Weighting gates in circuit complexity and holography

I. Akal
II. Institute for Theoretical Physics
University of Hamburg
D-22761 Hamburg, Germany∗
(Dated: March 15, 2019)
Motivated by recent studies of quantum computational complexity in quantum field theory and
holography, we discuss how weighting certain classes of gates building up a quantum circuit more
heavily than others does affect the complexity. Utilizing Nielsen’s geometric approach to circuit
complexity, we investigate the effects for a regulated field theory for which the optimal circuit is a
representation of GL(N, R). More precisely, we work out how a uniformly chosen weighting factor
acting on the entangling gates affects the complexity and, particularly, its divergent behavior. We
show that assigning a higher cost to the entangling gates increases the complexity. Employing the
arXiv:1903.06156v1 [hep-th] 14 Mar 2019

penalized and the unpenalized complexities for the Fκ=2 cost, we further find an interesting relation
between the latter and the one based on the unpenalized Fκ=1 cost. In addition, we exhibit how
imposing such penalties modifies the leading order UV divergence in the complexity. We show that
appropriately tuning the gate weighting eliminates the additional logarithmic factor, thus, resulting
in a simple power law scaling. We also compare the circuit complexity with holographic predictions,
specifically, based on the complexity=action conjecture, and relate the weighting factor to certain
bulk quantities. Finally, we comment on certain expectations concerning the role of gate penalties
in defining complexity in field theory and also speculate on possible implications for holography.

I. INTRODUCTION In the context of BHs, such an information/geometry

duality turns out to be even more astonishing. Namely, it
The anti de-Sitter (AdS)/conformal field theory (CFT) has been shown that while the holographic entanglement
correspondence [1] has substantially improved our un- entropy approaches a constant value with time during BH
derstanding of strongly coupled systems and black hole thermalization, certain bulk quantities such as the size
(BH) micro states. As an explicit realization of the holo- of the Einstein-Rosen bridge (ERB) for the eternal AdS
graphic principle [2, 3], it basically dictates how quan- BH keeps increasing [13]. In view of that conundrum,
tum gravitational physics can nonperturbatively be for- it has been proposed that the boundary quantity which
mulated within the language of a pure quantum field the- continues evolving after thermal equilibrium is quantum
ory (QFT). However, despite the enormous progress since computational complexity [14].
its first proposal, many aspects of the duality still remain Two separate conjectures have been proposed in
deeply mysterious. the holographic context. The first one, the complex-
In recent years, new concepts from quantum informa- ity=volume (CV) proposal, states that complexity is pro-
tion and quantum computation have further helped to portional to the volume of a maximal codimension one
advance our understanding of the mechanisms behind the surface in the bulk which extends to the boundary [15–
AdS/CFT correspondence [4–6]. The found relationships 17]. A second, more precise one, is known as the com-
may be traced back to the much celebrated Bekenstein- plexity=action (CA) proposal [18, 19]. It identifies the
Hawking entropy formula which relates an information complexity of the boundary CFT state with the gravita-
theoretic quantity to a geometric characteristic of a BH; tional action evaluated on a specific bulk region known as
the area of the event horizon. Motivated by this striking the Wheeler-DeWitt (WDW) patch. The WDW patch is
prediction, one could already expect that a more funda- the domain of dependence of a bulk Cauchy surface at-
mental information theoretic notion is related to other tached to a specific time slice. Also note that the notion
geometric properties of BHs. of complexity has recently been related to the spacetime
In fact, it has been shown that entanglement properties volume of the WDW patch [20].
of the boundary CFT are directly related to certain geo-
metric quantities in the bulk [7–9]. These findings have A sketch of the relevant geometric objects for the CA
provided surprising evidence that quantum entanglement and CV proposals is depicted in figure 1. In both cases
plays a profound role in the emergence of spacetime in a the corresponding quantity evolves with time even after
gravitational theory. Since then, understanding the rela- thermal equilibrium sets in. Thus, allowing to probe the
tion between quantum entanglement and the emergence interior region of BHs, the mentioned proposals add two
of semiclassical geometry is being actively worked on, see new classes of interesting gravitational observables with
e.g. [10–12]. connections to quantum information and computation.
Many aspects of the proposals and the related observ-
ables have recently been investigated, see e.g. [21–57].
Even though both the CV and CA conjectures are inter-
∗ [email protected] esting, much has to be done to advance our insights into
 2

