YITP-23-170

Entanglement Entropy via Double Cone Regularization

Taishi Kawamoto1, ∗ and Yu-ki Suzuki1, †

1
Center for Gravitational Physics and Quantum Information,
Yukawa Institute for Theoretical Physics, Kyoto University,
Kitashirakawa Oiwakecho, Sakyo-ku, Kyoto 606-8502, Japan
This paper proposes an alternative regularization method for handling the ultraviolet behavior
of entanglement entropy. Utilizing an iϵ prescription in the Euclidean double cone geometry, it
accurately reproduces the universal behavior of entanglement entropy. The method is demonstrated
in the free boson theory in arbitrary dimensions and two-dimensional conformal field theories. The
findings highlight the effectiveness of the iϵ regularization method in addressing ultraviolet issues in
quantum field theory and gravity, suggesting potential applications to other calculable quantities.

1.Introduction and Summary. In the realm of quantum iϵ prescription in the metric near the ZN orbifold
arXiv:2401.00219v2 [hep-th] 5 Jan 2024

field theory, particularly within the framework of
Feynman’s path integral [1, 2], the historical exploration
of the iϵ prescription for the propagator has yielded
valuable insights. This technique, also employed to
formulate the wave-functional of the vacuum state of the
universe [3], has more recently found in the regularization
of the tip of a double cone geometry, as observed in
the spectral form factor analysis [4–6]. This paper
delves into the utilization of the iϵ prescription in the
Euclidean version of the double cone geometry, offering
an alternative derivation of the universal behavior of the
entanglement entropy.
The entanglement entropy, introduced in quantum
information theories and also in quantum field theories
[7], measures the entanglement between subsystems.
A breakthrough in its calculation emerged with
the introduction of twist operators [8], providing a
pathway to derive the universal logarithmic behavior
of entanglement entropy in two-dimensional conformal
field theory (CFT). Significantly, entanglement entropy
diverges in quantum field theories due to the infinite
degrees of freedom inherent in the vacuum state,
necessitating the introduction of an ultra-violet (UV)
cut-off for its quantification.
For a d-dimensional quantum field theory, the
entanglement entropy of low-energy states is anticipated
to follow the form from the holographic calculation
[9, 10]:
( function evaluation and the non-Hermitian nature of
Cd−2 C0 log ξϵ + · · · , for even d
SA = d−2 + · · · + d−1
ϵ (−1) 2 F + · · · . for odd d
(1)
The initial term illustrates the area law of entanglement
entropy, prevalent at low energies. Remarkably, the
logarithmic term in even dimensions is proven to be
universal. Also in odd dimensions the constant term
exhibits universality. The term ”universal” is employed
to convey its independence of the regularization schemes.
This paper introduces an approach to extract the
universal term of entanglement entropy through the
Euclidean double cone calculations. By employing the
singularity (where 1/N represents the number of replica
sheets), we establish the efficacy of this regularization
method as an UV cut-off to entanglement entropy. This
assertion aligns with [11], emphasizing the necessity of
imposing boundary conditions around the entangling
surface (the end points of the subsystem) to define
entanglement entropy properly, given a division of the
Hilbert space into subsystems. A distinctive advantage of
our method lies in the compatibility with the Heat kernel
method, even with the introduction of a regulator. In
contrast, approaches involving a cut-off in the geometry,
such as the brick wall case [12], face challenges in solving
analytically via the method of images. Moreover, what
kind of physically relevant boundary conditions should be
favored remains unclear. To circumvent these difficulties,
we extract the half cone contribution by halving the
result in evaluating the partition function.
We demonstrated in the free scalar theory on a
flat space-time in any dimensions and arbitrary two-
dimensional CFTs. Despite successfully reproducing
universal terms, our approach comes with a trade-
off. Specifically, in our calculation of the entanglement
entropy for the free boson theory in even dimensions,
we encounter an imaginary constant term. Similarly, in
odd-dimensional cases, the non-universal part becomes
purely imaginary. This outcome aligns with the inherent
complexity of the Schwinger parameter in partition
the modular Hamiltonian for the sub-algebra, as noted
in [6]. While our double cone regularization method
yields the same universal terms compared to other
methods such as the momentum cut-off or the lattice
regularization, it operates through a totally different
manner. In particular, the spectrum of modular
Hamiltonian is drastically changed if we turn on the
regularization parameter. It is important to stress
that the modified modular Hamiltonian have complex
spectrum and quasi normal modes, which capture the
universal anomaly terms though the spectrum itself
does not depend on regularization parameter in an
explicit way. Understanding how to extract physically
 2

