Introduction
The question of whether the Kerr family of black holes is stable under gravitational perturbations has organized much of mathematical relativity since Teukolsky’s decoupling of the linearized Einstein equations into a single master equation for the extreme-helicity Weyl scalars [1 , 2 ] . For subextremal Kerr the picture is by now well developed: mode stability holds, and quantitative decay estimates for solutions of the wave and Teukolsky equations have been established by a combination of vector field, energy and integrated local energy decay methods. The extremal case a = M a = M a = M is markedly different. Aretakis showed [3 ] that transverse derivatives of a scalar field do not decay along the event horizon of an extremal black hole and that, at sufficiently high order, they grow without bound; the same instability was subsequently shown to afflict linearized gravitational and electromagnetic perturbations of extreme Kerr [4 ] , and to develop even from data supported arbitrarily far from the horizon [5 ] . In the axisymmetric sector, which is the one we shall work in, the wave equation on extreme Kerr admits energy estimates that give boundedness and power-law decay up to and including the horizon, while leaving the transversal derivatives uncontrolled [6 ] . This horizon instability is not detected by the mode analysis, coexists with decay in the exterior, and is tied to the degeneracy of the horizon rather than to any particular matter content.
Two developments make it natural to attribute the effect to the near-horizon region rather than to the extremal black hole as a whole. At near-extremality, Gralla, Zimmerman and Zimmerman [7 ] found that scalar, electromagnetic and gravitational perturbations of a rapidly rotating Kerr black hole produce a transient growth of the energy density and of the tidal forces measured by infalling observers, lasting arbitrarily long as extremality is approached, and attributed it to the near-horizon geometry. More directly, Gralla and Zimmerman [8 ] showed that the decay rates and the growth of high-order transverse derivatives follow from the emergent near-horizon conformal symmetry alone, through the A d S 2 \mathrm{AdS}_{2} AdS 2 structure of the throat.
The near-horizon region of extremal Kerr is captured by the NHEK geometry of Bardeen and Horowitz [9 ] , obtained from Kerr by a corotating scaling limit. NHEK is not asymptotically flat: it is a warped fibration of a U(1) over an A d S 2 \mathrm{AdS}_{2} AdS 2 base, and the two Killing vectors of Kerr are enhanced to an S L ( 2 , R ) × U ( 1 ) \mathrm{SL}(2,\mathbb{R})\times\mathrm{U}(1) SL ( 2 , R ) × U ( 1 ) isometry algebra. This enhancement underlies the Kerr/CFT correspondence [10 ] and ties the throat to the holography of two-dimensional dilaton gravity, whose study goes back to the A d S 2 \mathrm{AdS}_{2} AdS 2 fragmentation of extremal throats [11 ] . In a nearly-A d S 2 \mathrm{AdS}_{2} AdS 2 spacetime the reparametrization symmetry of the boundary time is broken both spontaneously and explicitly, and the resulting soft sector is governed by a Schwarzian action [12 , 13 , 14 , 15 ] . That this sector controls the low-temperature dynamics of near-extremal black holes has been established for charged solutions [16 , 17 ] and for rotating ones [18 ] , while the Schwarzian mode of the extremal Kerr throat in particular is the object studied in [19 , 20 ] . Linearized gravitational perturbations of NHEK have been analyzed in [21 , 22 ] , and it was shown in [22 ] that NHEK is stable at the level of modes. Set against the Aretakis result, this raises the question posed sharply by Durkee and Reall [23 ] : to what extent does the stability, or instability, of a near-horizon geometry control that of the parent black hole? Because the near-horizon geometry does not share the asymptotic regions of the full solution, the answer cannot be read off from decay at null infinity, and the statement one proves depends delicately on the norm, the foliation and the boundary conditions employed.
The present work addresses a version of this question in which the background itself is allowed to vary. Rather than treating NHEK and near-NHEK as two isolated geometries, we work with the family of solutions [19 , 20 ] obtained by acting on NHEK with the improper coordinate transformation generated by a reparametrization f ( t ) f(t) f ( t ) , whose only imprint on the metric is through the Schwarzian derivative. It makes manifest the conformal symmetry of the throat and the different scaling limits that resolve it at different depths, and it converts the comparison between backgrounds into the study of a single function,the Schwarzian, entering in the perturbation equation as a coefficient. That such a family exists at all is worth a remark, since it was argued in [21 ] that every solution asymptotic to NHEK is diffeomorphic to NHEK itself, so that the throat carries no nontrivial dynamics.
Revisiting the symmetries of the family, we realize that the Schwarzian L = { f ( t ) , t } \mathcal{L}=\{f(t),t\} L = { f ( t ) , t } is not free. We allow the gauge parameter to be field dependent, ϵ = ϵ [ L ] \epsilon=\epsilon[\mathcal{L}] ϵ = ϵ [ L ] , while requiring it to remain local in the background, a differential polynomial in L \mathcal{L} L and its derivatives. This is natural once one notices that the transformation law of L \mathcal{L} L under a reparametrization is generated by the symplectic operator of the Korteweg–de Vries hierarchy, so that the Gelfand–Dikii polynomials organize the Killing equation level by level. The first level returns NHEK and near-NHEK; the second is the stationary KdV equation, whose solitons are pulses that depart from one of these geometries and return to it. The criterion is purely kinematic, locality of the Killing parameter, and it constrains the Schwarzian mode rather than generating a flow for it, singling out a distinguished set of backgrounds on which to test stability. The role of KdV in deforming the geometry has been noted before in three dimensions [24 , 25 , 26 , 27 ] and in nearly-AdS2 _2 2 , where the throat is deformed by the KdV charges [28 ] . Most recently, this has been extended to higher dimensions, where genuinely four-dimensional perturbations were constructed, carrying the KdV dynamics and modifying the thermodynamics [29 ] .
We then study linearized gravitational perturbations of this family of metrics. We adopt the axisymmetric ansatz of [30 ] , which for a constant Schwarzian mode is related by a near-horizon limit to the Teukolsky master equation, but we retain the Schwarzian as a background field.
The analysis of this equation is carried out by energy methods rather than by mode expansion, in the spirit of the vector field method introduced by Klainerman and of the virial estimates that describe decay in settings where scattering theory is unavailable [32 , 33 ] . These tools have been used to control the long-time dynamics of solitons of the Einstein equations on Belinski–Zakharov backgrounds [34 ] . The advantage here is that the conclusions can be tracked directly as functions of the background profile L \mathcal{L} L . Taken together, the results give a concrete answer to the question raised above, at least within the axisymmetric sector: the stability statement one obtains is not a property of the near-horizon region as such, but of the background profile of the Schwarzian mode and of the norm in which decay is measured. The same geometry that is asymptotically stable in the weighted energy norm carries conserved and growing transverse derivatives on its degenerate horizon, while the thermal deformation damps them; and moving along the KdV family from NHEK to near-NHEK changes a conservation law into a Lyapunov inequality.
Finally, we compute the conserved charges associated with the symmetries. The charges of near-horizon extremal geometries have been studied extensively, both as the asymptotic Virasoro charges of the Kerr/CFT construction [10 , 35 ] and as symplectic charges on the phase space generated by finite deformations of the throat [36 , 37 , 38 ] . Our setting differs from the latter in that the deformation acts along the time direction rather than the angular one, so that the charges depend on the surface at which they are evaluated. Since the backgrounds are non-stationary, we use a radial Regge–Teitelboim foliation [39 , 40 ] with leaves of Lorentzian signature and evaluate the charge variation at a late-time corner of the throat. The soliton backgrounds asymptoting to NHEK and to near-NHEK carry different charges, the difference being controlled by the amplitude ω \omega ω of the pulse and by the residual temperature κ \kappa κ .
The paper is organized as follows. Section 2 reviews the Kerr solution and the scaling limits to NHEK and near-NHEK, and Section 2.1 develops the field-dependent extension of the symmetry algebra and its relation to the KdV hierarchy. Section 3 sets up the axisymmetric gravitational perturbations, establishes the Petrov type and the gauge invariance of χ \chi χ , and derives the wave equation. It contains the energy and virial estimates for NHEK, for general L \mathcal{L} L and for the KdV soliton, and Section 3.5 the horizon instability. Section 4 constructs the conserved charges, and Section 5 collects our conclusions and open questions. Three appendices contain, respectively, the reduction to Teukolsky’s master equation, the proofs of the energy estimates, and the conformal diagrams of the twisted fibration.
Kerr black hole and the limits to extremality
We consider the Kerr family of stationary black holes in 4 dimensions, that in natural units G = c = 1 G=c=1 G = c = 1 and using the Boyer–Lindquist coordinates is given by
d s 2 = − Σ Δ ( r ~ 2 + a 2 ) 2 − Δ a 2 sin 2 ( θ ) d t ~ 2 + Σ ( d r ~ 2 Δ + d θ 2 ) + sin 2 ( θ ) Σ ( ( r ~ 2 + a 2 ) 2 − Δ a 2 sin 2 ( θ ) ) ( d ϕ ~ − 2 a M r ~ ( r ~ 2 + a 2 ) 2 − Δ a 2 sin 2 ( θ ) d t ~ ) 2 (1) ds^2=-\frac{\Sigma\Delta}{(\tilde r^2+a^2)^2-\Delta a^2\sin^2(\theta)}d\tilde t^2+\Sigma\left(\frac{d\tilde r^2}{\Delta}+d\theta^2\right)+\frac{\sin^2(\theta)}{\Sigma}\left((\tilde r^2+a^2)^2-\Delta a^2\sin^2(\theta)\right)\left(d\tilde\phi-\frac{2aM\tilde r}{(\tilde r^2+a^2)^2-\Delta a^2\sin^2(\theta)}d\tilde t\right)^2 \tag*{(1)} d s 2 = − ( r ~ 2 + a 2 ) 2 − Δ a 2 sin 2 ( θ ) ΣΔ d t ~ 2 + Σ ( Δ d r ~ 2 + d θ 2 ) + Σ sin 2 ( θ ) ( ( r ~ 2 + a 2 ) 2 − Δ a 2 sin 2 ( θ ) ) ( d ϕ ~ − ( r ~ 2 + a 2 ) 2 − Δ a 2 sin 2 ( θ ) 2 a M r ~ d t ~ ) 2 (1)
where
Σ = r ~ 2 + a 2 cos 2 ( θ ) , Δ ( r ~ ) = ( r ~ − r − ) ( r ~ − r + ) , r ± = M ± M 2 − a 2 . \Sigma=\tilde r^2+a^2\cos^2(\theta),\qquad\Delta(\tilde r)=(\tilde r-r_-)(\tilde r-r_+),\qquad r_\pm=M\pm\sqrt{M^2-a^2}. Σ = r ~ 2 + a 2 cos 2 ( θ ) , Δ ( r ~ ) = ( r ~ − r − ) ( r ~ − r + ) , r ± = M ± M 2 − a 2 .
The solution is parametrized by its mass M M M and its angular momentum J J J by means of the rotation parameter a = J / M a=J/M a = J / M , generated by the Killing symmetries ∂ t \partial_t ∂ t and ∂ ϕ \partial_\phi ∂ ϕ respectively. The inner and outer horizons are located at r − r_- r − and r + r_+ r + , the angular velocity of the horizon Ω = a / 2 M r + \Omega=a/2Mr_+ Ω = a /2 M r + . Its temperature is given by
T = r + − r − 8 π M r + . (2) T=\frac{r_+-r_-}{8\pi Mr_+}. \tag*{(2)} T = 8 π M r + r + − r − . (2)
If a > M a>M a > M , there is no event horizon as the Kerr metric describes a naked singularity and if a = M a=M a = M , the black hole is said to be extremal as it is rotating at its maximum angular momentum and its temperature vanishes. We will place ourselves in this particular case and take the co-rotating change of coordinates,
r ~ = M + λ R , t ~ = 2 M 2 λ T , ϕ ~ = Φ + M λ T . (3) \tilde r=M+\lambda R,\qquad\tilde t=\frac{2M^2}{\lambda}T,\qquad\tilde\phi=\Phi+\frac{M}{\lambda}T. \tag*{(3)} r ~ = M + λ R , t ~ = λ 2 M 2 T , ϕ ~ = Φ + λ M T . (3)
In the limit λ → 0 \lambda\to0 λ → 0 , holding ( T , R , θ , Φ ) (T,R,\theta,\Phi) ( T , R , θ , Φ ) fixed, one obtains the so-called Near-Horizon Extreme Kerr (NHEK) metric [9 ] ,
d s 2 = Λ ( θ ) [ − R 2 d T 2 + d R 2 R 2 + d θ 2 ] + Γ ( θ ) [ d Φ + R d T ] 2 , (4) ds^2=\Lambda(\theta)\left[-R^2dT^2+\frac{dR^2}{R^2}+d\theta^2\right]+\Gamma(\theta)\left[d\Phi+RdT\right]^2, \tag*{(4)} d s 2 = Λ ( θ ) [ − R 2 d T 2 + R 2 d R 2 + d θ 2 ] + Γ ( θ ) [ d Φ + R d T ] 2 , (4)
for
Λ ( θ ) = M 2 ( 1 + cos 2 ( θ ) ) and Γ ( θ ) = 4 M 2 sin 2 ( θ ) 1 + cos 2 ( θ ) . (5) \Lambda(\theta)=M^2(1+\cos^2(\theta)) \qquad\text{and} \qquad\Gamma(\theta)=\frac{4M^2\sin^2(\theta)}{1+\cos^2(\theta)}. \tag*{(5)} Λ ( θ ) = M 2 ( 1 + cos 2 ( θ )) and Γ ( θ ) = 1 + cos 2 ( θ ) 4 M 2 sin 2 ( θ ) . (5)
The geometry (4) exhibits an AdS2 _2 2 factor associated to the T − R T-R T − R plane and axial symmetry, related with the metric coefficients being independent on Φ \Phi Φ . Surfaces of constant θ \theta θ are warped AdS3 _3 3 geometries [41 ] . This metric has an enhanced SL( 2 , R ) (2,\mathbb{R}) ( 2 , R ) symmetry with Killing vectors
ζ = ζ ( ϵ ) = ( ϵ ( T ) + ϵ ′ ′ ( T ) 2 R 2 ) ∂ T − R ϵ ′ ( T ) ∂ R − ϵ ′ ′ ( T ) R ∂ Φ + c Φ ∂ Φ , (6) \zeta=\zeta(\epsilon)=\left(\epsilon(T)+\frac{\epsilon''(T)}{2R^2}\right)\partial_T-R\epsilon'(T)\partial_R-\frac{\epsilon''(T)}{R}\partial_\Phi+c_\Phi\partial_\Phi, \tag*{(6)} ζ = ζ ( ϵ ) = ( ϵ ( T ) + 2 R 2 ϵ ′′ ( T ) ) ∂ T − R ϵ ′ ( T ) ∂ R − R ϵ ′′ ( T ) ∂ Φ + c Φ ∂ Φ , (6)
Here c Φ c_\Phi c Φ is an arbitrary constant associated to the U(1) symmetry and the gauge parameter obeys the condition ϵ ( T ) ′ ′ ′ = 0 → ϵ ( T ) = c − 1 T 2 + c 0 T + c + 1 \epsilon(T)'''=0\to\epsilon(T)=c_{-1}T^2+c_0T+c_{+1} ϵ ( T ) ′′′ = 0 → ϵ ( T ) = c − 1 T 2 + c 0 T + c + 1 where c n c_n c n are integration constants labeling the generators L n L_n L n of the s l ( 2 , R ) \mathfrak{sl}(2,\mathbb{R}) sl ( 2 , R ) algebra,
[ L n , L m ] = ( n − m ) L n + m for n , m = − 1 , 0 , 1. (7) [L_n,L_m]=(n-m)L_{n+m}\qquad\text{for}\qquad n,m=-1,0,1. \tag*{(7)} [ L n , L m ] = ( n − m ) L n + m for n , m = − 1 , 0 , 1. (7)
In particular, c + 1 c_{+1} c + 1 is connected to the time translation symmetry, and
L − 1 = ( 1 R 2 + T 2 ) ∂ T − 2 T R ∂ R − 2 R ∂ Φ , L 0 = T ∂ T − R ∂ R , L 1 = ∂ T . (8) L_{-1}=\left(\frac{1}{R^2}+T^2\right)\partial_T-2TR\partial_R-\frac{2}{R}\partial_\Phi,\qquad L_0=T\partial_T-R\partial_R,\qquad L_1=\partial_T. \tag*{(8)} L − 1 = ( R 2 1 + T 2 ) ∂ T − 2 T R ∂ R − R 2 ∂ Φ , L 0 = T ∂ T − R ∂ R , L 1 = ∂ T . (8)
They close the Lie algebra,
{ ζ [ ϵ 1 ] , ζ [ ϵ 2 ] } μ = ζ μ [ ϵ 3 ] for ϵ 3 = ϵ 1 ϵ 2 ′ − ϵ 2 ϵ 1 ′ . (9) \{\zeta[\epsilon_1],\zeta[\epsilon_2]\}^{\mu}=\zeta^{\mu}[\epsilon_3]\qquad\text{for}\qquad\epsilon_3=\epsilon_1\epsilon_2'-\epsilon_2\epsilon_1'. \tag*{(9)} { ζ [ ϵ 1 ] , ζ [ ϵ 2 ] } μ = ζ μ [ ϵ 3 ] for ϵ 3 = ϵ 1 ϵ 2 ′ − ϵ 2 ϵ 1 ′ . (9)
Near-NHEK, extensions of the symmetry group and KdV
To explore enhancements of the global conformal isometry group SL( 2 , R ) (2,\mathbb{R}) ( 2 , R ) , we review the study of the symmetries of nearly-AdS2 _2 2 geometries, as they are responsible for driving black holes away from the extremal zone. These metrics have been studied extensively, as they represent a closer zoom into the horizon, and are recovered by taking small deviations from extremality [42 ] , which captures several possible coordinate re-scalings, labelling different deviations that characterize time-like orbits. Specifically, the near-NHEK is obtained by applying the coordinate change
r ~ = r + + r − 2 + λ ( r + κ 2 4 r ) , t ~ = 2 M 2 t λ , ϕ = Φ + M t λ . (10) \tilde r=\frac{r_++r_-}{2}+\lambda\left(r+\frac{\kappa^2}{4r}\right),\qquad\tilde t=2M^2\frac{t}{\lambda},\qquad\phi=\Phi+\frac{Mt}{\lambda}. \tag*{(10)} r ~ = 2 r + + r − + λ ( r + 4 r κ 2 ) , t ~ = 2 M 2 λ t , ϕ = Φ + λ M t . (10)
on the Kerr metric (2.1). With the additional condition that r ± = M ± λ κ r_\pm=M\pm\lambda\kappa r ± = M ± λκ , so that a 2 = r + r − = M 2 − λ 2 κ 2 a^2=r_+r_-=M^2-\lambda^2\kappa^2 a 2 = r + r − = M 2 − λ 2 κ 2 , the decoupling limit λ → 0 \lambda\to0 λ → 0 gives
d s 2 = Λ ( θ ) [ − ( r − κ 2 4 r ) 2 d t 2 + d r 2 r 2 + d θ 2 ] + Γ ( θ ) [ d ϕ + ( r + κ 2 4 r ) d t ] 2 . (11) ds^2=\Lambda(\theta)\left[-\left(r-\frac{\kappa^2}{4r}\right)^2dt^2+\frac{dr^2}{r^2}+d\theta^2\right]+\Gamma(\theta)\left[d\phi+\left(r+\frac{\kappa^2}{4r}\right)dt\right]^2. \tag*{(11)} d s 2 = Λ ( θ ) [ − ( r − 4 r κ 2 ) 2 d t 2 + r 2 d r 2 + d θ 2 ] + Γ ( θ ) [ d ϕ + ( r + 4 r κ 2 ) d t ] 2 . (11)
Similarly to the previous limit, it yields a metric that does not depend on λ \lambda λ , but does depend on κ \kappa κ which captures a power law deviation from extremality of the Hawking temperature (2) as,
T = λ κ 4 π M 2 − λ 2 κ 2 4 π M 3 + O ( λ 3 κ 3 ) . (12) T = \frac{\lambda\kappa}{4\pi M^2} - \frac{\lambda^2\kappa^2}{4\pi M^3} + O(\lambda^3\kappa^3). \tag*{(12)} T = 4 π M 2 λκ − 4 π M 3 λ 2 κ 2 + O ( λ 3 κ 3 ) . (12)
To study the previous metrics in a more general setup, we consider a set of geometries proposed in [19 ] and revisited in [20 ] , that acknowledges the existence of several possible scalings while approaching the near horizon region of an extremal black hole. We take the improper coordinate transformation 6 ,
T = f ( t ) + 2 ( f ′ ) 2 f ′ ′ 4 r 2 f 2 − f ′ ′ , R = r f ′ − f ′ ′ 4 r ( f ′ ) 3 , Φ = ϕ + log [ 2 r f ′ − f ′ ′ 2 r f ′ + f ′ ′ ] , (13) T = f(t) + \frac{2(f')^2 f''}{4r^2 f^2 - f''}, \qquad R = \frac{r}{f'} - \frac{f''}{4r(f')^3}, \qquad\Phi= \phi+ \log\left[\frac{2r f' - f''}{2r f' + f''}\right], \tag*{(13)} T = f ( t ) + 4 r 2 f 2 − f ′′ 2 ( f ′ ) 2 f ′′ , R = f ′ r − 4 r ( f ′ ) 3 f ′′ , Φ = ϕ + log [ 2 r f ′ + f ′′ 2 r f ′ − f ′′ ] , (13)
where f ( t ) f(t) f ( t ) is the mode function associated to the time reparametrization symmetry. Here and throughout the paper we use the convention ∂ t = ( ) ′ \partial_t = ()' ∂ t = ( ) ′ for functions depending on the t t t -coordinate only. The change of coordinates (13) is applied in [20 ] and one obtains
d s 2 = Λ ( θ ) ( − ( r + L ( t ) 2 r ) 2 d t 2 + d r 2 r 2 + d θ 2 ) + Γ ( θ ) ( d ϕ + ( r − L ( t ) 2 r ) d t ) 2 , (14) ds^2 = \Lambda(\theta)\left(-\left(r + \frac{\mathcal{L}(t)}{2r}\right)^2dt^2 + \frac{dr^2}{r^2} + d\theta^2\right) + \Gamma(\theta)\left(d\phi+ \left(r - \frac{\mathcal{L}(t)}{2r}\right)dt\right)^2, \tag*{(14)} d s 2 = Λ ( θ ) ( − ( r + 2 r L ( t ) ) 2 d t 2 + r 2 d r 2 + d θ 2 ) + Γ ( θ ) ( d ϕ + ( r − 2 r L ( t ) ) d t ) 2 , (14)
with
L ( t ) = { f , t } = f ′ ′ ′ f ′ − 3 2 ( f ′ ′ f ′ ) 2 . (15) \mathcal{L}(t) = \{f,t\} = \frac{f'''}{f'} - \frac{3}{2}\left(\frac{f''}{f'}\right)^2. \tag*{(15)} L ( t ) = { f , t } = f ′ f ′′′ − 2 3 ( f ′ f ′′ ) 2 . (15)
This is the Schwarzian derivative { f , t } \{f,t\} { f , t } , which is invariant function under the action of the S L ( 2 , R ) \mathrm{SL}(2,\mathbb{R}) SL ( 2 , R ) transformation
f ( t ) → f ~ ( t ) = a f ( t ) + b c f ( t ) + d with a d − b c = 1. f(t) \to\tilde{f}(t) = \frac{a f(t) + b}{c f(t) + d} \quad\text{with} \quad ad - bc = 1. f ( t ) → f ~ ( t ) = c f ( t ) + d a f ( t ) + b with a d − b c = 1.
This metric contains the NHEK geometry by the choice f ( t ) = t f(t) = t f ( t ) = t , giving L = 0 \mathcal{L} = 0 L = 0 , and the near-NHEK metric by the choice f ( t ) = exp ( κ t ) f(t) = \exp(\kappa t) f ( t ) = exp ( κ t ) , and crucially, makes manifest the conformal symmetry. Indeed, computing the group symmetries generated by the vector ζ μ = { ζ t , ζ r , ζ θ , ζ ϕ } \zeta^\mu= \{\zeta^t,\zeta^r,\zeta^\theta,\zeta^\phi\} ζ μ = { ζ t , ζ r , ζ θ , ζ ϕ } , that preserves the form of the metric (14) ,
δ g μ ν = L ζ g μ ν , (16) \delta g_{\mu\nu} = \mathcal{L}_{\zeta}g_{\mu\nu}, \tag*{(16)} δ g μν = L ζ g μν , (16)
one finds
ζ = ( ϵ ( t ) + ϵ ′ ′ ( t ) 2 r 2 + L ( t ) ) ∂ t − r ϵ ′ ( t ) ∂ r − 2 r ϵ ′ ′ ( t ) 2 r 2 + L ( t ) ∂ ϕ + c ϕ ∂ ϕ , (17) \zeta= \left(\epsilon(t) + \frac{\epsilon''(t)}{2r^2 + \mathcal{L}(t)}\right)\partial_t - r\epsilon'(t)\partial_r - \frac{2r\epsilon''(t)}{2r^2 + \mathcal{L}(t)}\partial_\phi+ c_\phi\partial_\phi, \tag*{(17)} ζ = ( ϵ ( t ) + 2 r 2 + L ( t ) ϵ ′′ ( t ) ) ∂ t − r ϵ ′ ( t ) ∂ r − 2 r 2 + L ( t ) 2 r ϵ ′′ ( t ) ∂ ϕ + c ϕ ∂ ϕ , (17)
where ϵ \epsilon ϵ is an arbitrary function of time related with the time reparametrization symmetry and c ϕ c_\phi c ϕ is an arbitrary constant associated to rotational invariance. From (16) , we also obtain the additional symmetry transformation 7 ,
δ L = ϵ ( t ) L ′ ( t ) + 2 ϵ ′ ( t ) L ( t ) + ϵ ′ ′ ′ ( t ) . (18) \delta\mathcal{L}=\epsilon(t)\mathcal{L}'(t)+2\epsilon'(t)\mathcal{L}(t)+\epsilon'''(t). \tag*{(18)} δ L = ϵ ( t ) L ′ ( t ) + 2 ϵ ′ ( t ) L ( t ) + ϵ ′′′ ( t ) . (18)
Solving the Killing equation
An isometry of (14) is a vector (17) whose action leaves the Schwarzian mode inert, δ ϵ L = 0 \delta_\epsilon\mathcal{L}=0 δ ϵ L = 0 . By (18) this is the linear third-order equation
D 1 ϵ ≡ ϵ ′ ′ ′ + 2 L ϵ ′ + L ′ ϵ = 0 , (19) D_1\epsilon\equiv\epsilon'''+2\mathcal{L}\epsilon'+\mathcal{L}'\epsilon=0, \tag*{(19)} D 1 ϵ ≡ ϵ ′′′ + 2 L ϵ ′ + L ′ ϵ = 0 , (19)
is the classical third-order equation associated with the Hill operator. Indeed, if ψ 1 , ψ 2 \psi_1,\psi_2 ψ 1 , ψ 2 are any two solutions of
ψ ′ ′ + 1 2 L ψ = 0 , (20) \psi''+\frac{1}{2}\mathcal{L}\psi=0, \tag*{(20)} ψ ′′ + 2 1 L ψ = 0 , (20)
then each of the three products ψ 1 2 \psi_1^2 ψ 1 2 , ψ 1 ψ 2 \psi_1\psi_2 ψ 1 ψ 2 , ψ 2 2 \psi_2^2 ψ 2 2 solves (19) , as follows by direct substitution, and these span its solution space. Since (20) is of second order, the space of isometry parameters is Sym 2 \operatorname{Sym}^2 Sym 2 of a two-dimensional space, hence three-dimensional for every profile L \mathcal{L} L . Together with c ϕ ∂ ϕ c_\phi\partial_\phi c ϕ ∂ ϕ this is the algebra
s l ( 2 , R ) ⊕ u ( 1 ) , (21) \mathfrak{sl}(2,\mathbb{R})\oplus\mathfrak{u}(1), \tag*{(21)} sl ( 2 , R ) ⊕ u ( 1 ) , (21)
the composite parameter of (9) being the symmetric-square representation. Then, every member of the family (14) is obtained from NHEK by the coordinate change (13) , being all locally isometric where no choice of f f f can enlarge or reduce the algebra.
What does change with the background is the realization. For L = 0 \mathcal{L}=0 L = 0 one has ψ ∈ { 1 , t } \psi\in\{1,t\} ψ ∈ { 1 , t } and the familiar parameters ϵ ∈ { 1 , t , t 2 } \epsilon\in\{1,t,t^2\} ϵ ∈ { 1 , t , t 2 } of (6) . For near-NHEK, where f = e κ t f=e^{\kappa t} f = e κ t gives L = { e κ t , t } = − κ 2 / 2 \mathcal{L}=\{e^{\kappa t},t\}=-\kappa^2/2 L = { e κ t , t } = − κ 2 /2 , the Hill equation has ψ ∈ { e ± κ t / 2 } \psi\in\{e^{\pm\kappa t/2}\} ψ ∈ { e ± κ t /2 } and the parameters are ϵ ∈ { 1 , e κ t , e − κ t } \epsilon\in\{1,e^{\kappa t},e^{-\kappa t}\} ϵ ∈ { 1 , e κ t , e − κ t } , where the real exponentials appear because L < 0 \mathcal{L}<0 L < 0 .
Indeed, the sign of L \mathcal{L} L is geometrically meaningful. The metric factor r + L / 2 r r+\mathcal{L}/2r r + L /2 r of (14) vanishes at 2 r 2 = − L 2r^2=-\mathcal{L} 2 r 2 = − L , so for L < 0 \mathcal{L}<0 L < 0 a Killing horizon sits at finite radius, at r = κ / 2 r=\kappa/2 r = κ /2 in the near-NHEK case, whereas for L ≥ 0 \mathcal{L}\geq0 L ≥ 0 the coordinate r r r ranges over all of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) without encountering one. This distinction reappears in § 3 , where the weight ( L + 2 r 2 ) / r 2 (\mathcal{L}+2r^2)/r^2 ( L + 2 r 2 ) / r 2 controlling the energy degenerates precisely at that radius. In this case, L \mathcal{L} L acts as deformation of the background, where this assumption will act favourably in the study of the asymptotic stability of axisymmetric perturbations.