for free theories, it is remarkable that they give rise to

similar leading order UV divergences as obtained in holo-
graphic computations. In addition, we also note that,
motivated by the tensor network representation of the
partition function, a notion of complexity has been pro-
posed from the path integral point of view [77–79] which
has given rise to interesting insights [80].
In the present paper, we focus on the notion of circuit
complexity in QFT. More precisely, we investigate how
weighting certain classes of gates building up a quantum
FIG. 1. The extended Penrose diagrams illustrate the relevant circuit more heavily than others does affect the optimal
geometric objects for the CV (left) and CA (right) proposals circuit depth (i.e. complexity1 ). Introducing such gate
for the two sided eternal AdS BH which is dual to the ther- penalties may be motivated from different perspectives.
mofield double state (TFD) formed by entangling two copies Being already important concepts in quantum informa-
of a CFT. In the left diagram the blue solid curve represents tion and computation, such gate penalties may have in-
the maximal bulk hypersurface extending to the AdS bound- teresting implications for complexity related aspects in
ary, asymptoting to the indicated tL and tR time slices on QFT itself, but as well as for better understanding cer-
which the CFT state is defined. It connects the latter through tain characteristics of quantum condensed matter sys-
the ERB. The shaded blue region in the right diagram repre-
tems. Moreover, it may shed light on recent observations
sents the WDW patch where the CFT state is again evaluated
on the mentioned time slices on the left and right boundaries.
resulting from holographic approaches to complexity. To
be more precise, one may, for instance, ask for the con-
sequence when complexity, if taken to be a physical at-
tribute of a QFT, incorporates the notion of locality. In
the deep connection between quantum information and
fact, if it is so, gates which entangle far separated points
the structure of spacetime. For establishing the geomet-
should require much higher costs in the geometric dis-
ric dual of complexity, a precise definition of this quan-
tance in the space of unitaries than those which operate
tity in strongly coupled theories or even in more general
as entanglers of less separated points. In the simple case,
QFTs is for sure necessary.
namely, for a pair of coupled harmonic oscillators, it has
The concept of compuational complexity is rooted in
been shown that introducing such penalties has a drastic
the field of theoretical computer science. Generally, an
effect on the complexity [59]. To be noted, there, the
implemented algorithm for mapping an input (reference)
optimal circuit acts as a representation of GL(2, R).
quantum state |ψR i for a number of qubits to an output
(target) quantum state |ψT i is determined by some func-

{ { } }
tion which is a unitary operation U , i.e. |ψT i = U |ψR i.
In a quantum circuit model, such a unitary operation
(or circuit) is constructed from elementary gates selected
from a fixed set of universal gates, see figure 2. Accord-
ingly, the circuit complexity can be defined as the min-
imal number of elementary gates required to construct FIG. 2. A quantum circuit constructing the target state
the circuit. To define the circuit complexity for states in |ψT i from the initial reference state |ψR i. The unitary
the boundary field theory [30, 58], the corresponding task U = gin gin−1 . . . gi2 gi1 is a sequence of elementary gates.
would then be identifying the minimum number of ele-
mentary gates required to prepare a desired target state Here, we work out how this change is reflected when
by starting from a simpler reference state. For a con- similar penalties are introduced in the case of a field the-
struction built by a certain number of discrete elemen- ory regulated by a lattice. In such a situation, the cor-
tary gates, the complexity of the target state explicitly responding circuit becomes extended in form of a rep-
depends on a specific reference, a choice of the gate set resentation of GL(N, R). We should note that instead
and an error tolerance. of using distance dependent weighting factors, we work
Preliminary progress towards quantifying circuit com- with entangling gates which are uniformly weighted. Of
plexity in field theory has recently been made. For in- course, this does not entirely correspond to the situa-
stance, the formulation in [59] is based on earlier studies tion described above, i.e. concerning the notion of local-
in quantum information showing that finding the optimal ity. However, doing so generalizes previous findings for
circuit is equivalent to finding the minimal geodesic in the states in field theory. This will enable us to make inter-
space of unitaries [60–62]. This approach has also been esting comparisons with previous findings. Beyond that,
generalized to fermionic [63, 64] as well as interacting the-
ories [65]; for other recent studies see e.g. [66–74]. An-
other field theory proposal is based on the Fubini-Study 1 We would like to note that the terms optimal circuit depth and
metric approach [75] recently applied in e.g. [70, 76]. (circuit) complexity are meant to mean the same. We use them
Even though both proposals have originally been made interchangeably.
 3

