Conformal Field Theory Interpretation of Black Hole Quasi-normal Modes

Danny Birmingham ∗ , Ivo Sachs † and Sergey N. Solodukhin †

∗
Department of Mathematical Physics, University College Dublin, Belfield, Dublin 4, Ireland.
†
Theoretische Physik, Ludwig-Maximilians Universität, Theresienstrasse 37, D-80333, München, Germany.
We obtain exact expressions for the quasi-normal modes of various spin for the BTZ black hole.
These modes determine the relaxation time of black hole perturbations. Exact agreement is
found between the quasi-normal frequencies and the location of the poles of the retarded cor-
relation function of the corresponding perturbations in the dual conformal field theory. This
then provides a new quantitative test of the AdS/CFT correspondence.

PACS: 04.70Dy, [Link], [Link]

arXiv:hep-th/0112055v2 15 Feb 2002

The problem of how a perturbed thermodynamical times have been presented in [6]- [11]. For related
system returns to equilibrium is an important issue in discussions in the context of black hole formation see
statistical mechanics and finite temperature field the- [12]. Qualitative agreement was found with the re-
ory [1]. For a small perturbation, this process is de- sults expected from the conformal field theory side.
scribed by linear response theory [1,2]. The relaxation However, a quantitative test of such a correspondence
process is then completely determined by the poles, between quasi-normal modes and the linear response
in the momentum representation, of the retarded cor- of the dual conformal field theory is lacking so far.
relation function of the perturbation. On the other In this paper, we consider the 2 + 1 dimensional AdS
hand, black holes also constitute a thermodynamical black hole [13], and show that there is a precise quanti-
system. At equilibrium, the various thermodynami- tative agreement between its quasi-normal frequencies
cal quantities, such as the temperature and the en- and the location of the poles of the retarded corre-
tropy, are determined in terms of the mass, charge lation function describing the linear response on the
and angular momentum of the black hole. The decay conformal field theory side. Both computations are
of small perturbations of a black hole at equilibrium performed analytically. As a result, we can identify
are described by the so-called quasi-normal modes [3]. not just the lowest, but the complete (infinite) set of
For asymptotically flat black hole space-times, quasi- frequencies on both sides of the AdS/CFT correspon-
normal modes are analysed by solving the wave equa- dence. In spite of its simplicity, this model plays an
tion for matter or gravitational perturbations, subject important role also for black holes in higher dimen-
to the conditions that the flux at the horizon is in- sions whose near-horizon geometry is AdS2+1 (see e.g.
going, with out-going flux at asymptotic infinity. The [14] for a review and references).
wave equation, subject to these boundary conditions,
The metric of the BTZ black hole is given by
admits only a discrete set of solutions with complex
frequencies. The imaginary part of these quasi-normal 2
ds2 = − sinh2 µ (r+ dt − r− dφ) + dµ2
frequencies then determine the decay time of small 2
perturbations, or equivalently, the relaxation of the + cosh2 µ (−r− dt + r+ dφ) . (1)
system back to thermal equilibrium.
On another front, over the last few years increasing The angular coordinate φ has period 2π, and the radii
evidence has accumulated that there is a correspon- of the inner and outer horizons are denoted by r+
dence between gravity and quantum field theory in and r− , respectively. We have also set to unity the
flat space-time (for a review, see [4]). In particular, radius of the anti-de Sitter space, ℓ ≡ 1. The dual
this duality has led to important progress in our un- conformal field theory on the boundary is 1 + 1 di-
derstanding of the microscopic physics of a class of mensional, the conformal symmetry being generated
near-extremal black holes. The purpose of this let- by two copies of the Virasoro algebra acting separately
ter is to analyse whether such a correspondence exists on left- and right-moving sectors. Consequently, the
between quasi-normal modes in anti-de Sitter (AdS) conformal field theory splits into two independent sec-
black holes and linear response theory in scale invari- tors at thermal equilibrium with temperatures
ant finite temperature field theory. A correspondence TL = (r+ − r− )/2π , TR = (r+ + r− )/2π . (2)
between quasi-normal modes and the decay of per-
turbations in the dual conformal field theory (CFT)
was first suggested in [5]. The analysis of [5] is based According to the AdS3 /CFT2 correspondence, to each
on the numerical computation of quasi-normal modes field of spin s propagating in AdS3 there corresponds
for AdS-Schwarzschild black holes in four, five, and an operator O in the dual conformal field theory char-
seven dimensions. Further numerical computations acterised by conformal weights (hL , hR ) with [4]
of quasi-normal modes in asymptotically AdS space-
hR + hL = ∆ , hR − hL = ±s , (3)