Locality as a selection principle and Korteweg–de Vries
The question we answer now is for which profiles of L \mathcal{L} L , the isometry parameter ϵ \epsilon ϵ is expressible locally in terms of the background, that is, as a differential polynomial ϵ = ϵ [ L , L ′ , … ] \epsilon=\epsilon[\mathcal{L},\mathcal{L}',\ldots] ϵ = ϵ [ L , L ′ , … ] . Imposing locality converts (19) from an ordinary differential equation for ϵ \epsilon ϵ on a fixed background into a condition on L \mathcal{L} L itself. To start, we notice that the Korteweg–de Vries hierarchy is the structure that organizes the local solutions. The transformation law (18) is generated by the first Hamiltonian operator of KdV, D 1 = 2 L ∂ t + L ′ + ∂ t 3 D_1=2\mathcal{L}\partial_t+\mathcal{L}'+\partial_t^3 D 1 = 2 L ∂ t + L ′ + ∂ t 3 [44 ] , and together with the second structure D 2 = ∂ t D_2=\partial_t D 2 = ∂ t it defines the Lenard recursion
D 2 ϵ l + 1 = D 1 ϵ l , ϵ 0 = 1 , (22) D_2\epsilon_{l+1}=D_1\epsilon_l,\qquad\epsilon_0=1, \tag*{(22)} D 2 ϵ l + 1 = D 1 ϵ l , ϵ 0 = 1 , (22)
whose solvability within differential polynomials is the statement that D 1 ϵ l D_1\epsilon_l D 1 ϵ l is a total time derivative. The resulting Gelfand–Dikii polynomials8 are
ϵ 0 = 1 , ϵ 1 = L + ϵ ˉ 1 , ϵ 2 = 3 2 L 2 + L ′ ′ + ϵ ˉ 1 L + ϵ ˉ 2 , … (23) \epsilon_0=1,\qquad\epsilon_1=\mathcal{L}+\bar{\epsilon}_1,\qquad\epsilon_2=\frac{3}{2}\mathcal{L}^2+\mathcal{L}''+\bar{\epsilon}_1\mathcal{L}+\bar{\epsilon}_2,\qquad\ldots \tag*{(23)} ϵ 0 = 1 , ϵ 1 = L + ϵ ˉ 1 , ϵ 2 = 2 3 L 2 + L ′′ + ϵ ˉ 1 L + ϵ ˉ 2 , … (23)
Demanding that ϵ l \epsilon_l ϵ l be an isometry parameter, D 1 ϵ l = 0 D_1\epsilon_l=0 D 1 ϵ l = 0 , then selects a class of backgrounds at each level. The first level gives the condition
L ′ = 0 , (24) \mathcal{L}'=0, \tag*{(24)} L ′ = 0 , (24)
whose solutions are exactly NHEK and near-NHEK. The second gives
L ′ ′ ′ + 3 L L ′ + ϵ ˉ 1 L ′ = 0 , (25) \mathcal{L}'''+3\mathcal{L}\mathcal{L}'+\bar{\epsilon}_1\mathcal{L}'=0, \tag*{(25)} L ′′′ + 3 L L ′ + ϵ ˉ 1 L ′ = 0 , (25)
the stationary KdV equation, whose solitonic solutions describe a transient excursion away from extremality, as it will be discussed in detail along § 3 and § 4.1 . Each further level yields a higher stationary equation and hence a further class of admissible profiles.
By the recursion (22) , the stationary condition9 at level l l l can be stated equivalently as
D 1 ϵ l = 0 ⟺ ϵ l + 1 ′ = 0 , (26) D_1\epsilon_l=0\qquad\Longleftrightarrow\qquad\epsilon_{l+1}'=0, \tag*{(26)} D 1 ϵ l = 0 ⟺ ϵ l + 1 ′ = 0 , (26)
so that the polynomial following the one that generates the isometry is constant on the corresponding backgrounds. At the first level the statement is stronger: since L ′ = 0 \mathcal{L}'=0 L ′ = 0 , every Gelfand–Dikii polynomial reduces to a constant, and the whole tower is conserved. This is special to the constant L \mathcal{L} L case; already at the second level one has ϵ 1 ′ = L ′ ≠ 0 \epsilon_1'=\mathcal{L}'\ne0 ϵ 1 ′ = L ′ = 0 , while ϵ 2 \epsilon_2 ϵ 2 collapses to the constant $\bar{\epsilon}_2.
We emphasize that this is a statement about backgrounds, not about evolution. In the 1 + 1 1+1 1 + 1 dimensional hierarchy the polynomials generate flows ∂ s L = D 1 ϵ i \partial_s\mathcal{L}=\mathcal{D}_1\epsilon_i ∂ s L = D 1 ϵ i in an auxiliary time s s s ; here L \mathcal{L} L depends on t t t alone and the geometry is radially independent, so the flow parameter is absent and only the stationary sector survives. This is also why the KdV potential is a function of time rather than of a spatial coordinate. In the Euclidean section, t = i τ t=i\tau t = i τ with τ ∼ τ + β \tau\sim\tau+\beta τ ∼ τ + β , that coordinate becomes a genuine circle and the hierarchy acquires its usual periodic setting, with β \beta β the inverse temperature.
Example: a soliton background geometry
We can now return to (19) on one of the profiles singled out above and obtain the local generators for this case. Take the soliton of (25) , L = 4 b 2 sech 2 ( b t ) \mathcal{L}=4b^2\operatorname{sech}^2(bt) L = 4 b 2 sech 2 ( b t ) with ϵ ˉ 1 = − 4 b 2 \bar{\epsilon}_1=-4b^2 ϵ ˉ 1 = − 4 b 2 . The Hill equation (20) reads ψ ′ ′ + 2 b 2 sech 2 ( b t ) ψ = 0 \psi''+2b^2\operatorname{sech}^2(bt)\psi=0 ψ ′′ + 2 b 2 sech 2 ( b t ) ψ = 0 and is solved by
ψ 1 = tanh ( b t ) , ψ 2 = t tanh ( b t ) − 1 b , (27) \psi_1=\tanh(bt),\qquad\psi_2=t\tanh(bt)-\frac{1}{b}, \tag*{(27)} ψ 1 = tanh ( b t ) , ψ 2 = t tanh ( b t ) − b 1 , (27)
the second obtained from the first by reduction of order, with unit Wronskian. The three isometry parameters are the corresponding products,
ϵ = c 1 tanh 2 ( b t ) + c 2 ( t tanh 2 ( b t ) − tanh ( b t ) b ) + c 3 ( t tanh ( b t ) − 1 b ) 2 , (28) \epsilon=c_1\tanh^2(bt)+c_2\left(t\tanh^2(bt)-\frac{\tanh(bt)}{b}\right)+c_3\left(t\tanh(bt)-\frac{1}{b}\right)^2, \tag*{(28)} ϵ = c 1 tanh 2 ( b t ) + c 2 ( t tanh 2 ( b t ) − b tanh ( b t ) ) + c 3 ( t tanh ( b t ) − b 1 ) 2 , (28)
and c 1 , c 2 , c 3 c_1,c_2,c_3 c 1 , c 2 , c 3 are the three constants of (21) . As t → ± ∞ t\to\pm\infty t → ± ∞ the soliton decays and (28) approaches c 1 + c 2 t + c 3 t 2 c_1+c_2t+c_3t^2 c 1 + c 2 t + c 3 t 2 , recovering the NHEK parameters { 1 , t , t 2 } \{1,t,t^2\} { 1 , t , t 2 } , as it must since L → 0 \mathcal{L}\to0 L → 0 there.
The algebra these generate is again s l ( 2 , R ) \mathfrak{sl}(2,\mathbb{R}) sl ( 2 , R ) , and the products of Hill solutions provide a canonical basis for it. Normalizing the Wronskian of (20) to unity, ψ 1 ψ 2 ′ − ψ 2 ψ 1 ′ = 1 \psi_1\psi'_2-\psi_2\psi'_1=1 ψ 1 ψ 2 ′ − ψ 2 ψ 1 ′ = 1 , the composite parameter of (9) gives
[ ψ 1 2 , ψ 1 ψ 2 ] = ψ 1 2 , [ ψ 1 2 , ψ 2 2 ] = 2 ψ 1 ψ 2 , [ ψ 1 ψ 2 , ψ 2 2 ] = ψ 2 2 , (29) [\psi_1^2,\psi_1\psi_2]=\psi_1^2,\qquad[\psi_1^2,\psi_2^2]=2\psi_1\psi_2,\qquad[\psi_1\psi_2,\psi_2^2]=\psi_2^2, \tag*{(29)} [ ψ 1 2 , ψ 1 ψ 2 ] = ψ 1 2 , [ ψ 1 2 , ψ 2 2 ] = 2 ψ 1 ψ 2 , [ ψ 1 ψ 2 , ψ 2 2 ] = ψ 2 2 , (29)
which are the relations [ L n , L m ] = ( n − m ) L n + m [L_n,L_m]=(n-m)L_{n+m} [ L n , L m ] = ( n − m ) L n + m of (8) under the identification ψ 1 2 ↔ L 1 \psi_1^2\leftrightarrow L_1 ψ 1 2 ↔ L 1 , ψ 1 ψ 2 ↔ L 0 \psi_1\psi_2\leftrightarrow L_0 ψ 1 ψ 2 ↔ L 0 , ψ 2 2 ↔ L − 1 \psi_2^2\leftrightarrow L_{-1} ψ 2 2 ↔ L − 1 . The structure constants in (29) involve only the Wronskian, so they are the same for every profile L \mathcal{L} L : on NHEK the basis is { 1 , t , t 2 } \{1,t,t^2\} { 1 , t , t 2 } and on the soliton it is the one appearing in (28) .
Only the first of the three is local in the background: ψ 1 2 = 1 − L / 4 b 2 \psi_1^2=1-\mathcal{L}/4b^2 ψ 1 2 = 1 − L /4 b 2 , so that c 1 c_1 c 1 reproduces the Gelfand–Dikii parameter ϵ 1 = L + ϵ ˉ 1 \epsilon_1=\mathcal{L}+\bar{\epsilon}_1 ϵ 1 = L + ϵ ˉ 1 of (23) up to normalization. The other two cannot be written as differential polynomials in L \mathcal{L} L : any such polynomial is a bounded function of t t t once L \mathcal{L} L and all its derivatives decay, whereas ψ 1 ψ 2 \psi_1\psi_2 ψ 1 ψ 2 and ψ 2 2 \psi_2^2 ψ 2 2 grow linearly and quadratically.
The integrability of the Killing equation under the requirement of locality, provides a purely kinematic criterion so that it endows the Schwarzian mode with dynamics and singles out a distinguished set of backgrounds. We analyze in the remainder of the paper the stability of the geometries, with L ( t ) \mathcal{L}(t) L ( t ) entering the perturbation equation as an explicit time-dependent coefficient.
Gravitational perturbations
In this section we study axisymmetric gravitational perturbations of the family (14) . We adopt the ansatz of [30 ] , which for constant L \mathcal{L} L , and in particular for L = 0 \mathcal{L}=0 L = 0 , is related by a near-horizon limit to the axisymmetric Teukolsky master equation of the Kerr black hole. A similar ansatz was proposed in [46 ] . Here, we retain the Schwarzian as a background field, so that the profiles selected in § 2.1 enter the perturbation equation as an explicit time-dependent coefficient, and the stability analysis can be carried out as a function of the background.
Our main tool is the implementation of vector field methods adapted to the background. This a robust geometric approach (see [48 , 49 , 50 ] for a complete introduction of the methodology) that allows to say that, for functions that are sufficiently well-behaved , an L 2 L^2 L 2 -norm with weights decays. In its classical form [51 ] , vector fields are used as multipliers, contracted with the energy-momentum tensor, and produce energy currents whose divergence is governed by the deformation tensor of the field. When the field is a Killing field, the current is conserved and coincides with the Noether current of the corresponding symmetry; when it is not, the failure of the symmetry appears as an explicit bulk term, which may either spoil conservation or, if it has a definite sign, yield decay.
We first derive the wave-type equation from the linearized Einstein equations for the scalar that quantifies the size of the perturbation and relate it to the Weyl scalars in § 3.1 . Then, in § 3.2 we derive the currents and charges for a general vector field and its relation with the background field L \mathcal{L} L . We apply these general results to different backgrounds in § 3.3 . § 3.4 specifies to the pure NHEK solution and obtains decay of solutions from a virial argument. Finally, § 3.5 returns to the Near-NHEK family and obtains the divergence of suitable norms near the horizon.
Linearized Einstein's equations for axisymmetric perturbations
We start by deriving the equations for axisymmetric perturbations of a general near-NHEK solution from the spacetime
d s 2 = M 2 ( 1 + cos 2 θ + δ χ ( x a , θ ) ) [ g a b d x a d x b + d θ 2 ] + 4 M 2 sin 2 θ 1 + cos 2 θ + δ χ ( x a , θ ) [ d ϕ + A a d x a + δ T ] 2 . (30) ds^2 = M^2(1+\cos^2\theta+\delta\chi(x^a,\theta))\left[g_{ab}dx^a dx^b+d\theta^2\right]
+4M^2\frac{\sin^2\theta}{1+\cos^2\theta+\delta\chi(x^a,\theta)}[d\phi+A_a dx^a+\delta\mathcal{T}]^2 . \tag*{(30)} d s 2 = M 2 ( 1 + cos 2 θ + δ χ ( x a , θ )) [ g ab d x a d x b + d θ 2 ] + 4 M 2 1 + cos 2 θ + δ χ ( x a , θ ) sin 2 θ [ d ϕ + A a d x a + δ T ] 2 . (30)
Recall that the near-NHEK metric depends on the background L \mathcal{L} L through
g a b d x a d x b = − ( r + L ( t ) 2 r ) 2 d t 2 + d r 2 r 2 and A a d x a = ( r − L ( t ) 2 r ) d t . (31) g_{ab}dx^a dx^b=-\left(r+\frac{\mathcal{L}(t)}{2r}\right)^2dt^2+\frac{dr^2}{r^2}
\qquad\text{and}\qquad
A_a dx^a=\left(r-\frac{\mathcal{L}(t)}{2r}\right)dt . \tag*{(31)} g ab d x a d x b = − ( r + 2 r L ( t ) ) 2 d t 2 + r 2 d r 2 and A a d x a = ( r − 2 r L ( t ) ) d t . (31)
The deviation from the background solution is driven by the scalar χ \chi χ and the one-form T = T a ( x a , θ ) d x a \mathcal{T}=\mathcal{T}_a(x^a,\theta)dx^a T = T a ( x a , θ ) d x a , multiplied by the perturbative constant δ \delta δ .
Throughout the paper we use the shorthand
B ( t , r ) : = 2 r 2 + L ( t ) , (32) B(t,r):=2r^2+\mathcal{L}(t), \tag*{(32)} B ( t , r ) := 2 r 2 + L ( t ) , (32)
so that the lapse of the AdS2 _2 2 base is r + L / 2 r = B / 2 r r+\mathcal{L}/2r=B/2r r + L /2 r = B /2 r and g t t = − B 2 / 4 r 2 g_{tt}=-B^2/4r^2 g tt = − B 2 /4 r 2 . We also recall that the only nonvanishing component of the connection in (31) is
A t ( t , r ) = r − L ( t ) 2 r . (33) A_t(t,r)=r-\frac{\mathcal{L}(t)}{2r}. \tag*{(33)} A t ( t , r ) = r − 2 r L ( t ) . (33)
Weyl scalars
To analyze the nature of these perturbations we use the Newman–Penrose formalism [45 , 47 ] , in which the metric is written as g μ ν = − l μ n ν − l ν n μ + m μ m ˉ ν + m ˉ μ m ν g_{\mu\nu}=-l_\mu n_\nu-l_\nu n_\mu+m_\mu\bar{m}_\nu+\bar{m}_\mu m_\nu g μν = − l μ n ν − l ν n μ + m μ m ˉ ν + m ˉ μ m ν in terms of a complex null tetrad normalized as − l ⋅ n = m ⋅ m ˉ = 1 -l\cdot n=m\cdot\bar{m}=1 − l ⋅ n = m ⋅ m ˉ = 1 for signature ( − , + , + , + ) (-,+,+,+) ( − , + , + , + ) , with all remaining inner products vanishing, and with l l l and n n n tangent to affinely parametrized null geodesics. A tetrad adapted to (14) is, in terms of (5) , (32) and (33) ,
l = 2 r 2 A t Λ ( θ ) B 2 ( − ∂ t + B 2 ∂ r + A t ∂ ϕ ) , n = 1 A t ( − ∂ t − B 2 ∂ r + A t ∂ ϕ ) , m = 1 2 ( 1 Λ ( θ ) ∂ θ + i Γ ( θ ) ∂ ϕ ) . (34) \begin{aligned}
l&=\frac{2r^2A_t}{\Lambda(\theta)B^2}\left(-\partial_t+\frac{B}{2}\partial_r+A_t\partial_\phi\right),\\
n&=\frac{1}{A_t}\left(-\partial_t-\frac{B}{2}\partial_r+A_t\partial_\phi\right),\\
m&=\frac{1}{\sqrt{2}}\left(\frac{1}{\sqrt{\Lambda(\theta)}}\partial_\theta+\frac{i}{\sqrt{\Gamma(\theta)}}\partial_\phi\right). \tag*{(34)}
\end{aligned} l n m = Λ ( θ ) B 2 2 r 2 A t ( − ∂ t + 2 B ∂ r + A t ∂ ϕ ) , = A t 1 ( − ∂ t − 2 B ∂ r + A t ∂ ϕ ) , = 2 1 ( Λ ( θ ) 1 ∂ θ + Γ ( θ ) i ∂ ϕ ) . (34)
The vectors l l l and n n n differ only in the sign of their radial component: both are built on the horizontal combination − ∂ t + A t ∂ ϕ -\partial_t + A_t\partial_\phi − ∂ t + A t ∂ ϕ , orthogonal to the U ( 1 ) U(1) U ( 1 ) fiber, while m m m is the orthonormal frame of the fiber block Λ d θ 2 + Γ ( d ϕ + A t d t ) 2 \Lambda\,d\theta^2 + \Gamma(d\phi+ A_tdt)^2 Λ d θ 2 + Γ ( d ϕ + A t d t ) 2 . Of the five Weyl scalars built by contracting the Weyl curvature C μ ν α β C_{\mu\nu\alpha\beta} C μν α β with this basis we shall need the two of helicity ± 2 \pm2 ± 2 ,
Ψ 0 = C μ ν α β l μ m ˉ ν l α m ˉ β , Ψ 4 = C μ ν α β n μ m ˉ ν n α m ˉ β , (35) \Psi_0 = C_{\mu\nu\alpha\beta}l^\mu\bar{m}^\nu l^\alpha\bar{m}^\beta,\qquad\Psi_4 = C_{\mu\nu\alpha\beta}n^\mu\bar{m}^\nu n^\alpha\bar{m}^\beta, \tag*{(35)} Ψ 0 = C μν α β l μ m ˉ ν l α m ˉ β , Ψ 4 = C μν α β n μ m ˉ ν n α m ˉ β , (35)
together with Ψ 2 \Psi_2 Ψ 2 . On the unperturbed geometry (14) only the latter survives,
Ψ 0 = Ψ 1 = Ψ 3 = Ψ 4 = 0 , Ψ 2 = − i M 2 ( cos θ − i ) 3 , (36) \Psi_0 = \Psi_1 = \Psi_3 = \Psi_4 = 0,\qquad\Psi_2 = -\frac{i}{M^2(\cos\theta-i)^3}, \tag*{(36)} Ψ 0 = Ψ 1 = Ψ 3 = Ψ 4 = 0 , Ψ 2 = − M 2 ( cos θ − i ) 3 i , (36)
so that (14) is of Petrov type D, as is the Kerr geometry from which it descends. Two standard consequences will be used. By the Goldberg–Sachs theorem the principal null directions of a vacuum type D spacetime are shear-free geodesic congruences; and Ψ 0 \Psi_0 Ψ 0 and Ψ 4 \Psi_4 Ψ 4 are the only irreducible components of the Weyl tensor carrying two transverse directions and helicity ± 2 \pm2 ± 2 , hence the only ones that propagate. A perturbation therefore carries gravitational degrees of freedom precisely when it renders them non-vanishing.
At linear order in the perturbations, only Ψ 0 \Psi_0 Ψ 0 and Ψ 4 \Psi_4 Ψ 4 are invariant under tetrad rotations and diffeomorphisms. It is obtained that
Ψ 0 = M 2 A t 2 ( i + cos θ ) 2 8 Λ ( θ ) 3 sin 2 θ δ l a l b ∇ a ∇ b χ + O ( δ 2 ) = M 2 r 4 A t 4 ( i + cos θ ) 2 2 Λ ( θ ) 5 B 4 sin 2 θ δ k + a k + b ∇ a ∇ b χ + O ( δ 2 ) , (37) \Psi_0 = \frac{M^2 A_t^2(i+\cos\theta)^2}{8\Lambda(\theta)^3\sin^2\theta}\,\delta\,l^a l^b\nabla_a\nabla_b\chi+ O(\delta^2) = \frac{M^2 r^4 A_t^4(i+\cos\theta)^2}{2\Lambda(\theta)^5 B^4\sin^2\theta}\,\delta\,k_+^a k_+^b\nabla_a\nabla_b\chi+ O(\delta^2), \tag*{(37)} Ψ 0 = 8Λ ( θ ) 3 sin 2 θ M 2 A t 2 ( i + cos θ ) 2 δ l a l b ∇ a ∇ b χ + O ( δ 2 ) = 2Λ ( θ ) 5 B 4 sin 2 θ M 2 r 4 A t 4 ( i + cos θ ) 2 δ k + a k + b ∇ a ∇ b χ + O ( δ 2 ) , (37)
Ψ 4 = 2 M 2 ( i + cos θ ) 2 Λ ( θ ) A t 2 sin 2 θ δ n a n b ∇ a ∇ b χ + O ( δ 2 ) = 2 M 2 ( i + cos θ ) 2 Λ ( θ ) A t 4 sin 2 θ δ k − a k − b ∇ a ∇ b χ + O ( δ 2 ) , (38) \Psi_4 = \frac{2M^2(i+\cos\theta)^2}{\Lambda(\theta)A_t^2\sin^2\theta}\,\delta\,n^a n^b\nabla_a\nabla_b\chi+ O(\delta^2) = \frac{2M^2(i+\cos\theta)^2}{\Lambda(\theta)A_t^4\sin^2\theta}\,\delta\,k_-^a k_-^b\nabla_a\nabla_b\chi+ O(\delta^2), \tag*{(38)} Ψ 4 = Λ ( θ ) A t 2 sin 2 θ 2 M 2 ( i + cos θ ) 2 δ n a n b ∇ a ∇ b χ + O ( δ 2 ) = Λ ( θ ) A t 4 sin 2 θ 2 M 2 ( i + cos θ ) 2 δ k − a k − b ∇ a ∇ b χ + O ( δ 2 ) , (38)
where in the second equalities k ± : = − ∂ t ± B 2 ∂ r k_\pm:= -\partial_t \pm\frac{B}{2}\partial_r k ± := − ∂ t ± 2 B ∂ r are the null vectors of the base g a b g_{ab} g ab obtained by projecting l l l and n n n , so that l a = 2 r 2 A t Λ B 2 k + a l^a = \frac{2r^2 A_t}{\Lambda B^2}k_+^a l a = Λ B 2 2 r 2 A t k + a and n a = A t − 1 k − a n^a = A_t^{-1}k_-^a n a = A t − 1 k − a . Both scalars are proportional to χ \chi χ , meaning that in (42) , χ \chi χ is gauge invariant. In both cases, we can relate to the standard separation of variables used to solve the Teukolsky master equation presented in the Appendix § 7 , and identify U l ( t , r ) U_l(t,r) U l ( t , r ) .
Linearized Einstein's equations
We solve Einstein’s equations perturbatively, expanding the Ricci tensor as
R μ ν = R μ ν ( 0 ) + δ R μ ν ( 1 ) + O ( δ 2 ) R_{\mu\nu} = R_{\mu\nu}^{(0)} + \delta R_{\mu\nu}^{(1)} + O(\delta^2) R μν = R μν ( 0 ) + δ R μν ( 1 ) + O ( δ 2 )
and solve the equations at linear order in δ \delta δ , so that
R μ ν ( 1 ) = 0. R_{\mu\nu}^{(1)} = 0. R μν ( 1 ) = 0.
Inserting the metric (30) gives the following system of equation
∂ θ ( ∇ b T t b ) = 0 (39) \partial_\theta\left(\nabla_b T^{tb}\right)=0 \tag*{(39)} ∂ θ ( ∇ b T t b ) = 0 (39)
sin 2 θ ∂ θ ( ϵ a b T b ) + cot θ ∇ a χ = 0 (40) \sin^2\theta\,\partial_\theta\left(\epsilon_{ab}T^b\right)+\cot\theta\,\nabla_a\chi=0 \tag*{(40)} sin 2 θ ∂ θ ( ϵ ab T b ) + cot θ ∇ a χ = 0 (40)
ϵ a b ∇ a T b − cos 2 θ sin 3 θ ∂ θ ( χ cos θ ) = 0. (41) \epsilon^{ab}\nabla_aT_b-\frac{\cos^2\theta}{\sin^3\theta}\partial_\theta\left(\frac{\chi}{\cos\theta}\right)=0. \tag*{(41)} ϵ ab ∇ a T b − sin 3 θ cos 2 θ ∂ θ ( cos θ χ ) = 0. (41)
Here ϵ a b \epsilon_{ab} ϵ ab is the antisymmetric pseudo-tensor, where ϵ t r = − det g a b \epsilon_{tr}=\sqrt{-\det g_{ab}} ϵ t r = − det g ab and all the indices are lowered and raised with the 2-dimensional metric g a b g_{ab} g ab .
Acting with ∂ θ \partial_\theta ∂ θ on (41) and inserting (40) gives the scalar equation
□ 2 χ + sin 3 θ cos θ ∂ θ [ cos 2 θ sin 3 θ ∂ θ ( χ cos θ ) ] = 0. (42) \Box_2\chi+\frac{\sin^3\theta}{\cos\theta}\partial_\theta\left[\frac{\cos^2\theta}{\sin^3\theta}\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\right]=0. \tag*{(42)} □ 2 χ + cos θ sin 3 θ ∂ θ [ sin 3 θ cos 2 θ ∂ θ ( cos θ χ ) ] = 0. (42)
Here and in the following, □ 2 \Box_2 □ 2 is the nearly-AdS2 _2 2 Laplacian,
□ 2 = 4 r 2 L ′ B 3 ∂ t − 4 r 2 B 2 ∂ t 2 + 4 r 3 B ∂ r + r 2 ∂ r 2 . (43) \Box_2=\frac{4r^2\mathcal{L}'}{B^3}\partial_t-\frac{4r^2}{B^2}\partial_t^2+\frac{4r^3}{B}\partial_r+r^2\partial_r^2. \tag*{(43)} □ 2 = B 3 4 r 2 L ′ ∂ t − B 2 4 r 2 ∂ t 2 + B 4 r 3 ∂ r + r 2 ∂ r 2 . (43)
Since the scalar χ \chi χ determines the Weyl scalars Ψ 0 \Psi_0 Ψ 0 and Ψ 4 \Psi_4 Ψ 4 , we focus on the scalar equation (42) in the following.
The angular part
The coefficients in (42) are singular near the poles θ = 0 , π \theta=0,\pi θ = 0 , π . This forces nonsingular solutions χ \chi χ to vanish suitably at these points, which guarantees that the metric (30) is free of conical singularities at these points.
One way to see this is to separate variables as in
χ ( x a , θ ) = sin 2 θ S ( θ ) χ ~ ( x a ) , \chi(x^a,\theta)=\sin^2\theta S(\theta)\widetilde{\chi}(x^a), χ ( x a , θ ) = sin 2 θ S ( θ ) χ ( x a ) ,
which leads to the equation
∂ θ 2 S + cot θ ∂ θ S + ( K − 4 sin 2 θ ) S = 0. (44) \partial_\theta^2S+\cot\theta\,\partial_\theta S+\left(K-\frac{4}{\sin^2\theta}\right)S=0. \tag*{(44)} ∂ θ 2 S + cot θ ∂ θ S + ( K − sin 2 θ 4 ) S = 0. (44)
which reduces to Legendre’s equation with x = cos ( θ ) x=\cos(\theta) x = cos ( θ ) . For the solution to be free of conical singularities, we need that S ( θ ) S(\theta) S ( θ ) is bounded for θ \theta θ near 0 , π 0,\pi 0 , π , such that χ ( x a , θ ) \chi(x^a,\theta) χ ( x a , θ ) tends to zero as θ \theta θ approaches 0 , π 0,\pi 0 , π . The Legendre equation for S S S has bounded solutions as soon as K = l ( l + 1 ) K=l(l+1) K = l ( l + 1 ) for some l ∈ N l\in\mathbb{N} l ∈ N with l ≥ 2 l\ge2 l ≥ 2 . They are of the form
S ( θ ) = c 1 P l ( 2 ) ( cos θ ) (45) S(\theta)=c_1P_l^{(2)}(\cos\theta) \tag*{(45)} S ( θ ) = c 1 P l ( 2 ) ( cos θ ) (45)
with P l ( m ) P_l^{(m)} P l ( m ) the associated Legendre function of the first kind.
