearing in the above proof are actually almost equivalent
to the non-vanishing of ℵ. First, given the definitions of ℘0 , ℘1 , ℘2 and ℵ (or also using the radial Teukolsky-
Starobinsky inversion formulas and (3.13)), it is easy to show that ℘20 + ℘1 ℘2 = ℵ. It is also easy to check
from their definitions that ℘1 = 0 implies ℘2 = 0. Thus, the conditions (℘0 , Z± + 12iM Ξω, ℘1 , ℘2 ) = (0, 0, 0, 0)
actually reduce to ℘1 = 0 and ℵ = 0. Moreover, the vanishing of ℘1 itself is not independent of the vanishing of
ℵ: one can check that ℘1 = 0 when the triplets (m, λ, Ξω) take the seven special values (3.3), (3.5) of the proof
of Lemma 3.1.

18
5 The stationary case ω − mω+ = 0
Theorem 1.8 splits into the following two propositions.
m`
Proposition 5.1 (The m = 0 case). Let m = 0 and ω = mω+ = 0. For all radial functions R[±2],m ,
solutions to the Teukolsky equations (2.3) satisfying the boundary conditions at infinity (2.4) and the regularity
m` m`
conditions (2.5c) at the horizon, we have R[+2],m = R[−2],m = 0.

Proposition 5.2 (The general m case and the Hawking-Reall bound). If the black hole parameters are such that
the Hawking-Reall bound (1.2) and either assumption (1.5a) or (1.5b) are satisfied, then the following holds.
m`
For all m ∈ Z and for all radial functions R[±2],ω , solutions to the Teukolsky equations (2.3) with ω = mω+
and satisfying the boundary conditions at infinity (2.4) and the regularity conditions (2.5c) at the horizon, we
m` m`
have R[+2],ω = R[−2],ω = 0.
Remark 5.3. The positivity of the potentials crucially used in the proof of Proposition 5.2 dramatically fails
to hold if the Hawking-Reall bound (1.2) is violated. In that case, the potentials can indeed be made arbitrarily
negative provided that the azimuthal number |m| is sufficiently large (see Remarks 5.5 and 5.10). In the case of
the wave equation, the same feature was the key to construct stationary (and growing) mode solutions in [Dol17].
The proof of Propositions 5.1 and 5.2 is postponed to Sections 5.4 and 5.5 respectively. It is based on an
energy identity and uses estimates on the angular eigenvalues which are respectively obtained in the following
Sections 5.1 and 5.3. In some specific case, we will also need Hardy estimates which are obtained in the following
Section 5.2.

5.1 Energy identity

We have the following energy identity.
m`
Lemma 5.4 (Energy identity). For all radial functions R = R[±2],ω , solutions to the Teukolsky equations (2.3)
with ω = mω+ satisfying the boundary conditions at infinity (2.4) and the regularity conditions (2.5c) at the
horizon, we have
Z π/2  
∆ 
mω+

e 2 + Vstat
|∂ R|
r∗
m e2+
[λ̃]|R| λ̃m` − λ̃ |R| dr∗ = 0,
e 2 (5.1)
−∞ (r2 + a2 )2

for all λ̃ ∈ R, where

2 2 2
e := (r + a ) R,
R
∆
and where
2
∆ − k 2 (r − r+ )2 (r + r+ )2

m ∆ Ξω+ m
Vstat [λ̃] := λ̃ +
(r2 + a2 )2 k (r2 + a2 )2
(∂r ∆)2 ∆ ∆∂r ∆ ∆2
2 + a2 k 2 − 6k 2 r2 − ∂r2 ∆ + r 2 − (2r2 − a2 ) 2


+ 2 2 2
+ 2 2 2 2 3
,
(r + a ) (r + a ) (r + a ) (r + a2 )4

and
 2
mω mω 2 2 Ξω+ m
λ̃m` + := λm` + −2−a k − .
k

Proof. Using the expressions and notations of Lemma 2.6, we have

(r2 + a2 )2 em,ω ω
 
∆
0= I[+2] [λm` ] R
e
∆ (r2 + a2 )2 (5.2)
 
= ∂r2∗ R m
e − Vstat [λ̃ωm` ] − iIm (V
m,ω ω
[λm` ]) R.
e

19
From (5.2) we infer
 
0 = ∂r∗ R∂ e∗ + R
e r∗ R e∗ ∂r∗ R e 2 − 2V m [λ̃ω ]|R|
e − 2|∂r∗ R| e 2. (5.3)
stat m`

Integrating (5.3), using the conditions at the horizon (2.5c) and the Robin conditions at infinity (4.1), we obtain
the desired (5.1) and this finishes the proof of the lemma.
Remark 5.5. From a direct computation, one can check that the renormalised angular eigenvalue λ̃ is the
leading order term of the stationary potential at infinity, i.e.
m mω r→+∞ mω
Vstat [λ̃ = λ̃m` + ] −−−−−→ k 2 λ̃m` + . (5.4)

The key to establishing the positivity of the potential (up to using Hardy estimates) is therefore to obtain a
mω
suitable lower bound on λ̃m` + . See Section 5.3 and Remark 5.10.