1
and ∆ is determined in terms of the mass m of the lying in the lower half-plane are the same as the poles
field. In particular, we have of the retarded correlation function (6). Restricting
p the poles of (7) to the lower half-plane, we find two
∆ = 1 + 1 + m2 , (4) sets of poles
ωL = k − 4πiTL(n + hL ) ,
for scalar fields, and
ωR = −k − 4πiTR (n + hR ) . (8)
∆ = 1 + |m| , (5)
Here, and in the following, n takes the integer values
for both fermionic and vector fields. For a small per- (n = 0, 1, 2, ...). This set of poles characterises the
turbation, the manner in which the field theory re- decay of the perturbation on the CFT side, and co-
laxes back to thermal equilibrium can then be anal- incides precisely with the quasi-normal frequencies of
ysed within linear response theory [2]. One expects the BTZ black hole, as we shall now show for fields of
that at late times the perturbed system will approach various spin.
equilibrium exponentially with a characteristic time- Scalar Perturbation (s = 0). Scalar perturbations are
scale. This time-scale is inversely proportional to the described by the wave equation
imaginary part of the poles, in momentum space, of
∇2 − m2 Φ = 0 .


the correlation function of the perturbation operator (9)
O. In the present case, according to our proposal, the
relevant correlation function is the retarded real time We use the ansatz
correlation function +
Φ = e−i(k+ x +k− x− )
R(µ) , (10)
Dret (x, x′ ) = iθ(t − t′ ) h[O(x), O(x′ )]iT
= iθ(t − t′ )D̄(x, x′ ) , (6) where x+ = r+ t − r− φ, x− = r+ φ − r− t, and
(k+ + k− )(r+ − r− ) = ω − k ,
where D̄(x, x′ ) = D+ (x, x′ ) − D− (x, x′ ) is the commu-
tator evaluated in the equilibrium canonical ensemble. (k+ − k− )(r+ + r− ) = ω + k . (11)
For a conformal field theory at zero temperature, the
2-point correlation functions can be determined, up to Here, ω and k are the energy and angular momen-
a normalisation, from conformal invariance. At finite tum of the perturbation. Changing variables to z =
temperature T , one has to take into account the infi- tanh2 µ, we end up with the hypergeometric equation
nite sum over images to render the correlation func-
tion periodic in imaginary time, with period 1/T . The d2 R dR
z(1 − z) 2
+ (1 − z)
result of this summation in two dimensions has been dz dz
 2 2
k+ k− m2

determined in [15], and depends only on the confor-
+ − − R = 0 . (12)
mal dimensions (hL , hR ) of the perturbation operator. 4z 4 4(1 − z)
We have (x± = t ± σ),
The solution which is in-going at the horizon is given
(πTR )2hR (πTL )2hL by
D+ (x) =
sinh2hR (πTR x− −iǫ) sinh2hL (πTL x+ −iǫ)
R(z) = z α (1 − z)βs F (as , bs , cs , z) , (13)
and a similar expression for D− (x) with ǫ → −ǫ. In √
where α = − ik2+ , βs = 1


order to determine the location of the poles, we need 2 1− 1 + m2 , and
to compute the Fourier transform of (6). This is com-
plicated by the presence of the θ function. We can, (k+ − k− ) (k+ + k− )
as = + βs , bs = + βs ,
however determine the location of the poles indirectly. 2i 2i
For this we first consider the Fourier transform of the cs = 1 + 2α . (14)
commutator D̄(x). This integral can be evaluated us-
ing contour techniques, leading to [16] The quasi-normal modes for the scalar perturbations
    were found in [17] by imposing the vanishing Dirichlet
p+ p− condition at infinity. Here, we re-evaluate these modes
D̄(k+ , k− ) ∝ Γ hL + i Γ hR + i ×
2πTL 2πTR using the condition that the flux given by
   
p+ p− √ 1
Γ hL − i Γ hR − i , (7) F= g (R∗ ∂µ R − R∂µ R∗ ) (15)
2πTL 2πTR 2i

where p± = 12 (ω ∓ k). This function has poles in both vanishes at asymptotic infinity. For m2 > 0, the
the upper and lower half of the ω-plane. The poles asymptotic flux has a set of divergent terms, with the

2
leading term of order (1 − z)2βs . Each of these terms where α = − ik2+ , βf = − 12 (m + 21 ), cf = 1
2 + 2α, and
is proportional to [18]
k+ − k− 1 k+ + k−
af = + βf + , bf = + βf . (22)
 Γ(cs )Γ(cs − as − bs ) 2
 
2i 2 2i
 Γ(cs − as )Γ(cs − bs )  . (16)
 
The asymptotic form (z → 1) of ψ1 and ψ2 can now
Thus, the asymptotic flux vanishes if cs − as = −n, or be determined explicitly. In analogy with the scalar
cs − bs = −n, i.e. perturbations, we then impose the condition that the
flux [19]
i 1 p 
√
(k+ ± k− ) = n + 1 + 1 + m2 . (17) F = g Ψ̄γµ Ψ ≃ (1 − z)−1 (|ψ1 |2 − |ψ2 |2 ) (23)
2 2