For later use, we recall that
P l ( 2 ) ( cos ( θ ) ) = sin 2 ( θ ) p l ( cos ( θ ) ) (46) P_l^{(2)}(\cos(\theta)) = \sin^2(\theta)p_l(\cos(\theta)) \tag*{(46)} P l ( 2 ) ( cos ( θ )) = sin 2 ( θ ) p l ( cos ( θ )) (46)
with p l p_l p l some polynomial of degree l − 2 l-2 l − 2 . Thus, the general solution has the form
χ ( x a , θ ) = sin 4 ( θ ) ∑ l = 2 ∞ χ l ( x a ) p l ( cos ( θ ) ) . \chi(x^a,\theta) = \sin^4(\theta)\sum_{l=2}^{\infty}\chi_l(x^a)p_l(\cos(\theta)). χ ( x a , θ ) = sin 4 ( θ ) l = 2 ∑ ∞ χ l ( x a ) p l ( cos ( θ )) .
where each mode χ l \chi_l χ l satisfies the wave equation
□ 2 χ ~ ( x ) = l ( l + 1 ) χ ~ ( x ) (47) \Box_2\widetilde{\chi}(x) = l(l+1)\widetilde{\chi}(x) \tag*{(47)} □ 2 χ ( x ) = l ( l + 1 ) χ ( x ) (47)
for the nearly-AdS2 _2 2 Laplacian (3.14). For the cases L \mathcal{L} L constant and equal to zero, the equation (3.18) becomes the radial Teukolsky equation for axisymmetric perturbations.
Killing Vector Fields and Energy Formalism
In this section, we aim to obtain identities on the relation between a vector field ζ \zeta ζ parametrized by a time-dependent function ϵ \epsilon ϵ as in Section 2.1, and the perturbed AdS2 _2 2 -metric that contains the free function L \mathcal{L} L . This first part is key to develop the energy formalism for the equation (3.13).
We will use a vector field ζ \zeta ζ parametrized as in Section 2.1 by a t t t -dependent function ϵ \epsilon ϵ . Recall that we took
ζ = ( ϵ ( t ) + ϵ ′ ′ ( t ) 2 r 2 + L ( t ) ) ∂ t − r ϵ ′ ( t ) ∂ r − 2 r ϵ ′ ′ ( t ) 2 r 2 + L ( t ) ∂ ϕ + c ϕ ∂ ϕ , (48) \zeta= \left(\epsilon(t) + \frac{\epsilon^{\prime\prime}(t)}{2r^2+\mathcal{L}(t)}\right)\partial_t-r\epsilon^\prime(t)\partial_r-\frac{2r\epsilon^{\prime\prime}(t)}{2r^2+\mathcal{L}(t)}\partial_\phi+c_\phi\partial_\phi, \tag*{(48)} ζ = ( ϵ ( t ) + 2 r 2 + L ( t ) ϵ ′′ ( t ) ) ∂ t − r ϵ ′ ( t ) ∂ r − 2 r 2 + L ( t ) 2 r ϵ ′′ ( t ) ∂ ϕ + c ϕ ∂ ϕ , (48)
where ϵ ( t ) \epsilon(t) ϵ ( t ) is an arbitrary function of time related to the time reparametrization symmetry and c ϕ c_\phi c ϕ is an arbitrary constant associated with the angular invariance.
The deformation tensor
First, we begin by computing the components of the Lie derivative along the vector field ζ \zeta ζ , we obtain through direct computation that the components of the deformation tensor π \pi π are given by
π r r = ζ r ∂ r g r r + 2 g r r ∂ r ζ r = − r ϵ ′ ∂ r ( 1 r 2 ) + 2 r 2 ∂ r ( − r ϵ ′ ) = 0 , (49) \pi_{rr}=\zeta^r\partial_r g_{rr}+2g_{rr}\partial_r\zeta^r=-r\epsilon^\prime\partial_r\left(\frac{1}{r^2}\right)+\frac{2}{r^2}\partial_r(-r\epsilon^\prime)=0, \tag*{(49)} π r r = ζ r ∂ r g r r + 2 g r r ∂ r ζ r = − r ϵ ′ ∂ r ( r 2 1 ) + r 2 2 ∂ r ( − r ϵ ′ ) = 0 , (49)
π t r = g r r ∂ t ζ r + g t t ∂ r ζ t = − ϵ ′ ′ r − B 2 4 r 2 ∂ r ( ϵ + ϵ ′ ′ B ) = 0 , (50) \pi_{tr}=g_{rr}\partial_t\zeta^r+g_{tt}\partial_r\zeta^t=-\frac{\epsilon^{\prime\prime}}{r}-\frac{B^2}{4r^2}\partial_r\left(\epsilon+\frac{\epsilon^{\prime\prime}}{B}\right)=0, \tag*{(50)} π t r = g r r ∂ t ζ r + g tt ∂ r ζ t = − r ϵ ′′ − 4 r 2 B 2 ∂ r ( ϵ + B ϵ ′′ ) = 0 , (50)
and
π t t = ζ r ∂ r g t t + ζ t ∂ t g t t + 2 g t t ∂ t ζ t = − r ε ′ ( − 32 B r 3 + 8 B 2 r 16 r 4 ) − ( ε + ε ′ ′ B ) B C ′ 2 r 2 − B 2 2 r 2 ( ε ′ + ε ′ ′ ′ B − ε ′ ′ C ′ B 2 ) = B ε ′ ( 2 − B r 2 ) − B 2 r 2 ( ε C ′ + ε ′ ′ ′ ) = − B 2 r 2 δ ε L . (51) \begin{aligned}
\pi_{tt}=\zeta^{r}\partial_{r}g_{tt}+\zeta^{t}\partial_{t}g_{tt}+2g_{tt}\partial_{t}\zeta^{t} \\
&=-r\varepsilon'\left(\frac{-32Br^{3}+8B^{2}r}{16r^{4}}\right)-\left(\varepsilon+\frac{\varepsilon''}{B}\right)\frac{BC'}{2r^{2}}-\frac{B^{2}}{2r^{2}}\left(\varepsilon'+\frac{\varepsilon'''B-\varepsilon''C'}{B^{2}}\right) \\
&=B\varepsilon'\left(2-\frac{B}{r^{2}}\right)-\frac{B}{2r^{2}}\left(\varepsilon C'+\varepsilon'''\right)=-\frac{B}{2r^{2}}\delta_{\varepsilon}\mathcal{L}. \tag*{(51)}
\end{aligned} π tt = ζ r ∂ r g tt + ζ t ∂ t g tt + 2 g tt ∂ t ζ t = − r ε ′ ( 16 r 4 − 32 B r 3 + 8 B 2 r ) − ( ε + B ε ′′ ) 2 r 2 B C ′ − 2 r 2 B 2 ( ε ′ + B 2 ε ′′′ B − ε ′′ C ′ ) = B ε ′ ( 2 − r 2 B ) − 2 r 2 B ( ε C ′ + ε ′′′ ) = − 2 r 2 B δ ε L . (51)
Here B B B is defined in (3.3) and, using the definition of δ ε L \delta_{\varepsilon}\mathcal{L} δ ε L in (2.18),
δ ε L = ε ′ ′ ′ + 2 L ε ′ + L ′ ε . \delta_{\varepsilon}\mathcal{L}=\varepsilon'''+2\mathcal{L}\varepsilon'+\mathcal{L}'\varepsilon. δ ε L = ε ′′′ + 2 L ε ′ + L ′ ε .
The last component of the deformation tensor π t t \pi_{tt} π tt reveals the conditional character of ζ \zeta ζ ; that is, for a given ε \varepsilon ε , the field ζ [ ε ] \zeta[\varepsilon] ζ [ ε ] is a Killing field of g a b g_{ab} g ab if and only if δ ε L = 0 \delta_{\varepsilon}\mathcal{L}=0 δ ε L = 0 . For every profile L ( t ) \mathcal{L}(t) L ( t ) this third-order linear equation has a three-dimensional space of solutions 2.1, so Killing fields always exist.
Using the previous results, a direct computation shows that the commutator of the wave operator □ 2 \Box_{2} □ 2 with the vector field ζ \zeta ζ can also be characterized in terms of the symmetry term δ ε L \delta_{\varepsilon}\mathcal{L} δ ε L , as follows
[ □ 2 , ζ ] = δ ε L ( − 8 r 2 B 3 ∂ t 2 + 12 r 2 L ′ B 4 ∂ t + 4 r 3 B 2 ∂ r ) − 4 r 2 B 3 ( δ ε L ) ′ ∂ t . [\Box_{2},\zeta]=\delta_{\varepsilon}\mathcal{L}\left(-\frac{8r^{2}}{B^{3}}\partial_{t}^{2}+\frac{12r^{2}\mathcal{L}'}{B^{4}}\partial_{t}+\frac{4r^{3}}{B^{2}}\partial_{r}\right)-\frac{4r^{2}}{B^{3}}(\delta_{\varepsilon}\mathcal{L})'\partial_{t}. [ □ 2 , ζ ] = δ ε L ( − B 3 8 r 2 ∂ t 2 + B 4 12 r 2 L ′ ∂ t + B 2 4 r 3 ∂ r ) − B 3 4 r 2 ( δ ε L ) ′ ∂ t .
Thus, δ ε L = 0 \delta_{\varepsilon}\mathcal{L}=0 δ ε L = 0 implies that [ □ 2 , ζ ] = 0 [\Box_{2},\zeta]=0 [ □ 2 , ζ ] = 0 . Moreover, since ζ \zeta ζ is independent of θ \theta θ and
ρ θ [ ⋅ ] : = sin 3 θ cos θ ∂ θ [ cos 2 θ sin 3 θ ∂ θ ( ⋅ cos θ ) ] \rho_{\theta}[\cdot]:=\frac{\sin^{3}\theta}{\cos\theta}\partial_{\theta}\left[\frac{\cos^{2}\theta}{\sin^{3}\theta}\partial_{\theta}\left(\frac{\cdot}{\cos\theta}\right)\right] ρ θ [ ⋅ ] := cos θ sin 3 θ ∂ θ [ sin 3 θ cos 2 θ ∂ θ ( cos θ ⋅ ) ]
is independent of ( r , t ) (r,t) ( r , t ) , we also have
[ ρ θ , ζ ] = 0. [\rho_{\theta},\zeta]=0. [ ρ θ , ζ ] = 0.
The goal is now to develop the energy formalism for equation (3.13), and to characterize δ ε L \delta_{\varepsilon}\mathcal{L} δ ε L in such a way that allows us to establish a proper coercivity criterion. In the following section, we establish a unified identity for the energy density, depending on ε \varepsilon ε , and derive an exact coercivity criterion for the case in which ζ \zeta ζ is a Killing vector field.
Stability from energy conservation
Recall that the full perturbation equation for the nearly-AdS2 _2 2 Laplacian (3.14) for χ \chi χ depending on t t t , r r r , θ \theta θ is given by
□ 2 χ + sin 3 θ cos θ ∂ θ [ cos 2 θ sin 3 θ ∂ θ ( χ cos θ ) ] = 0 , ( t , r , θ ) ∈ R × ( 0 , ∞ ) × ( 0 , π ) . (52) \Box_{2}\chi+\frac{\sin^{3}\theta}{\cos\theta}\partial_{\theta}\left[\frac{\cos^{2}\theta}{\sin^{3}\theta}\partial_{\theta}\left(\frac{\chi}{\cos\theta}\right)\right]=0,\qquad(t,r,\theta)\in\mathbb{R}\times(0,\infty)\times(0,\pi). \tag*{(52)} □ 2 χ + cos θ sin 3 θ ∂ θ [ sin 3 θ cos 2 θ ∂ θ ( cos θ χ ) ] = 0 , ( t , r , θ ) ∈ R × ( 0 , ∞ ) × ( 0 , π ) . (52)
Expanding the angular part gives
□ 2 χ + χ θ θ − 3 cot θ χ θ − 2 χ = □ 2 χ + 1 s ( θ ) ∂ θ ( s ( θ ) χ θ ) − 2 χ = 0 , (53) \Box_2\chi+\chi_{\theta\theta}-3\cot\theta\chi_\theta-2\chi=\Box_2\chi+\frac{1}{s(\theta)}\partial_\theta(s(\theta)\chi_\theta)-2\chi=0, \tag*{(53)} □ 2 χ + χ θ θ − 3 cot θ χ θ − 2 χ = □ 2 χ + s ( θ ) 1 ∂ θ ( s ( θ ) χ θ ) − 2 χ = 0 , (53)
with s ( θ ) = sin − 3 ( θ ) s(\theta)=\sin^{-3}(\theta) s ( θ ) = sin − 3 ( θ ) , and ( t , r , θ ) ∈ ( 0 , ∞ ) 2 × ( 0 , π ) (t,r,\theta)\in(0,\infty)^2\times(0,\pi) ( t , r , θ ) ∈ ( 0 , ∞ ) 2 × ( 0 , π ) .
Writing its angular part in Sturm–Liouville form ρ θ χ = 1 s ( θ ) ∂ θ ( s ( θ ) χ θ ) − 2 χ \rho_\theta\chi=\frac{1}{s(\theta)}\partial_\theta(s(\theta)\chi_\theta)-2\chi ρ θ χ = s ( θ ) 1 ∂ θ ( s ( θ ) χ θ ) − 2 χ , we can identify the weight that makes the wave-type operator formally self-adjoint, given by
s ( θ ) − g ˉ d θ d t d r : = sin − 3 ( θ ) B 2 r 2 d θ d t d r . s(\theta)\sqrt{-\bar{g}}\,d\theta\,dt\,dr:=\sin^{-3}(\theta)\frac{B}{2r^2}\,d\theta\,dt\,dr. s ( θ ) − g ˉ d θ d t d r := sin − 3 ( θ ) 2 r 2 B d θ d t d r .
Here and in the following, g ˉ \bar{g} g ˉ denotes the determinant of the t , r t,r t , r -metric g a b g_{ab} g ab .
If χ \chi χ and its derivatives vanish suitably at the poles, the equation (53) matches the Euler-Lagrange equation of the Lagrangian
L [ χ ] = − g ˉ sin 3 ( θ ) [ − 1 2 g a b ∂ a χ ∂ b χ − 1 2 ( χ θ 2 + 2 χ 2 ) ] , a , b ∈ { t , r } . (54) \mathcal{L}[\chi]=\frac{\sqrt{-\bar{g}}}{\sin^3(\theta)}\left[-\frac{1}{2}g^{ab}\partial_a\chi\partial_b\chi-\frac{1}{2}(\chi_\theta^2+2\chi^2)\right],\qquad a,b\in\{t,r\}. \tag*{(54)} L [ χ ] = sin 3 ( θ ) − g ˉ [ − 2 1 g ab ∂ a χ ∂ b χ − 2 1 ( χ θ 2 + 2 χ 2 ) ] , a , b ∈ { t , r } . (54)
The associated energy-momentum tensor on the ( t , r ) (t,r) ( t , r ) -sector is
T a b = ∂ a χ ∂ b χ − 1 2 g a b ( g c d ∂ c χ ∂ d χ + χ θ 2 + 2 χ 2 ) , a , b ∈ { t , r } . (55) T_{ab}=\partial_a\chi\,\partial_b\chi-\frac{1}{2}g_{ab}\left(g^{cd}\partial_c\chi\,\partial_d\chi+\chi_\theta^2+2\chi^2\right),\qquad a,b\in\{t,r\}. \tag*{(55)} T ab = ∂ a χ ∂ b χ − 2 1 g ab ( g c d ∂ c χ ∂ d χ + χ θ 2 + 2 χ 2 ) , a , b ∈ { t , r } . (55)
and the explicit components are given by
T t t = 1 2 χ t 2 + B 2 8 χ r 2 + B 2 8 r 2 ( χ θ 2 + 2 χ 2 ) , T_{tt}=\frac{1}{2}\chi_t^2+\frac{B^2}{8}\chi_r^2+\frac{B^2}{8r^2}\left(\chi_\theta^2+2\chi^2\right), T tt = 2 1 χ t 2 + 8 B 2 χ r 2 + 8 r 2 B 2 ( χ θ 2 + 2 χ 2 ) ,
T r r = 1 2 χ r 2 + 2 B 2 χ t 2 − 1 2 r 2 ( χ θ 2 + 2 χ 2 ) , T_{rr}=\frac{1}{2}\chi_r^2+\frac{2}{B^2}\chi_t^2-\frac{1}{2r^2}\left(\chi_\theta^2+2\chi^2\right), T r r = 2 1 χ r 2 + B 2 2 χ t 2 − 2 r 2 1 ( χ θ 2 + 2 χ 2 ) ,
T t r = χ t χ r . (56) T_{tr}=\chi_t\chi_r. \tag*{(56)} T t r = χ t χ r . (56)
We regard (54) as the variational functional associated with equation (53) . The resulting stress tensor should therefore be understood as an analytic energy functional for the perturbation equation.
In order to propose the energy functional, recall the generalized Killing vector field
ζ = ( ϵ + ϵ ′ ′ 2 r 2 + L ) ∂ t − r ϵ ′ ∂ r + ( − 2 r ϵ ′ ′ 2 r 2 + L + c ϕ ) ∂ ϕ . (57) \zeta=\left(\epsilon+\frac{\epsilon''}{2r^2+\mathcal{L}}\right)\partial_t-r\epsilon'\partial_r+\left(-\frac{2r\epsilon''}{2r^2+\mathcal{L}}+c_\phi\right)\partial_\phi. \tag*{(57)} ζ = ( ϵ + 2 r 2 + L ϵ ′′ ) ∂ t − r ϵ ′ ∂ r + ( − 2 r 2 + L 2 r ϵ ′′ + c ϕ ) ∂ ϕ . (57)
Assuming axisymmetry, ∂ ϕ χ = 0 \partial_{\phi}\chi= 0 ∂ ϕ χ = 0 , the ϕ \phi ϕ -component does not contribute dynamically. We define the component of the current as
J a = − T b a ζ b , and J θ : = − χ θ ζ ( χ ) , J^{a} = -T^{a}_{\ b}\zeta^{b}, \qquad\text{and} \qquad J^{\theta} := -\chi_{\theta}\zeta(\chi), J a = − T b a ζ b , and J θ := − χ θ ζ ( χ ) ,
then the explicit components are given by
J t = 4 r 2 B 2 ( T t t ζ t + T t r ζ r ) , J^{t} = \frac{4r^{2}}{B^{2}}\left(T_{tt}\zeta^{t} + T_{tr}\zeta^{r}\right), J t = B 2 4 r 2 ( T tt ζ t + T t r ζ r ) ,
J r = − r 2 ( T r t ζ t + T r r ζ r ) , J^{r} = -r^{2}\left(T_{rt}\zeta^{t} + T_{rr}\zeta^{r}\right), J r = − r 2 ( T r t ζ t + T r r ζ r ) ,
J θ = − χ θ ( B ϵ + ϵ ′ ′ B χ t − r ϵ ′ χ r ) , J^{\theta} = -\chi_{\theta}\left(\frac{B\epsilon+\epsilon''}{B}\chi_{t} - r\epsilon'\chi_{r}\right), J θ = − χ θ ( B B ϵ + ϵ ′′ χ t − r ϵ ′ χ r ) ,
and the divergence satisfies
∇ a J a = − ( ∇ a T a b ) ζ b − 1 2 T a b π a b . (58) \nabla_{a}J^{a} = -(\nabla^{a}T_{ab})\zeta^{b} - \frac{1}{2}T^{ab}\pi_{ab}. \tag*{(58)} ∇ a J a = − ( ∇ a T ab ) ζ b − 2 1 T ab π ab . (58)
For any χ \chi χ we have
s ( θ ) ∇ a T a b = s ( θ ) ( □ 2 χ + ρ θ χ ) ∂ b χ − ∂ θ ( s ( θ ) χ θ ∂ b χ ) , b ∈ { t , r } , s(\theta)\nabla^{a}T_{ab} = s(\theta)(\Box_{2}\chi+\rho_{\theta}\chi)\partial_{b}\chi- \partial_{\theta}(s(\theta)\chi_{\theta}\partial_{b}\chi), \qquad b\in\{t,r\}, s ( θ ) ∇ a T ab = s ( θ ) ( □ 2 χ + ρ θ χ ) ∂ b χ − ∂ θ ( s ( θ ) χ θ ∂ b χ ) , b ∈ { t , r } ,
furthermore, since
∇ a J a = 1 − g ∂ a ( − g J a ) , \nabla_{a}J^{a} = \frac{1}{\sqrt{-g}}\partial_{a}\left(\sqrt{-g}J^{a}\right), ∇ a J a = − g 1 ∂ a ( − g J a ) ,
hence, the equation (3.29) gives rise to the following identity
∇ a ( s ( θ ) J a ) + ∂ θ ( s ( θ ) J θ ) = − s ( θ ) ( □ 2 χ + ρ θ χ ) ζ ( χ ) − 1 2 s ( θ ) T a b π a b , (59) \nabla_{a}\left(s(\theta)J^{a}\right)+\partial_{\theta}\left(s(\theta)J^{\theta}\right) = -s(\theta)(\Box_{2}\chi+\rho_{\theta}\chi)\zeta(\chi)-\frac{1}{2}s(\theta)T^{ab}\pi_{ab}, \tag*{(59)} ∇ a ( s ( θ ) J a ) + ∂ θ ( s ( θ ) J θ ) = − s ( θ ) ( □ 2 χ + ρ θ χ ) ζ ( χ ) − 2 1 s ( θ ) T ab π ab , (59)
using (3.22), the last term above is given by
T a b π a b = ( g t t ) 2 T t t π t t = 16 r 4 B 4 T t t ( − B 2 r 2 δ ϵ L ) = − 8 r 2 B 3 δ ϵ L T t t . (60) T^{ab}\pi_{ab} = (g^{tt})^{2}T_{tt}\pi_{tt} = \frac{16r^{4}}{B^{4}}T_{tt}\left(-\frac{B}{2r^{2}}\delta_{\epsilon}\mathcal{L}\right) = -\frac{8r^{2}}{B^{3}}\delta_{\epsilon}\mathcal{L}T_{tt}. \tag*{(60)} T ab π ab = ( g tt ) 2 T tt π tt = B 4 16 r 4 T tt ( − 2 r 2 B δ ϵ L ) = − B 3 8 r 2 δ ϵ L T tt . (60)
The divergence equation (3.30) gives us the following key quasi-continuity equation that describe the divergence identity with angular flux
∂ t ( − g s ( θ ) J t ) + ∂ r ( − g s ( θ ) J r ) + ∂ θ ( − g s ( θ ) J θ ) = − − g s ( θ ) ( □ 2 χ + ρ θ χ ) ζ ( χ ) + 2 s ( θ ) B 2 δ ϵ L T t t . (61) \begin{aligned}
\partial_{t}\left(\sqrt{-g}s(\theta)J^{t}\right)+\partial_{r}\left(\sqrt{-g}s(\theta)J^{r}\right)+\partial_{\theta}\left(\sqrt{-g}s(\theta)J^{\theta}\right)
&= -\sqrt{-g}s(\theta)(\Box_{2}\chi+\rho_{\theta}\chi)\zeta(\chi) \\
&\quad+ \frac{2s(\theta)}{B^{2}}\delta_{\epsilon}\mathcal{L}T_{tt}. \tag*{(61)}
\end{aligned} ∂ t ( − g s ( θ ) J t ) + ∂ r ( − g s ( θ ) J r ) + ∂ θ ( − g s ( θ ) J θ ) = − − g s ( θ ) ( □ 2 χ + ρ θ χ ) ζ ( χ ) + B 2 2 s ( θ ) δ ϵ L T tt . (61)
Consequently, whenever δ ϵ L = 0 \delta_{\epsilon}\mathcal{L}=0 δ ϵ L = 0 , ζ \zeta ζ is a Killing vector of g a b g_{ab} g ab , π a b = 0 \pi_{ab}=0 π ab = 0 , and the right-hand side of (3.30) vanishes on solutions of (3.23). Now, consider the spacetime regions given by
D t 0 : = { ( t , r , θ ) : 0 ≤ t ≤ t 0 , r ∈ ( 0 , ∞ ) , θ ∈ ( 0 , π ) } , D_{t_{0}} := \{(t,r,\theta): 0\leq t\leq t_{0},\ r\in(0,\infty),\ \theta\in(0,\pi)\}, D t 0 := {( t , r , θ ) : 0 ≤ t ≤ t 0 , r ∈ ( 0 , ∞ ) , θ ∈ ( 0 , π )} ,
and the time slice,
Σ t 0 = { ( t 0 , r , θ ) ∈ D t 0 } , \Sigma_{t_0}=\{(t_0,r,\theta)\in D_{t_0}\}, Σ t 0 = {( t 0 , r , θ ) ∈ D t 0 } ,
then the future unit normal vector is n = 2 r B ∂ t n=\frac{2r}{B}\partial_t n = B 2 r ∂ t , and we get in particular that
− g ˉ J t = − − g ˉ g ˉ t t T t b ζ b = B 2 r 2 4 r 2 B 2 T t b ζ b = 2 B ( T t t ζ t + T t r ζ r ) . \sqrt{-\bar{g}}\,J^t=-\sqrt{-\bar{g}}\,\bar{g}^{tt}T_{tb}\zeta^b=\frac{B}{2r^2}\frac{4r^2}{B^2}T_{tb}\zeta^b=\frac{2}{B}(T_{tt}\zeta^t+T_{tr}\zeta^r). − g ˉ J t = − − g ˉ g ˉ tt T t b ζ b = 2 r 2 B B 2 4 r 2 T t b ζ b = B 2 ( T tt ζ t + T t r ζ r ) .
Integrating (61) over ( r , θ ) ∈ ( 0 , ∞ ) × ( 0 , π ) (r,\theta)\in(0,\infty)\times(0,\pi) ( r , θ ) ∈ ( 0 , ∞ ) × ( 0 , π ) , the ∂ r \partial_r ∂ r and ∂ θ \partial_\theta ∂ θ terms become boundary terms that vanish under suitable assumptions, and the quantity whose time derivative remains defines the following energy formulation
E ϵ ( t ) : = ∫ Σ t T a b n a ζ b d r r s ( θ ) d θ = ∫ 0 π ∫ 0 ∞ − g ˉ s ( θ ) J t d r d θ = ∫ 0 π ∫ 0 ∞ s ( θ ) 2 B [ ζ t T t t + ζ r T t r ] d r d θ . (62) \begin{aligned}
E_\epsilon(t):=\int_{\Sigma_t}T_{ab}n^a\zeta^b\frac{\mathrm{d}r}{r}s(\theta)\mathrm{d}\theta\\
&=\int_0^\pi\int_0^\infty\sqrt{-\bar{g}}s(\theta)J^t\mathrm{d}r\mathrm{d}\theta\\
&=\int_0^\pi\int_0^\infty s(\theta)\frac{2}{B}\left[\zeta^tT_{tt}+\zeta^rT_{tr}\right]\mathrm{d}r\mathrm{d}\theta. \tag*{(62)}
\end{aligned} E ϵ ( t ) := ∫ Σ t T ab n a ζ b r d r s ( θ ) d θ = ∫ 0 π ∫ 0 ∞ − g ˉ s ( θ ) J t d r d θ = ∫ 0 π ∫ 0 ∞ s ( θ ) B 2 [ ζ t T tt + ζ r T t r ] d r d θ . (62)
The equation (62) is precisely the geometric version of an energy law for a hyperbolic equation. Inserting the components of the stress tensor given in (56) we get that the functional of energy formulation is given by
E ϵ [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) [ B ϵ + ϵ ′ ′ B 2 χ t 2 + B ϵ + ϵ ′ ′ 4 χ r 2 + B ϵ + ϵ ′ ′ 4 r 2 ( χ θ 2 + 2 χ 2 ) − 2 r ϵ ′ B χ t χ r ] d r d θ . (63) E_\epsilon[\chi]=\int_0^\pi\int_0^\infty s(\theta)\left[\frac{B\epsilon+\epsilon''}{B^2}\chi_t^2+\frac{B\epsilon+\epsilon''}{4}\chi_r^2+\frac{B\epsilon+\epsilon''}{4r^2}(\chi_\theta^2+2\chi^2)-\frac{2r\epsilon'}{B}\chi_t\chi_r\right]\mathrm{d}r\mathrm{d}\theta. \tag*{(63)} E ϵ [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) [ B 2 B ϵ + ϵ ′′ χ t 2 + 4 B ϵ + ϵ ′′ χ r 2 + 4 r 2 B ϵ + ϵ ′′ ( χ θ 2 + 2 χ 2 ) − B 2 r ϵ ′ χ t χ r ] d r d θ . (63)
If we assume finite-energy boundary conditions, namely
{ − g ˉ sin − 3 θ J r → 0 as r → 0 + ∧ r → ∞ , − g ˉ sin − 3 θ J θ → 0 as θ → 0 , π , (64) \left\{
\begin{aligned}
\sqrt{-\bar{g}}\sin^{-3}\theta J^r&\to0&&\text{as }r\to0^+\wedge r\to\infty,\\
\sqrt{-\bar{g}}\sin^{-3}\theta J^\theta&\to0&&\text{as }\theta\to0,\pi,
\end{aligned}
\right. \tag*{(64)} { − g ˉ sin − 3 θ J r − g ˉ sin − 3 θ J θ → 0 → 0 as r → 0 + ∧ r → ∞ , as θ → 0 , π , (64)
then, for each choice of the vector field ζ \zeta ζ parametrized by ϵ ( t ) \epsilon(t) ϵ ( t ) , from the continuity equation (61) we get
d d t E ϵ [ χ ] ( t ) = δ ϵ L ( t ) ∫ 0 π ∫ 0 ∞ s ( θ ) [ χ t 2 B 2 + χ r 2 4 + 1 4 r 2 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (65) \frac{\mathrm{d}}{\mathrm{d}t}E_\epsilon[\chi](t)=\delta_\epsilon\mathcal{L}(t)\int_0^\pi\int_0^\infty s(\theta)\left[\frac{\chi_t^2}{B^2}+\frac{\chi_r^2}{4}+\frac{1}{4r^2}(\chi_\theta^2+2\chi^2)\right]\mathrm{d}r\mathrm{d}\theta. \tag*{(65)} d t d E ϵ [ χ ] ( t ) = δ ϵ L ( t ) ∫ 0 π ∫ 0 ∞ s ( θ ) [ B 2 χ t 2 + 4 χ r 2 + 4 r 2 1 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (65)
Both boundary conditions are natural if χ \chi χ is a perturbation of the metric. For the first one, if L ≠ 0 \mathcal{L}\ne0 L = 0 the limit r → 0 + r\to0^+ r → 0 + is a conformal boundary and these are reflecting conditions; for L = 0 \mathcal{L}=0 L = 0 it is the Poincaré horizon and they are decay assumptions on the initial data. As pointed out before, the second boundary condition corresponds to the absence of conical singularities.