it has recently been argued that holographic complexity which implements a specific unitary operation U . This
might be nonlocal, means that the gate set in the CFT can be approached as a control problem of finding the
contains bilocal gates acting at arbitraryly distant points right Hamiltonian which constructs the desired unitary
[42]. This additionally motivates us to penalize only one or circuit. For doing so, one specifies a space of unitaries
specific class of gates without implementing any distance where the interesting paths satisfy the boundary condi-
dependencies. tions U (t = 0) = 1 and U (t = 1) ≡ U .3 In addition,
Another motivation for considering the setup detailed one also defines the corresponding cost function, call it
above goes back to the found similarities between holo- F . Afterwards, by minimizing the following functional
graphic complexity and circuit complexity in field the- Z 1  
ory. Particularly, we would like to understand how in- D(U (t)) = dt F U (t), U̇ (t) (1)
troducing gate penalties, even if only acting uniformly, 0
will affect the leading order UV divergence. Such studies
one obtains the optimal path (or circuit depth) which
may be helpful for further comparisons with predictions
yields the optimal quantum circuit. Here, the mentioned
from the gravity side. Closely related, introducing gate
cost F (U, V ) is a local function of U in the space of uni-
penalties may also be of particular importance for ten-
taries and V is a vector in the tangent space at the point
sor network descriptions such as MERA [81] or cMERA
U . It is argued that F must be (I) continuous, (II) pos-
[82, 83]. It has been argued that these constructions pro-
itive, (III) homogeneous and (IV) should satisfy the tri-
vide a representation of a time slice of the AdS space
angle inequality. In the case when (I) is replaced with
[84, 85].
the criteria of smoothness, D(U ) defines a length in a
The paper is structured as follows: in section II, we Finsler manifold which corresponds to a class of differ-
start with a short description of the geometric approach ential manifolds with a quasimetric and the measure as
to circuit complexity proposed by Nielsen and collabora- defined above. Thus, a Finsler geometry generalizes the
tors [60–62]. In section III, we first introduce the setup notion of a Riemannian manifold. According to this in-
under consideration and some basic notations. We then struction, the problem of finding the optimal circuit be-
comment on the general procedure and summarize some comes the problem of finding the geodesics the resulting
of the earlier results. We would like to note that even geometry. What remains to be done is fixing the form of
though much of the content in this part has already been the cost F . Once F is chosen appropriately, the complex-
presented in [59], we recap the relevant findings which ity is simply defined as the length of the optimal path in
are needed for later computations. We also discuss some the corresponding geometry.
additional aspects which are important at the later stage.
In section IV, we go through previous findings concern-
ing the effect of implementing gate penalties in the case III. SETUP
of two coupled oscillators. Afterwards, we comment on
some further details and extensions related to other cost
In the following, we specify the system we would like
functions. Section V is the main part of this paper. We
to focus on. We consider a free d dimensional scalar field
generalize the computations in penalized geometry for
theory described by the Hamiltonian H given by
the regulated field theory and work out the differences Z
to the unpenalized case. We then compare our results 1  
~ 2 (x) + m2 φ2 (x) . (2)
with holographic complexity computations, particularly, H= dd−1 x π 2 (x) + ∇φ
2
by referring to the findings based on the CA proposal. We
also briefly comment on certain expectations concerning We regulate this theory by putting it on a square lat-
the implementation of distance dependent gates and on tice with spacing δ which describes an infinite number of
earlier findings which may share certain similarities with coupled harmonic oscillators. The corresponding Hamil-
the presented results. In section VI, we finalize with a tonian then takes the form
brief conclusion. In the remaining part, we mostly stick 1 X P 2 (~n)
to the notations in [59] to allow for a direct comparison H= + M ω 2 X 2 (~n)
2 M
with the unpenalized case. ~
n
X (3)
2
+ M Ω2 (X(~n) − X(~n − x̂j ))
j
II. CIRCUIT COMPLEXITY
where the introduced definitions in H above read as fol-
lows
Let us begin by introducing the main idea of gemeotriz-
ing circuit complexity à la Nielsen and collaborators.2 X(~n) = δ d/2 φ(~n), P (~n) = δ −d/2 p(~n),
The basic task is finding the optimal quantum circuit (4)
M = 1/δ, ω = m, Ω = 1/δ.