(n)
meaningful quantities amid the renormalization of the SA by ϵ2 . The universal term is given by the
complex parameters represents an interesting avenue for constant term and logarithmic term for odd and even
future exploration. For the scope of this paper, our focus d, respectively
remains on the universal terms. 
(n+1) π 2−d


It is pertinent to draw attention to the parallelism 
 6n d Γ 2 · Vd−2 md−2 , d ∈ odd
(n) (4π) 2
with pseudo entropy [13, 14]. We anticipate that the SA ∼ (n+1) d −1
2π (−1) 2 1


specific details of the complex contour deformation will

 6n d · Vd−2 md−2 log mϵ . d ∈ even
(4π) 2 ( d
2 −1 )!
yield distinct signs and values for the constant terms.
Pursuing this avenue, we find it intriguing to investigate Above we manually introduced the cut-off scale in (5)
its applications to the holographic entanglement entropy through dimensional analysis. A natural question to ask
[9, 10] and also to de Sitter holography [15]. The utility is whether we can interpret this ϵ as a geometric cut-off.
of the iϵ prescription in the metric extends to various The iϵ prescription provides a lucid understanding of this.
situations where analytical solutions are favored while Below, we demonstrate that this prescription effectively
introducing a cut-off. resolves the orbifold singularity.
2.Entanglement entropy in free scalar fields. In the In this instance, we shape the Euclidean double cone
subsequent section, we revisit the orbifold method geometry by extending the radial direction to negative
employed in deriving entanglement entropy as outlined values and slightly deforming it into the imaginary
in [16]. Our focus centers on the entanglement entropy in direction, as detailed in [4–6]
the free scalar field in d-dimensional flat space, denoted d
as Rd . We choose a subsystem A, which is defined as
X
ds2 = dr2 + (r − iϵ)2 dθ2 + (dxi )2 , (6)
x1 > 0. For ease of representation, we amalgamate τ i=2
and x1 into a single complex plane C. Employing the
orbifold method, we calculate the n-th Rényi entropy for where −∞ < r < ∞. Replacing iϵ with ϵ might appear to
the half-infinite region. yield similar outcomes naively. However, it is crucial to
Utilizing the orbifold method, the n-th Rényi entropy note that the metric becomes non-invertible at r = ϵ. We
for a semi-infinite subsystem is derived as follows: claim that the metric, utilizing the iϵ prescription, not
only satisfies the Einstein equation but also offers a more
 
(n) 1  d−2
 1  d−2
 flexible method for regulating the UV divergence though
SA = Z C/ZN × R − Z C×R , initially we do not necessarily introduce an imaginary
1−n N N= n 1

(2) regulator. We assert that by anticipating the survival of

where the action h of ZN follows the Z2 symmetry, which exchanges r ↔ −r, we divide
the partition function by two, extracting a half cone
2πi
h : X = x1 + iτ → Xe . N (3) contribution, akin to the approach in [18, 19].
We now delve into the details of calculating the
In this context, the partition function corresponds to the entanglement entropy. Specifically, we apply the orbifold
one at the first quantization. Employing the heat kernel method and conduct a volume integral of the heat kernel.
method, as extensively reviewed in [17], especially we use As a result, we obtain:
the expression for the heat kernel in flat space Rd ;
1 
Z C/ZN × Rd−2 − Z C × Rd−2
  
1 2
− r4t −tm2 N
KRd (x, x′ ; t) = d e , r = |x − x′ |, (4) Z ∞
dt
Z
√
(4πt) d


2
= d x g KC/ZN ×Rd−2 (x, x; t) − KC×Rd−2 (x, x; t)
0 2t
where t is the Schwinger parameter. Also the expression N −1 Z Z √
2πVd−2 X dt 1 2 r2 2 πk
for the heat kernel for the orbifold and the n-th Rényi = d dr r2 e−tm − t sin N ,
entropy is obtained via method of images as N 2t (4πt) 2 γ
k=1 Γ
Z ∞ where g is the determinant of the metric. We choose a
√
Z
(n) dt
SA = dd x g KC/ZN ×Rd−2 (x, x; t) contour of radius r, denoted as γ, in the complex plane
ϵ2 2t
of Fig.1. The integration with respect to r can be carried
−KC×Rd−2 (x, x; t)) |N =1/n
out as follows: let us deform γ to r −iϵ, where −∞ < r <
(n + 1)πVd−2 ∞ dt
Z
= −m2 t ∞, to evaluate√ the integral. It is crucial to handle the
d e
6n ϵ 2 (4πt) 2 branch of r 2 . To obtain a non-zero result, we select the