Using (11), one sees that these are the quasi-normal vanishes at infinity. The resulting quasi-normal modes
modes found in [17].√For the
scalar bulk field, we have are then obtained as follows. For m > 0, the lead-
hL = hR = 21 1 + 1 + m2 . Thus, we observe that ing divergent term in the asymptotic flux is of order
(17) exactly reproduces (8). (1 − z)2β+1 . Vanishing flux then requires that the
coefficient
In AdS space-time, a negative mass squared for a
scalar field is consistent, as long as −1 < m2 < 0. Γ(cf )Γ(cf − af − bf )
, (24)
A detailed analysis shows that in this case there is Γ(cf − af )Γ(cf − bf )
a second set of modes with √ as = −n, or bs = −n,
that is hL = hR = 21 1 − 1 + m2 . This is in fact


vanishes, that is,
expected as for −1 < m2 < 0 there √ are two sets of
dual√ operators with ∆+ = 1 + 1 + m2 and ∆− = i 1 m
(k+ + k− ) = n + + or
1− 1 + m2 , respectively [4]. The second set of quasi- 2 4 2
normal frequencies in this range then matches exactly i 3 m
(k+ − k− ) = n + + . (25)
the dual operators with ∆ = ∆− . We note in passing 2 4 2
that the Dirichlet boundary condition suggested in [5]
leads to the same quasi-normal modes for m2 > 0 but The above conditions imply that all coefficients of
does not lead to any quasi-normal modes for m2 < 0. the sub-leading, asymptotically non-vanishing, con-
Fermion Perturbation (s = 1/2). In [9], the quasi- tributions to the flux also vanish. Thus, we have
normal fermionic perturbation in the BTZ background precise agreement with (8), where the left and right
has been analysed numerically. However, as expected, conformal weights are given by hL = 41 + 21 m, and
it is possible to find analytic solutions in this simple hR = 43 + 12 m. For m < 0, one obtains a similar
case [19]. We begin with the Dirac equation result with hL = 34 − 12 m, and hR = 14 − 21 m. Note
  again that imposing Dirichlet boundary conditions for
−i(k+ x+ +k− x− ) ψ1 ψ1 and ψ2 at infinity would lead to the absence of
(D/ + m)Ψ = 0 , Ψ = e . (18)
ψ2 quasi-normal modes for −1 < m < 1. We can think
of no physical reason for the absence of quasi-normal
modes in this range of masses. Thus, we take this
Following [19], we make the substitutions
as another motivation for imposing vanishing flux at
s infinity, rather that Dirichlet conditions for asymp-
cosh µ ± sinh µ ′ totically AdS space-times. One should also note that
ψ1 ± ψ2 = (ψ1 ± ψ2′ ) , (19)
cosh µ sinh µ for positive mass the spinor perturbation is asymp-
totically left-handed, whereas it is right-handed for
to obtain negative mass (see also [4]).
Vector Perturbation (s = 1). The massive Maxwell
2(1 − z)z 1/2 ∂z ψ1′ + i(k+ z −1/2 + k− z 1/2 )ψ1′ = field in AdS3 is described by the first order equation
 
1 ′
− i(k+ + k− ) + m + ψ , (20)
2 2 ǫλαβ ∂α Aβ = −mAλ . (26)

and a similar equation where ψ1′ , ψ2′ , and k± and −k± Let
are interchanged. The solutions of these equations +

with in-going flux at the horizon are given by Ai = e−i(k+ x +k− x− )

Ai (µ) , (27)

ψ1′ = z α (1 − z)βf F (af , bf , cf , z) , (21) where A1,2 = A+ ±A− . Then, after changing variables
a f − c f as before, we recover the scalar equation (12) for A1
ψ2′ = ( )z α+1/2 (1 − z)βf F (af , bf + 1, cf + 1, z),
cf and A2 [19], where the scalar mass squared is replaced

3
by m2 + 2εi m, with ε1 = −ε2 = 1. As in the scalar relaxation time for the decay of the black hole pertur-
case, the solutions with in-going flux at the horizon bation is determined by the imaginary part of the low-
are then given by est quasi-normal mode. Our analysis thus establishes
a direct relation between this relaxation time and the
A1 = e1 z α (1 − z)βv +1 F (av + 1, bv + 1, cv , z) , time-scale for return to equilibrium of the dual con-
A2 = e2 z α (1 − z)βv F (av , bv , cv , z) , (28) formal field theory. This result also provides a new
quantitative test of the AdS/CFT correspondence.
where α = − ik2+ , βv = m
2, cv = 1 + 2α, and We would like to thank T.S. Evans, R. Myers, A.
(k+ − k− ) (k+ + k− ) Peet and H. Wagner for helpful discussions. D.B.
av = + βv , bv = + βv . (29) would like to thank the CERN Theory Division for
2i 2i
hospitality during the initial stages of this work. S.S.
Note that the two “scalar modes” in (27) are not in- would like to thank the Aspen Center for Physics for
dependent. The first order equation (26) relates the hospitality while this work was in progress. The work
two coefficients e1 and e2 by of D.B. and I.S. was partially supported by Enterprise
Ireland grant IC/2001/004. S.S. is supported by the
e2 i(k+ − k− ) + m cv − b v − 1 grant DFG-SPP 1096, Stringtheorie.
= = . (30)
e1 i(k+ + k− ) − m bv