2 For more details we refer the interested reader to [59]. 3 The notation 1 stands for the identity matrix.
 4

Let us first proceed with the simplest case, i.e. a system that for our purpose the relevant one are the entangling
of two coupled oscillators. and scaling gates, i.e.

Qab = eixa pb ∀ a 6= b (entangling),

A. Pair of oscillators /2 ixa pa
(11)
Qaa = e e (scaling),
1. Position space noting that a, b ∈ {1, 2}. The parameter  appearing in
(11) is chosen to be infinitesimal5 in order ensure only
The system of two coupled harmonic oscillators is de- small changes when the gates act on the wave function.
scribed by the following Hamiltonian Next, we may express the circuit U in path ordered
form,
1 2
p1 + p22 + ω(x21 + x22 ) + Ω2 (x1 − x2 )2


H= (5)
2 ←
Z s 
U (s) = P exp ds̃ Y I (s̃)OI . (12)
which is written in terms of the physical variables where 0
xj denote the spatial positions and Mj = 1 is set for sim-
plicity. The eigenstates and eigenenergies of (5) can be This representation will be used instead of a discrete
solved by rewriting the problem in terms of two decou- gate representation as considered in the original formu-
pled oscillators via changing from the original position lation. The operators OI precisely correspond to the
basis to the normal mode basis expressed in terms of the previous gates from (11) with Qab = exp (Oab ) and
following normal mode coordinates and frequencies Oab = (ixa pb + δab /2) where ab ≡ I ∈ {11, 12, 21, 22}.
Accordingly, the path ordered6 exponential (12) parame-
1 2 terizes the product of gates where the function Y I decides
x̃± := √ (x1 ± x2 ), ω̃+ := ω 2 ,
2 (6) whether the gate of type I is switched on or off. Note
2 that the differential ds̃ behaves analogous to the infinites-
ω̃− := ω 2 + 2Ω2 .
imal parameter . By proceeding in this way, any circuit
The normalized4 ground state wave function can be writ- follows a particular trajectory determined by Y I through
ten as the product of the ground state wave functions for the space of unitaries such that ψT = U (s = 1)ψR and
the two individual oscillators. In terms of the physical U (s = 0) = 1. In the following, we choose the cost F to
position coordinates, the resulting wave function reads be set by the following two norm
sX
(ω 2 − β 2 )1/4 − ω (x21 +x22 )−βx1 x2 F = F2 ≡ (Y I )2 . (13)
ψ0 (x1 , x2 ) = √ e 2 (7)
π I