Returning to (65) , if we choose ϵ \epsilon ϵ such that δ ϵ L = 0 \delta_\epsilon\mathcal{L}=0 δ ϵ L = 0 , from (65) we obtain that the functional E ϵ E_\epsilon E ϵ is a conserved energy, indeed
E ϵ [ χ ] ( t ) = E ϵ [ χ ] ( 0 ) for all t . (66) E_\epsilon[\chi](t)=E_\epsilon[\chi](0)\qquad\text{for all }t. \tag*{(66)} E ϵ [ χ ] ( t ) = E ϵ [ χ ] ( 0 ) for all t . (66)
Remark 3.2 If δ ϵ L = 0 \delta_\epsilon\mathcal{L}=0 δ ϵ L = 0 , then ζ \zeta ζ commutes with □ 2 + ρ θ \Box_2+\rho_\theta □ 2 + ρ θ , so ζ χ \zeta\chi ζ χ is again a solution and E ϵ [ ϵ ~ χ ] E_\epsilon[\tilde{\epsilon}\chi] E ϵ [ ϵ ~ χ ] is conserved for any pair of Killing parameters ϵ , ϵ ~ \epsilon,\tilde{\epsilon} ϵ , ϵ ~ . For fixed L = { f , t } \mathcal{L}=\{f,t\} L = { f , t } , the Killing parameters form the three-dimensional space ϵ = ( a + b f + c f 2 ) / f ′ \epsilon=(a+bf+cf^2)/f' ϵ = ( a + b f + c f 2 ) / f ′ , so this procedure yields conserved energies of higher order in derivatives (a finite family at each order), rather than new symmetries.
Coercivity of the energy
We now check that the conservation of E ϵ [ χ ] E_\epsilon[\chi] E ϵ [ χ ] does indeed guarantee that the perturbation χ \chi χ remains bounded in a suitable norm. For this, we have to control the last term in the definition of E ϵ E_\epsilon E ϵ , which has no definite sign. To do so, we collect all the terms of (3.34) containing χ t \chi_t χ t or χ r \chi_r χ r into a single quadratic form. Set
A ϵ ( t , r ) : = B ϵ + ϵ ′ ′ = 2 r 2 ϵ + L ϵ + ϵ ′ ′ , v : = ( 2 χ t B , χ r ) T (67) A_\epsilon(t,r):=B\epsilon+\epsilon''=2r^2\epsilon+\mathcal{L}\epsilon+\epsilon'',\qquad\boldsymbol{v}:=\left(\frac{2\chi_t}{B},\chi_r\right)^T \tag*{(67)} A ϵ ( t , r ) := B ϵ + ϵ ′′ = 2 r 2 ϵ + L ϵ + ϵ ′′ , v := ( B 2 χ t , χ r ) T (67)
so that the energy (3.34) takes the form
E ϵ [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) [ Q ϵ [ χ ] + A ϵ 4 r 2 ( χ t 2 + 2 χ 2 ) ] d r d θ , (68) E_\epsilon[\chi]=\int_0^\pi\int_0^\infty s(\theta)\left[Q_\epsilon[\chi]+\frac{A_\epsilon}{4r^2}(\chi_t^2+2\chi^2)\right]\,drd\theta, \tag*{(68)} E ϵ [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) [ Q ϵ [ χ ] + 4 r 2 A ϵ ( χ t 2 + 2 χ 2 ) ] d r d θ , (68)
where
Q ϵ [ χ ] : = A ϵ B 2 χ t 2 − 2 r ϵ ′ B χ t χ r + A ϵ 4 χ r 2 = 1 4 v T M ϵ v , M ϵ ( t , r ) : = ( A ϵ − 2 r ϵ ′ − 2 r ϵ ′ A ϵ ) . (69) Q_\epsilon[\chi]:=\frac{A_\epsilon}{B^2}\chi_t^2-\frac{2r\epsilon'}{B}\chi_t\chi_r+\frac{A_\epsilon}{4}\chi_r^2=\frac{1}{4}\boldsymbol{v}^T M_\epsilon\boldsymbol{v},\qquad M_\epsilon(t,r):=\begin{pmatrix} A_\epsilon& -2r\epsilon' \\ -2r\epsilon' & A_\epsilon\end{pmatrix}. \tag*{(69)} Q ϵ [ χ ] := B 2 A ϵ χ t 2 − B 2 r ϵ ′ χ t χ r + 4 A ϵ χ r 2 = 4 1 v T M ϵ v , M ϵ ( t , r ) := ( A ϵ − 2 r ϵ ′ − 2 r ϵ ′ A ϵ ) . (69)
The matrix M ϵ M_\epsilon M ϵ has eigenvectors ( 1 , 1 ) T (1,1)^T ( 1 , 1 ) T and ( 1 , − 1 ) T (1,-1)^T ( 1 , − 1 ) T for every ( t , r ) (t,r) ( t , r ) , with eigenvalues
λ ± : = A ϵ ± 2 r ϵ ′ , (70) \lambda_\pm:=A_\epsilon\pm2r\epsilon', \tag*{(70)} λ ± := A ϵ ± 2 r ϵ ′ , (70)
respectively. Therefore Q ϵ Q_\epsilon Q ϵ is positive definite at ( t , r ) (t,r) ( t , r ) if and only if
λ min ( t , r ) : = min { λ + , λ − } = A ϵ − 2 r ∣ ϵ ′ ∣ > 0 , (71) \lambda_{\min}(t,r):=\min\{\lambda_+,\lambda_-\}=A_\epsilon-2r|\epsilon'|>0, \tag*{(71)} λ m i n ( t , r ) := min { λ + , λ − } = A ϵ − 2 r ∣ ϵ ′ ∣ > 0 , (71)
or equivalently, in terms of the trace and the determinant of M ϵ M_\epsilon M ϵ ,
A ϵ > 0 and A ϵ 2 − 4 r 2 ( ϵ ′ ) 2 > 0. (72) A_\epsilon>0\qquad\text{and}\qquad A_\epsilon^2-4r^2(\epsilon')^2>0. \tag*{(72)} A ϵ > 0 and A ϵ 2 − 4 r 2 ( ϵ ′ ) 2 > 0. (72)
In the orthonormal frame n = 2 r B ∂ t n=\frac{2r}{B}\partial_t n = B 2 r ∂ t , e = r ∂ r e=r\partial_r e = r ∂ r , the ( t , r ) (t,r) ( t , r ) part of ζ \zeta ζ reads ζ = A ϵ 2 r n − ϵ ′ e \zeta=\frac{A_\epsilon}{2r}n-\epsilon'e ζ = 2 r A ϵ n − ϵ ′ e , so that
g ( ζ , n ) = − A ϵ 2 r , g ( ζ , ζ ) = − A ϵ 2 − 4 r 2 ( ϵ ′ ) 2 4 r 2 = − λ + λ − 4 r 2 . g(\zeta,n)=-\frac{A_\epsilon}{2r},\qquad g(\zeta,\zeta)=-\frac{A_\epsilon^2-4r^2(\epsilon')^2}{4r^2}=-\frac{\lambda_+\lambda_-}{4r^2}. g ( ζ , n ) = − 2 r A ϵ , g ( ζ , ζ ) = − 4 r 2 A ϵ 2 − 4 r 2 ( ϵ ′ ) 2 = − 4 r 2 λ + λ − .
Hence (72) are precisely the conditions that ζ \zeta ζ be timelike and future directed with respect to n n n . Moreover, for any sign of λ min \lambda_{\min} λ m i n ,
Q ε [ χ ] ≥ 1 4 λ min ∣ v ∣ 2 = λ min ( χ t 2 B 2 + χ r 2 4 ) . (73) Q_{\varepsilon}[\chi] \ge\frac{1}{4}\lambda_{\min}\lvert v\rvert^{2} = \lambda_{\min}\left(\frac{\chi_{t}^{2}}{B^{2}}+\frac{\chi_{r}^{2}}{4}\right). \tag*{(73)} Q ε [ χ ] ≥ 4 1 λ m i n ∣ v ∣ 2 = λ m i n ( B 2 χ t 2 + 4 χ r 2 ) . (73)
The dependence on r r r is made explicit by completing the square in (70) :
2 ε λ ± = ( 2 r ε ± ε ′ ) 2 + C ε , C ε : = 2 ε ε ′ ′ − ( ε ′ ) 2 + 2 L ε 2 . (74) 2\varepsilon\lambda_{\pm}=(2r\varepsilon\pm\varepsilon')^{2}+C_{\varepsilon},\qquad C_{\varepsilon}:=2\varepsilon\varepsilon''-(\varepsilon')^{2}+2\mathcal{L}\varepsilon^{2}. \tag*{(74)} 2 ε λ ± = ( 2 r ε ± ε ′ ) 2 + C ε , C ε := 2 ε ε ′′ − ( ε ′ ) 2 + 2 L ε 2 . (74)
Since λ min ∼ 2 r 2 ε \lambda_{\min}\sim2r^{2}\varepsilon λ m i n ∼ 2 r 2 ε as r → ∞ r\to\infty r → ∞ , the condition ε ≥ 0 \varepsilon\ge0 ε ≥ 0 is necessary. For ε > 0 \varepsilon>0 ε > 0 , (74) gives 2 ε λ min = ( 2 r ε − ∣ ε ′ ∣ ) 2 + C ε 2\varepsilon\lambda_{\min}=(2r\varepsilon-\lvert\varepsilon'\rvert)^{2}+C_{\varepsilon} 2 ε λ m i n = ( 2 r ε − ∣ ε ′ ∣ ) 2 + C ε , whose minimum over r > 0 r>0 r > 0 is attained at r ∗ : = ∣ ε ′ ∣ / ( 2 ε ) r_{*}:=\lvert\varepsilon'\rvert/(2\varepsilon) r ∗ := ∣ ε ′ ∣ / ( 2 ε ) . Thus, on a slice where ε > 0 \varepsilon>0 ε > 0 :
If C ε = 0 C_{\varepsilon}=0 C ε = 0 and ε ′ ≠ 0 \varepsilon'\ne0 ε ′ = 0 , then Q ε Q_{\varepsilon} Q ε is non-negative but degenerates at r = r ∗ ( t ) r=r_{*}(t) r = r ∗ ( t ) . If C ε = 0 C_{\varepsilon}=0 C ε = 0 and ε ′ = 0 \varepsilon'=0 ε ′ = 0 , then λ min = 2 r 2 ε \lambda_{\min}=2r^{2}\varepsilon λ m i n = 2 r 2 ε , which degenerates only as r → 0 r\to0 r → 0 .
If C ε < 0 C_{\varepsilon}<0 C ε < 0 , then λ min < 0 \lambda_{\min}<0 λ m i n < 0 at r = r ∗ r=r_{*} r = r ∗ when ε ′ ≠ 0 \varepsilon'\ne0 ε ′ = 0 , and for small r r r when ε ′ = 0 \varepsilon'=0 ε ′ = 0 . There ζ \zeta ζ is not future timelike and Q ε Q_{\varepsilon} Q ε takes negative values.
In the case ε > 0 \varepsilon>0 ε > 0 , C ε ≥ 0 C_{\varepsilon}\ge0 C ε ≥ 0 , ( ε ′ ) 2 + C ε > 0 (\varepsilon')^{2}+C_{\varepsilon}>0 ( ε ′ ) 2 + C ε > 0 , we also compare Q ε Q_{\varepsilon} Q ε with its diagonal part 1 4 A ε ∣ v ∣ 2 \frac{1}{4}A_{\varepsilon}\lvert v\rvert^{2} 4 1 A ε ∣ v ∣ 2 . Here A ε > 0 A_{\varepsilon}>0 A ε > 0 , and the ratio 2 r ∣ ε ′ ∣ / A ε 2r\lvert\varepsilon'\rvert/A_{\varepsilon} 2 r ∣ ε ′ ∣ / A ε is maximal at r M = ( L ε + ε ′ ′ ) / ( 2 ε ) r_{M}=\sqrt{(\mathcal{L}\varepsilon+\varepsilon'')/(2\varepsilon)} r M = ( L ε + ε ′′ ) / ( 2 ε ) , which gives
C 0 ( t ) : = inf r > 0 λ min ( t , r ) A ε ( t , r ) = 1 − ∣ ε ′ ∣ 2 ε ( L ε + ε ′ ′ ) = 1 − ∣ ε ′ ∣ ( ε ′ ) 2 + C ε ∈ [ 0 , 1 ] . (75) C_{0}(t):=\inf_{r>0}\frac{\lambda_{\min}(t,r)}{A_{\varepsilon}(t,r)}=1-\frac{\lvert\varepsilon'\rvert}{\sqrt{2\varepsilon(\mathcal{L}\varepsilon+\varepsilon'')}}=1-\frac{\lvert\varepsilon'\rvert}{\sqrt{(\varepsilon')^{2}+C_{\varepsilon}}}\in[0,1]. \tag*{(75)} C 0 ( t ) := r > 0 inf A ε ( t , r ) λ m i n ( t , r ) = 1 − 2 ε ( L ε + ε ′′ ) ∣ ε ′ ∣ = 1 − ( ε ′ ) 2 + C ε ∣ ε ′ ∣ ∈ [ 0 , 1 ] . (75)
Thus C 0 ( t ) > 0 C_{0}(t)>0 C 0 ( t ) > 0 whenever C ε > 0 C_{\varepsilon}>0 C ε > 0 , and C 0 ( t ) = 0 C_{0}(t)=0 C 0 ( t ) = 0 when C ε = 0 C_{\varepsilon}=0 C ε = 0 and ε ′ ≠ 0 \varepsilon'\ne0 ε ′ = 0 , in which case r M = r ∗ r_{M}=r_{*} r M = r ∗ .
Remark 3.3 Note that C ε ′ = 2 ε δ ε L C'_{\varepsilon}=2\varepsilon\,\delta_{\varepsilon}\mathcal{L} C ε ′ = 2 ε δ ε L , so C ε C_{\varepsilon} C ε is constant for Killing parameters; the bounds below are pointwise in t t t and do not require δ ε L = 0 \delta_{\varepsilon}\mathcal{L}=0 δ ε L = 0 .
Since the angular part in (68) is non-negative whenever A ε ≥ 0 A_{\varepsilon}\ge0 A ε ≥ 0 , using (73) in (68) we obtain the pointwise coercivity
E ε [ χ ] ( t ) ≥ ∫ 0 π ∫ 0 ∞ s ( θ ) [ 1 4 λ min ∣ v ∣ 2 + A ε 4 r 2 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (76) E_{\varepsilon}[\chi](t)\ge\int_{0}^{\pi}\int_{0}^{\infty}s(\theta)\left[\frac{1}{4}\lambda_{\min}\lvert v\rvert^{2}+\frac{A_{\varepsilon}}{4r^{2}}\left(\chi_{\theta}^{2}+2\chi^{2}\right)\right]\,drd\theta. \tag*{(76)} E ε [ χ ] ( t ) ≥ ∫ 0 π ∫ 0 ∞ s ( θ ) [ 4 1 λ m i n ∣ v ∣ 2 + 4 r 2 A ε ( χ θ 2 + 2 χ 2 ) ] d r d θ . (76)
Using the uniform bound C 0 ( t ) C_{0}(t) C 0 ( t ) in r r r , we obtain the criterion
E ε [ χ ] ( t ) ≥ C 0 ( t ) ∫ 0 π ∫ 0 ∞ s ( θ ) A ε [ 1 4 ∣ v ∣ 2 + 1 4 r 2 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (77) E_{\varepsilon}[\chi](t)\ge C_{0}(t)\int_{0}^{\pi}\int_{0}^{\infty}s(\theta)A_{\varepsilon}\left[\frac{1}{4}\lvert v\rvert^{2}+\frac{1}{4r^{2}}\left(\chi_{\theta}^{2}+2\chi^{2}\right)\right]\,drd\theta. \tag*{(77)} E ε [ χ ] ( t ) ≥ C 0 ( t ) ∫ 0 π ∫ 0 ∞ s ( θ ) A ε [ 4 1 ∣ v ∣ 2 + 4 r 2 1 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (77)
It is worth emphasizing that, given the general framework in which the energy formalism has been presented up to this point, estimates (76) and (77) will provide useful criteria across various problem settings. In particular, assuming δ ϵ L = 0 \delta_{\epsilon}\mathcal{L}=0 δ ϵ L = 0 and ϵ > 0 \epsilon>0 ϵ > 0 . If C ϵ > 0 C_{\epsilon}>0 C ϵ > 0 , the perturbation equation admits a conserved energy which is coercive in the sense of (77) , with C 0 ( t ) > 0 C_{0}(t)>0 C 0 ( t ) > 0 . If C ϵ = 0 C_{\epsilon}=0 C ϵ = 0 , the energy is conserved and non-negative but degenerate on the curve r = r M ( t ) r=r_{M}(t) r = r M ( t ) ; this is the case of ϵ = 1 \epsilon=1 ϵ = 1 for L = 0 \mathcal{L}=0 L = 0 , where r M = 0 r_{M}=0 r M = 0 is the horizon, which suggests that horizon instabilities of Aretakis type may still be present. If C ϵ < 0 C_{\epsilon}<0 C ϵ < 0 , the energy is conserved but not sign-definite: for instance, for L = − κ 2 / 2 \mathcal{L}=-\kappa^{2}/2 L = − κ 2 /2 and ϵ = 1 \epsilon=1 ϵ = 1 one has C 1 = − κ 2 C_{1}=-\kappa^{2} C 1 = − κ 2 and B B B changes sign at r = κ / 2 r=\kappa/2 r = κ /2 . Positivity of ϵ \epsilon ϵ alone is therefore not sufficient. The conservation law is tied to the Killing symmetries of the Schwarzian background, and through (65) the same construction quantifies the failure of conservation when δ ϵ L ≠ 0 \delta_{\epsilon}\mathcal{L}\ne0 δ ϵ L = 0 . The aim of the following sections is precisely to identify what characterizations we can obtain from this formalism, depending on the background under consideration, starting with the near-NHEK case, but also assuming more general settings for the background L \mathcal{L} L .
Applications to specific backgrounds
We now use (65) and its consequences on several classes of backgrounds: the near-NHEK solution, the case of general L \mathcal{L} L , and the KdV soliton-solution.
The Thermal Case: near-NHEK
Taking into account that the energy defined in the previous section is, in a suitable sense, a general density in terms of the fields L ( t ) \mathcal{L}(t) L ( t ) , ϵ ( t ) \epsilon(t) ϵ ( t ) , let us consider the thermal background L ( t ) = − κ 2 / 2 \mathcal{L}(t)=-\kappa^{2}/2 L ( t ) = − κ 2 /2 , with ϵ = cosh ( κ t ) \epsilon=\cosh(\kappa t) ϵ = cosh ( κ t ) is one of the Killing parameters. In this setting, it follows that δ ϵ L = 0 \delta_{\epsilon}\mathcal{L}=0 δ ϵ L = 0 , implying that ζ \zeta ζ is indeed a Killing vector field for the nearly-AdS2 _{2} 2 Laplacian, therefore by (65) the energy E ϵ [ χ ] ( t ) E_{\epsilon}[\chi](t) E ϵ [ χ ] ( t ) defined in (63) is conserved. Furthermore, a direct computation yields the quantities appearing in the coercivity estimate (76) , as shown below
B = 2 r 2 − κ 2 / 2 , A ϵ = 2 cosh ( κ t ) ( r 2 + κ 2 / 4 ) > 0 , B 2 = 4 ( r − κ / 2 ) 2 ( r + κ / 2 ) 2 , (78) \begin{aligned}
B &= 2r^{2}-\kappa^{2}/2, \qquad A_{\epsilon}=2\cosh(\kappa t)(r^{2}+\kappa^{2}/4)>0,\\
B^{2} &= 4(r-\kappa/2)^{2}(r+\kappa/2)^{2}, \tag*{(78)}
\end{aligned} B B 2 = 2 r 2 − κ 2 /2 , A ϵ = 2 cosh ( κ t ) ( r 2 + κ 2 /4 ) > 0 , = 4 ( r − κ /2 ) 2 ( r + κ /2 ) 2 , (78)
meanwhile
λ min = 2 cosh ( κ t ) ( r 2 + κ 2 / 4 ) − 2 r κ ∣ sinh ( κ t ) ∣ ≥ 2 cosh ( κ t ) ( r − κ / 2 ) 2 (79) \lambda_{\min}=2\cosh(\kappa t)(r^{2}+\kappa^{2}/4)-2r\kappa|\sinh(\kappa t)|\ge2\cosh(\kappa t)(r-\kappa/2)^{2} \tag*{(79)} λ m i n = 2 cosh ( κ t ) ( r 2 + κ 2 /4 ) − 2 r κ ∣ sinh ( κ t ) ∣ ≥ 2 cosh ( κ t ) ( r − κ /2 ) 2 (79)
Simplifying, we obtain
λ min B 2 ≥ cosh ( κ t ) 2 ( r + κ / 2 ) 2 , λ min 4 ≥ 1 2 cosh ( κ t ) ( r − κ / 2 ) 2 , \frac{\lambda_{\min}}{B^2} \ge\frac{\cosh(\kappa t)}{2(r+\kappa/2)^2}, \qquad\frac{\lambda_{\min}}{4} \ge\frac{1}{2}\cosh(\kappa t)(r-\kappa/2)^2, B 2 λ m i n ≥ 2 ( r + κ /2 ) 2 cosh ( κ t ) , 4 λ m i n ≥ 2 1 cosh ( κ t ) ( r − κ /2 ) 2 ,
A ϵ 4 r 2 = cosh ( κ t ) ( r 2 + κ 2 / 4 2 r 2 ) , \frac{A_{\epsilon}}{4r^2}=\cosh(\kappa t)\left(\frac{r^2+\kappa^2/4}{2r^2}\right), 4 r 2 A ϵ = cosh ( κ t ) ( 2 r 2 r 2 + κ 2 /4 ) ,
therefore using (3.47), for finite-energy data on the whole slice,
E ϵ [ χ ] ( t ) ≥ cosh ( κ t ) ∥ χ ( t ) ∥ 2 , (80) E_{\epsilon}[\chi](t)\ge\cosh(\kappa t)\lVert\chi(t)\rVert^2, \tag*{(80)} E ϵ [ χ ] ( t ) ≥ cosh ( κ t ) ∥ χ ( t ) ∥ 2 , (80)
where the norm in this case is define by
∥ χ ∥ 2 : = ∫ 0 π ∫ 0 ∞ sin − 3 θ [ 1 2 ( r + κ / 2 ) 2 χ t 2 + ( r − κ / 2 ) 2 2 χ r 2 + r 2 + κ 2 / 4 2 r 2 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (81) \lVert\chi\rVert^2:=\int_0^\pi\int_0^\infty\sin^{-3}\theta\left[\frac{1}{2(r+\kappa/2)^2}\chi_t^2+\frac{(r-\kappa/2)^2}{2}\chi_r^2+\frac{r^2+\kappa^2/4}{2r^2}(\chi_\theta^2+2\chi^2)\right]\,dr\,d\theta. \tag*{(81)} ∥ χ ∥ 2 := ∫ 0 π ∫ 0 ∞ sin − 3 θ [ 2 ( r + κ /2 ) 2 1 χ t 2 + 2 ( r − κ /2 ) 2 χ r 2 + 2 r 2 r 2 + κ 2 /4 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (81)
Consequently, since δ L = 0 \delta L=0 δ L = 0 , by (3.37) and (3.51),
∥ χ ( t ) ∥ 2 ≤ E ϵ [ χ ] ( 0 ) cosh ( κ t ) \lVert\chi(t)\rVert^2\le\frac{E_{\epsilon}[\chi](0)}{\cosh(\kappa t)} ∥ χ ( t ) ∥ 2 ≤ cosh ( κ t ) E ϵ [ χ ] ( 0 )
Thus, in the thermal background, the norm ∥ χ ∥ \lVert\chi\rVert ∥ χ ∥ decays exponentially in t t t , at least at the rate κ \kappa κ set by the temperature. Only the weight of χ r 2 \chi_r^2 χ r 2 degenerates at r = κ 2 r=\frac{\kappa}{2} r = 2 κ ; in particular, the weighted norms of χ t \chi_t χ t and of the angular terms decay at this rate also at r = κ 2 r=\frac{\kappa}{2} r = 2 κ , and the full local energy decays at the same rate on any region { ∣ r − κ / 2 ∣ ≥ δ , r ≤ R } \{|r-\kappa/2|\ge\delta,\ r\le R\} { ∣ r − κ /2∣ ≥ δ , r ≤ R } .
Remark 3.4 The estimate above uses the point-wise bound on λ min \lambda_{\min} λ m i n , and the positivity of A ϵ A_\epsilon A ϵ is checked directly; the constant C ϵ C_\epsilon C ϵ plays no role in it. It enters only if one uses instead the infimum in r r r ,
C 0 ( t ) = 1 − ∣ κ ∣ ∣ sinh ( κ t ) ∣ κ 2 sinh 2 ( κ t ) + κ 2 , C ϵ = κ 2 , C_0(t)=1-\frac{|\kappa||\sinh(\kappa t)|}{\sqrt{\kappa^2\sinh^2(\kappa t)+\kappa^2}},\qquad C_\epsilon=\kappa^2, C 0 ( t ) = 1 − κ 2 sinh 2 ( κ t ) + κ 2 ∣ κ ∣∣ sinh ( κ t ) ∣ , C ϵ = κ 2 ,
in (3.48). This gives weights that do not degenerate at r = κ / 2 r=\kappa/2 r = κ /2 , but since C 0 ( t ) cosh ( κ t ) = e − κ ∣ t ∣ C_0(t)\cosh(\kappa t)=e^{-\kappa|t|} C 0 ( t ) cosh ( κ t ) = e − κ ∣ t ∣ the resulting bound,
∫ 0 π ∫ 0 ∞ sin − 3 θ ( r 2 + κ 2 / 4 ) [ 2 χ t 2 B 2 + χ r 2 2 ] d r d θ ≤ e κ ∣ t ∣ E ϵ [ χ ] ( 0 ) , \int_0^\pi\int_0^\infty\sin^{-3}\theta(r^2+\kappa^2/4)\left[\frac{2\chi_t^2}{B^2}+\frac{\chi_r^2}{2}\right]\,dr\,d\theta\le e^{\kappa|t|}E_\epsilon[\chi](0), ∫ 0 π ∫ 0 ∞ sin − 3 θ ( r 2 + κ 2 /4 ) [ B 2 2 χ t 2 + 2 χ r 2 ] d r d θ ≤ e κ ∣ t ∣ E ϵ [ χ ] ( 0 ) ,
allows growth of that stronger norm. The point-wise bound is therefore the one that yields stability and decay, while the infimum is the one to use when non-degenerate control at r = κ / 2 r=\kappa/2 r = κ /2 is needed. The infimum is attained precisely at r = κ / 2 r=\kappa/2 r = κ /2 .
Time-dependent Schwarzian background
We now consider (3.13) on a general background, with L = L ( t ) \mathcal{L}=\mathcal{L}(t) L = L ( t ) a suitable function and B = 2 r 2 + L B=2r^{2}+\mathcal{L} B = 2 r 2 + L . Since the equation is not time invariant, we do not expect to find a conserved energy in general. To simplify the expressions, set
D [ χ ] : = χ t 2 B 2 + χ r 2 4 + 1 4 r 2 ( χ θ 2 + 2 χ 2 ) , E m o d ( t ) : = ∫ 0 π ∫ 0 ∞ s ( θ ) D [ χ ] d r d θ , (82) D[\chi]:=\frac{\chi_{t}^{2}}{B^{2}}+\frac{\chi_{r}^{2}}{4}+\frac{1}{4r^{2}}\left(\chi_{\theta}^{2}+2\chi^{2}\right),\qquad E_{\mathrm{mod}}(t):=\int_{0}^{\pi}\int_{0}^{\infty}s(\theta)D[\chi]\,\mathrm{d}r\mathrm{d}\theta, \tag*{(82)} D [ χ ] := B 2 χ t 2 + 4 χ r 2 + 4 r 2 1 ( χ θ 2 + 2 χ 2 ) , E mod ( t ) := ∫ 0 π ∫ 0 ∞ s ( θ ) D [ χ ] d r d θ , (82)
so that (3.34) and (3.36) read
E ϵ [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) [ A ϵ D [ χ ] − 2 r ϵ ′ B χ t χ r ] d r d θ , d d t E ϵ [ χ ] = δ ϵ L E m o d ( t ) . E_{\epsilon}[\chi]=\int_{0}^{\pi}\int_{0}^{\infty}s(\theta)\left[A_{\epsilon}D[\chi]-\frac{2r\epsilon'}{B}\chi_{t}\chi_{r}\right]\mathrm{d}r\mathrm{d}\theta,\qquad\frac{\mathrm{d}}{\mathrm{d}t}E_{\epsilon}[\chi]=\delta_{\epsilon}\mathcal{L}E_{\mathrm{mod}}(t). E ϵ [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) [ A ϵ D [ χ ] − B 2 r ϵ ′ χ t χ r ] d r d θ , d t d E ϵ [ χ ] = δ ϵ L E mod ( t ) .
We now take the simplest choice ϵ ≡ 1 \epsilon\equiv1 ϵ ≡ 1 , for which ζ = ∂ t \zeta=\partial_{t} ζ = ∂ t (up to the term in ∂ φ \partial_{\varphi} ∂ φ , which does not contribute), A 1 = B A_{1}=B A 1 = B , δ 1 L = L ′ \delta_{1}\mathcal{L}=\mathcal{L}' δ 1 L = L ′ , and the cross term vanishes. The energy E : = E 1 [ χ ] E:=E_{1}[\chi] E := E 1 [ χ ] of this section is therefore
E ( t ) = ∫ 0 π ∫ 0 ∞ s ( θ ) B D [ χ ] d r d θ , (83) E(t)=\int_{0}^{\pi}\int_{0}^{\infty}s(\theta)B\,D[\chi]\,\mathrm{d}r\mathrm{d}\theta, \tag*{(83)} E ( t ) = ∫ 0 π ∫ 0 ∞ s ( θ ) B D [ χ ] d r d θ , (83)
and it satisfies
d d t E = L ′ E m o d ( t ) . (84) \frac{\mathrm{d}}{\mathrm{d}t}E=\mathcal{L}'E_{\mathrm{mod}}(t). \tag*{(84)} d t d E = L ′ E mod ( t ) . (84)
Precisely, this last identity reveals that E E E is a Lyapunov functional. If L ≥ 0 \mathcal{L}\geq0 L ≥ 0 , then B > 0 B>0 B > 0 for all r > 0 r>0 r > 0 and E ≥ 0 E\geq0 E ≥ 0 ; this agrees with Section 3.2.2, since λ ± = B \lambda_{\pm}=B λ ± = B and C 1 = 2 L C_{1}=2\mathcal{L} C 1 = 2 L . For L < 0 \mathcal{L}<0 L < 0 , E E E is not sign-definite, see for instance Section 3.3.1.