5.2 Hardy estimate

The following Hardy estimate is used in the proof of Proposition 5.2. It is designed to compensate the potentially
negative leading coefficient of the potential while keeping the lower coefficients positive (see also Remark 5.5).
2 2
Lemma 5.6 (Hardy estimate). For all radial functions R e such that R := (r +a ) R e satisfies the boundary
∆
conditions at infinity (2.4) and the regularity conditions (2.5c) at the horizon, we have
Z π/2 Z π/2
e 2 dr∗ ≥
|∂r∗ R| e 2 dr∗ .
VHardy [λ̃]|R| (5.5)
−∞ −∞

for all λ̃ ∈ R, where

   2
∆ d r − r+ r − r+
VHardy [λ̃] := λ̃ 2 2
− λ̃2 .
r + a dr r2 + a2 r2 + a2
Proof. Let define
r∗
!
r − r+
Z
∗ ∗ ∗,0 ∗,0
y(r ) := λ̃ 2 , f (r ) := exp y(r ) dr .
r + a2 −∞

We have
Z π/2   2
 −1 e[+2]  dr∗
0≤ f ∂r∗ f R
−∞
Zπ/2 Z π/2
e[+2] |2 dr∗ + e[+2] 2 dr∗
 
= |∂r∗ R ∂r∗ (log f )R
−∞ −∞
Z π/2  
+ e∗ ∂r∗ R
∂r∗ (log f ) R e[+2] + R e∗
e[+2] ∂r∗ R ∗
[+2] [+2] dr
−∞
Z π/2 Z π/2   h iπ/2
e[+2] |2 dr∗ + 2 e[+2] |2 dr∗ + ∂r∗ (log f )|R
= |∂ R
r∗ −∂r2∗ (log f ) + (∂r∗ (log f )) |R e[+2] |2 .
−∞ −∞ −∞

Using that y(r∗ ) → 0 when r∗ → π/2 and r∗ → −∞, and using the limits given by the conditions (2.4)
and (2.5c), the boundary terms in the above computation vanish and the bound (5.5) follows.

5.3 Estimates for the angular eigenvalues

Define the following two parameters
δX
X := ak, aω+ =: .
1+X
We have the following bounds on the angular eigenvalues.

20
Lemma 5.7. Let m ∈ Z. For all admissible Kerr-adS black hole parameters (M, a, k) and for all ` ≥ 2, the
following holds.
1. For m = 0, we have
mω
λ̃m` + ≥ 0. (5.6)

2. If the Hawking-Reall bound (1.2) holds, we have for |m| = 1

mω
λ̃m` + ≥ −(1 + X)2 − 2X(1 − X), (5.7)

and, for |m| ≥ 2,

mω
λ̃m` + ≥ −4(1 + X)2 + 4Ξ(1 − δ 2 )(1 − X)2 . (5.8)

3. If the Hawking-Reall bound (1.2) and the rotation restriction (1.5a) hold, we have for |m| = 1
mω
λ̃m` + ≥ 0, (5.9)

and, for |m| ≥ 2,

mω 5δ
λ̃m` + ≥ − . (5.10)
2

Remark 5.8. As it will be clear in the proof of Item 3, the positive bound (5.9) also holds for |m| ≥ 5. However,
in that case, the bound (5.10) is sufficient for the proof of Proposition 5.2. Also the bound (5.10) is not sharp
but sufficient for the proof of Proposition 5.2.
Before turning to the proof of Lemma 5.7, we have the following preliminary estimate.

Lemma 5.9. Let m ∈ Z. For all admissible Kerr-adS black hole parameters (M, a, k) and for all ` ≥ 2, we
have
Z π Z π
mω mω mω
λ̃m` + ≥ Ξ |∂ϑ Sm` + |2 sin ϑdϑ + F (ϑ)|Sm` + |2 sin ϑdϑ, (5.11)
0 0

where
Ξ2 Ξ
4m cos ϑ + (4 + m2 ) cos2 ϑ +


F (ϑ) := 2 X(1 + X)4m cos ϑ
∆ϑ sin ϑ ∆ϑ
Ξ
(1 − δ)(1 − X) (1 + X + δ(1 − X))m2 − 4mX cos ϑ .


+
∆ϑ
Proof. We drop the indices for simplicity. Multiplying (2.1) by S sin ϑ, integrating for ϑ ∈ (0, π) and using the
regularity conditions (2.2) at ϑ = 0, π, we have
Z π Z π p
− ∆ϑ L†−1 ∆ϑ L2 S + 6aΞω cos ϑ − 6a2 k 2 cos2 ϑ |S|2 sin ϑdϑ
p
|S|2 sin ϑdϑ =


λ
0
Z0 π (5.12)
∆ϑ |∂ϑ S|2 + G(ϑ)|S|2 sin ϑdϑ,


=
0

where
2
Ξ2

2 2 Ξω+ m
− a2 k 2 cos(2θ) + 4m cos ϑ + 4 + m2 cos2 ϑ



G(ϑ) := 2 + a k + 2
k ∆ϑ sin ϑ
Ξ Ξ
(1 − δ)(1 − X) (1 + X + δ(1 − X))m2 − 4mX cos ϑ .