The remaining component Aµ is related to A± by

1
Aµ = ∂[+ A−] . (31)
m cosh µ sinh µ
[1] R. Kubo, M. Toda and N. Hashitsume, “Statistical
For a real vector field, the particle flux is not defined. Physics II”, Springer Verlag, Berlin (1985).
One way to avoid this difficulty is to consider a com- [2] A.L. Fetter and J.D. Walecka, “Quantum Theory
plex vector field. Alternatively, one can consider the of Many-particle Systems,” McGraw-Hill Book Com-
energy flux divided by the red-shifted frequency [16]. pany (1971).
Both approaches lead to the same conditions, namely [3] See e.g. V.P. Frolov and I.D. Novikov, “Black Hole
that A1 and A2 vanish at infinity. Thus, we impose Physics: Basic Concepts And New Developments,”
Dirichlet boundary condition for A1,2 . Using (30), one Dordrecht, Netherlands: Kluwer Academic (1998).
finds the leading asymptotic behaviour (z → 1) of the [4] [Link], S.S. Gubser, J. Maldacena, H. Ooguri and
solutions (28) for positive m is given by Y. Oz, Phys. Rep. 323, 183 (2000).
[5] G.T. Horowitz and V.E. Hubeny, Phys. Rev. D 62
Γ(cv )Γ(av + bv − cv + 2)
m (2000) 024027.
A1 ≃ (1 − z)− 2 , (32) [6] J.S. Chan and R.B. Mann, Phys. Rev. D 55, 7546
Γ(av + 1)Γ(bv )
(1997).
m Γ(cv )Γ(av + bv − cv )
A2 ≃ (1 − z)1− 2 (cv − bv − 1) . [7] B. Wang, C.Y. Lin, and E. Abdalla, Phys. Lett. B
Γ(av )Γ(bv ) 481, 79 (2000).
[8] T.R. Govindarajan and V. Suneeta, Class. Quantum
By imposing the vanishing Dirichlet condition at infin- Grav. 18, 265 (2001).
ity for the components A1 and A2 , we find the quasi- [9] V. Cardoso and J.P.S. Lemos, Phys. Rev. D 63,
normal modes av + 1 = −n, or bv = −n, i.e. 124015 (2001).
i m [10] V. Cardoso and J.P.S. Lemos, Phys. Rev. D 64,
(k+ − k− ) = n + 1 + or 084017 (2001).
2 2 [11] B. Wang, E. Abdalla, and R.B. Mann, “Scalar Wave
i m Propagation in Topological Black Hole Backgrounds”,
(k+ + k− ) = n + . (33)
2 2 [arXiv:hep-th/0107243].
[12] U. H. Danielsson, E. Keski-Vakkuri and M. Kruczen-
Now, for spin s = 1, the conformal weights are either ski, Nucl. Phys. B 563 (1999) 279; JHEP 0002 (2000)
|m|/2 or 1 + |m|/2. Thus, we again find agreement 039.
with (8), where the left and right conformal weights [13] M. Bañados, C. Teitelboim, and J. Zanelli, Phys. Rev.
are given by hL = 21 m, and hR = 1 + 12 m. For neg- Lett. 69, 1849 (1992).
ative m, the situation is analogous to the fermionic [14] D. Birmingham, I. Sachs and S. Sen, “Exact results
perturbations. One finds the same conditions with for the BTZ black hole,” [arXiv:hep-th/0102155].
hL = 1 − 21 m, and hR = − 12 m. [15] J.L. Cardy, Nucl. Phys. B 270, 186 (1986).
[16] S. Gubser, Phys. Rev. D 56, 7854 (1997).
In conclusion, we have shown that there is a quan- [17] D. Birmingham, Phys. Rev. D 64, 064024 (2001).
titative agreement between the quasi-normal frequen- [18] M. Abramowitz and I.A. Stegun, “Handbook of Math-
cies of the BTZ black hole and the poles of the re- ematical Functions,” Dover, New York, (1970).
tarded correlation function of the corresponding per- [19] S. Das and A. Dasgupta, JHEP 9910:025 (1999).
turbations of the dual conformal field theory. The