We define E \mathcal{E} E as the energy space of functions where E E E in (3.54) is well-defined and E E E satisfies (3.55). By (3.35) with ζ = ∂ t \zeta=\partial_{t} ζ = ∂ t , this space requires
E E E well-defined in ( t , r , θ ) ∈ ( 0 , ∞ ) × ( 0 , ∞ ) × ( 0 , π ) (t,r,\theta)\in(0,\infty)\times(0,\infty)\times(0,\pi) ( t , r , θ ) ∈ ( 0 , ∞ ) × ( 0 , ∞ ) × ( 0 , π ) ,
s ( θ ) B ∂ t χ ∂ r χ s(\theta)B\,\partial_{t}\chi\,\partial_{r}\chi s ( θ ) B ∂ t χ ∂ r χ vanishing at r = 0 + r=0^{+} r = 0 + and as r → + ∞ r\to+\infty r → + ∞ , with θ , t \theta,t θ , t free parameters,
B 2 r 2 s ( θ ) ∂ t χ ∂ θ χ \frac{B}{2r^{2}}s(\theta)\,\partial_{t}\chi\,\partial_{\theta}\chi 2 r 2 B s ( θ ) ∂ t χ ∂ θ χ vanishing at θ = 0 \theta=0 θ = 0 and θ = π \theta=\pi θ = π , with r , t r,t r , t free parameters.
The previous conditions impose decay properties on χ \chi χ and its derivatives at the boundaries. Precisely, these conditions are set to have a well-defined energy. Weaker conditions can be assumed, however the existence of a well-defined energy is not clear to us. We shall assume these finite energy conditions from now on.
There are several interesting consequences obtained directly from (84) . We shall always assume that L ( t ) ≥ 0 \mathcal{L}(t) \ge0 L ( t ) ≥ 0 for t ≥ 0 t \ge0 t ≥ 0 .
Notice first that we recover the case L ≡ 0 \mathcal{L} \equiv0 L ≡ 0 , where E E E is constant in time. Moreover, the sign of the derivative of E E E only depends on the sign of L ′ \mathcal{L}' L ′ . Therefore, in the case L \mathcal{L} L decreasing in time, we get a decreasing energy E E E .
Profiles with a positive lower bound
Now we assume that L \mathcal{L} L increases towards a nonzero positive value L 0 \mathcal{L}_0 L 0 . First of all, (84) reads
d d t E = ∣ L ′ ( t ) ∣ E m o d ( t ) . \frac{d}{dt}E = |\mathcal{L}'(t)| E_{\mathrm{mod}}(t). d t d E = ∣ L ′ ( t ) ∣ E mod ( t ) .
Since L \mathcal{L} L is increasing, we have B ≥ L ( 0 ) B \ge\mathcal{L}(0) B ≥ L ( 0 ) . Therefore, from (84) ,
d d t E = ∣ L ′ ∣ ∫ 0 π ∫ 0 ∞ 1 B s ( θ ) B D [ χ ] d r d θ ≤ ∣ L ′ ( t ) ∣ L ( 0 ) E , \frac{d}{dt}E = |\mathcal{L}'| \int_0^\pi\int_0^\infty\frac{1}{B}s(\theta)BD[\chi]\,drd\theta\le\frac{|\mathcal{L}'(t)|}{\mathcal{L}(0)}E, d t d E = ∣ L ′ ∣ ∫ 0 π ∫ 0 ∞ B 1 s ( θ ) B D [ χ ] d r d θ ≤ L ( 0 ) ∣ L ′ ( t ) ∣ E ,
leading to
E ( t ) ≤ e L ( 0 ) − 1 ∫ 0 t ∣ L ′ ( s ) ∣ d s E 0 . E(t) \le e^{\mathcal{L}(0)^{-1}\int_0^t |\mathcal{L}'(s)|ds}E_0. E ( t ) ≤ e L ( 0 ) − 1 ∫ 0 t ∣ L ′ ( s ) ∣ d s E 0 .
In the case where one has lim t → + ∞ ∫ 0 t ∣ L ′ ( s ) ∣ d s < ∞ \lim_{t\to+\infty}\int_0^t |\mathcal{L}'(s)|ds < \infty lim t → + ∞ ∫ 0 t ∣ L ′ ( s ) ∣ d s < ∞ , one has that E E E stays bounded for all times.
In the general case where L ′ ( t ) \mathcal{L}'(t) L ′ ( t ) has any sign, but L ( t ) ≥ L 0 > 0 \mathcal{L}(t) \ge\mathcal{L}_0 > 0 L ( t ) ≥ L 0 > 0 , from (84) one has
d d t E ≤ ∣ L ′ ∣ L 0 E . \frac{d}{dt}E \le\frac{|\mathcal{L}'|}{\mathcal{L}_0}E. d t d E ≤ L 0 ∣ L ′ ∣ E .
Consequently, if ∫ 0 ∞ ∣ L ′ ( t ) ∣ d t < C L < + ∞ \int_0^\infty|\mathcal{L}'(t)|dt < C_{\mathcal{L}} < +\infty ∫ 0 ∞ ∣ L ′ ( t ) ∣ d t < C L < + ∞ , we will have the stability bound E ( t ) ≤ E 0 e C L / L 0 E(t) \le E_0e^{C_{\mathcal{L}}/\mathcal{L}_0} E ( t ) ≤ E 0 e C L / L 0 .
The KdV case
We continue describing the perturbed NHEK case, but this time we assume that L \mathcal{L} L satisfies the KdV equation (25) .
First of all, notice that from (25) L ( t ) \mathcal{L}(t) L ( t ) satisfies
3 2 L 2 + L ′ ′ + ϵ ˉ 1 L = 0 , (85) \frac{3}{2}\mathcal{L}^2+\mathcal{L}''+\bar{\epsilon}_1\mathcal{L}=0, \tag*{(85)} 2 3 L 2 + L ′′ + ϵ ˉ 1 L = 0 , (85)
where we have assumed that L ( + ∞ ) = 0 \mathcal{L}(+\infty)=0 L ( + ∞ ) = 0 , as well as its derivatives. Regular solutions decaying at infinity require ϵ ˉ 1 : = − ω 2 < 0 \bar{\epsilon}_1:=-\omega^2<0 ϵ ˉ 1 := − ω 2 < 0 , then one has soliton solutions of the form
L ( t ) = ω 2 cosh 2 ( ω t − t 0 2 ) , (86) \mathcal{L}(t)=\frac{\omega^2}{\cosh^2\left(\omega\frac{t-t_0}{2}\right)}, \tag*{(86)} L ( t ) = cosh 2 ( ω 2 t − t 0 ) ω 2 , (86)
where t 0 t_0 t 0 is any free parameter. With no loss of generality, we assume t 0 = 0 t_0=0 t 0 = 0 . Therefore, from (84) one has that E E E is a decreasing quantity and for all t ≥ 0 t\geq0 t ≥ 0 ,
E ( t ) ≤ E 0 . E(t)\leq E_0. E ( t ) ≤ E 0 .
Notice that in this case B B B vanishes as r → 0 r\to0 r → 0 and t → ± ∞ t\to\pm\infty t → ± ∞ , so the coefficients in the energy are singular. Moreover,
d d t E = − ω 3 sinh ( ω t 2 ) cosh 3 ( ω t 2 ) E m o d ( t ) . (87) \frac{d}{dt}E=-\frac{\omega^3\sinh\left(\omega\frac{t}{2}\right)}{\cosh^3\left(\omega\frac{t}{2}\right)}E_{\mathrm{mod}}(t). \tag*{(87)} d t d E = − cosh 3 ( ω 2 t ) ω 3 sinh ( ω 2 t ) E mod ( t ) . (87)
From (87) , we are in a case where L ′ < 0 \mathcal{L}'<0 L ′ < 0 and
∫ 0 ∞ ∣ L ′ ∣ E m o d ≤ E 0 . \int_0^\infty\lvert\mathcal{L}'\rvert E_{\mathrm{mod}}\leq E_0. ∫ 0 ∞ ∣ L ′ ∣ E mod ≤ E 0 .
However, since the weight induced by the soliton is very strong, only a particular portion of the energy density is converging to zero. Indeed,
d d t E = − ω sinh ( ω t 2 ) cosh ( ω t 2 ) ∫ L B e ( t ) , \frac{d}{dt}E=-\frac{\omega\sinh\left(\omega\frac{t}{2}\right)}{\cosh\left(\omega\frac{t}{2}\right)}\int\frac{\mathcal{L}}{B}e(t), d t d E = − cosh ( ω 2 t ) ω sinh ( ω 2 t ) ∫ B L e ( t ) ,
where e : = s ( θ ) B D [ χ ] e:=s(\theta)BD[\chi] e := s ( θ ) B D [ χ ] is the energy density. Therefore, proceeding as in the previous computations, since the energy is bounded and − ω sinh ( ω t 2 ) cosh ( ω t 2 ) -\frac{\omega\sinh\left(\omega\frac{t}{2}\right)}{\cosh\left(\omega\frac{t}{2}\right)} − c o s h ( ω 2 t ) ω s i n h ( ω 2 t ) does not integrate in time, we can ensure that there is a sequence t n → + ∞ t_n\to+\infty t n → + ∞ such that the mild decay property lim n → + ∞ ∫ L ( t n ) B ( t n , r ) e ( t n ) = 0 \lim_{n\to+\infty}\int_{\mathcal{L}(t_n)}^{B(t_n,r)}e(t_n)=0 lim n → + ∞ ∫ L ( t n ) B ( t n , r ) e ( t n ) = 0 holds.
A conserved energy on the soliton background
The previous energy E E E defined in (83) corresponds to the energy E 1 E_1 E 1 in (63) with ϵ = 1 \epsilon=1 ϵ = 1 , for which δ L = L ′ ≠ 0 \delta\mathcal{L}=\mathcal{L}'\neq0 δ L = L ′ = 0 . The energy formalism presented in § 3.2.2 allows another choice. Using the same background
L ( t ) = ω 2 cosh 2 ( ω t 2 ) , \mathcal{L}(t)=\frac{\omega^2}{\cosh^2\left(\frac{\omega t}{2}\right)}, L ( t ) = cosh 2 ( 2 ω t ) ω 2 ,
we have that ϵ = ω 2 − L \epsilon=\omega^2-\mathcal{L} ϵ = ω 2 − L , which is local in L \mathcal{L} L , yields a Killing vector for the equation (53) , precisely because L \mathcal{L} L solves the KdV equation with ϵ ˉ = − ω 2 \bar{\epsilon}=-\omega^2 ϵ ˉ = − ω 2 . Therefore, the energy E ε E_{\varepsilon} E ε associated with this parameter is conserved. Note that for this setting, the terms appearing in the quasi-coercivity estimate (76) have a definite sign; indeed, since L \mathcal{L} L satisfies (85) , then
L ε − L ′ ′ = L ( ω 2 − L ) − ( ω 2 L − 3 2 L 2 ) = 1 2 L 2 , \mathcal{L}\varepsilon-\mathcal{L}''=\mathcal{L}(\omega^2-\mathcal{L})-\left(\omega^2\mathcal{L}-\frac{3}{2}\mathcal{L}^2\right)=\frac{1}{2}\mathcal{L}^2, L ε − L ′′ = L ( ω 2 − L ) − ( ω 2 L − 2 3 L 2 ) = 2 1 L 2 ,
ε = ω 2 ( 1 − sech 2 ( ω t 2 ) ) = ω 2 tanh 2 ( ω t 2 ) , \varepsilon=\omega^2\left(1-\operatorname{sech}^2\left(\frac{\omega t}{2}\right)\right)=\omega^2\tanh^2\left(\frac{\omega t}{2}\right), ε = ω 2 ( 1 − sech 2 ( 2 ω t ) ) = ω 2 tanh 2 ( 2 ω t ) ,
therefore, the principal term A ε A_{\varepsilon} A ε is given by
A ε = 2 r 2 ε + L ε − L ′ ′ = 2 r 2 ω 2 tanh 2 ( ω t 2 ) + 1 2 L 2 > 0 , A_{\varepsilon}=2r^2\varepsilon+\mathcal{L}\varepsilon-\mathcal{L}''=2r^2\omega^2\tanh^2\left(\frac{\omega t}{2}\right)+\frac{1}{2}\mathcal{L}^2>0, A ε = 2 r 2 ε + L ε − L ′′ = 2 r 2 ω 2 tanh 2 ( 2 ω t ) + 2 1 L 2 > 0 ,
meanwhile C ε = 0 C_{\varepsilon}=0 C ε = 0 , and for λ min \lambda_{\min} λ m i n we get
λ min = 2 r 2 ω 2 tanh 2 ( ω t / 2 ) + 1 2 L 2 − 2 r ∣ L ′ ∣ = 1 2 ( 2 r ω ∣ tanh ( ω t / 2 ) ∣ − L ) 2 ≥ 0 , (88) \lambda_{\min}=2r^2\omega^2\tanh^2(\omega t/2)+\frac{1}{2}\mathcal{L}^2-2r|\mathcal{L}'|=\frac{1}{2}\left(2r\omega|\tanh(\omega t/2)|-\mathcal{L}\right)^2\geq0, \tag*{(88)} λ m i n = 2 r 2 ω 2 tanh 2 ( ω t /2 ) + 2 1 L 2 − 2 r ∣ L ′ ∣ = 2 1 ( 2 r ω ∣ tanh ( ω t /2 ) ∣ − L ) 2 ≥ 0 , (88)
so that, by (76) the energy E ε E_{\varepsilon} E ε described in (63) is nonnegative; here C 0 ( t ) = 0 C_0(t)=0 C 0 ( t ) = 0 for t ≠ 0 t\ne0 t = 0 , thus, the energy becomes degenerate when r = 1 2 cosh ( ω t / 2 ) ∣ sinh ( ω t / 2 ) ∣ r=\frac{1}{2\cosh(\omega t/2)|\sinh(\omega t/2)|} r = 2 c o s h ( ω t /2 ) ∣ s i n h ( ω t /2 ) ∣ 1 , and r → 0 r\to0 r → 0 when t → ± ∞ t\to\pm\infty t → ± ∞ . At the instant t = 0 t=0 t = 0 , we have ε = ε ′ = 0 \varepsilon=\varepsilon'=0 ε = ε ′ = 0 and A ε = ω 4 2 A_{\varepsilon}=\frac{\omega^4}{2} A ε = 2 ω 4 , hence by (66) , for all t t t
E ε [ χ ] ( t ) = E ε [ χ ] ( 0 ) = ω 4 2 E m o d ( 0 ) , E_{\varepsilon}[\chi](t)=E_{\varepsilon}[\chi](0)=\frac{\omega^4}{2}E_{\mathrm{mod}}(0), E ε [ χ ] ( t ) = E ε [ χ ] ( 0 ) = 2 ω 4 E mod ( 0 ) ,
where E m o d E_{\mathrm{mod}} E mod is the functional defined in (82) .
Virial methods for the NHEK case L = 0 \mathcal{L}=0 L = 0
In this case, from (65) with ε = 1 \varepsilon=1 ε = 1 and δ ε L = 0 \delta_{\varepsilon}\mathcal{L}=0 δ ε L = 0 , taking the expressions from Subsection 3.3.1 with κ = 0 \kappa=0 κ = 0 , so that A 1 = B = 2 r 2 A_1=B=2r^2 A 1 = B = 2 r 2 , the conserved energy reads
E 1 [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) 2 r 2 D [ χ ] d r d θ = ∫ 0 π ∫ 0 ∞ s ( θ ) [ χ t 2 2 r 2 + r 2 χ r 2 2 + 1 2 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (89) E_1[\chi]=\int_0^\pi\int_0^\infty s(\theta)\,2r^2\mathcal{D}[\chi]\,drd\theta=\int_0^\pi\int_0^\infty s(\theta)\left[\frac{\chi_t^2}{2r^2}+\frac{r^2\chi_r^2}{2}+\frac{1}{2}\left(\chi_\theta^2+2\chi^2\right)\right]drd\theta. \tag*{(89)} E 1 [ χ ] = ∫ 0 π ∫ 0 ∞ s ( θ ) 2 r 2 D [ χ ] d r d θ = ∫ 0 π ∫ 0 ∞ s ( θ ) [ 2 r 2 χ t 2 + 2 r 2 χ r 2 + 2 1 ( χ θ 2 + 2 χ 2 ) ] d r d θ . (89)
To simplify the radial part of the wave equation, define, as usual ψ : = r χ \psi:=r\chi ψ := r χ , r > 0 r>0 r > 0 . Then ψ \psi ψ satisfies
1 r 4 ψ t t = ψ r r + 1 r 2 ( ψ θ θ − 3 cot θ ψ θ − 2 ψ ) . (90) \frac{1}{r^4}\psi_{tt}=\psi_{rr}+\frac{1}{r^2}\left(\psi_{\theta\theta}-3\cot\theta\psi_\theta-2\psi\right). \tag*{(90)} r 4 1 ψ tt = ψ r r + r 2 1 ( ψ θ θ − 3 cot θ ψ θ − 2 ψ ) . (90)
Since χ = ψ / r \chi=\psi/r χ = ψ / r , one has χ t = ψ t / r \chi_t=\psi_t/r χ t = ψ t / r , χ θ = ψ θ / r \chi_\theta=\psi_\theta/r χ θ = ψ θ / r and r 2 χ r 2 = ψ r 2 − ∂ r ( ψ 2 / r ) r^2\chi_r^2=\psi_r^2-\partial_r(\psi^2/r) r 2 χ r 2 = ψ r 2 − ∂ r ( ψ 2 / r ) , so that (89) becomes
E ~ 1 [ ψ ] : = E 1 [ r − 1 ψ ] = 1 2 ∫ 0 π ∫ 0 ∞ s ( θ ) ( ψ t 2 r 4 + 2 ψ 2 r 2 + ψ r 2 + ψ θ 2 r 2 ) d r d θ (91) \widetilde{E}_1[\psi]:=E_1[r^{-1}\psi]=\frac{1}{2}\int_0^\pi\int_0^\infty s(\theta)\left(\frac{\psi_t^2}{r^4}+\frac{2\psi^2}{r^2}+\psi_r^2+\frac{\psi_\theta^2}{r^2}\right)drd\theta \tag*{(91)} E 1 [ ψ ] := E 1 [ r − 1 ψ ] = 2 1 ∫ 0 π ∫ 0 ∞ s ( θ ) ( r 4 ψ t 2 + r 2 2 ψ 2 + ψ r 2 + r 2 ψ θ 2 ) d r d θ (91)
This conserved energy gives a form of orbital stability. In order to get decay, we need to obtain a virial functional: a quadratic form that is finite in the energy space and whose time derivative controls a localized norm of the perturbation. To this end, using the strategy from [34 ] .
In the following computations, we shall always assume that the boundary terms are suitably made zero by finite energy assumptions. Following the lines in [34 ] , we need the continuity equation for the energy density in (91) : for r > 0 r > 0 r > 0 , θ ∈ ( 0 , π ) \theta\in(0,\pi) θ ∈ ( 0 , π ) ,
∂ t ( sin ( θ ) 2 ( ψ t 2 r 4 + 2 ψ 2 r 2 + ψ r 2 + ψ θ 2 r 2 ) ) = ∂ r ( sin ( θ ) ψ r ψ t ) + ∂ θ ( sin ( θ ) r 2 ψ θ ψ t ) (92) \partial_t\left(\frac{\sin(\theta)}{2}\left(\frac{\psi_t^2}{r^4}+\frac{2\psi^2}{r^2}+\psi_r^2+\frac{\psi_\theta^2}{r^2}\right)\right)=\partial_r\left(\sin(\theta)\psi_r\psi_t\right)+\partial_\theta\left(\frac{\sin(\theta)}{r^2}\psi_\theta\psi_t\right) \tag*{(92)} ∂ t ( 2 sin ( θ ) ( r 4 ψ t 2 + r 2 2 ψ 2 + ψ r 2 + r 2 ψ θ 2 ) ) = ∂ r ( sin ( θ ) ψ r ψ t ) + ∂ θ ( r 2 sin ( θ ) ψ θ ψ t ) (92)
we obtain the distorted conservation law
∂ t ( sin ( θ ) r 2 ψ r ψ t ) = ∂ r ( sin ( θ ) ψ t 2 2 r 2 ) + sin ( θ ) ψ t 2 r 3 + 1 2 sin ( θ ) ∂ r ( r 2 ψ r 2 ) − sin ( θ ) r ψ r 2 + ∂ θ ( sin ( θ ) ψ θ ψ r ) − ∂ r ( sin ( θ ) ψ θ 2 2 ) − ∂ r ( sin ( θ ) ψ 2 ) . (93) \partial_t\left(\frac{\sin(\theta)}{r^2}\psi_r\psi_t\right)=\partial_r\left(\sin(\theta)\frac{\psi_t^2}{2r^2}\right)+\sin(\theta)\frac{\psi_t^2}{r^3}+\frac{1}{2}\sin(\theta)\partial_r(r^2\psi_r^2)-\sin(\theta)r\psi_r^2+\partial_\theta\left(\sin(\theta)\psi_\theta\psi_r\right)-\partial_r\left(\sin(\theta)\frac{\psi_\theta^2}{2}\right)-\partial_r\left(\sin(\theta)\psi^2\right). \tag*{(93)} ∂ t ( r 2 sin ( θ ) ψ r ψ t ) = ∂ r ( sin ( θ ) 2 r 2 ψ t 2 ) + sin ( θ ) r 3 ψ t 2 + 2 1 sin ( θ ) ∂ r ( r 2 ψ r 2 ) − sin ( θ ) r ψ r 2 + ∂ θ ( sin ( θ ) ψ θ ψ r ) − ∂ r ( sin ( θ ) 2 ψ θ 2 ) − ∂ r ( sin ( θ ) ψ 2 ) . (93)
Additionally, we shall need the identity
∂ t ( sin ( θ ) r 3 ψ ψ t ) = sin ( θ ) ψ t 2 r 3 + sin ( θ ) ( ∂ r ( r ψ ψ r ) − ψ ψ r − r ψ r 2 ) + ∂ θ ( sin ( θ ) r ψ ψ θ ) − sin ( θ ) r ψ θ 2 − 2 sin ( θ ) r ψ 2 . (94) \partial_t\left(\frac{\sin(\theta)}{r^3}\psi\psi_t\right)=\sin(\theta)\frac{\psi_t^2}{r^3}+\sin(\theta)\left(\partial_r(r\psi\psi_r)-\psi\psi_r-r\psi_r^2\right)+\partial_\theta\left(\frac{\sin(\theta)}{r}\psi\psi_\theta\right)-\frac{\sin(\theta)}{r}\psi_\theta^2-\frac{2\sin(\theta)}{r}\psi^2. \tag*{(94)} ∂ t ( r 3 sin ( θ ) ψ ψ t ) = sin ( θ ) r 3 ψ t 2 + sin ( θ ) ( ∂ r ( r ψ ψ r ) − ψ ψ r − r ψ r 2 ) + ∂ θ ( r sin ( θ ) ψ ψ θ ) − r sin ( θ ) ψ θ 2 − r 2 sin ( θ ) ψ 2 . (94)
The previous identities are motivated by the natural d’Alambertian operator, that even under a twisted geometry “preserves” a suitable energy and momentum density.
As before, we consider the energy space E \mathcal{E} E as precisely the set of data ( ψ , ∂ t ψ ) ∣ t = 0 (\psi,\partial_t\psi)|_{t=0} ( ψ , ∂ t ψ ) ∣ t = 0 and evolutions ( ψ , ∂ t ψ ) ( t ) (\psi,\partial_t\psi)(t) ( ψ , ∂ t ψ ) ( t ) , t > 0 t>0 t > 0 under which both ψ r ψ t sin − 3 θ \psi_r\psi_t\sin^{-3}\theta ψ r ψ t sin − 3 θ and 1 r 2 sin − 3 θ ψ θ ψ t \frac{1}{r^2}\sin^{-3}\theta\psi_\theta\psi_t r 2 1 sin − 3 θ ψ θ ψ t vanish at the boundary, and E ~ 1 \widetilde{E}_1 E 1 is well-defined. We shall prove in Appendix § 8 that one has the integrated local energy bound
∫ 0 t ∫ r , θ sin ( θ ) ( 1 r ( ψ θ 2 + ψ 2 ) ) + ∫ 0 t ∫ r , θ sin ( θ ) 1 + r 2 ( r 2 ψ r 2 + ψ t 2 r 2 ) ≤ C ( E ) , (95) \int_0^t\int_{r,\theta}\sin(\theta)\left(\frac{1}{r}\left(\psi_\theta^2+\psi^2\right)\right)+\int_0^t\int_{r,\theta}\frac{\sin(\theta)}{1+r^2}\left(r^2\psi_r^2+\frac{\psi_t^2}{r^2}\right)\le C(E), \tag*{(95)} ∫ 0 t ∫ r , θ sin ( θ ) ( r 1 ( ψ θ 2 + ψ 2 ) ) + ∫ 0 t ∫ r , θ 1 + r 2 sin ( θ ) ( r 2 ψ r 2 + r 2 ψ t 2 ) ≤ C ( E ) , (95)
revealing mild decay of finite energy solutions. Finally, one has the asymptotic stability property for finite energy solutions:
lim t → + ∞ ∫ r , θ r 2 sin ( θ ) 1 + r 2 ( ψ t 2 r 4 + ψ 2 r 2 + ψ r 2 + ψ θ 2 r 2 ) = 0. (96) \lim_{t\to+\infty}\int_{r,\theta}\frac{r^2\sin(\theta)}{1+r^2}\left(\frac{\psi_t^2}{r^4}+\frac{\psi^2}{r^2}+\psi_r^2+\frac{\psi_\theta^2}{r^2}\right)=0. \tag*{(96)} t → + ∞ lim ∫ r , θ 1 + r 2 r 2 sin ( θ ) ( r 4 ψ t 2 + r 2 ψ 2 + ψ r 2 + r 2 ψ θ 2 ) = 0. (96)
Therefore, the solutions converge to zero in a weighted norm (note that (96) differs from the norm in (91) by the additional factor r 2 / ( 1 + r 2 ) r^2/(1+r^2) r 2 / ( 1 + r 2 ) ). This weight vanishes as r r r approaches the horizon at zero, so it only gives information about the decay away from this boundary. Finally, we stress that these computations are valid when all boundary terms vanish, so we need a suitable decay of ψ \psi ψ near r = 0 r=0 r = 0 and as r → ∞ r\to\infty r → ∞ .
Averaged and Aretakis instability in the Near-NHEK
Now we analyze averaged and Aretakis horizon instabilities [3 ] for the near-NHEK family of metrics (11) . We shall provide an analysis in terms of growth of local energy type norms based on coordinates adapted to the event horizon. These local energy norms are less demanding than the ones introduced in previous sections, precisely to be able to capture a more detailed behavior at the near horizon itself. For the moment we shall assume L ( t ) = − κ 2 2 \mathcal{L}(t)=-\frac{\kappa^2}{2} L ( t ) = − 2 κ 2 . We define ρ = r − L 2 r = r + κ 2 4 r \rho=r-\frac{\mathcal{L}}{2r}=r+\frac{\kappa^2}{4r} ρ = r − 2 r L = r + 4 r κ 2 , which maps r ∈ ( κ / 2 , ∞ ) r\in(\kappa/2,\infty) r ∈ ( κ /2 , ∞ ) onto ρ ∈ ( κ , ∞ ) \rho\in(\kappa,\infty) ρ ∈ ( κ , ∞ ) , and set r ˉ = ρ − κ \bar{r}=\rho-\kappa r ˉ = ρ − κ , so that r ˉ ≥ 0 \bar{r}\geq0 r ˉ ≥ 0 , with the event horizon located at r ˉ = 0 \bar{r}=0 r ˉ = 0 . We shall now use ingoing Eddington–Finkelstein coordinates. Define the tortoise coordinate
d r ∗ d r ˉ = ( r ˉ ( r ˉ + 2 κ ) ) − 1 , r ∗ ( r ˉ ) = 1 2 κ log ∣ r ˉ r ˉ + 2 κ ∣ . \frac{dr_*}{d\bar{r}}=\left(\bar{r}(\bar{r}+2\kappa)\right)^{-1},\qquad r_*(\bar{r})=\frac{1}{2\kappa}\log\left|\frac{\bar{r}}{\bar{r}+2\kappa}\right|. d r ˉ d r ∗ = ( r ˉ ( r ˉ + 2 κ ) ) − 1 , r ∗ ( r ˉ ) = 2 κ 1 log r ˉ + 2 κ r ˉ .
Denote
v = t + r ∗ ( r ˉ ) , ϕ ~ = ϕ − 1 2 log ∣ r ˉ ( r ˉ + 2 κ ) ∣ . v=t+r_*(\bar{r}),\qquad\tilde{\phi}=\phi-\frac{1}{2}\log\left|\bar{r}(\bar{r}+2\kappa)\right|. v = t + r ∗ ( r ˉ ) , ϕ ~ = ϕ − 2 1 log ∣ r ˉ ( r ˉ + 2 κ ) ∣ .