+ X(1 + X)4m cos ϑ +
∆ϑ ∆ϑ

21
 We have
Z π Z π
∆ϑ |∂ϑ S|2 sin ϑ = Ξ|∂ϑ S|2 + X 2 sin2 ϑ|∂ϑ S|2 sin ϑdϑ


0 0
Z π
=Ξ |∂ϑ S|2 sin ϑdϑ
0
Z π  (5.13)
2 2
+X |∂ϑ (sin ϑS)| − 2 sin ϑ cos ϑS∂ϑ S − cos2 ϑ|S|2 sin ϑdϑ
Z π 0 Z π
2 2
≥Ξ |∂ϑ S| sin ϑdϑ + X cos(2ϑ)|S|2 sin ϑdϑ.
0 0

Combining (5.12) and (5.13) we obtain (5.11) and this finishes the proof of the lemma.
Proof of Lemma 5.7. In the m = 0 case, the proof of (5.6) follows directly from (5.11). We turn to the proof of
Item 2 and we assume that the Hawking-Reall bound (1.2) holds. This bound together with the definition of
the admissible Kerr-adS black hole parameter (see Definition 1.1), rewrites as

0 ≤ δ ≤ 1. (5.14)

Note also that that same definition imposes

0 ≤ X < 1. (5.15)

For m = 1 (the m = −1 case is obtained along the same lines), we have

2
Ξ2 X sin2 ϑ

F (ϑ) = 2 cos ϑ + 1 + − (1 + X)2
∆ϑ sin2 ϑ 1−X
Ξ
+ (1 − δ)(1 − X) ((1 + X + δ(1 − X)) − 4X cos ϑ)
∆ϑ (5.16)
Ξ
≥ −(1 + X)2 + (1 − δ)(1 − X) ((1 + X + δ(1 − X)) − 4X cos ϑ)
∆ϑ
≥ −(1 + X)2 − 2X(1 − X),

where in the last line we used (5.14) and (5.15). For m ≥ 2 (the m ≤ −2 case is obtained along the same lines),
we have
2
Ξ2 2X sin2 ϑ

F (ϑ) = m cos ϑ + 2 + − 4(1 + X)2
∆ϑ sin2 ϑ 1−X
Ξ
(1 − δ)(1 − X) (1 + X + δ(1 − X))m2 − 4mX cos ϑ


+
∆ϑ (5.17)
Ξ
≥ −4(1 + X)2 + (1 − δ)(1 − X) (4(1 + X + δ(1 − X)) − 8X cos ϑ)
∆ϑ
≥ −4(1 + X)2 + 4Ξ(1 − δ 2 )(1 − X)2 ,

where in the two last lines we used (5.14) and (5.15). Combining (5.16) and (5.17) with (5.11) this yields (5.7)
and (5.8) respectively and finishes the proof of Item 2.

We now turn to the proof of Item 3 and assume that (1.5a) holds, which together with the definition of
admissible Kerr-adS black hole parameters rewrite
1
0≤X≤ . (5.18)
20
Define
Ξ Ξ
(1 − δ)(1 − X) (1 + X + δ(1 − X))m2 − 4mX cos ϑ .


Fe(ϑ) := X(1 + X)4m cos ϑ +
∆ϑ ∆ϑ

22
Let us consider m = 1 (the m = −1 case is obtained along the same lines). Using (5.14) and (5.15), we have
Fe(ϑ) ≥ −4X(1 + X) − 2X(1 − X). (5.19)
We have
π Z π
Ξ2
Z
|∂ϑ S|2 sin ϑdϑ + 4m cos ϑ + (4 + m2 ) cos2 ϑ |S|2 sin ϑdϑ


Ξ 2
0 0 ∆ϑ sin ϑ
Z π Z π 
2 2 1 2 2

2
≥Ξ |∂ϑ S| sin ϑdϑ + 2 4m cos ϑ + (4 + m ) cos ϑ |S| sin ϑdϑ
0 sin ϑ
 0 Z π 
2 −2 −1 −2

2 3
=Ξ 1+ sin ϑ∂ϑ (sin ϑS) − 2 sin ϑ(sin ϑS) sin ϑ dϑ
0
≥ Ξ2 ,
where the third line is obtained by integration by parts, using the limits at ϑ → 0, π given by (2.2). Plugging
the above estimate and (5.19) into (5.11), we obtain that
λ̃ ≥ (1 − X 2 )2 − 4X(1 + X) − 2X(1 − X).
This polynomial is positive under the bound (5.18), and estimate (5.9) follows.

Let us consider m ≥ 2. Using (5.14) and (5.15), we have

Fe(ϑ) ≥ −4X(1 + X)m + (1 − δ)(1 − X) (1 + X + δ(1 − X))m2 − 4mX .


(5.20)
We have
π π
Ξ2
Z Z
2
4m cos ϑ + (4 + m2 ) cos2 ϑ |S|2 sin ϑdϑ


Ξ |∂ϑ S| sin ϑdϑ + 2
0 0 ∆ϑ sin ϑ
Z π Z π 
2 2 1 2 2

2
≥Ξ |∂ϑ S| sin ϑdϑ + 2 4m cos ϑ + (4 + m ) cos ϑ |S| sin ϑdϑ
0 sin ϑ
 0 Z π 
2 −m −1 −m

2 2m
=Ξ m−4+ sin ϑ∂ϑ (sin ϑS) − 2 sin ϑ(sin ϑS) sin ϑ dϑ
0
≥ Ξ2 (m − 4),
where the third line is obtained by integration by parts, using the limits at ϑ → 0, π given by (2.2). Plugging
the above estimate and (5.20) into (5.11), we obtain
λ̃ ≥ (1 − X 2 )2 (m − 4) − 4X(1 + X)m + (1 − δ)(1 − X) (1 + X + δ(1 − X))m2 − 4mX