(n + 1) π Vd−2 negative branch for Re[r] < 0 part. Since the integrand
2 2


= d E d ϵ m , (5) is Z2 invariant, we consider only the part Re[r] > 0.
6n (4π) 2 ϵd−2 2
Consequently, we obtain:
where Eν (z) is the ν-th order exponential integral similar 1 
Z C/ZN × Rd−2 − Z C × Rd−2
  
to (1) and we introduce a cut-off ϵ. We can expand N
 3

N −1
Γ′ t′
Z
πVd−2 X 1 d 2 2
+ ϵt sin2 ( kπ
N )
Γ t
= dt · t− 2 e−tm .
sin2
d kπ


(4π) 2 N k=1 N Γ

We observe that if we choose the contour for the

γ
FIG. 2. (Left):The contour for the Heat kernel. (Right):The
contour for the Hankel function.

FIG. 1. The contour γ for our double cone regularization. for Re[z] > 0 [22]. Γ′ is illustrated in Fig.2. Let us set
z = imr, where Re[r] > 0 and Im[r] < 0. Then we see
Re[imr] > 0 and Im[imr] > 0. By combining these
Schwinger parameter as Γ = [0, ∞), the integral diverges.
iν dt′ imr
Z Z
Therefore, we must consider Γ in the complex plane. (t′ − t1′ ) 1  r ν dt −m2 t− r2
Kν (mr) = e 2 = e 4t .
This implies that we need to involve complex Schwinger 2 Γ′ t′ν+1 2 2m Γ t ν+1

parameters when dealing with the complex metric space- 2

time [20]. By setting ν = d2 − 1 and t′ = − 2m imr t, we obtain (10)
In determining the integral contour for the complex for r such that Re[r] > 0 and Im[r] < 0 so that Γ is the
Schwinger parameter, we remind a fundamental identity correct contour for this region.
as described in [21], Next we move to the evaluation of the entanglement
entropy. By choosing contour Γ in Fig.2, we obtain the
Z
finite expression for the Rényi entropy,
dtKRd (x, x′ ; t) = GRd (x, x′ ), (7)
1
Z[C/ZN × Rd−2 ] − Z[C × Rd−2 ]
where GRd (x, x′ ) is a Green’s function, satisfying N
N −1
! d2 −1
iπ 2 Vd−2 X
 
1 1 m (1) kπ
(□x − m2 )GRd (x, x′ ) = p δ(x − x′ ). (8) = 2 − H d −1 2ϵm sin .
(4π) 2 N k=1 sin kπ
d
g(x) N ϵ sin kπ
N
2 N

We can calculate the Green’s function and the heat As with the case for the momentum cut-off, we can
kernel separately. From this identity, we can specify the explore the ϵ expansion. Let’s focus on the scenario when
appropriate contour for the Schwinger parameter. The d is even. In this case, the n-th Rényi entropy can be
Green’s function for real positive r is also known expanded as follows:
d

N −1
1  m  d2 −1 (n) 2 −1 ! d Vd−2 1 X 1
GRd (x, x′ ) = K d −1 (mr), (9) SA = (−1) 2 −1 d−2 kπ d−2
+ ···
2π 2πr 2
2 ϵ N (sin N )
k=1
d  
where Kν (z) is the modified Bessel function of the second n + 1 2π (−1) 2 d−2 1
+
· Vd−2 m log
kind. For the real r > 0 and t > 0 case we can show that 3n (4π) d2 d2 − 1 ! mϵ
the previous identity holds n + 1 2π (−1) 2 −1
d  
d

iπ

+ ψ −1 −γ+ · Vd−2 md−2 · · · .
3n (4π) d2 d2 − 1 !