The metric (14) in this case becomes
d s 2 = Λ ( θ ) ( − r ( r + 2 κ ) d v 2 + 2 d v d r + d θ 2 ) + Γ ( θ ) ( d ϕ ~ + ( r + κ ) d v ) 2 . ds^2=\Lambda(\theta)\left(-r(r+2\kappa)dv^2+2dvdr+d\theta^2\right)+\Gamma(\theta)\left(d\tilde{\phi}+(r+\kappa)dv\right)^2. d s 2 = Λ ( θ ) ( − r ( r + 2 κ ) d v 2 + 2 d v d r + d θ 2 ) + Γ ( θ ) ( d ϕ ~ + ( r + κ ) d v ) 2 .
In what follows we drop the bar and write r r r for r ˉ \bar{r} r ˉ , so that r ∈ [ 0 , ∞ ) r\in[0,\infty) r ∈ [ 0 , ∞ ) and the event horizon is located at r = 0 r=0 r = 0 . Then (53) becomes
□ 2 χ + 1 s ( θ ) ∂ θ ( s ( θ ) χ θ ) − 2 χ = 0 (97) \Box_2\chi+\frac{1}{s(\theta)}\partial_\theta(s(\theta)\chi_\theta)-2\chi=0 \tag*{(97)} □ 2 χ + s ( θ ) 1 ∂ θ ( s ( θ ) χ θ ) − 2 χ = 0 (97)
where the nearly-AdS2 _2 2 Laplacian is given now by
∂ r ( ( r 2 + 2 κ r ) ∂ r + 2 ∂ v ) . \partial_r\left((r^2+2\kappa r)\partial_r+2\partial_v\right). ∂ r ( ( r 2 + 2 κ r ) ∂ r + 2 ∂ v ) .
Averaged instability
As previously mentioned, we shall work in a space slightly larger than the energy space determined in (83) . This will be expressed in terms of the weighted norms used to show growth of local norms. First, assume κ > 0 \kappa>0 κ > 0 . Since r 2 + 2 κ r r^2+2\kappa r r 2 + 2 κ r vanishes at r = 0 r=0 r = 0 , integrating (97) in r r r from the horizon gives
( r 2 + 2 κ r ) ∂ r χ + 2 ∂ v χ = 2 ∂ v χ ∣ r = 0 − ∂ r − 1 ( 1 s ( θ ) ∂ θ ( s ( θ ) χ θ ) − 2 χ ) , ∂ r − 1 f : = ∫ 0 r f d r ′ . (98) (r^2+2\kappa r)\partial_r\chi+2\partial_v\chi=2\partial_v\chi\big|_{r=0}-\partial_r^{-1}\left(\frac{1}{s(\theta)}\partial_\theta(s(\theta)\chi_\theta)-2\chi\right),\qquad\partial_r^{-1}f:=\int_0^r f\,dr'. \tag*{(98)} ( r 2 + 2 κ r ) ∂ r χ + 2 ∂ v χ = 2 ∂ v χ r = 0 − ∂ r − 1 ( s ( θ ) 1 ∂ θ ( s ( θ ) χ θ ) − 2 χ ) , ∂ r − 1 f := ∫ 0 r f d r ′ . (98)
The first term on the right-hand side is the value of 2 ∂ v χ 2\partial_v\chi 2 ∂ v χ on the horizon; it does not vanish in general. Integrating on ( r , θ ) ∈ [ 0 , ∞ ) × ( 0 , π ) (r,\theta)\in[0,\infty)\times(0,\pi) ( r , θ ) ∈ [ 0 , ∞ ) × ( 0 , π ) and using (98)
d d v 1 2 ∫ 1 r 1 2 s ( θ ) χ 2 = − 1 2 ∫ 1 r 1 2 χ ∂ r − 1 ( ∂ θ ( s ( θ ) χ θ ) − 2 s ( θ ) χ ) + 1 4 ∫ s ( θ ) ∂ r ( r 3 2 + 2 κ r 1 2 ) χ 2 + ∫ 1 r 1 2 s ( θ ) χ ∂ v χ ∣ r = 0 . (99) \frac{d}{dv}\frac{1}{2}\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2=-\frac{1}{2}\int\frac{1}{r^{\frac{1}{2}}}\chi\partial_r^{-1}\left(\partial_\theta(s(\theta)\chi_\theta)-2s(\theta)\chi\right)+\frac{1}{4}\int s(\theta)\partial_r\left(r^{\frac{3}{2}}+2\kappa r^{\frac{1}{2}}\right)\chi^2+\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi\partial_v\chi\big|_{r=0}. \tag*{(99)} d v d 2 1 ∫ r 2 1 1 s ( θ ) χ 2 = − 2 1 ∫ r 2 1 1 χ ∂ r − 1 ( ∂ θ ( s ( θ ) χ θ ) − 2 s ( θ ) χ ) + 4 1 ∫ s ( θ ) ∂ r ( r 2 3 + 2 κ r 2 1 ) χ 2 + ∫ r 2 1 1 s ( θ ) χ ∂ v χ r = 0 . (99)
Notice that the weight r − 1 2 r^{-\frac{1}{2}} r − 2 1 represents a concentrated operator that localizes measurements around r = 0 r=0 r = 0 .
As a first approach, let us assume that the term ∂ v χ ∣ r = 0 \left.\partial_v\chi\right|_{r=0} ∂ v χ ∣ r = 0 vanishes, then, integrating by parts and simplifying we get,
d d v 1 2 ∫ 1 r 1 2 s ( θ ) χ 2 = 1 2 ∫ 1 r 1 2 s ( θ ) cos 2 θ ∂ θ ( χ cos θ ) ∂ r − 1 ( ∂ θ ( χ cos θ ) ) + 1 8 ∫ s ( θ ) ( 3 r 3 2 + 2 κ r 1 2 ) χ 2 ≥ 1 8 ∫ 1 r 1 2 s ( θ ) cos 2 θ ( ∂ r − 1 ∂ θ ( χ cos θ ) ) 2 + 1 4 κ ∫ 1 r 1 2 s ( θ ) χ 2 ≥ 1 4 κ ∫ 1 r 1 2 s ( θ ) χ 2 . \begin{aligned}
\frac{d}{dv}\frac{1}{2}\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2
={}&\frac{1}{2}\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\cos^2\theta\,\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\partial_r^{-1}\left(\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\right)\\
&+\frac{1}{8}\int s(\theta)\left(3r^{\frac{3}{2}}+\frac{2\kappa}{r^{\frac{1}{2}}}\right)\chi^2\\
\ge{}&\frac{1}{8}\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\cos^2\theta\left(\partial_r^{-1}\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\right)^2+\frac{1}{4}\kappa\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2\\
\ge{}&\frac{1}{4}\kappa\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2.
\end{aligned} d v d 2 1 ∫ r 2 1 1 s ( θ ) χ 2 = ≥ ≥ 2 1 ∫ r 2 1 1 s ( θ ) cos 2 θ ∂ θ ( cos θ χ ) ∂ r − 1 ( ∂ θ ( cos θ χ ) ) + 8 1 ∫ s ( θ ) ( 3 r 2 3 + r 2 1 2 κ ) χ 2 8 1 ∫ r 2 1 1 s ( θ ) cos 2 θ ( ∂ r − 1 ∂ θ ( cos θ χ ) ) 2 + 4 1 κ ∫ r 2 1 1 s ( θ ) χ 2 4 1 κ ∫ r 2 1 1 s ( θ ) χ 2 .
We have used here that the boundary terms s ( θ ) ( r 3 2 + 2 κ r 1 2 ) χ 2 ∣ r = 0 s(\theta)\left(r^{\frac{3}{2}}+2\kappa r^{\frac{1}{2}}\right)\chi^2\big|_{r=0} s ( θ ) ( r 2 3 + 2 κ r 2 1 ) χ 2 r = 0 and very importantly the term 1 r 1 2 χ ∂ r − 1 ( 1 cos θ ( s ( θ ) cos 2 θ ∂ θ ( χ cos θ ) ) ) ∣ r = 0 \frac{1}{r^{\frac{1}{2}}}\chi\partial_r^{-1}\left(\frac{1}{\cos\theta}\left(s(\theta)\cos^2\theta\,\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\right)\right)\big|_{r=0} r 2 1 1 χ ∂ r − 1 ( c o s θ 1 ( s ( θ ) cos 2 θ ∂ θ ( c o s θ χ ) ) ) r = 0 are zero. This is ensured if χ \chi χ has sufficient decay to zero at r = 0 r=0 r = 0 . Therefore,
∫ 1 r 1 2 s ( θ ) χ 2 ( v ) ≥ e 1 2 κ v ∫ 1 r 1 2 s ( θ ) χ 2 ( v = 0 ) . \int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2(v)\ge e^{\frac{1}{2}\kappa v}\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2(v=0). ∫ r 2 1 1 s ( θ ) χ 2 ( v ) ≥ e 2 1 κ v ∫ r 2 1 1 s ( θ ) χ 2 ( v = 0 ) .
We can conclude that there is exponential instability and growth of this energy weighted norm as v v v tends to infinity. We emphasize that the allowed data is important. Indeed, the previous assumption is not satisfied by finite energy solutions in the sense presented above. For instance, using the modes of § 3.1.3
χ = sin 4 θ p l ( cos θ ) e − ( l + 1 ) κ v ( r + 2 κ ) − ( l + 1 ) , \chi=\sin^4\theta p_l(\cos\theta)e^{-(l+1)\kappa v}(r+2\kappa)^{-(l+1)}, χ = sin 4 θ p l ( cos θ ) e − ( l + 1 ) κ v ( r + 2 κ ) − ( l + 1 ) ,
solves (97) , is regular at r = 0 r=0 r = 0 and decays as r → ∞ r\to\infty r → ∞ , while ∫ r − 1 / 2 s ( θ ) χ 2 \int r^{-1/2}s(\theta)\chi^2 ∫ r − 1/2 s ( θ ) χ 2 decays like e − 2 ( l + 1 ) κ v e^{-2(l+1)\kappa v} e − 2 ( l + 1 ) κ v ; This implies that, for this solution, the last term in (99) is negative and dominates, in concordance with the absence of a classical Aretakis instability in the case κ > 0 \kappa>0 κ > 0 .
Without any assumption on the term ∂ v χ ∣ r = 0 \left.\partial_v\chi\right|_{r=0} ∂ v χ ∣ r = 0 , we recover (99) , again, integrating by parts and simplifying we get,
d d v 1 2 ∫ 1 r 1 2 s ( θ ) χ 2 = 1 8 ∫ s ( θ ) r 3 2 ( ( ∂ θ ∂ r − 1 χ ) 2 + 2 ( ∂ r − 1 χ ) 2 ) + 1 8 ∫ s ( θ ) cos 2 θ ( 3 r 3 2 + 2 κ r 1 2 ) χ 2 + ∫ 1 r 1 2 s ( θ ) χ ∂ v χ ∣ r = 0 . (100) \begin{aligned}
\frac{d}{dv}\frac{1}{2}\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi^2
={}&\frac{1}{8}\int\frac{s(\theta)}{r^{\frac{3}{2}}}\left((\partial_\theta\partial_r^{-1}\chi)^2+2(\partial_r^{-1}\chi)^2\right)+\frac{1}{8}\int s(\theta)\cos^2\theta\left(3r^{\frac{3}{2}}+\frac{2\kappa}{r^{\frac{1}{2}}}\right)\chi^2\\
&+\int\frac{1}{r^{\frac{1}{2}}}s(\theta)\chi\partial_v\chi\big|_{r=0}. \tag*{(100)}
\end{aligned} d v d 2 1 ∫ r 2 1 1 s ( θ ) χ 2 = 8 1 ∫ r 2 3 s ( θ ) ( ( ∂ θ ∂ r − 1 χ ) 2 + 2 ( ∂ r − 1 χ ) 2 ) + 8 1 ∫ s ( θ ) cos 2 θ ( 3 r 2 3 + r 2 1 2 κ ) χ 2 + ∫ r 2 1 1 s ( θ ) χ ∂ v χ r = 0 . (100)
The first two terms are nonnegative, but the last one is a flux through the horizon and has no definite sign. In fact, for κ > 0 \kappa> 0 κ > 0 the horizon is non-degenerate and the red-shift effect damps the transverse derivatives at r = 0 r = 0 r = 0 , so no Aretakis instability is expected in the near-NHEK background.
The case κ = 0 \kappa= 0 κ = 0 requires another argument. Indeed, it is not difficult to check that
d d v 1 2 ∫ 1 r 2 s ( θ ) χ 2 ≥ 1 2 ∫ 1 r 2 s ( θ ) cos 2 θ ( ∂ r − 1 ∂ θ ( χ cos θ ) ) 2 ≥ 0 , \frac{\mathrm{d}}{\mathrm{d}v}\frac{1}{2}\int\frac{1}{r^2}s(\theta)\chi^2 \geq\frac{1}{2}\int\frac{1}{r^2}s(\theta)\cos^2\theta\left(\partial_r^{-1}\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\right)^2 \geq0, d v d 2 1 ∫ r 2 1 s ( θ ) χ 2 ≥ 2 1 ∫ r 2 1 s ( θ ) cos 2 θ ( ∂ r − 1 ∂ θ ( cos θ χ ) ) 2 ≥ 0 ,
and therefore ∫ 1 r 2 s ( θ ) χ 2 ( v ) \int\frac{1}{r^2}s(\theta)\chi^2(v) ∫ r 2 1 s ( θ ) χ 2 ( v ) is increasing in v v v , under χ ( v , 0 , θ ) ≡ 0 \chi(v,0,\theta)\equiv0 χ ( v , 0 , θ ) ≡ 0 . We call it a mild instability, in the sense that this local norm grows, but it might be the case that it finally converges to a final state. In the general case where κ = 0 \kappa= 0 κ = 0 but χ \chi χ depends on θ \theta θ , if ∫ v ∫ 1 r 2 s ( θ ) cos 2 θ ( ∂ r − 1 ∂ θ ( χ cos θ ) ) 2 \int_v\int\frac{1}{r^2}s(\theta)\cos^2\theta\left(\partial_r^{-1}\partial_\theta\left(\frac{\chi}{\cos\theta}\right)\right)^2 ∫ v ∫ r 2 1 s ( θ ) cos 2 θ ( ∂ r − 1 ∂ θ ( c o s θ χ ) ) 2 does not integrate in time v ∈ [ 0 , ∞ ) v\in[0,\infty) v ∈ [ 0 , ∞ ) (notice that this might be case even with the integrand converging to zero), there will be a sort of averaged instability for non finite energy solutions.
On Aretakis instability
The Aretakis instability [3 , 5 , 6 ] concerns in this case the behavior of the transverse derivatives ∂ r k χ \partial_r^k\chi ∂ r k χ on the horizon, as opposite to the averaged quantities found in the previous paragraph. By applying ∂ r k \partial_r^k ∂ r k to (98) and restricting to the horizon r = 0 r = 0 r = 0 gives, for each admissible mode χ l \chi_l χ l ,
2 ∂ v χ l ( k + 1 ) + 2 κ ( k + 1 ) χ l ( k + 1 ) = [ l ( l + 1 ) − k ( k + 1 ) ] χ l ( k ) , 2\partial_v\chi_l^{(k+1)}+2\kappa(k+1)\chi_l^{(k+1)}=\left[l(l+1)-k(k+1)\right]\chi_l^{(k)}, 2 ∂ v χ l ( k + 1 ) + 2 κ ( k + 1 ) χ l ( k + 1 ) = [ l ( l + 1 ) − k ( k + 1 ) ] χ l ( k ) ,
χ l ( k ) : = ∂ r k χ l ∣ r = 0 , k ≥ 0. \chi_l^{(k)}:=\left.\partial_r^k\chi_l\right|_{r=0},\qquad k\geq0. χ l ( k ) := ∂ r k χ l r = 0 , k ≥ 0.
For κ > 0 \kappa> 0 κ > 0 each χ l ( k + 1 ) \chi_l^{(k+1)} χ l ( k + 1 ) satisfies a linear ODE in v v v with damping rate κ ( k + 1 ) \kappa(k+1) κ ( k + 1 ) , so all transverse derivatives on the horizon decay once χ l ∣ r = 0 \left.\chi_l\right|_{r=0} χ l ∣ r = 0 does (redshift). There is no Aretakis instability on near-NHEK, as expected. For κ = 0 \kappa= 0 κ = 0 the right-hand side vanishes at k = l k = l k = l , so
H l [ χ ] : = ∂ r l + 1 χ l ∣ r = 0 = χ l ( l + 1 ) H_l[\chi]:=\left.\partial_r^{l+1}\chi_l\right|_{r=0}=\chi_l^{(l+1)} H l [ χ ] := ∂ r l + 1 χ l r = 0 = χ l ( l + 1 )
is conserved along the Poincaré horizon (r = 0 , v r = 0,v r = 0 , v ), and at k = l + 1 k = l+1 k = l + 1 one gets ∂ v χ l ( l + 2 ) = − ( l + 1 ) H l \partial_v\chi_l^{(l+2)}=-(l+1)H_l ∂ v χ l ( l + 2 ) = − ( l + 1 ) H l , i.e.
∂ r l + 2 χ l ∣ r = 0 ( v ) = ∂ r l + 2 χ l ∣ r = 0 ( 0 ) − ( l + 1 ) H l v , \left.\partial_r^{l+2}\chi_l\right|_{r=0}(v)=\left.\partial_r^{l+2}\chi_l\right|_{r=0}(0)-(l+1)H_l v, ∂ r l + 2 χ l r = 0 ( v ) = ∂ r l + 2 χ l r = 0 ( 0 ) − ( l + 1 ) H l v ,
which grows linearly for every datum with H l ≠ 0 H_l\ne0 H l = 0 . This is the Aretakis mechanism [3 , 5 ] for the gauge-invariant variable χ \chi χ on NHEK. Notice that it is compatible with the decay statements of previous sections, which control weighted norms away from r = 0 r = 0 r = 0 only.
Symmetries and conserved quantities
In this section, we return to the unperturbed metric (2.14). To investigate the role of the symmetries as large gauge transformations, we first note that the spacetimes under consideration depend explicitly on time and are therefore non-stationary. We analyze the problem within the Hamiltonian framework of Regge and Teitelboim [39 ] , taking the radial coordinate as the evolution parameter. As we will discuss later on, this choice is natural for studying the long-time behaviour of the system and will allow us to connect the present results with the stability analysis of § 3 . Most of the near-horizon charge literature is phrased in the covariant phase space language of Iyer and Wald rather than in the Hamiltonian one; the relation between the two formalisms is reviewed in [52 ] , and we appeal to it when contrasting our results with the covariant computations of [30 , 38 ] below.
The starting point is to write the metric in ADM form,
d s 2 = N 2 d r 2 + h i j ( d x i + N i d r ) ( d x j + N j d r ) , (101) ds^2 = N^2 dr^2 + h_{ij}\left(dx^i + N^i dr\right)\left(dx^j + N^j dr\right), \tag*{(101)} d s 2 = N 2 d r 2 + h ij ( d x i + N i d r ) ( d x j + N j d r ) , (101)
where r r r is the radial coordinate, h i j h_{ij} h ij is the induced metric on the three-dimensional timelike hypersurfaces Σ \Sigma Σ defined by constant slices of r r r , and N N N and N i N^i N i are the lapse and shift functions, respectively. Latin indices label the coordinates of h i j h_{ij} h ij , and ∂ Σ \partial\Sigma ∂ Σ denotes the boundary of Σ \Sigma Σ . The Hamiltonian is
H [ N , N i ] = ∫ Σ d 3 x ( N H + N i H i ) + B [ N , N i ] , (102) H[N,N^i] = \int_{\Sigma} d^3x\left(N\mathcal{H} + N^i\mathcal{H}_i\right) + B[N,N^i], \tag*{(102)} H [ N , N i ] = ∫ Σ d 3 x ( N H + N i H i ) + B [ N , N i ] , (102)
where H \mathcal{H} H and H i \mathcal{H}_i H i are the Hamiltonian and momentum constraints, which are proportional to Einstein’s equations and therefore vanish on the near-NHEK solutions. The boundary term B B B is added to ensure a well-posed variational problem.
The canonical generator of asymptotic symmetries is defined as [39 ]
Q [ ξ , ξ i ] = B [ ξ , ξ i ] , (103) Q[\xi,\xi^i] = B[\xi,\xi^i], \tag*{(103)} Q [ ξ , ξ i ] = B [ ξ , ξ i ] , (103)
where ( ξ , ξ i ) (\xi,\xi^i) ( ξ , ξ i ) are the gauge parameters associated to the asymptotic Killing vector η μ \eta^\mu η μ via the ADM decomposition,
ξ = N η r , ξ i = η i + N i η r . (104) \xi= N\eta^r,\qquad\xi^i = \eta^i + N^i\eta^r. \tag*{(104)} ξ = N η r , ξ i = η i + N i η r . (104)
Its variation takes the form
δ Q [ ξ , ξ i ] = ∫ d 2 S l [ G i j k l ( ξ ∇ k δ h i j − δ h i j ∇ k ξ ) + 2 ξ k δ π k l + ( 2 ξ k π j l − ξ l π k j ) δ h j k ] , (105) \delta Q[\xi,\xi^i] = \int d^2 S_l \left[G^{ijkl}\left(\xi\nabla_k\delta h_{ij} - \delta h_{ij}\nabla_k\xi\right) + 2\xi_k\delta\pi^{kl} + \left(2\xi^k\pi^{jl} - \xi^l\pi^{kj}\right)\delta h_{jk}\right], \tag*{(105)} δ Q [ ξ , ξ i ] = ∫ d 2 S l [ G ij k l ( ξ ∇ k δ h ij − δ h ij ∇ k ξ ) + 2 ξ k δ π k l + ( 2 ξ k π j l − ξ l π k j ) δ h j k ] , (105)
with G i j k l = 1 2 h ( h i k h j l + h i l h j k − 2 h i j h k l ) G^{ijkl} = \frac{1}{2}\sqrt{h}\left(h^{ik}h^{jl} + h^{il}h^{jk} - 2h^{ij}h^{kl}\right) G ij k l = 2 1 h ( h ik h j l + h i l h j k − 2 h ij h k l ) . Here π i j \pi^{ij} π ij is the momentum conjugate to h i j h_{ij} h ij ,
π i j = h 2 κ g ( K i j − h i j K ) , K i j = 1 2 N ( ∇ i N j + ∇ j N i − ∂ r h i j ) , (106) \pi^{ij} = \frac{\sqrt{h}}{2\kappa_{\mathrm{g}}}\left(K^{ij} - h^{ij}K\right),\qquad K_{ij} = \frac{1}{2N}\left(\nabla_iN_j + \nabla_jN_i - \partial_rh_{ij}\right), \tag*{(106)} π ij = 2 κ g h ( K ij − h ij K ) , K ij = 2 N 1 ( ∇ i N j + ∇ j N i − ∂ r h ij ) , (106)
where K i j K_{ij} K ij is the extrinsic curvature of Σ \Sigma Σ and κ g = 8 π G \kappa_g = 8\pi G κ g = 8 π G .
To carry over these expressions from the standard temporal foliation to the radial one, where h i j h_{ij} h ij carries Lorentzian signature ( − , + , + ) (-,+,+) ( − , + , + ) , we use the analytic continuation proposed in [40 ] : N → i N N \to iN N → i N , which induces K → − i K K \to-iK K → − i K and h → i h \sqrt{h} \to i\sqrt{h} h → i h .
Finally, we recall that H [ ξ , ξ i ] H[\xi,\xi^i] H [ ξ , ξ i ] is the generator of the gauge transformations (16) . Specifically, the field variations are obtained via Poisson brackets with the radial Hamiltonian,
δ h i j = { h i j ( x ) , H [ ξ , ξ i ] } , δ π i j = { π i j ( x ) , H [ ξ , ξ i ] } , (107) \delta h_{ij}=\{h_{ij}(x),H[\xi,\xi^i]\}, \qquad\delta\pi^{ij}=\{\pi^{ij}(x),H[\xi,\xi^i]\}, \tag*{(107)} δ h ij = { h ij ( x ) , H [ ξ , ξ i ]} , δ π ij = { π ij ( x ) , H [ ξ , ξ i ]} , (107)
which coincide with the Lie derivatives of the fields when evaluated on-shell [39 , 40 ] . When the foliation of the volume spacetime Σ \Sigma Σ is set-up so that the leaves of the ADM foliation are timelike, the conserved charges associated to the transformations laws (107) , should be constructed, considering the radial evolution as the gravitational Poisson bracket, which should occur for different times t t t and t ′ t' t ′ with fixed radius.
Radial ADM with Regge–Teitelboim boundary terms, charges at a late-time corner of a warped throat
To define the conserved charges associated to the asymptotic symmetries, one must specify the physical boundary on which they are evaluated. The surface loci should be defined for some asymptotic value of t t t , which as we will see later, define a natural physical boundary compatible with the chosen radial foliation of the volume spacetime Σ \Sigma Σ . In this setup, the conventional ADM picture is reversed and leaves of the foliation are timelike, and the “evolution” occurs along a spacelike direction. Radial foliations of this type are standard in asymptotically AdS holography [39 , 40 ] and have been employed in the near-horizon AdS2 _2 2 setting [20 , 42 ] ; here we adapt them to the warped geometry of the extremal Kerr throat. In this case, we deal with the absence of a Fefferman–Graham expansion by defining the charges directly by Regge–Teitelboim variation at a late-time cut.
Considering the Regge–Teitelboim expression (105) , we write the the NHEK geometry in the ADM radial form, and consider to express the symmetries in terms of the surface deformation vectors (17) , the variation of the surface charge is
δ Q [ ϵ ] ∣ ∂ Σ = 8 π c ϕ δ M 2 + 4 π ( − ( 4 r 0 2 ( 2 r 0 2 + L ( t ) ) 2 + 1 ) ϵ ′ ′ ( t ) r 0 + ϵ ( t ) ( 2 r 0 − L ( t ) r 0 ) ) δ M 2 , (108) \delta Q[\epsilon]\big|_{\partial\Sigma}=8\pi c_{\phi}\delta M^{2}+4\pi\left(-\frac{\left(\dfrac{4r_{0}^{2}}{\sqrt{(2r_{0}^{2}+\mathcal{L}(t))^{2}}}+1\right)\epsilon''(t)}{r_{0}}+\epsilon(t)\left(2r_{0}-\frac{\mathcal{L}(t)}{r_{0}}\right)\right)\delta M^{2}, \tag*{(108)} δ Q [ ϵ ] ∂ Σ = 8 π c ϕ δ M 2 + 4 π − r 0 ( ( 2 r 0 2 + L ( t ) ) 2 4 r 0 2 + 1 ) ϵ ′′ ( t ) + ϵ ( t ) ( 2 r 0 − r 0 L ( t ) ) δ M 2 , (108)
where r 0 r_0 r 0 highlights the fact that these charges are computed at constant radial hypersurfaces, and depend explicitly on ( r 0 , t ) (r_0,t) ( r 0 , t ) .
The charges are evaluated on the two-surface Σ r 0 , t 1 = { r = r 0 , t = t 1 } \Sigma_{r_0,t_1}=\{r=r_0,t=t_1\} Σ r 0 , t 1 = { r = r 0 , t = t 1 } , coordinatized by ( θ , ϕ ) (\theta,\phi) ( θ , ϕ ) and the boundary of a leaf of the radial foliation, and we then let t 1 → + ∞ t_1\to+\infty t 1 → + ∞ at fixed r 0 r_0 r 0 ; we write i + i^+ i + for the resulting cut. We show that such limit exists for a class of backgrounds considered down below, so that the explicit t t t -dependence of (108) enters through ϵ ( t ) \epsilon(t) ϵ ( t ) and L ( t ) \mathcal{L}(t) L ( t ) in a finite manner when evaluated at i + i^+ i + .
The geometric picture behind this cut is the familiar one for the extremal throat. Since the fiber block d θ 2 + Λ 2 ( d ϕ + r d t ) 2 d\theta^2+\Lambda^2(d\phi+r\,dt)^2 d θ 2 + Λ 2 ( d ϕ + r d t ) 2 of (14) is positive definite, every causal curve projects to a causal curve of the AdS2 _2 2 base and every causal base curve admits a causal corotating lift, so the causal relations of (near-)NHEK are those of AdS2 _2 2 and the conformal diagram of the base organizes the throat, each of its points representing a squashed ( θ , ϕ ) (\theta,\phi) ( θ , ϕ ) fiber [11 , 53 ] . In these terms NHEK is the Poincaré triangle of Figure 1 and near-NHEK the sub-wedge of Figure 2 , related by the exponential map (C.3); the same pair of charts, and the same map between them, is used in [54 ] to relate near-NHEK plunges to NHEK circular orbits. The curves of fixed r 0 r_0 r 0 run between the corners i ± i^\pm i ± of the diagram, which is the sense in which the t 1 → + ∞ t_1\to+\infty t 1 → + ∞ cut is a corner of the throat. Details of the construction are collected in Appendix § 9 .
Figure 1. The figure shows the constant-Z Z Z (R = 1 / Z R=1/Z R = 1/ Z ) worldlines of this diagram; each point of the picture is now a squashed ( θ , ϕ ) (\theta,\phi) ( θ , ϕ ) sphere rather than a spacetime point, and the frame dragging is encoded in the twisting of its U(1) fiber over the base, through the connection r d t r\,dt r d t . Worldlines of constant Z Z Z (static Poincaré observers), shown for Z = 0.3 Z=0.3 Z = 0.3 (dashed), Z = 1 Z=1 Z = 1 (solid) and Z = 5 Z=5 Z = 5 (dotted); each label sits just to the right of its curve. All curves interpolate between the two corners i ± i^\pm i ± of the triangle; the maximal depth into the bulk is σ ~ max = 2 arctan Z \widetilde{\sigma}_{\max}=2\arctan Z σ m a x = 2 arctan Z , reached at τ = 0 \tau=0 τ = 0 . As Z → ∞ Z\to\infty Z → ∞ the worldline degenerates onto the Poincaré horizons.
Figure 2. The black hole chart ( t ˉ , r ˉ ) (\bar{t},\bar{r}) ( t ˉ , r ˉ ) covers the dark wedge 0 < u < v 0 < u < v 0 < u < v : the causal diamond of the boundary interval T ∈ ( 0 , ∞ ) T \in(0,\infty) T ∈ ( 0 , ∞ ) . Its Killing horizons meet at the bifurcation point, where r ˉ = κ \bar{r} = \kappa r ˉ = κ at finite t ˉ \bar{t} t ˉ . A constant-r ˉ \bar{r} r ˉ trajectory is shown dashed.