= ((1 − δ)(1 − X)(1 + X + δ(1 − X))) m2 + (1 − X 2 )2 − 4X(1 + X) − 4X(1 − δ)(1 − X) m

(5.21)
− 4(1 − X 2 )2 .
Under the bounds (5.14) and (5.18), we have
(1 − δ)(1 − X)(1 + X + δ(1 − X)) ≥ 0, (1 − X 2 )2 − 4X(1 + X) − 4X(1 − δ)(1 − X) ≥ 0. (5.22)
With the bounds (5.22), the right-hand side of (5.21) is a polynomial of order 2 (or 1) increasing in m ≥ 2. We
thus have
λ̃ ≥ 4 ((1 − δ)(1 − X)(1 + X + δ(1 − X))) + 2 (1 − X 2 )2 − 4X(1 + X) − 4X(1 − δ)(1 − X)

− 4(1 − X 2 )2 (5.23)
2 2 2 2
= 4(1 − X) (1 − δ ) − 2(1 − X ) − 8X(1 + X).
Under the bound (5.18), we have
4(1 − X)2 ≥ 5/2, −2(1 − X 2 )2 − 8X(1 + X) ≥ −5/2. (5.24)
Using (5.24) in (5.23), we deduce
5 5δ
(1 − δ 2 ) − 1 ≥ − ,


λ̃ ≥
2 2
which concludes the proof of (5.10) and of the lemma.

23
Remark 5.10. Using the notations of the proof of Lemmas 5.7 and 5.9, we have by the min-max principle that
Z π 
mω+ 2 2


λ̃mm = min ∆ϑ (∂ϑ S) + G(ϑ)S sin ϑ dϑ
kSk=1 0
Z π 
(4m cos ϑ + (4 + m2 ) cos2 ϑ) 2
 
2
≤ min (∂ϑ S) + S sin ϑ dϑ
kSk=1 0 sin2 ϑ
Z π 
+ sup (Fe(ϑ) − a2 k 2 cos(2ϑ))S 2 sin ϑdϑ
kSk=1 0

≤ (|m| − 4) + sup (Fe(ϑ) − a2 k 2 cos(2ϑ)),

ϑ=0···π

where in the last line we used a comparison with spin-weighted harmonics in the a = 0 case. If the Hawking-Reall
bound is violated, i.e. δ > 1, we have

sup (Fe(ϑ) − a2 k 2 cos(2ϑ))

ϑ=0···π
 
Ξ Ξ 2

2
= sup X(1 + X)4m cos ϑ + (1 − δ)(1 − X) (1 + X + δ(1 − X))m − 4mX cos ϑ − X cos(2ϑ)
ϑ=0···π ∆ϑ ∆ϑ
≤ 4|m|X(1 + X) + 4|m|X(1 − X)(δ − 1) + (1 − δ)Ξ(1 − X)m2 + X 2 .

Thus, using that δ > 1, we infer

λ̃mω 2
mm ≤ (|m| − 4) + 4|m|X(1 + X) + 4|m|X(1 − X)(δ − 1) + (1 − δ)Ξ(1 − X)m + X
+ 2

|m|→+∞
(5.25)
−−−−−−→ −∞.
mω m mω
Since λ̃mm+ is the limit of the stationary potential Vstat [λ̃mm+ ] at infinity (see Remark 5.5), we deduce that – if
the Hawking-Reall bound is violated – the (limit of the) potential can be made arbitrarily negative provided that
|m| is sufficiently large. See also Remark 5.3. We emphasise that (5.25) holds for all admissible parameters
0 ≤ X < 1.

5.4 Proof of Proposition 5.1

From the energy identity of Lemma 5.4 and the estimates on the angular eigenvalues (5.6), the proof of Propo-
sition 5.1 follows provided that we can show that for all admissible Kerr-adS black hole parameters (M, a, k)
we have
m=0
Vstat [λ̃ = 0](r) > 0, for all r > r+ . (5.26)

Using the rescaling

r → M r, a → M a, k → k/M, (5.27)

the proof of (5.26) reduces to the M = 1 case. Let consider the a = 0 case. For all r ≥ r+ we have

(r2 + a2 )4 Vstat
m=0
[λ̃ = 0](r) = r5 −6 + 4r + 18k 2 r2

5
−6 + 4r+ + 18k 2 r+2


≥ r+
  
5 2
≥ r+ −6 + 4r+ + 18 −1
r+
> 0,

where in the last inequality we used that 0 < r+ < 2.

Let now consider the case a > 0. Let us define

x := r/a, x+ := r+ /a, X := ak.

24
From ∆(r+ ) = 0 we deduce
X(1 + x2+ )(1 + x2+ X 2 )
k = K(x+ , X) := , a = X/k. (5.28)
2x+
We can therefore swap the two black hole parameters (a, k) for the parameters (x+ , X). To determine the
domain of the parameters (x+ , X), let define

K0 (x+ , X) := −1 + 3x4+ X 2 + x2+ (1 + X 2 ).

We have the following lemma.

Lemma 5.11. Denote by A the set of admissible subextremal Kerr-adS black hole parameters (M, a, k) of
Definition 1.1. The map

θ : (x+ , X) 7→ (a = X/K(x+ , X), k = K(x+ , X))

is a bijection from B := {K0 ≥ 0} ∩ {0 < X < 1} onto the set of admissible Kerr-adS black hole parameters
A1 := {(a, k) : (1, a, k) ∈ A, a 6= 0}. We call B the set of admissible parameters (x+ , X).
Proof. By the definition of admissible parameters and the above definitions of x+ and X, we easily have

A1 = θ ({0 < X < 1} ∩ {x+ > 0}) .