Z ∞ 2 2
dtKRd (x, x′ ; t) = GRd (x, x′ ). (10)
0 It’s noteworthy that all terms can be divided by N − 1,
allowing us to safely define the entanglement entropy.
In our double cone regularization, we should ensure that When d is odd, we see:
the identity (10) holds for r such that Re[r] > 0 and
Im[r] < 0. To achieve this, we consider deforming the N −1 d
(−1) 2 −1 Γ d2 − 1 Vd−2


(n) 2π 1 X 1
contour of the integral away from the real axis. One SA = d
d
(4π) 2 N k=1 sin kπ 2 ϵd−2
possible contour satisfying these criteria is illustrated in N
 
Fig.2. This is justified as follows: it is known that the n+1 π 2−d
(1) + ··· + d Γ · Vd−2 md−2 + · · · .
Hankel function of the first kind Hν (z) has an integral 3n (4π) 2 2
representation,

dt′ z2 (t′ − 1′ )
Z
(1) 1 It is important to note that the divergent terms for odd
Hν (z) = e t , (11)
iπ Γ′ t′ν+1 d are purely imaginary.

One crucial observation is that if divided by two
(extract a half cone), the universal term of the Rényi
entropy obtained via our double cone regularization
matches the one from the momentum cut-off (6).
As is customary, the density matrix of the right Rindler
wedge in Minkowski space is naively expressed as
1
e−βK , Z(m2 , β) = Tr e−βK ,
 
ρ= (12)
Z(m2 , β)
where K represents a boost operator [23]. In the double
cone regularization, it is natural to replace the boost
operator K with the modified boost operator K e [6],
which is defined as the translation of θ in the deformed
metric (6) , yielding
1 h i
ρ= e−2πK , Z(m
e 2 , β) = Tr e−β Ke . (13)
e
e 2 , β)
Z(m
It is important to note that if this holds true, ρ is not
a Hermitian operator since K e is non-Hermitian. The
eigenvalues of the K
e are called quasi-normal modes (here
correspondingly they are complex valued). In Appendix
A, we derive the spectrum of these quasi-normal modes
and show that it is independent of our regularization
parameter ϵ > 0. These quasi-normal modes determine
the poles of the partition function Ze [24], leading to the
formula
i2r∞ ξ + 2 l − 41

!
Pol(m2 )
Y
2
Z(m , β) = e
e Γ 1+β , of the subsystem), we employ the iϵ prescription in
2π
l≥1,k
where r∞ is an IR cut-off,√ ⃗k are momentum for the
Rd−2 directions and ξ = k2 + m2 . β is 2π or 2πn
depending on assuming one-sheet or n-sheet geometry,
respectively. Here, we exclusively consider the quasi-
normal modes (l ≥ 1) as Im[K] e < 0 as discussed in [6].
Despite the absence of the UV parameter ϵ in the quasi-
normal modes, the analytic function Pol(m2 ) includes the
UV parameter. This discussion on non-Hermitian density
matrices assures us that the Renyi entropy is not obliged
to be a real value. Specifically, taking the limit n → 1
yields a complex-valued entanglement entropy. However,
in our scenario, entropic inequalities like subadditivity
or strong subadditivity are not naively applicable due to
the non-Hermitian nature of the density matrix. This
is similar to a case that von Neumann entropy defined
through transition matrix, called pseudo entropy, violates
entropic inequalities [13].
3.Double cone in two-dimensional CFTs. In two-
dimensional conformal field theories, the entanglement
entropy in the single interval case is universal
c L
SA = log , (14)
3 ϵ
where the c is the central charge of the CFT and L
is the interval length. As elucidated in the preceding
4
section, our computations in the free boson theory in two
dimensions are consistent with the behavior expressed in
(14), with c = 1.
Several methodologies exist for deriving the universal
behavior articulated in (14). Authors have employed
techniques such as twist operators, as documented in [8],
and boundary states, allowing for understanding from
the algebraic structure of the Hilbert space, as explained
in [11].
Now, we intend to employ the iϵ prescription in the
replica manifold. We shall focus on the scenario of a flat
two-dimensional plane, denoted as R2 , hosting a unitary
conformal field theory (CFT). It is noteworthy that
our considerations encompass arbitrary two-dimensional
CFTs, irrespective of their central charges. The metric
of interest is expressed as follows:
ds2 = dx2 + dτ 2 = dr2 + r2 dθ2 . (15)
We designate the subsystem A as x ∈ [0, L] at τ = 0. To
facilitate calculations, we introduce complex coordinates
as follows:
z = x + iτ, z̄ = x − iτ. (16)
We begin by constructing the double cone geometry
resembling a ”wormhole” with two throats attached
around z = 0 and z = L. In order to regulate the
physics around the entangling surface (the boundary
that region. The procedure is outlined as follows: We
consider the radial coordinate around the endpoints of
the interval, denoted as z = 0, L in Fig.3. Similar to
the flat plane case, we extend the radial direction into
negative values using the iϵ prescription
z = (r − iϵ)eiθ . (17)
One may wonder if this prescription is correct since we
can also construct a double cone geometry in an n-sheet
geometry. We can leverage a map
  n1
z−L
ζ= , (18)
z
to construct the n-sheet geometry. Since we are in a
flat space R2 , we can create the double cone geometry
using the iϵ prescription, as done previously. In this
scenario, the position of the ”wormhole” throat slightly
differs from z = 0 and z = L if we pull back to one-sheet
geometry. For instance, employing a map (see (18)),
we can construct a double cone geometry with the iϵ
prescription, which can then be pulled back to the one-
sheet geometry. The position of the throat around L is
determined by:
L
z= , (19)
1 − (−iϵeiθ )n