Let us recall that the KdV equation constrains the Schwarzian mode and emerges as a condition from the Killing equation (2.18)-(2.19). In particular, it appears under the
choice of symmetry generator
ϵ ( t ) = ϵ 1 ( t ) = L ( t ) + ϵ ˉ 1 , (109) \epsilon(t) = \epsilon_1(t) = \mathcal{L}(t) + \bar{\epsilon}_1, \tag*{(109)} ϵ ( t ) = ϵ 1 ( t ) = L ( t ) + ϵ ˉ 1 , (109)
where ϵ ˉ 1 \bar{\epsilon}_1 ϵ ˉ 1 is the integration constant of (23) , which labels the class of admissible backgrounds rather than a particular solution within it.
One relevant solution of (25) is the soliton (86) , obtained for ϵ ˉ 1 = − ω 2 \bar{\epsilon}_1 = -\omega^2 ϵ ˉ 1 = − ω 2 , which satisfies L ( ± ∞ ) = 0 \mathcal{L}(\pm\infty) = 0 L ( ± ∞ ) = 0 and therefore approaches the NHEK geometry. It depends on a single free parameter ω \omega ω , which fixes the geometry of the solution in two correlated ways. The peak value is L ( t 0 ) = ω 2 \mathcal{L}(t_0) = \omega^2 L ( t 0 ) = ω 2 and, since L = { f , t } \mathcal{L} = \{f,t\} L = { f , t } measures the deviation from pure NHEK, larger ω \omega ω corresponds to a stronger departure from extremality at the peak. The characteristic temporal width is Δ t ∼ 1 / ω \Delta t \sim1/\omega Δ t ∼ 1/ ω , as follows from the argument of the cosh. Larger ω \omega ω therefore produces a narrow, sharply peaked pulse, so that the deviation from extremality is intense but brief, while small ω \omega ω produces a broad, slowly varying profile in which the deviation is mild but persists over a long time. Amplitude and width are not independent, as is standard for KdV solitons.
The other case corresponds to the solution
L ( t ) = ω 2 cosh 2 ( ω ( t − t 0 ) 2 ) − κ 2 2 , (110) \mathcal{L}(t) = \frac{\omega^2}{\cosh^2\left(\frac{\omega(t-t_0)}{2}\right)} - \frac{\kappa^2}{2}, \tag*{(110)} L ( t ) = cosh 2 ( 2 ω ( t − t 0 ) ) ω 2 − 2 κ 2 , (110)
which solves (25) provided
ϵ ˉ 1 = 3 2 κ 2 − ω 2 , equivalently κ 2 = 2 3 ( ϵ ˉ 1 + ω 2 ) , (111) \bar{\epsilon}_1 = \frac{3}{2}\kappa^2-\omega^2,\qquad\text{equivalently}\qquad\kappa^2=\frac{2}{3}\left(\bar{\epsilon}_1+\omega^2\right), \tag*{(111)} ϵ ˉ 1 = 2 3 κ 2 − ω 2 , equivalently κ 2 = 3 2 ( ϵ ˉ 1 + ω 2 ) , (111)
with ω > 0 \omega>0 ω > 0 . It satisfies L ( ± ∞ ) = − κ 2 / 2 \mathcal{L}(\pm\infty)=-\kappa^2/2 L ( ± ∞ ) = − κ 2 /2 , which by (11) is the near-NHEK value of the Schwarzian mode, so that the solution is asymptotic to the near-NHEK geometry and κ \kappa κ is the same parameter appearing there. The case ϵ ˉ 1 = − ω 2 \bar{\epsilon}_1=-\omega^2 ϵ ˉ 1 = − ω 2 gives κ = 0 \kappa=0 κ = 0 and retrieves the NHEK scenario. The solution (110) thus corresponds to shifting the soliton by a constant background value, so that the geometry asymptotes to Rindler-AdS2 _2 2 rather than to pure NHEK, and its peak value is L ( t 0 ) = ω 2 − κ 2 / 2 \mathcal{L}(t_0)=\omega^2-\kappa^2/2 L ( t 0 ) = ω 2 − κ 2 /2 .
In both cases the soliton is a transient deformation: the profile departs from its asymptotic value over a time Δ t ∼ 1 / ω \Delta t\sim1/\omega Δ t ∼ 1/ ω and returns to the same value, approaching NHEK and near-NHEK respectively. The two solutions render two different asymptotic values of the symmetry generator (109) as t → + ∞ t\to+\infty t → + ∞ ,
N H E K ϵ ( + ∞ ) = − ω 2 , n e a r - N H E K ϵ ( + ∞ ) = κ 2 − ω 2 . (112) \begin{aligned}
\mathrm{NHEK}\qquad&\epsilon(+\infty)=-\omega^2,\\
\mathrm{near\text{-}NHEK}\qquad&\epsilon(+\infty)=\kappa^2-\omega^2. \tag*{(112)}
\end{aligned} NHEK near - NHEK ϵ ( + ∞ ) = − ω 2 , ϵ ( + ∞ ) = κ 2 − ω 2 . (112)
The general form of the charge variation (108) at the late-time cut, containing both cases, is
δ Q ∣ i + = 8 π c ϕ δ M 2 + 4 π [ ϵ ( + ∞ ) ( 2 r 0 − L ( + ∞ ) r 0 ) ] δ M 2 , (113) \left.\delta Q\right|_{i^+}=8\pi c_\phi\,\delta M^2+4\pi\left[\epsilon(+\infty)\left(2r_0-\frac{\mathcal{L}(+\infty)}{r_0}\right)\right]\delta M^2, \tag*{(113)} δ Q ∣ i + = 8 π c ϕ δ M 2 + 4 π [ ϵ ( + ∞ ) ( 2 r 0 − r 0 L ( + ∞ ) ) ] δ M 2 , (113)
since the term proportional to ϵ ′ ′ ( t ) = L ′ ′ ( t ) \epsilon''(t)=\mathcal{L}''(t) ϵ ′′ ( t ) = L ′′ ( t ) in (108) decays exponentially on both solutions.
For the soliton asymptotic to NHEK the generator is (112) and L ( + ∞ ) = 0 \mathcal{L}(+\infty)=0 L ( + ∞ ) = 0 , so that the only contribution is
N H E K Q ∣ i + = 8 π c ϕ J − 8 π ( ω 2 r 0 ) M 2 . (114) \mathrm{NHEK}\qquad\left.Q\right|_{i^+}=8\pi c_\phi J-8\pi(\omega^2r_0)M^2. \tag*{(114)} NHEK Q ∣ i + = 8 π c ϕ J − 8 π ( ω 2 r 0 ) M 2 . (114)
We have imposed the extremality relation J = M 2 J=M^2 J = M 2 to associate each generator with its corresponding conserved charge. The parameter ω \omega ω controls a transient pulse that decays away as t → ± ∞ t\to\pm\infty t → ± ∞ on the Schwarzian L \mathcal{L} L , but not on the gauge parameter (112) , which survives at the cut and is what the charge records.10
In the second case, where the soliton asymptotes to the near-NHEK geometry, the solution is governed by two independent scales. The background κ 2 / 2 \kappa^2/2 κ 2 /2 encodes a permanent
Q ∣ i + = 8 π c ϕ J + 2 π c ˉ + 1 M 2 . (115) \left.Q\right|_{i^+}=8\pi c_\phi J+2\pi\bar{c}_{+1}M^2. \tag*{(115)} Q ∣ i + = 8 π c ϕ J + 2 π c ˉ + 1 M 2 . (115)
deviation from extremality, so that the black hole does not return to zero temperature but asymptotes to a finite Hawking temperature T ∼ κ T \sim\kappa T ∼ κ , and there is an additional contribution coming from the peak value of the soliton. With ϵ ( + ∞ ) \epsilon(+\infty) ϵ ( + ∞ ) given by (4.13) and L ( + ∞ ) = − κ 2 / 2 \mathcal{L}(+\infty)=-\kappa^2/2 L ( + ∞ ) = − κ 2 /2 ,
near-NHEK Q ∣ i + = 8 π c ϕ J + 4 π ( κ 2 − ω 2 ) ( 2 r 0 + κ 2 2 r 0 ) M 2 , (116) \text{near-NHEK}\qquad Q|_{i^+}=8\pi c_{\phi}J+4\pi\left(\kappa^2-\omega^2\right)\left(2r_0+\frac{\kappa^2}{2r_0}\right)M^2, \tag*{(116)} near-NHEK Q ∣ i + = 8 π c ϕ J + 4 π ( κ 2 − ω 2 ) ( 2 r 0 + 2 r 0 κ 2 ) M 2 , (116)
Throughout this section we hold ω \omega ω and κ \kappa κ fixed. They are not gauge parameters but coordinates on the space of solutions, so that (4.15) and (4.17) are the integrals of (4.14) along the δ M 2 \delta M^2 δ M 2 direction alone, where integrability is immediate. Whether the charge remains integrable once δ ω \delta\omega δ ω and δ κ \delta\kappa δ κ are included we leave open; the generator (4.9) is field dependent, which is the situation in which obstructions are known to arise [55 ] .
The charge is well defined at any fixed value of the radial coordinate, but (4.15) and (4.17) diverge as r 0 → ∞ r_0\to\infty r 0 → ∞ because of the term linear in r 0 r_0 r 0 , and (4.17) diverges as r 0 → 0 r_0\to0 r 0 → 0 as well. One manner to regularize the first divergence is to take ϵ 1 ( + ∞ ) = ϵ ˉ 1 2 / r 0 \epsilon_1(+\infty)=\bar{\epsilon}_1^2/r_0 ϵ 1 ( + ∞ ) = ϵ ˉ 1 2 / r 0 , which in practice means reinterpreting the peak value of the soliton and the temperature as correlated with the distance to infinity. With that choice (4.15) becomes
Q ∣ i + = 8 π c ϕ J + 4 π ϵ ˉ 1 2 ( 2 − L ( + ∞ ) r 0 2 ) M 2 , (117) Q|_{i^+}=8\pi c_{\phi}J+4\pi\bar{\epsilon}_1^2\left(2-\frac{\mathcal{L}(+\infty)}{r_0^2}\right)M^2, \tag*{(117)} Q ∣ i + = 8 π c ϕ J + 4 π ϵ ˉ 1 2 ( 2 − r 0 2 L ( + ∞ ) ) M 2 , (117)
which converges as r 0 → ∞ r_0\to\infty r 0 → ∞ , since L ( + ∞ ) \mathcal{L}(+\infty) L ( + ∞ ) vanishes for the soliton without background and is finite in the near-NHEK case.
Alternatively, one may note that the radial dependence of (4.14) is stationary at a radius fixed by the geometry. Writing F ( r 0 ) = 2 r 0 − L ( + ∞ ) / r 0 F(r_0)=2r_0-\mathcal{L}(+\infty)/r_0 F ( r 0 ) = 2 r 0 − L ( + ∞ ) / r 0 , one has
d F d r 0 = 2 r 0 2 + L ( + ∞ ) r 0 2 , (118) \frac{dF}{dr_0}=\frac{2r_0^2+\mathcal{L}(+\infty)}{r_0^2}, \tag*{(118)} d r 0 d F = r 0 2 2 r 0 2 + L ( + ∞ ) , (118)
which vanishes exactly at the horizon radius r 0 = κ / 2 r_0=\kappa/2 r 0 = κ /2 identified in § 2.1 , where F = 2 κ F=2\kappa F = 2 κ . Evaluated there, the near-NHEK charge takes the finite value
Q ∣ i + , r 0 = κ / 2 = 8 π c ϕ J + 8 π κ ( κ 2 − ω 2 ) M 2 , (119) Q|_{i^+,\,r_0=\kappa/2}=8\pi c_{\phi}J+8\pi\kappa\left(\kappa^2-\omega^2\right)M^2, \tag*{(119)} Q ∣ i + , r 0 = κ /2 = 8 π c ϕ J + 8 π κ ( κ 2 − ω 2 ) M 2 , (119)
with no further prescription. This choice degenerates in the NHEK case, where L ( + ∞ ) = 0 \mathcal{L}(+\infty)=0 L ( + ∞ ) = 0 and the stationary radius collapses to r 0 = 0 r_0=0 r 0 = 0 .
It is instructive to contrast (4.15) and (4.17) with the charges of the Near-Horizon Extremal Geometries phase space of [37 , 38 ] , where the throat is deformed along the torus of U(1) isometry directions. Those deformations commute with the time translation and scaling isometries of the background, so their symplectic charges are independent of the surface on which they are evaluated and the symmetry algebra is an extension of the Virasoro algebra with the entropy as central term. Our deformation acts along time and breaks precisely those isometries, which is why (114) and (116) depend on r 0 r_0 r 0 and on the cut. For the same reason the S L ( 2 , R ) \mathrm{SL}(2,\mathbb{R}) SL ( 2 , R ) charges, which vanish identically on the phase space of [37 ] at fixed angular momenta and entropy, need not vanish on the family (14) . A direct comparison does remain in the U ( 1 ) \mathrm{U}(1) U ( 1 ) sector where the ∂ ϕ \partial_\phi ∂ ϕ charge of [30 ] is likewise proportional to J J J . In this case, nonetheless, they compute charges of the linear perturbation while here we restrain to charges of the background.
Conclusions
We have studied axisymmetric gravitational perturbations of the near-horizon region of (near-)extremal Kerr deformed by a Schwarzian mode, and two structures organize the results.
The first is kinematic. Allowing the gauge parameter to depend on the field, ϵ = ϵ [ L ] \epsilon= \epsilon[\mathcal{L}] ϵ = ϵ [ L ] , and requiring it to remain local, turns the Killing equation of the family (14) into the stationary sector of the Korteweg–de Vries hierarchy for L \mathcal{L} L . The first level gives NHEK and near-NHEK (L ′ = 0 \mathcal{L}' = 0 L ′ = 0 ); the second gives the stationary KdV equation (25) , whose solitons are pulses of amplitude ω 2 \omega^2 ω 2 that depart from and return to L ( ± ∞ ) = 0 \mathcal{L}(\pm\infty) = 0 L ( ± ∞ ) = 0 (NHEK) or − κ 2 / 2 -\kappa^2/2 − κ 2 /2 (near-NHEK). This is a selection of backgrounds rather than an evolution: the hierarchy imposes constraints instead of generating a flow, and every member of the family keeps the isometry algebra s l ( 2 , R ) ⊕ u ( 1 ) \mathfrak{sl}(2,\mathbb{R}) \oplus\mathfrak{u}(1) sl ( 2 , R ) ⊕ u ( 1 ) . What the hierarchy singles out is the subset of profiles on which one of the three generators is a local functional of L \mathcal{L} L .
The second structure is dynamical. The linearized Einstein equations in the axisymmetric sector reduce to a single gauge-invariant scalar χ \chi χ obeying a wave equation on the nearly-A d S 2 \mathrm{AdS}_2 AdS 2 base, with L ( t ) \mathcal{L}(t) L ( t ) as its only background-dependent coefficient; for constant L \mathcal{L} L this is the radial Teukolsky equation, recovering [30 ] . Using energy and virial methods, we found that the stability statement tracks the monotonicity of the background. For L ≡ 0 \mathcal{L} \equiv0 L ≡ 0 the energy (91) is conserved and the integrated local energy estimate (95) gives asymptotic stability in a weighted norm (96) . On near-NHEK the conserved energy associated with ϵ = cosh ( κ t ) \epsilon= \cosh(\kappa t) ϵ = cosh ( κ t ) yields exponential decay of the norm at a rate set by κ \kappa κ . For general L ≥ 0 \mathcal{L} \ge0 L ≥ 0 time-translation invariance is lost, but (83) remains a Lyapunov functional, E ˙ = L ′ × \dot{E} = \mathcal{L}' \times E ˙ = L ′ × (positive density): profiles with L ≥ L 0 > 0 \mathcal{L} \ge\mathcal{L}_0 > 0 L ≥ L 0 > 0 and L ′ ∈ L 1 \mathcal{L}' \in L^1 L ′ ∈ L 1 give the uniform bound E ( t ) ≤ E 0 e C L / L 0 E(t) \le E_0 e^{C\mathcal{L}/\mathcal{L}_0} E ( t ) ≤ E 0 e C L / L 0 , while along the KdV soliton, where the background weight is strong, only the part of the energy density supported where L ≳ r 2 \mathcal{L} \gtrsim r^2 L ≳ r 2 relaxes, and only along a sequence of times. In horizon-regular coordinates, the degenerate horizon of NHEK supports an Aretakis-type instability (the ( l + 1 ) (l+1) ( l + 1 ) -th transverse derivative of each mode is conserved and the ( l + 2 ) (l+2) ( l + 2 ) -th grows linearly in v v v ), whereas on near-NHEK the redshift damps all transverse derivatives. Finally, with a radial Regge–Teitelboim foliation adapted to the non-stationary backgrounds, the surface charges at a late-time corner of the throat distinguish the two solitonic families through the pulse amplitude ω \omega ω and the residual temperature κ \kappa κ .
The picture that emerges is that “stability of the near-horizon geometry” is not a single property. On NHEK, decay in a weighted energy norm coexists with growth of transverse derivatives on the horizon; moving along the KdV family to near-NHEK removes the latter and turns the conservation law for ϵ = 1 \epsilon= 1 ϵ = 1 into a one-sided inequality, without any change in the local geometry. This bears on the question of [23 ] , whether an instability of the near-horizon geometry implies one of the parent black hole: our results indicate that it is underdetermined as posed, since the answer depends on the norm and on the profile of the Schwarzzian mode, neither of which is fixed by the near-horizon limit. This complements [7 , 8 ] , where the transient growth at finite temperature and the decay rates and transverse-derivative growth follow from the near-horizon conformal symmetry alone: within that symmetry, the Schwarzian profile is a further datum.
Several directions remain open. First, perturbations beyond the axisymmetric sector: restricting to m = 0 m = 0 m = 0 removes superradiance and makes the energy method tractable, but for extremal Kerr the non-axisymmetric modes govern the late-time behaviour and grow faster [31 ] . Second, the quantum imprint of the background selection: the soft sector that labels these backgrounds controls the low-temperature partition function of near-extremal black holes [17 ] , and replacing Schwarzian by KdV asymptotics already modifies the temperature dependence of the entropy in Jackiw–Teitelboim gravity [28 , 29 ] . Since the Schwarzian term itself arises as a counterterm fixed by the parent theory [56 ] , computing the partition function over the family (2.14), level by level in the hierarchy, would show whether the classical sensitivity to the profile survives quantization. Third, the construction uses little beyond the AdS2 _2 2 factor and the warped fibration, so extremal Kerr–Newman, Reissner–Nordström and higher-dimensional extremal horizons should admit the same field-dependent symmetry analysis, indicating how much of what we found is general to AdS2 _2 2 throats and how much is specific to rotation.
Acknowledgements
The authors received support from the Center for Mathematical Modeling (Universidad de Chile & CNRS IRL 2807) through ANID/Basal project # FB210005: H.VDB received partial funding from ANID/Fondecyt project # 125-0596. This work is part of the CMM PRISMALab initiative. C.M. received funding via Fondecyt ANID 1231250. M.C. is grateful to I. J. Araya and R. Olea, for valuable discussions. Special thanks to R. Mancilla for many stimulating discussions and for insights that significantly improved this work.
Teukolsky's master equation
An important feature of the Einstein equations that arises when studying gravitational perturbations is that those equations contain, for the geometric part, only the metric, the Ricci scalar R R R , and the Ricci tensor R μ ν R_{\mu\nu} R μν , where the latter one is a trace of the Riemann tensor (R μ ν = g α β R α μ β ν R_{\mu\nu}=g^{\alpha\beta}R_{\alpha\mu\beta\nu} R μν = g α β R α μ β ν ). On the other hand, we have the Weyl tensor C μ ν α β C_{\mu\nu\alpha\beta} C μν α β , which is the traceless part of the Riemann tensor
C μ ν α β = R μ ν α β − g μ [ α R β ] ν + g ν [ α R β ] μ + 1 3 R g μ [ α g β ] ν . (120) C_{\mu\nu\alpha\beta}=R_{\mu\nu\alpha\beta}-g_{\mu[\alpha}R_{\beta]\nu}+g_{\nu[\alpha}R_{\beta]\mu}+\frac{1}{3}Rg_{\mu[\alpha}g_{\beta]\nu}. \tag*{(120)} C μν α β = R μν α β − g μ [ α R β ] ν + g ν [ α R β ] μ + 3 1 R g μ [ α g β ] ν . (120)
Due to this, the Weyl tensor is left unconstrained by the Einstein equations and matter fields, and therefore represents the purely gravitational field. This is why the Weyl tensor has crucial information about the gravitational perturbations. With this tensor, one can also build the Weyl scalars: five scalars Ψ i \Psi_i Ψ i that are invariant under diffeomorphisms,
Ψ 0 = C μ ν α β l μ m ν l α m β , (121) \Psi_0=C_{\mu\nu\alpha\beta}l^\mu m^\nu l^\alpha m^\beta, \tag*{(121)} Ψ 0 = C μν α β l μ m ν l α m β , (121)
Ψ 1 = C μ ν α β l μ n ν l α m β , (122) \Psi_1=C_{\mu\nu\alpha\beta}l^\mu n^\nu l^\alpha m^\beta, \tag*{(122)} Ψ 1 = C μν α β l μ n ν l α m β , (122)
Ψ 2 = C μ ν α β l μ m ν m ˉ α n β , (123) \Psi_2=C_{\mu\nu\alpha\beta}l^\mu m^\nu\bar{m}^\alpha n^\beta, \tag*{(123)} Ψ 2 = C μν α β l μ m ν m ˉ α n β , (123)
Ψ 3 = C μ ν α β l μ n ν m ˉ α n β , (124) \Psi_3=C_{\mu\nu\alpha\beta}l^\mu n^\nu\bar{m}^\alpha n^\beta, \tag*{(124)} Ψ 3 = C μν α β l μ n ν m ˉ α n β , (124)
Ψ 4 = C μ ν α β n μ m ˉ ν n α m ˉ β . (125) \Psi_4=C_{\mu\nu\alpha\beta}n^\mu\bar{m}^\nu n^\alpha\bar{m}^\beta. \tag*{(125)} Ψ 4 = C μν α β n μ m ˉ ν n α m ˉ β . (125)
Using these five scalars, it is possible to classify different spacetimes according to the value of said scalars. This classification is called the Petrov classification, and we will focus on Petrov type D spacetimes, which are characterized by having only 1 non-zero Weyl scalar [47 ] : Ψ 2 \Psi_2 Ψ 2 .
Keeping this idea in mind, it results natural to think of perturbations to the Weyl scalars Ψ = Ψ ˉ + δ Ψ \Psi=\bar{\Psi}+\delta\Psi Ψ = Ψ ˉ + δ Ψ , due to perturbations of the metric, but without the need to compute the perturbation h μ ν h_{\mu\nu} h μν . In fact, by studying δ Ψ \delta\Psi δ Ψ one can reconstruct h μ ν h_{\mu\nu} h μν . This is how [2 ] studied the previous works on the Newman-Penrose formalism and found his master equation:
[ ( r 2 + a 2 ) 2 Δ − a 2 sin 2 θ ] ∂ t 2 Ψ ( s ) + 4 M a r Δ ∂ t ∂ ϕ Ψ ( s ) + [ a 2 Δ − 1 sin 2 θ ] ∂ ϕ 2 Ψ ( s ) − Δ − s ∂ r ( Δ s + 1 ∂ r Ψ ( s ) ) − 1 sin θ ∂ θ ( sin θ ∂ θ Ψ ( s ) ) − 2 s [ a ( r − M ) Δ + i cos θ sin 2 θ ] ∂ ϕ Ψ ( s ) − 2 s [ M ( r 2 − a 2 ) Δ − r − i a cos θ ] ∂ t Ψ ( s ) + ( s 2 cot 2 θ − s ) Ψ ( s ) = 0. (126) \begin{aligned}
\left[\frac{(r^2+a^2)^2}{\Delta}-a^2\sin^2\theta\right]\partial_t^2\Psi^{(s)}
&+\frac{4Mar}{\Delta}\partial_t\partial_\phi\Psi^{(s)}
+\left[\frac{a^2}{\Delta}-\frac{1}{\sin^2\theta}\right]\partial_\phi^2\Psi^{(s)} \\
&-\Delta^{-s}\partial_r\left(\Delta^{s+1}\partial_r\Psi^{(s)}\right)
-\frac{1}{\sin\theta}\partial_\theta\left(\sin\theta\partial_\theta\Psi^{(s)}\right) \\
&-2s\left[\frac{a(r-M)}{\Delta}+\frac{i\cos\theta}{\sin^2\theta}\right]\partial_\phi\Psi^{(s)}
-2s\left[\frac{M(r^2-a^2)}{\Delta}-r-ia\cos\theta\right]\partial_t\Psi^{(s)} \\
&+(s^2\cot^2\theta-s)\Psi^{(s)}=0. \tag*{(126)}
\end{aligned} [ Δ ( r 2 + a 2 ) 2 − a 2 sin 2 θ ] ∂ t 2 Ψ ( s ) + Δ 4 M a r ∂ t ∂ ϕ Ψ ( s ) + [ Δ a 2 − sin 2 θ 1 ] ∂ ϕ 2 Ψ ( s ) − Δ − s ∂ r ( Δ s + 1 ∂ r Ψ ( s ) ) − sin θ 1 ∂ θ ( sin θ ∂ θ Ψ ( s ) ) − 2 s [ Δ a ( r − M ) + sin 2 θ i cos θ ] ∂ ϕ Ψ ( s ) − 2 s [ Δ M ( r 2 − a 2 ) − r − ia cos θ ] ∂ t Ψ ( s ) + ( s 2 cot 2 θ − s ) Ψ ( s ) = 0. (126)
This master equation stands for any spin s s s perturbation field Ψ ( s ) \Psi^{(s)} Ψ ( s ) , e. g., Ψ ( ± 2 ) \Psi^{(\pm2)} Ψ ( ± 2 ) for gravitational perturbations, Ψ ( ± 1 2 ) \Psi^{(\pm\frac{1}{2})} Ψ ( ± 2 1 ) for fermionic perturbations, etc. We will focus on perturbations to the Weyl scalars, with s = ± 2 s=\pm2 s = ± 2 . By performing the diffeomorphism (2.10), for a field with no explicit dependence in ϕ \phi ϕ , the resulting equation adapted to the near-NHEK geometry is
16 r 2 ( 4 r 2 − κ 2 ) 2 ∂ t 2 Ψ ( s ) − M 4 ( 4 r 2 − κ 2 ) 2 ( 4 M 2 − κ 2 ) 2 r 2 ∂ r 2 Ψ ( s ) − 2 M 2 ( 1 + s ) ( 4 r 2 + κ 2 ) ( 4 M 2 − κ 2 ) r ∂ r Ψ ( s ) − 1 sin θ ∂ θ ( sin θ ∂ θ Ψ ( s ) ) − 8 s r ( 4 r 2 + κ 2 ) ( 4 r 2 − κ 2 ) 2 ∂ t Ψ ( s ) + ( s 2 csc 2 θ − s 2 − s ) Ψ ( s ) = 0. (127) \frac{16r^2}{(4r^2-\kappa^2)^2}\partial_t^2\Psi^{(s)}-\frac{M^4(4r^2-\kappa^2)^2}{(4M^2-\kappa^2)^2r^2}\partial_r^2\Psi^{(s)}-\frac{2M^2(1+s)(4r^2+\kappa^2)}{(4M^2-\kappa^2)r}\partial_r\Psi^{(s)}
-\frac{1}{\sin\theta}\partial_\theta(\sin\theta\partial_\theta\Psi^{(s)})-\frac{8sr(4r^2+\kappa^2)}{(4r^2-\kappa^2)^2}\partial_t\Psi^{(s)}+(s^2\csc^2\theta-s^2-s)\Psi^{(s)}=0. \tag*{(127)} ( 4 r 2 − κ 2 ) 2 16 r 2 ∂ t 2 Ψ ( s ) − ( 4 M 2 − κ 2 ) 2 r 2 M 4 ( 4 r 2 − κ 2 ) 2 ∂ r 2 Ψ ( s ) − ( 4 M 2 − κ 2 ) r 2 M 2 ( 1 + s ) ( 4 r 2 + κ 2 ) ∂ r Ψ ( s ) − sin θ 1 ∂ θ ( sin θ ∂ θ Ψ ( s ) ) − ( 4 r 2 − κ 2 ) 2 8 sr ( 4 r 2 + κ 2 ) ∂ t Ψ ( s ) + ( s 2 csc 2 θ − s 2 − s ) Ψ ( s ) = 0. (127)
Performing the separation of variables
Ψ ( ± 2 ) = ∑ l U l ( t , r ) S l ( θ ) , (128) \Psi^{(\pm2)}=\sum_l U_l(t,r)S_l(\theta), \tag*{(128)} Ψ ( ± 2 ) = l ∑ U l ( t , r ) S l ( θ ) , (128)
the equation (A.8) splits into
S l ′ ′ + cot θ S l ′ = − l ( l + 1 ) S l , (129) S_l''+\cot\theta S_l'=-l(l+1)S_l, \tag*{(129)} S l ′′ + cot θ S l ′ = − l ( l + 1 ) S l , (129)
− 16 r 2 ( 4 r 2 − κ 2 ) 2 ∂ t 2 U l + M 4 ( 4 r 2 − κ 2 ) 2 r 2 ( 4 M 2 − κ 2 ) 2 ∂ r 2 U l + 2 M 2 ( 1 + s ) ( 4 r 2 + κ 2 ) r ( 4 M 2 − κ 2 ) ∂ r U l + 8 s r ( 4 r 2 + κ 2 ) ( 4 r 2 − κ 2 ) 2 ∂ t U l − ( l ( l + 1 ) − s − s 2 + s 2 csc 2 θ ) U l = 0. (130) -\frac{16r^2}{(4r^2-\kappa^2)^2}\partial_t^2U_l+\frac{M^4(4r^2-\kappa^2)^2}{r^2(4M^2-\kappa^2)^2}\partial_r^2U_l+\frac{2M^2(1+s)(4r^2+\kappa^2)}{r(4M^2-\kappa^2)}\partial_rU_l
+\frac{8sr(4r^2+\kappa^2)}{(4r^2-\kappa^2)^2}\partial_tU_l-(l(l+1)-s-s^2+s^2\csc^2\theta)U_l=0. \tag*{(130)} − ( 4 r 2 − κ 2 ) 2 16 r 2 ∂ t 2 U l + r 2 ( 4 M 2 − κ 2 ) 2 M 4 ( 4 r 2 − κ 2 ) 2 ∂ r 2 U l + r ( 4 M 2 − κ 2 ) 2 M 2 ( 1 + s ) ( 4 r 2 + κ 2 ) ∂ r U l + ( 4 r 2 − κ 2 ) 2 8 sr ( 4 r 2 + κ 2 ) ∂ t U l − ( l ( l + 1 ) − s − s 2 + s 2 csc 2 θ ) U l = 0. (130)
Proofs of energy estimates
Proof of (3.63). We multiply (3.61) by s ( θ ) ψ t s(\theta)\psi_t s ( θ ) ψ t to obtain
1 r 4 ψ t t ψ t s ( θ ) = ψ r r ψ t s ( θ ) + 1 r 2 ( s ( θ ) ψ t ψ θ θ − 3 s ( θ ) cot ( θ ) ψ t ψ θ − 2 s ( θ ) ψ ψ t ) . \frac{1}{r^4}\psi_{tt}\psi_ts(\theta)=\psi_{rr}\psi_ts(\theta)+\frac{1}{r^2}\left(s(\theta)\psi_t\psi_{\theta\theta}-3s(\theta)\cot(\theta)\psi_t\psi_\theta-2s(\theta)\psi\psi_t\right). r 4 1 ψ tt ψ t s ( θ ) = ψ r r ψ t s ( θ ) + r 2 1 ( s ( θ ) ψ t ψ θ θ − 3 s ( θ ) cot ( θ ) ψ t ψ θ − 2 s ( θ ) ψ ψ t ) .