Let 0 < X < 1 be fixed. A direct computation gives

X 0
∂x+ K(x+ , X) = K (x+ , X). (5.29)
2x2+

The map x+ 7→ K0 (x+ , X) is strictly increasing and K0 (0, X) = −1 < 0 thus there exists a unique xc (X) > 0
such that K0 (x+ , X) > 0 for all x+ > xc (X) and K0 (x+ , X) < 0 for all x+ < xc (X). Therefore, the map
φ : x+ 7→ K(x+ , X) strictly decreases on (0, xc (X)) and strictly increases on (xc (X), +∞). Since φ(x+ ) → +∞,
we deduce that φ is a bijection from [xc (X), +∞) onto φ ((0, +∞)) and this finishes the proof of the lemma.

Figure 4: The set of admissible parameters (x+ , X).

Using the new variables X = ak, x = r/a, we rewrite

a−4 k 2 (r2 + a2 )4 Vstat

m=0
[λ̃ = 0](r) = 4k 2 (1 + 4x2 ) + (1 + x2 )2 X 2 (1 − X 2 ) 1 + x2 (4 − 3X 2 )

+ 2kx(1 + x2 )X 9x4 X 2 − 7 − 3X 2 + x2 (−3 + 2X 2 )

(5.30)
=: Ve0 ,

25
which is a polynomial of degree 7 in x. The strict positivity of Ve0 (x) for x > x+ will follow if

∂xj Ve0 (k = K(x+ , X), x = x+ ) ≥ 0, for all admissible parameters (x+ , X), (5.31)

for all 0 ≤ j ≤ 7 and if at least one of these quantities is strictly positive. It is clear from the expression (5.30)
of Ve0 that ∂x6 Ve0 , ∂x7 Ve0 are strictly positive if 0 < X < 1. Using (5.28), a computation for the lower derivatives
gives

2
Ve0 (x = x+ ) = x−2 2 2 2
+ X (1 + x+ ) −1 + 3x4+ X 2 + x2+ (1 + X 2 ) ,

∂x Ve0 (x = x+ ) = x−1 2 2
− 1 + 3x4+ X 2 + x2+ (1 + X 2 ) − 1 + 9x2+ + 3X 2 + 14x2+ X 2 + 21x4+ X 2 ,



+ (1 + x+ )X

−2 2
∂x2 Ve0 (x = x+ ) = 2x+ X 4 − 16x2+ − 2x4+ + 30x6+ − 4x2+ X 2 + 3x4+ X 2 + 142x6+ X 2
+ 159x8+ X 2 + 3x2+ X 4 + 37x4+ X 4 + 160x6+ X 4 + 303x8+ X 4 + 189x10 4


+X ,

−1 2
∂x3 Ve0 (x = x+ ) = 6x+ X − 10 − 4x2+ + 50x4+ − X 2 + 39x2+ X 2 + 245x4+ X 2 + 285x6+ X 2
+ 23x2+ X 4 + 169x4+ X 4 + 425x6+ X 4 + 315x8+ X 4 ,

∂x4 Ve0 (x = x+ ) = 24X 2 − 6 + 45x2+ + 40X 2 + 250x2+ X 2 + 300x4+ X 2 + 6X 4 + 100x2+ X 4 + 370x4+ X 4 + 315x6+ X 4 ,

∂x5 Ve0 (x = x+ ) = 120x−1 2

− 3 + 21x2+ + 11X 2 + 155x2+ X 2 + 186x4+ X 2 + 29x2+ X 4 + 200x4+ X 4 + 189x6+ X 4 .


+ X
√
It is immediate that Ve0 (x = x+ ) is non-negative. Using that x+ ≥ 1/ 3, we easily obtain that ∂x Ve0 (x =
x+ ), ∂x4 Ve0 (x = x+ ), ∂x5 Ve0 (x = x+ ) are non-negative for all admissible parameters (x+ , X). Let define

P2 (x+ , X) := 4 − 16x2+ − 2x4+ + 30x6+ − 4x2+ X 2 + 3x4+ X 2 + 142x6+ X 2

+ 159x8+ X 2 + 3x2+ X 4 + 37x4+ X 4 + 160x6+ X 4 + 303x8+ X 4 + 189x10 4
+X .

For all admissible parameters (x+ , X) we have the following sequence of inequalities

P2 (x+ , X) ≥ 4 − 16x2+ − 2x4+ + 30x6+ + − 4x2+ + 3x4+ + 142x6+ X 2

115 2
≥ 4 − 16x2+ − 2x4+ + 30x6+ + X
27
1 − x2+
 
115
≥ 4 − 16x2+ − 2x4+ + 30x6+ +
27 x2+ + 3x4+
≥ 0.

Thus, ∂x2 Ve0 (x = x+ ) = 2x−2 2

+ X P2 (x+ , X) is non-negative. Arguing along the same lines we obtain that
∂x3 Ve0 (x = x+ ) is non-negative. This finishes the proof of (5.26) and of Proposition 5.1.