and even in the n → 1 limit this position differs
from the position, where we originally introduced the iϵ
prescription
z = L − iϵeiθ .
Although the details of the throat positions may differ
(and sometimes the shape of the throat may also differ
in different regularization schemes), they do not change
the leading terms in the entanglement entropy, as we will
see below. The discussion below parallels the argument
in [11].
After introducing the double cone geometry with the
iϵ prescription for the one-sheet geometry, we can map it
into the torus
z
w = log . (21)
L−z
Note that, since we are dealing with the complex
z w
Conformal
& the iϵ prescription applied to the Euclidean double cone
U U
map
I I
& I
& we successfully reproduced the universal logarithmic and
“Wormholes”
FIG. 3. Construction of the ”torus” geometry connected by
”wormholes” via the iϵ prescription. The left figure shows the
double cone geometry through ”wormholes”, which is made
of two planes. Each planes describes Re[r] > 0 and Re[r] <
0 regions respectively. The right one describes the ”torus”
geometry connected by the ”wormhole” throats.
manifold, the originally anti-holomorphic part is
completely independent. Here and below, we consider the
holomorphic part, but we find that the anti-holomorphic
part gives the same contribution. After using this map,
the length and circumference of the torus become l =
4 log Lϵ + · · · and 2π, respectively. The length of the
torus varies depending on the details of the regularization
described above, though it only affects the sub-leading
terms in the Lϵ expansion. We can also construct the
n-sheet geometry and obtain a torus whose length and
circumference are given by l = 4 log Lϵ + · · · and 2πn,
respectively. Using a modular transformation, we can
rescale the length and the circumference as l = n4 log Lϵ +
· · · and 2π. We also comment that the sub-leading
corrections to length can be imaginary valued in general
depending on the detailed choice of the complex contour.
It also depends on which manifold (one-sheet or n-sheet)
we construct the double cone geometry.
Then, the entanglement entropy reads
 
1 Zn
SA = lim log , (22)
n→1 1 − n (Z1 )n
5
where Zn and Z1 denote the partition function of the
n and one-sheet geometries. In the long length limit
(log Lϵ → ∞), the partition function can be approximated
by truncating the vacuum state propagation
(20)
1 c 1 − n2 L c L
SA = lim log = log . (23)
n→1 1 − n 6 n ϵ 3 ϵ
We need to extract the half cone contribution, divide
the result by two, and take into account the contribution
from the anti-holomorphic part. Therefore, we obtain
c L
SA = log , (24)
3 ϵ
where we ignore the sub-leading corrections in the Lϵ
expansions, which are complex valued in general.
4.Conclusion. We presented an alternative derivation of
the universal component of entanglement entropy using
geometry. Our calculations were demonstrated in the
free boson theory in arbitrary dimensions and in two-
dimensional CFTs. In the case of the free boson theory,
constant terms in even and odd dimensions, respectively.
Notably, the constant terms in even dimensions were
found to be imaginary.
For two-dimensional CFTs, we derived the universal
logarithmic term in the entanglement entropy. While
entanglement entropy is typically expected to be real-
valued, our results included imaginary terms. The
interpretation of these imaginary terms requires further
exploration, understanding the connection with complex
geometries as well as their implications for the de Sitter
holography.
Throughout this paper, our focus has been on
elucidating universal behavior, and while we have
encountered imaginary contributions in non-universal
components, we anticipate that there may be a suitable
method to define real-valued entanglement entropy even
with the deformed complex contour. Additionally,
applying our approach to different metrics and problems
where careful control of the cut-off scale is necessary
could provide further insights.
Acknowledgments.—We are grateful for Y.Taki and
M.Watanabe in useful discussions at the early stage
of this work and ongoing collaborations related to this
work. We thank S.M.Ruan and T.Takayanagi for
fruitful discussions. We also thank K.Doi, W.Zixia and
ChatGPT 3.5 for revising our writings and H.Liu for
his insightful comments on the quasi-normal modes and
non-Hermitian modular operators. TK is supported
by Grant-in-Aid for JSPS Fellows No. 23KJ1315.
YS is supported by Grant-in-Aid for JSPS Fellows
No.23KJ1337.
 YITP-23-170