Therefore,
d d t ( 1 2 r 2 ψ t 2 s ( θ ) ) = s ( θ ) ψ r r ψ t − 2 r ψ r ψ t s ( θ ) + d d t ( ψ r 2 r 2 s ( θ ) ) + r ψ t s ( θ ) ψ θ θ − 3 r sin 2 θ cos θ ψ θ ψ t − d d t ( r ψ 2 s ( θ ) ) . \frac{d}{dt}\left(\frac{1}{2r^2}\psi_t^2s(\theta)\right)=s(\theta)\psi_{rr}\psi_t-\frac{2}{r}\psi_r\psi_ts(\theta)+\frac{d}{dt}\left(\frac{\psi_r^2}{r^2}s(\theta)\right)+r\psi_ts(\theta)\psi_{\theta\theta}-3r\sin^2\theta\cos\theta\psi_\theta\psi_t-\frac{d}{dt}\left(r\psi^2s(\theta)\right). d t d ( 2 r 2 1 ψ t 2 s ( θ ) ) = s ( θ ) ψ r r ψ t − r 2 ψ r ψ t s ( θ ) + d t d ( r 2 ψ r 2 s ( θ ) ) + r ψ t s ( θ ) ψ θ θ − 3 r sin 2 θ cos θ ψ θ ψ t − d t d ( r ψ 2 s ( θ ) ) .
Thus
d d t ( 1 2 r 4 ψ t 2 s ( θ ) ) = ∂ r ( ψ r ψ t s ( θ ) ) − d d t ( 1 2 ψ r 2 s ( θ ) ) + ∂ θ ( 1 r 2 s ( θ ) ψ θ ψ t ) − d d t ( 1 2 r 2 s ( θ ) ψ θ 2 ) − d d t ( 1 r 2 ψ 2 s ( θ ) ) . \frac{d}{dt}\left(\frac{1}{2r^4}\psi_t^2s(\theta)\right)=\partial_r\left(\psi_r\psi_ts(\theta)\right)-\frac{d}{dt}\left(\frac{1}{2}\psi_r^2s(\theta)\right)+\partial_\theta\left(\frac{1}{r^2}s(\theta)\psi_\theta\psi_t\right)-\frac{d}{dt}\left(\frac{1}{2r^2}s(\theta)\psi_\theta^2\right)-\frac{d}{dt}\left(\frac{1}{r^2}\psi^2s(\theta)\right). d t d ( 2 r 4 1 ψ t 2 s ( θ ) ) = ∂ r ( ψ r ψ t s ( θ ) ) − d t d ( 2 1 ψ r 2 s ( θ ) ) + ∂ θ ( r 2 1 s ( θ ) ψ θ ψ t ) − d t d ( 2 r 2 1 s ( θ ) ψ θ 2 ) − d t d ( r 2 1 ψ 2 s ( θ ) ) .
This gives (92) .
Proof of (3.64). We compute and use (90) :
∂ t ( 1 r 2 s ( θ ) ψ r ψ t ) = 1 r 2 s ( θ ) ψ t ψ t + 1 r 2 s ( θ ) ψ r ψ t t = ∂ r ( s ( θ ) ψ t 2 2 r 2 ) + s ( θ ) ψ t 2 r 3 + s ( θ ) ψ r ( r 2 ψ r r + ψ θ θ − 3 cot θ ψ θ − 2 ψ ) = ∂ r ( s ( θ ) ψ t 2 2 r 2 ) + s ( θ ) ψ t 2 r 3 + 1 2 s ( θ ) ∂ r ( r 2 ψ r 2 ) − s ( θ ) r ψ r 2 + ∂ θ ( s ( θ ) ψ θ ψ r ) − ∂ r ( s ( θ ) ψ θ 2 2 ) − ∂ r ( sin − 3 θ ψ 2 ) . \begin{aligned}
\partial_t\left(\frac{1}{r^2}s(\theta)\psi_r\psi_t\right)
&=\frac{1}{r^2}s(\theta)\psi_t\psi_t+\frac{1}{r^2}s(\theta)\psi_r\psi_{tt}\\
&=\partial_r\left(s(\theta)\frac{\psi_t^2}{2r^2}\right)+s(\theta)\frac{\psi_t^2}{r^3}+s(\theta)\psi_r\left(r^2\psi_{rr}+\psi_{\theta\theta}-3\cot\theta\psi_\theta-2\psi\right)\\
&=\partial_r\left(s(\theta)\frac{\psi_t^2}{2r^2}\right)+s(\theta)\frac{\psi_t^2}{r^3}+\frac{1}{2}s(\theta)\partial_r(r^2\psi_r^2)-s(\theta)r\psi_r^2\\
&\quad+\partial_\theta\left(s(\theta)\psi_\theta\psi_r\right)-\partial_r\left(s(\theta)\frac{\psi_\theta^2}{2}\right)-\partial_r\left(\sin^{-3}\theta\psi^2\right).
\end{aligned} ∂ t ( r 2 1 s ( θ ) ψ r ψ t ) = r 2 1 s ( θ ) ψ t ψ t + r 2 1 s ( θ ) ψ r ψ tt = ∂ r ( s ( θ ) 2 r 2 ψ t 2 ) + s ( θ ) r 3 ψ t 2 + s ( θ ) ψ r ( r 2 ψ r r + ψ θ θ − 3 cot θ ψ θ − 2 ψ ) = ∂ r ( s ( θ ) 2 r 2 ψ t 2 ) + s ( θ ) r 3 ψ t 2 + 2 1 s ( θ ) ∂ r ( r 2 ψ r 2 ) − s ( θ ) r ψ r 2 + ∂ θ ( s ( θ ) ψ θ ψ r ) − ∂ r ( s ( θ ) 2 ψ θ 2 ) − ∂ r ( sin − 3 θ ψ 2 ) .
This proves (3.64).
Proof of (3.65). Simply compute and use (90) ; one gets
∂ t ( 1 r 3 s ( θ ) ψ ψ t ) = 1 r 3 s ( θ ) ψ t 2 + 1 r 3 s ( θ ) ψ ψ t t = 1 r 3 s ( θ ) ψ t 2 + s ( θ ) ψ ( r ψ r r + 1 r ψ θ θ − 3 r cot θ ψ θ − 2 r ψ ) = s ( θ ) ψ t 2 r 3 + s ( θ ) ( ∂ r ( r ψ ψ r ) − ψ ψ r − r ψ r 2 ) + ∂ θ ( 1 r s ( θ ) ψ ψ θ ) − 1 r s ( θ ) ψ θ 2 − 2 r s ( θ ) ψ 2 . \begin{aligned}
\partial_t\left(\frac{1}{r^3}s(\theta)\psi\psi_t\right)
&=\frac{1}{r^3}s(\theta)\psi_t^2+\frac{1}{r^3}s(\theta)\psi\psi_{tt}\\
&=\frac{1}{r^3}s(\theta)\psi_t^2+s(\theta)\psi\left(r\psi_{rr}+\frac{1}{r}\psi_{\theta\theta}-\frac{3}{r}\cot\theta\psi_\theta-\frac{2}{r}\psi\right)\\
&=s(\theta)\frac{\psi_t^2}{r^3}+s(\theta)\left(\partial_r(r\psi\psi_r)-\psi\psi_r-r\psi_r^2\right)\\
&\quad+\partial_\theta\left(\frac{1}{r}s(\theta)\psi\psi_\theta\right)-\frac{1}{r}s(\theta)\psi_\theta^2-\frac{2}{r}s(\theta)\psi^2.
\end{aligned} ∂ t ( r 3 1 s ( θ ) ψ ψ t ) = r 3 1 s ( θ ) ψ t 2 + r 3 1 s ( θ ) ψ ψ tt = r 3 1 s ( θ ) ψ t 2 + s ( θ ) ψ ( r ψ r r + r 1 ψ θ θ − r 3 cot θ ψ θ − r 2 ψ ) = s ( θ ) r 3 ψ t 2 + s ( θ ) ( ∂ r ( r ψ ψ r ) − ψ ψ r − r ψ r 2 ) + ∂ θ ( r 1 s ( θ ) ψ ψ θ ) − r 1 s ( θ ) ψ θ 2 − r 2 s ( θ ) ψ 2 .
Proof of (3.66). First of all, from (3.64) and (3.65),
d d t ∫ s ( θ ) ( 1 r 2 ψ r ψ t − 1 r 3 ψ ψ t ) = ∫ s ( θ ) ( ( 1 2 + 1 r ) ψ θ 2 + 2 r ψ 2 ) . \frac{d}{dt}\int s(\theta)\left(\frac{1}{r^2}\psi_r\psi_t-\frac{1}{r^3}\psi\psi_t\right)=\int s(\theta)\left(\left(\frac{1}{2}+\frac{1}{r}\right)\psi_\theta^2+\frac{2}{r}\psi^2\right). d t d ∫ s ( θ ) ( r 2 1 ψ r ψ t − r 3 1 ψ ψ t ) = ∫ s ( θ ) ( ( 2 1 + r 1 ) ψ θ 2 + r 2 ψ 2 ) .
Notice that
∣ ∫ s ( θ ) ( 1 r 2 ψ r ψ t ± 1 r 3 ψ ψ t ) ∣ ≲ ∫ s ( θ ) ( ψ r 2 + ψ t 2 r 4 + ψ 2 r 2 ) ≲ E . \left|\int s(\theta)\left(\frac{1}{r^2}\psi_r\psi_t\pm\frac{1}{r^3}\psi\psi_t\right)\right|\lesssim\int s(\theta)\left(\psi_r^2+\frac{\psi_t^2}{r^4}+\frac{\psi^2}{r^2}\right)\lesssim E. ∫ s ( θ ) ( r 2 1 ψ r ψ t ± r 3 1 ψ ψ t ) ≲ ∫ s ( θ ) ( ψ r 2 + r 4 ψ t 2 + r 2 ψ 2 ) ≲ E .
This proves that
∫ 0 t ∫ r , θ s ( θ ) r ( ψ θ 2 + 2 ψ 2 ) ≤ C ( E ) . (131) \int_0^t\int_{r,\theta}\frac{s(\theta)}{r}\left(\psi_\theta^2+2\psi^2\right)\le C(E). \tag*{(131)} ∫ 0 t ∫ r , θ r s ( θ ) ( ψ θ 2 + 2 ψ 2 ) ≤ C ( E ) . (131)
To get (3.66), let m = m ( r ) m=m(r) m = m ( r ) be a bounded smooth function, with m ( 0 ) = 0 m(0)=0 m ( 0 ) = 0 . Again using (3.64) and (3.65),
d d t ∫ m ( r ) s ( θ ) ( ψ r ψ t r 2 − ψ ψ t r 3 ) = − 1 2 ∫ m ′ ( r ) s ( θ ) ( ψ t 2 r 2 + r 2 ψ r 2 ) + 1 2 ∫ ( m ′ ( r ) + 2 m ( r ) r ) s ( θ ) ψ θ 2 + ∫ ( 2 m ( r ) r − 1 2 m ′ ′ ( r ) r ) s ( θ ) ψ 2 . (132) \frac{d}{dt}\int m(r)s(\theta)\left(\frac{\psi_r\psi_t}{r^2}-\frac{\psi\psi_t}{r^3}\right)=-\frac{1}{2}\int m'(r)s(\theta)\left(\frac{\psi_t^2}{r^2}+r^2\psi_r^2\right)
+\frac{1}{2}\int\left(m'(r)+\frac{2m(r)}{r}\right)s(\theta)\psi_\theta^2
+\int\left(\frac{2m(r)}{r}-\frac{1}{2}m''(r)r\right)s(\theta)\psi^2. \tag*{(132)} d t d ∫ m ( r ) s ( θ ) ( r 2 ψ r ψ t − r 3 ψ ψ t ) = − 2 1 ∫ m ′ ( r ) s ( θ ) ( r 2 ψ t 2 + r 2 ψ r 2 ) + 2 1 ∫ ( m ′ ( r ) + r 2 m ( r ) ) s ( θ ) ψ θ 2 + ∫ ( r 2 m ( r ) − 2 1 m ′′ ( r ) r ) s ( θ ) ψ 2 . (132)
Taking m ( r ) = arctan r m(r)=\arctan r m ( r ) = arctan r , we get ∣ m ′ ( r ) + 2 m ( r ) r ∣ ≲ 1 + 1 r \left|m'(r)+\frac{2m(r)}{r}\right|\lesssim1+\frac{1}{r} m ′ ( r ) + r 2 m ( r ) ≲ 1 + r 1 , and ∣ 2 m ( r ) r − 1 2 m ′ ′ ( r ) r ∣ ≲ 1 r \left|\frac{2m(r)}{r}-\frac{1}{2}m''(r)r\right|\lesssim\frac{1}{r} r 2 m ( r ) − 2 1 m ′′ ( r ) r ≲ r 1 , and integrating in (B.2) and using (B.1),
∫ 0 t ∫ s ( θ ) 1 + r 2 ( ψ t 2 r 2 + r 2 ψ r 2 ) ≲ E + ∫ 0 t ∫ s ( θ ) ( ( 1 + 1 r ) ψ θ 2 + 1 r ψ 2 ) ≲ C ( E ) . \int_0^t\int\frac{s(\theta)}{1+r^2}\left(\frac{\psi_t^2}{r^2}+r^2\psi_r^2\right)\lesssim E+\int_0^t\int s(\theta)\left(\left(1+\frac{1}{r}\right)\psi_\theta^2+\frac{1}{r}\psi^2\right)\lesssim C(E). ∫ 0 t ∫ 1 + r 2 s ( θ ) ( r 2 ψ t 2 + r 2 ψ r 2 ) ≲ E + ∫ 0 t ∫ s ( θ ) ( ( 1 + r 1 ) ψ θ 2 + r 1 ψ 2 ) ≲ C ( E ) .
This proves (3.66). Consequently, there exists an increasing sequence of time t n → + ∞ t_n\to+\infty t n → + ∞ such that
lim n → + ∞ ∫ r , θ s ( θ ) 1 + r 2 ( ( ψ θ 2 + 2 ψ 2 ) + r 2 ψ r 2 + ψ t 2 r 2 ) ( t n ) \lim_{n\to+\infty}\int_{r,\theta}\frac{s(\theta)}{1+r^2}\left((\psi_\theta^2+2\psi^2)+r^2\psi_r^2+\frac{\psi_t^2}{r^2}\right)(t_n) n → + ∞ lim ∫ r , θ 1 + r 2 s ( θ ) ( ( ψ θ 2 + 2 ψ 2 ) + r 2 ψ r 2 + r 2 ψ t 2 ) ( t n )
≤ 2 lim n → + ∞ ∫ r , θ s ( θ ) ( ( 1 + 1 r ) ψ θ 2 + 1 r ψ 2 + 1 1 + r 2 ( r 2 ψ r 2 + ψ t 2 r 2 ) ) ( t n ) = 0. \leq2\lim_{n\to+\infty}\int_{r,\theta}s(\theta)\left(\left(1+\frac{1}{r}\right)\psi_\theta^2+\frac{1}{r}\psi^2+\frac{1}{1+r^2}\left(r^2\psi_r^2+\frac{\psi_t^2}{r^2}\right)\right)(t_n)=0. ≤ 2 n → + ∞ lim ∫ r , θ s ( θ ) ( ( 1 + r 1 ) ψ θ 2 + r 1 ψ 2 + 1 + r 2 1 ( r 2 ψ r 2 + r 2 ψ t 2 ) ) ( t n ) = 0.
From (3.63), we get
∣ ∫ r 2 sin − 3 θ 1 + r 2 ( ψ t 2 r 4 + 2 ψ 2 r 2 + ψ r 2 + ψ θ 2 r 2 ) ( t ) − ∫ r 2 sin − 3 θ 1 + r 2 ( ψ t 2 r 4 + 2 ψ 2 r 2 + ψ r 2 + ψ θ 2 r 2 ) ( t n ) ∣ ≤ ∫ t t n r ( 1 + r 2 ) 2 ∫ sin − 3 θ ∣ ψ r ψ t ∣ ≲ ∫ t t n ∫ sin − 3 θ 1 + r 2 ( ψ t 2 r 2 + r 2 ψ r 2 ) ≤ C ( E ) . \begin{aligned}
\left|\int\frac{r^2\sin^{-3}\theta}{1+r^2}\left(\frac{\psi_t^2}{r^4}+\frac{2\psi^2}{r^2}+\psi_r^2+\frac{\psi_\theta^2}{r^2}\right)(t)-\int\frac{r^2\sin^{-3}\theta}{1+r^2}\left(\frac{\psi_t^2}{r^4}+\frac{2\psi^2}{r^2}+\psi_r^2+\frac{\psi_\theta^2}{r^2}\right)(t_n)\right| \\
&\leq\int_t^{t_n}\frac{r}{(1+r^2)^2}\int\sin^{-3}\theta|\psi_r\psi_t|\lesssim\int_t^{t_n}\int\frac{\sin^{-3}\theta}{1+r^2}\left(\frac{\psi_t^2}{r^2}+r^2\psi_r^2\right)\leq C(E).
\end{aligned} ∫ 1 + r 2 r 2 sin − 3 θ ( r 4 ψ t 2 + r 2 2 ψ 2 + ψ r 2 + r 2 ψ θ 2 ) ( t ) − ∫ 1 + r 2 r 2 sin − 3 θ ( r 4 ψ t 2 + r 2 2 ψ 2 + ψ r 2 + r 2 ψ θ 2 ) ( t n ) ≤ ∫ t t n ( 1 + r 2 ) 2 r ∫ sin − 3 θ ∣ ψ r ψ t ∣ ≲ ∫ t t n ∫ 1 + r 2 sin − 3 θ ( r 2 ψ t 2 + r 2 ψ r 2 ) ≤ C ( E ) .
Sending n → + ∞ n\to+\infty n → + ∞ and then t → + ∞ t\to+\infty t → + ∞ , we get the desired result. This proves (3.67).
The twisted fibration over AdS2 and its conformal diagrams
In the near-horizon extreme Kerr geometry (2.4), the ( T , R ) (T,R) ( T , R ) block is Poincaré AdS2 _2 2 with R = 1 / Z R=1/Z R = 1/ Z , and the full geometry is a twisted U(1) fibration over it, warped over θ \theta θ . Because the fiber block d θ 2 + Λ 2 ( d Φ + R d T ) 2 d\theta^2+\Lambda^2(d\Phi+R\,dT)^2 d θ 2 + Λ 2 ( d Φ + R d T ) 2 is positive definite, the projection of any causal curve onto the ( T , R ) (T,R) ( T , R ) base is causal in the AdS2 _2 2 metric, and any causal base curve lifts to a causal NHEK curve (take θ ˙ = 0 \dot{\theta}=0 θ ˙ = 0 and the co-rotating lift Φ ˙ = − R T ˙ \dot{\Phi}=-R\dot{T} Φ ˙ = − R T ˙ ). The causal relation between fibers of NHEK is therefore exactly the causal relation between points of AdS2 _2 2 , so the conformal diagram of the base organizes the throat, each of its points representing a squashed, twisted ( θ , Φ ) (\theta,\Phi) ( θ , Φ ) sphere. This is what allows the two-dimensional diagrams used in § 4.1 to describe a warped four-dimensional geometry.
The construction of those diagrams is standard [11 , 53 ] . In the Poincaré chart d s 2 = Z − 2 ( − d T 2 + d Z 2 ) ds^2=Z^{-2}(-dT^2+dZ^2) d s 2 = Z − 2 ( − d T 2 + d Z 2 ) , with T ∈ R T\in\mathbb{R} T ∈ R and Z ∈ ( 0 , ∞ ) Z\in(0,\infty) Z ∈ ( 0 , ∞ ) , the null coordinates u = T − Z u=T-Z u = T − Z , v = T + Z v=T+Z v = T + Z and the compactification U = arctan u U=\arctan u U = arctan u , V = arctan v V=\arctan v V = arctan v bring the metric to the global form
d s 2 = 1 sin 2 σ ~ ( − d τ 2 + d σ ~ 2 ) , τ = U + V , σ ~ = V − U , (133) ds^2=\frac{1}{\sin^2\widetilde{\sigma}}\left(-d\tau^2+d\widetilde{\sigma}^2\right),\qquad\tau=U+V,\qquad\widetilde{\sigma}=V-U, \tag*{(133)} d s 2 = sin 2 σ 1 ( − d τ 2 + d σ 2 ) , τ = U + V , σ = V − U , (133)
while the patch condition v − u = 2 Z > 0 v-u=2Z>0 v − u = 2 Z > 0 becomes the triangle σ ~ ∈ ( 0 , π ) \widetilde{\sigma}\in(0,\pi) σ ∈ ( 0 , π ) , ∣ τ ∣ < π − σ ~ |\tau|<\pi-\widetilde{\sigma} ∣ τ ∣ < π − σ inside the global strip. Its vertical base is a segment of one conformal boundary, carrying all of Poincaré time through T = tan ( τ / 2 ) T=\tan(\tau/2) T = tan ( τ /2 ) ; its two null edges are the past and future Poincaré horizons, which are the Killing horizons of ∂ T \partial_T ∂ T ; and its apex is the single point ( τ , σ ~ ) = ( 0 , π ) (\tau,\widetilde{\sigma})=(0,\pi) ( τ , σ ) = ( 0 , π ) at which the patch touches the second boundary. The two base corners are i ± i^\pm i ± and the apex is i 0 i^0 i 0 .
The curves relevant to the charges of § 4.1 are the worldlines of constant Z Z Z , equivalently of constant R R R , which are the orbits of ∂ T \partial_T ∂ T . Writing
τ = arctan ( T + Z ) + arctan ( T − Z ) , σ ~ = arctan ( T + Z ) − arctan ( T − Z ) , (134) \tau=\arctan(T+Z)+\arctan(T-Z),\qquad\widetilde{\sigma}=\arctan(T+Z)-\arctan(T-Z), \tag*{(134)} τ = arctan ( T + Z ) + arctan ( T − Z ) , σ = arctan ( T + Z ) − arctan ( T − Z ) , (134)
every such curve runs from the bottom corner of the triangle at T → − ∞ T\to-\infty T → − ∞ to the top corner at T → + ∞ T\to+\infty T → + ∞ , whatever the value of Z Z Z , and reaches its maximum depth σ ~ max = 2 arctan Z \widetilde{\sigma}_{\max}=2\arctan Z σ m a x = 2 arctan Z at τ = 0 \tau=0 τ = 0 (Figure 1 ). Small Z Z Z hugs the boundary, Z = 1 Z=1 Z = 1 reaches the middle of the strip, and as Z → ∞ Z\to\infty Z → ∞ the curve degenerates onto the pair of null horizons. This is the sense in which the late-time cut used in § 4.1 is a corner of the diagram.
From NHEK to near-NHEK
To describe near-NHEK in the same terms we use the coordinate transformation
T = e κ t ρ κ ρ 2 − κ 2 , R = e − κ t ρ 2 − κ 2 , Φ = ϕ + 1 2 ln ρ − κ ρ + κ , (135) T=\frac{e^{\kappa t}\rho}{\kappa\sqrt{\rho^2-\kappa^2}},\qquad R=e^{-\kappa t}\sqrt{\rho^2-\kappa^2},\qquad\Phi=\phi+\frac{1}{2}\ln\frac{\rho-\kappa}{\rho+\kappa}, \tag*{(135)} T = κ ρ 2 − κ 2 e κ t ρ , R = e − κ t ρ 2 − κ 2 , Φ = ϕ + 2 1 ln ρ + κ ρ − κ , (135)
whose inverse reads
T ± Z = e κ t κ ρ ± κ ρ ∓ κ , (136) T\pm Z=\frac{e^{\kappa t}}{\kappa}\sqrt{\frac{\rho\pm\kappa}{\rho\mp\kappa}}, \tag*{(136)} T ± Z = κ e κ t ρ ∓ κ ρ ± κ , (136)
and which covers the sub-wedge 0 < u < v 0<u<v 0 < u < v of the Poincaré patch, with horizon at ρ = κ \rho=\kappa ρ = κ . Applied to (4) , the full map ( T , R , θ , Φ ) → ( t , ρ , θ , ϕ ) (T,R,\theta,\Phi)\to(t,\rho,\theta,\phi) ( T , R , θ , Φ ) → ( t , ρ , θ , ϕ ) yields
d s 2 = Λ ( θ ) [ − ( ρ 2 − κ 2 ) d t 2 + d ρ 2 ρ 2 − κ 2 + d θ 2 ] + Γ ( θ ) [ d ϕ + ρ d t ] 2 , (137) ds^2=\Lambda(\theta)\left[-(\rho^2-\kappa^2)\,dt^2+\frac{d\rho^2}{\rho^2-\kappa^2}+d\theta^2\right]+\Gamma(\theta)[d\phi+\rho\,dt]^2, \tag*{(137)} d s 2 = Λ ( θ ) [ − ( ρ 2 − κ 2 ) d t 2 + ρ 2 − κ 2 d ρ 2 + d θ 2 ] + Γ ( θ ) [ d ϕ + ρ d t ] 2 , (137)
which is (2.11) after the change r = 1 2 ( ρ + ρ 2 − κ 2 ) r = \frac{1}{2}\left(\rho+ \sqrt{\rho^{2}-\kappa^{2}}\right) r = 2 1 ( ρ + ρ 2 − κ 2 ) , equivalently ρ = r + κ 2 / 4 r \rho= r+\kappa^{2}/4r ρ = r + κ 2 /4 r . This is the same intermediate coordinate used in § 3.5 , and the horizon ρ = κ \rho= \kappa ρ = κ corresponds to r = κ / 2 r = \kappa/2 r = κ /2 , the radius identified in § 2.1 .
With the tortoise coordinate ρ ∗ = 1 2 κ ln ρ − κ ρ + κ \rho_* = \frac{1}{2\kappa}\ln\frac{\rho-\kappa}{\rho+\kappa} ρ ∗ = 2 κ 1 ln ρ + κ ρ − κ , which tends to − ∞ -\infty − ∞ as ρ → κ \rho\to\kappa ρ → κ , the null coordinates are
u = 1 κ e κ ( t + ρ ∗ ) , v = 1 κ e κ ( t − ρ ∗ ) . (138) u = \frac{1}{\kappa}e^{\kappa(t+\rho_*)}, \qquad v = \frac{1}{\kappa}e^{\kappa(t-\rho_*)}. \tag*{(138)} u = κ 1 e κ ( t + ρ ∗ ) , v = κ 1 e κ ( t − ρ ∗ ) . (138)
Sending ρ → κ \rho\to\kappa ρ → κ at fixed t t t gives u → 0 u \to0 u → 0 and v → ∞ v \to\infty v → ∞ , so that the horizon at finite time is the single point
( U , V ) = ( 0 , π 2 ) , i.e. ( τ , σ ~ ) = ( π 2 , π 2 ) , (139) (U,V) = \left(0,\frac{\pi}{2}\right), \qquad\text{i.e.} \qquad(\tau,\widetilde{\sigma}) = \left(\frac{\pi}{2},\frac{\pi}{2}\right), \tag*{(139)} ( U , V ) = ( 0 , 2 π ) , i.e. ( τ , σ ) = ( 2 π , 2 π ) , (139)
exactly where the bifurcation point sits in Figure 2 ; the two full horizon branches are reached only as t → ± ∞ t \to\pm\infty t → ± ∞ along the edges. At fixed ρ \rho ρ both u u u and v v v are proportional to e κ t e^{\kappa t} e κ t , so that v = ρ + κ ρ − κ u v = \frac{\rho+\kappa}{\rho-\kappa}u v = ρ − κ ρ + κ u is a straight ray through the origin of the ( u , v ) (u,v) ( u , v ) plane, whose image is the curve tan V = c tan U \tan V = c\tan U tan V = c tan U with c = ρ + κ ρ − κ c = \frac{\rho+\kappa}{\rho-\kappa} c = ρ − κ ρ + κ , drawn dashed in Figure 2 for c = 3 c = 3 c = 3 , that is ρ = 2 κ \rho= 2\kappa ρ = 2 κ . Using (C.2) one checks that t → ∞ t \to\infty t → ∞ at fixed ρ \rho ρ approaches σ ~ = 0 \widetilde{\sigma} = 0 σ = 0 and τ = π \tau= \pi τ = π .