5.5 Proof of Proposition 5.2

5.5.1 The (1.5a), |m| = 1 case
Let us first treat the (1.5a), |m| = 1 case. Using the positivity of the potential (5.26) obtained in the previous
section, we have
2
∆ − k 2 (r − r+ )2 (r + r+ )2

m=±1 Ξω+ m=0 m=0
Vstat [λ̃ = 0](r) = + Vstat [λ̃ = 0](r) ≥ Vstat [λ̃ = 0](r) > 0.
k (r2 + a2 )2

Plugging this estimate and the positivity of the angular eigenvalue (5.9) into the energy identity (5.1), we deduce
Proposition 5.2 in the (1.5a) case in for |m| = 1.

26
5.5.2 The a = 0 case
2
From Lemma 2.2, and the fact that ω+ = a/(r+ + a2 ) = 0, we have

λ̃0m` = λ0m` − 2 = `(` + 1) − 4 ≥ 0.

since ` ≥ 2. Moreover, using that ω+ = 0 and the positivity (5.26), we have

m m=0 m=0
Vstat [λ̃ = 0](r) = Vstat [λ̃ = 0](r) ≥ Vstat [λ̃ = 0](r) > 0.

As in the previous section, this finishes the proof of Proposition 5.2 if a = 0.

5.5.3 The remaining cases

m m=2
From the energy identity of Lemma 5.4 – using that for |m| ≥ 2, Vstat ≥ Vstat –, the Hardy estimate of
Lemma 5.6 and the estimates on the angular eigenvalues of Lemma 5.7, the proof of Proposition 5.2 in the
remaining cases follows from the following lemma.

Lemma 5.12. Let a > 0. If the Hawking-Reall bound (1.2) is satisfied, we have
m=2
V−5δ/2 := Vstat [λ̃ = −5δ/2](r) + VHardy [λ̃ = −5δ/2](r) > 0, (5.32)

for all r > r+ . Moreover, if the bound (1.5b) is satisfied, we have

m=1
Vfar,1 := Vstat [λ̃ = −(1 + X)2 − 2X(1 − X)](r)
+ VHardy [λ̃ = −(1 + X)2 − 2X(1 − X)](r) (5.33)
> 0,

and
m=2
Vfar,2 := Vstat [λ̃ = −4(1 + X)2 + 4Ξ(1 − δ 2 )(1 − X)2 ](r)
+ VHardy [λ̃ = −4(1 + X)2 + 4Ξ(1 − δ 2 )(1 − X)2 ](r) (5.34)
> 0,

for all r > r+ .

Proof. The proofs of (5.32) and (5.33) go along the same lines (and are easier) as the proof of (5.34), and we
refer the reader to the Mathematica notebook for the specific computations. Using the rescaling (5.27), the
proof of (5.34) reduces to the M = 1 case. We refer to Section 5.4 for the definitions of the parameters and
variables used in this proof. From an inspection of the expression of Vfar,2 , the quantity

X 2 (1 + x2+ )8 (1 + X 2 x2+ )6
 
Vefar,2 := (r2 + a2 )4 Vfar,2
16x4+

is a polynomial of order 8 in x = r/a, and (5.34) follows provided that for 0 ≤ i ≤ 8 we have ∂xi Vefar,2 (x =
x+ ) ≥ 0 (and that at least one is strictly positive). Using formula (5.28) for k and that, by definition, we have
1+X
δ = X(1+x 2 ) , a direct computation gives for the first coefficient
+

2
Vefar,2 (x = x+ ) = 4X 2 (1 + x2+ )4 − 1 + (1 + X 2 )x2+ + 3X 2 x4+ ≥ 0.

A computation of the next coefficient gives

5
!
X
i
∂x Vefar,2 (x = x+ ) = 4x+ (1 + x2+ )2 X −1 Pi (X) Xx2+ −4 ,
i=0

27
where

P0 (X) := 560 + 10036X + 41932X 2 + 13345X 3 + 11380X 4 − 455X 5 − 2356X 6 − 192X 7 ,

P1 (X) := 428 + 10648X + 57683X 2 + 16376X 3 + 12181X 4 − 1472X 5 − 2057X 6 − 112X 7 ,
P2 (X) := 108 + 4211X + 31540X 2 + 7323X 3 + 4728X 4 − 1042X 5 − 592X 6 − 16X 7 ,
P3 (X) := 9 + 736X + 8583X 2 + 1424X 3 + 790X 4 − 272X 5 − 56X 6 ,
P4 (X) := 48X + 1164X 2 + 102X 3 + 48X 4 − 24X 5 ,
P5 (X) := 63X 2 .

We easily check that Pi (X) > 0 for all 0 < X < 1 and all 0 ≤ i ≤ 5. Thus, using that the bound (1.5b) rewrites
Xx2+ − 4 ≥ 0, we have that ∂x Vefar,2 (x = x+ ) > 0. Along the same lines, we check that all the coefficients
∂xi Vefar,2 (x = x+ ) are positive for 2 ≤ i ≤ 8 (the computations are displayed in the companion Mathematica
file). This finishes the proof of the lemma.

References
[ABBM19] L. Andersson, T. Bäckdahl, P. Blue, S. Ma, Stability for linearized gravity on the Kerr spacetime,
arXiv:1903.03859 (2019), 99 pp.