and Z. Wei, Phys. Rev. D 103, 026005 (2021),

arXiv:2005.13801 [hep-th].
[14] A. Mollabashi, N. Shiba, T. Takayanagi, K. Tamaoka,
∗
taishi.kawamoto@yukawa.kyoto-u.ac.jp and Z. Wei, Phys. Rev. Lett. 126, 081601 (2021),
†
yu-ki.suzuki@yukawa.kyoto-u.ac.jp arXiv:2011.09648 [hep-th].
[1] R. P. Feynman, Rev. Mod. Phys. 20, 367 (1948). [15] A. Strominger, JHEP 10, 034 (2001), arXiv:hep-
[2] R. P. Feynman, Phys. Rev. 76, 769 (1949). th/0106113.
[3] J. B. Hartle and S. W. Hawking, Phys. Rev. D 28, 2960 [16] T. Nishioka and T. Takayanagi, JHEP 01, 090 (2007),
(1983). arXiv:hep-th/0611035.
[4] P. Saad, S. H. Shenker, and D. Stanford, (2018), [17] D. V. Vassilevich, Phys. Rept. 388, 279 (2003),
arXiv:1806.06840 [hep-th]. arXiv:hep-th/0306138.
[5] E. Witten, (2021), arXiv:2111.06514 [hep-th]. [18] T. Anegawa, N. Iizuka, and D. Kabat, Phys. Rev. D
[6] Y. Chen, V. Ivo, and J. Maldacena, (2023), 105, 085003 (2022), arXiv:2111.03886 [hep-th].
arXiv:2310.11617 [hep-th]. [19] T. Anegawa, N. Iizuka, and D. Kabat, Phys. Rev. D
[7] C. Holzhey, F. Larsen, and F. Wilczek, Nucl. Phys. B 106, 085010 (2022), arXiv:2205.01137 [hep-th].
424, 443 (1994), arXiv:hep-th/9403108. [20] E. Witten, JHEP 04, 055 (2015), arXiv:1307.5124 [hep-
[8] P. Calabrese and J. L. Cardy, J. Stat. Mech. 0406, th].
P06002 (2004), arXiv:hep-th/0405152. [21] R. B. Mann and S. N. Solodukhin, Phys. Rev. D 55, 3622
[9] S. Ryu and T. Takayanagi, Phys. Rev. Lett. 96, 181602 (1997), arXiv:hep-th/9609085.
(2006), arXiv:hep-th/0603001. [22] M. Shinichi, Iwanami Sugaku Kousiki, Formulas in
[10] S. Ryu and T. Takayanagi, JHEP 08, 045 (2006), Mathematics, Vol. III (Iwanami, 2002).
arXiv:hep-th/0605073. [23] W. G. Unruh, Phys. Rev. D 14, 870 (1976).
[11] K. Ohmori and Y. Tachikawa, J. Stat. Mech. 1504, [24] F. Denef, S. A. Hartnoll, and S. Sachdev, Class. Quant.
P04010 (2015), arXiv:1406.4167 [hep-th]. Grav. 27, 125001 (2010), arXiv:0908.2657 [hep-th].
[12] G. ’t Hooft, Nucl. Phys. B 256, 727 (1985). [25] I. Bah, Y. Chen, and J. Maldacena, JHEP 04, 061
[13] Y. Nakata, T. Takayanagi, Y. Taki, K. Tamaoka, (2023), arXiv:2212.08668 [hep-th].
 2