[AMPW17] L. Andersson, S. Ma, C. Paganini, B. F. Whiting, Mode stability on the real axis, J. Math. Phys.
58 (2017), no. 7, 072501.
[BR11] P. Bizoń, A. Rostworowski, Weakly Turbulent Instability of Anti–de Sitter Spacetime, Phys. Rev.
Lett. 107 (2011), no. 3, 031102.
[CDH+ 14] V. Cardoso, Ó. J. C. Dias, G. S. Hartnett, L. Lehner, J. E. Santos, Holographic thermalization,
quasinormal modes and superradiance in Kerr-AdS, J. High Energ. Phys. 2014 (2014), no. 4, 183.
[CL01] V. Cardoso, J. P. S. Lemos, Quasinormal modes of Schwarzschild–anti-de Sitter black holes: Elec-
tromagnetic and gravitational perturbations, Phys. Rev. D 64 (2001), no. 8, 084017.
[CM94] C. M. Chambers, I. G. Moss, Stability of the Cauchy horizon in Kerr–de Sitter spacetimes, Class.
Quantum Grav. 11 (1994), no. 4, 1035–1054.

[CS22] A. Chatzikaleas, J. Smulevici, Non-linear periodic waves on the Einstein cylinder, arXiv:2201.05447
(2022), 90 pp.
[CT21a] M. Casals, R. Teixeira da Costa, Hidden spectral symmetries and mode stability of subextremal
Kerr(-dS) black holes, arXiv:2105.13329 (2021), 24 pp.

[CT21b] M. Casals, R. Teixeira da Costa, The Teukolsky–Starobinsky constants: Facts and fictions, Class.
Quantum Grav. 38 (2021), no. 16, 165016.
[DHR19a] M. Dafermos, G. Holzegel, I. Rodnianski, Boundedness and decay for the Teukolsky equation on
Kerr spacetimes I: The case a << M, Ann. PDE 5 (2019), no. 1, 118 pp.

[DHR19b] M. Dafermos, G. Holzegel, I. Rodnianski, The linear stability of the Schwarzschild solution to
gravitational perturbations, Acta Math. 222 (2019), no. 1, 1–214.
[DHRT21] M. Dafermos, G. Holzegel, I. Rodnianski, M. Taylor, The non-linear stability of the Schwarzschild
family of black holes, arXiv:2104.08222 (2021), 519 pp.
[Dol17] D. Dold, Unstable mode solutions to the Klein–Gordon equation in Kerr-anti-de Sitter spacetimes,
Comm. Math. Phys. 350 (2017), no. 2, 639–697.
[DR09] M. Dafermos, I. Rodnianski, The red-shift effect and radiation decay on black hole spacetimes,
Comm. Pure Appl. Math. 62 (2009), no. 7, 859–919.

28
[DRS09] Ó. J. Dias, H. S. Reall, J. E. Santos, Kerr-CFT and gravitational perturbations, J. High Energy
Phys. 2009 (2009), no. 08, 101–101.
[DRS16] M. Dafermos, I. Rodnianski, Y. Shlapentokh-Rothman, Decay for solutions of the wave equation
on Kerr exterior spacetimes III: The full subextremal case a < M, Ann. Math. 183 (2016), no. 3,
787–913.
[DS13] Ó. J. C. Dias, J. E. Santos, Boundary conditions for Kerr-AdS perturbations, J. High Energ. Phys.
2013 (2013), no. 10, 156.

[DSW15] Ó. J. C. Dias, J. E. Santos, B. Way, Black holes with a single Killing vector field: Black resonators,
J. High Energ. Phys. 2015 (2015), no. 12, 1–10.
[Dya16] S. Dyatlov, Spectral gaps for normally hyperbolic trapping, Ann. inst. Fourier 66 (2016), no. 1,
55–82.
[DZ19] S. Dyatlov, M. Zworski, Mathematical Theory of Scattering Resonances, volume 200 of Graduate
Studies in Mathematics, American Mathematical Society, Providence, Rhode Island (2019).
[EK19] A. Enciso, N. Kamran, Lorentzian Einstein metrics with prescribed conformal infinity, J. Differential
Geom. 112 (2019), no. 3.
[Fan21] A. J. Fang, Nonlinear stability of the slowly-rotating Kerr-de Sitter family, arXiv:2112.07183 (2021),
49 pp.
[Fan22] A. J. Fang, Linear stability of Kerr-de Sitter, in preparation (2022).
[Fri95] H. Friedrich, Einstein equations and conformal structure: Existence of anti-de Sitter-type space-
times, Journal of Geometry and Physics 17 (1995), no. 2, 125–184.

[Gan14] O. Gannot, Quasinormal Modes for Schwarzschild–AdS Black Holes: Exponential Convergence to
the Real Axis, Comm. Math. Phys. 330 (2014), no. 2, 771–799.
[Gan18] O. Gannot, Elliptic boundary value problems for Bessel operators, with applications to anti-de Sitter
spacetimes, Comptes Rendus Mathematique 356 (2018), no. 10, 988–1029.
[GHIW16] S. R. Green, S. Hollands, A. Ishibashi, R. M. Wald, Superradiant instabilities of asymptotically
anti-de Sitter black holes, Class. Quantum Grav. 33 (2016), no. 12, 125022.
[HHV21] D. Häfner, P. Hintz, A. Vasy, Linear stability of slowly rotating Kerr black holes, Invent. math. 223
(2021), no. 3, 1227–1406.
[Hin21] P. Hintz, Mode stability and shallow quasinormal modes of Kerr-de Sitter black holes away from
extremality, arXiv:2112.14431 (2021), 91 pp.
[HLSW20] G. Holzegel, J. Luk, J. Smulevici, C. Warnick, Asymptotic properties of linear field equations in
anti-de Sitter space, Comm. Math. Phys. 374 (2020), no. 2, 1125–1178.
[Hol09] G. Holzegel, On the massive wave equation on slowly rotating Kerr-AdS spacetimes, Comm. Math.
Phys. 294 (2009), no. 1, 169–197.