APPENDIX A: QUASI-NORMAL MODES ON THE DOUBLE CONE

In this appendix, we explore the massive free field theory within the framework of a double cone geometry in
Minkowski space. While our approach parallels that of [6, 25], we explicitly provide the exact solution. The scalar
fields ϕl (r) and ϕr (r) are considered for Re[r] > 0 and Re[r] < 0, respectively. The wave equation to be solved is
given by
d
!
1 √ 1 2 X 2 2
√ ∂r r ∂r + 2 ∂τ +
2 ∂i − m ϕ = 0, (25)
r2 r i=2

with the metric

d
X
ds2 = dr2 + r2 dτ 2 + (dxi )2 . (26)
i=2

Assuming Z2 symmetry, we propose the following ansatz for the equations:

⃗ ⃗
ϕr (r) = e−ωτ eik·⃗x Fω (r), for r > 0, ϕl (r) = Λ(ω)e−ωτ eik·⃗x Fω (|r|), for r < 0,(27)
p
where r is a real variable, ⃗k, ⃗x denote the momentum and coordinate for the vertical direction and ξ = ⃗k 2 + m2 .
The solutions for the wave equation take the form of a linear combination of Bessel functions:

Fω (r) = A(ω)Jiω (−irξ ) + B(ω)Yiω (−irξ). (28)

To ensure the analytic continuation of the right field to the lower half-plane matches the left field, as expressed in
Eq. (27), we impose the conditions:

Jiω (e−iπ R) = eπω Jiω (R),

(29)
Yiω (e−iπ R) = e−πω Yiω (R) − 2i cosh πωJiω (R).

For the case where iω ∈

/ Z, this leads to

ϕl (e−iπ r) = e−iωt (eπω A(ω) − 2i cosh πωB(ω))Jiω (−irξ) + e−πω B(ω)Yiω (−irξ) ,


(30)
ϕr (−r) = e−iωt Λ(ω)(A(ω)Jiω (−irξ) + B(ω)Yiω (−irξ)).

Given the independence of the two Bessel functions for all orders, we deduce the relations:
eπω A(ω) − 2i cosh πωB(ω) = Λ(ω)A(ω),
(31)
e−πω B(ω) = Λ(ω)B(ω).

Suppose B(ω) ̸= 0, leading to Λ(ω) = e−πω . The first equation becomes

sinh πωA(ω) − i cosh πωB(ω) = 0. (32)

Introducing the IR cutoff r = r∞ and demanding the Dirichlet boundary condition ϕ(r = r∞ ) = 0 yields the
solution:

Fω (r) = c(ω)(Yiω (−ir∞ ξ)Jiω (−irξ) − Jiω (−ir∞ ξ)Yiω (−irξ)). (33)

We find B(ω) = − YJiω (−ir∞ ξ)

iω (−ir∞ ξ)
A(ω). Substituting this into Eq. (32), we obtain

sinh πωYiω (−ir∞ ξ) + i cosh πωJiω (−ir∞ ξ) = 0. (34)

For the case where the IR cutoff is very large r∞ ξ ≫ 1, and using the asymptotic form of the Bessel functions
r  
1 iωπ π
Jiω (z) ∼ cos z − − + ··· ,
πz 2 4
r   (35)
1 iωπ π
Yiω (z) ∼ sin z − − + ··· ,
πz 2 4
 3

we obtain
   
iωπ π iωπ π
sin iπω sin −ir∞ ξ − − − cos iπω cos −ir∞ ξ − − = 0. (36)
2 4 2 4

This implies
 
ωπ π 1
i −ir∞ ξ − = l − π, l ∈ Z,
2 4 2
  (37)
2r∞ ξ 1
ωl = − 2i l − , l ∈ Z.
π 4

Thus, we have obtained the quasi-normal mode. In the case where iω ∈ Z, the analytic continuation of the Bessel
functions of the second kind is slightly modified:

Yl (e−iπ R) = (−1)l (Yl (R) − 2iJl (R)). (38)

From the junction condition (27), we deduce

(−1)l (Al − 2iBl ) = Λl Al ,

(39)
(−1)l Bl = Λl Bl .

Considering the fall-off condition Bl = − YJll (−ir ∞ ξ)

(−ir∞ ξ) Al and aiming for nontrivial solutions with Al ̸= 0, Bl ̸= 0, we find
Λl = (−1)l . However, this implies Bl = 0, and consequently, we do not have a non-trivial solution in this case.