[HR99] S. W. Hawking, H. S. Reall, Charged and rotating AdS black holes and their CFT duals, Phys. Rev.
D 61 (1999), no. 2, 024014.
[HS13] G. Holzegel, J. Smulevici, Decay properties of Klein-Gordon fields on Kerr-adS spacetimes, Comm.
Pure Appl. Math. 66 (2013), no. 11, 1751–1802.

[HS14] G. Holzegel, J. Smulevici, Quasimodes and a lower bound on the uniform energy decay rate for
Kerr–AdS spacetimes, Anal. PDE 7 (2014), no. 5, 1057–1090.
[HV18] P. Hintz, A. Vasy, The global non-linear stability of the Kerr–de Sitter family of black holes, Acta
Mathematica 220 (2018), no. 1, 1–206.

29
[Kha83] U. Khanal, Rotating black hole in asymptotic de Sitter space: Perturbation of the space-time with
spin fields, Phys. Rev. D 28 (1983), no. 6, 1291–1297.
[KS20] S. Klainerman, J. Szeftel, Global nonlinear stability of Schwarzschild spacetime under polarized
perturbations, Annals of Math. Studies 210 (2020), xviii+856 pp.
[KS21] S. Klainerman, J. Szeftel, Kerr stability for small angular momentum, arXiv:2104.11857 (2021),
799 pp.
[Mav21a] G. Mavrogiannis, Morawetz estimates without relative degeneration and exponential decay on
Schwarzschild-de Sitter spacetimes, arXiv:2111.09494 (2021), 29 pp.
[Mav21b] G. Mavrogiannis, Quasilinear wave equations on Schwarzschild-de Sitter, arXiv:2111.09495 (2021),
40 pp.
[Mil21] P. Millet, Geometric background for the Teukolsky equation revisited, arXiv:2111.03347 (2021), 40
pp.
[MN02] I. G. Moss, J. P. Norman, Gravitational quasinormal modes for anti-de Sitter black holes, Class.
Quantum Grav. 19 (2002), no. 8, 2323–2332.
[Mos18] G. Moschidis, A proof of the instability of AdS for the Einstein–massless Vlasov system,
arXiv:1812.04268 (2018), 132 pp.
[Mos20] G. Moschidis, A proof of the instability of AdS for the Einstein-null dust system with an inner
mirror, Analysis & PDE 13 (2020), no. 6, 1671–1754.
[NP62] E. Newman, R. Penrose, An approach to gravitational radiation by a method of spin coefficients,
Journal of Mathematical Physics 3 (1962), no. 3, 566–578.
[SC74] A. A. Starobinsky, S. M. Churilov, Amplification of electromagnetic and gravitational waves scat-
tered by a rotating ”black hole”, J. Exp. Theor. Phys. 38 (1) (1974), 3–11.
[Sch16] V. Schlue, Decay of the Weyl curvature in expanding black hole cosmologies, arXiv:1610.04172
(2016), 124 pp.
[Shl14] Y. Shlapentokh-Rothman, Exponentially Growing Finite Energy Solutions for the Klein–Gordon
Equation on Sub-Extremal Kerr Spacetimes, Comm. Math. Phys. 329 (2014), no. 3, 859–891.
[Shl15] Y. Shlapentokh-Rothman, Quantitative Mode Stability for the Wave Equation on the Kerr Space-
time, Ann. Henri Poincaré 16 (2015), no. 1, 289–345.
[ST20] Y. Shlapentokh-Rothman, R. Teixeira da Costa, Boundedness and decay for the Teukolsky equation
on Kerr in the full subextremal range a, arXiv:2007.07211 (2020), 125 pp.
[Tei20] R. Teixeira da Costa, Mode stability for the Teukolsky equation on extremal and subextremal Kerr
spacetimes, Comm. Math. Phys. 378 (2020), no. 1, 705–781.
[Teu72] S. A. Teukolsky, Rotating Black Holes: Separable Wave Equations for Gravitational and Electro-
magnetic Perturbations, Phys. Rev. Lett. 29 (1972), no. 16, 1114–1118.
[TP74] S. A. Teukolsky, W. H. Press, Perturbations of a rotating black hole. III - Interaction of the hole
with gravitational and electromagnetic radiation, ApJ 193 (1974), 443.
[Vas13] A. Vasy, Microlocal analysis of asymptotically hyperbolic and Kerr-de Sitter spaces (with an appendix
by Semyon Dyatlov), Invent. math. 194 (2013), no. 2, 381–513.
[Wal73] R. M. Wald, On perturbations of a Kerr black hole, J. Math. Phys. 14 (1973), no. 10, 10 pp.
[Wal78] R. M. Wald, Construction of Solutions of Gravitational, Electromagnetic, or Other Perturbation
Equations from Solutions of Decoupled Equations, Phys. Rev. Lett. 41 (1978), no. 4, 203–206.
[War15] C. M. Warnick, On quasinormal modes of asymptotically anti-de Sitter black holes, Comm. Math.
Phys. 333 (2015), no. 2, 959–1035.
[Whi89] B. F. Whiting, Mode stability of the Kerr black hole, J. Math. Phys. 30 (1989), 6.

30