where For this choice, note that the optimal path is a geodesic
ω̃+ + ω̃− ω̃+ − ω̃− in usual Riemannian geometry. So the functional for the
ω= , β= . (8) circuit depth from (1) can be written as
2 2
Z 1
Following the same motivation as described in [59], we
q
D(U ) = ds̃ GIJ Y I (s̃)Y J (s̃) (14)
choose a factorized Gaussian reference state where both 0
oscillators are disentangled, i.e.
√ where we assume that the metric7 is fixed by setting
ω0 ω0 2 2 GIJ = δIJ . Now, one can in principle express the func-
ψR (x1 , x2 ) = √ e− 2 (x1 +x2 ) . (9)
π tion Y I which specifies the velocity vector tangent to the
optimal trajectory in the space of unitaries in terms of the
Here, ω0 denotes some arbitrary frequency characterizing operators OI . However, it turns out to be more conve-
the reference state. Its explicit form will be discussed nient to work with unitary matrices instead of operators.
later. In order to construct U such that Accordingly, one just needs to reformulate the problem
in matrix representation. In case of Gaussian states, as
|ψ0 i ≡ |ψT i = U |ψR i, (10)
it is the case for the present setup, one may think of the
we need to fix the set of appropriate gates. A set of
elementary gates which implement a unitary transforma-
tion as prescribed in (10) can be found in the literature.
5 Note that due to   1 we only need to find the circuit which
Without going into the details, here we will only mention
minimizes the coefficient of the leading 1/ term in the complex-
ity.
6 The operators at smaller s̃ act prior to those at larger s̃.
7 In general, the metric GIJ allows to weight particular gates in
4 d2 x |ψ0 |2 = 1.
R
The normalization is chosen such that the circuit. This will be discussed in the sections below.
 5

space of states as the space

of positive quadratic forms, By proceeding as described, it has been found that the
i.e. ψ ∼ exp − 21 xa Aab xb where shortest geodesic is given by [59]

y = y1 s, ρ = ρ1 s, θ=τ =0 (21)
 
ω β
AR = ω0 1, AT = . (15)
β ω
where
It can be shown that the corresponding gate matrices p !
take the form 1 ω2 − β 2
y1 = log ,
2 ω0
QI = exp (MI ) (16) ! (22)
1 ω
ρ1 = cosh−1 .
where Mab,cd = δac δbd denotes a 2 × 2 matrix. For the
p
2 ω2 − β 2
present setup, the explicit generators MI are
    Substituting this straight line geodesic—since the metric
1 0 0 0 is flat—into the general expression for U ≡ U (τ, θ, ρ, y) ∈
M11 = , M22 = ,
0 0 0 1 GL(2, R) leads to the following optimal (straight line)
(17)
circuit
   
0 1 0 0
M12 = , M21 = .
0 0 1 0 
y1 −ρ1
 
U0 (s) = exp s , 0 ≤ s ≤ 1. (23)
−ρ1 y1
In terms of the gates from (17), it becomes clear that
the circuits form a representation of GL(2, R). Using the Here, we have expressed U0 from (23) in exponential form
generator matrices in (17), the path ordered exponential which is more convenient for later purpose. For instance,
in (12) can be replaced by using (23), we can immediately identify the correspond-
←
Z s ing velocity vector components, i.e. Y I ’s. Having deter-
U (s) = P ds̃ Y I (s̃)MI (18) mined the latter, we just need to insert them into (14)
0 which finally yields the complexity
where q
C2 ≡ D2 (U0 ) = 2y12 + 2ρ21 (24)
Y I (s) = tr(∂s U (s)U −1 (s)MIT ). (19)
for the F2 cost from (13) as we have indicated by the
The expression in (19) drastically simplifies the form subscript.
of the velocity vector Y I which determines the opti-
mal circuit depth. However, in order use the expres-
sion from (19), it is required to construct an explicit pa- 2. Normal mode subspace
rameterization of the mentioned transformations. After
having done this, the main task reduces to finding the Of course, one can express the complexity C2 from (24)
minimal8 geodesic in the resulting metric on GL(2, R) in terms of the physical parameters. For this, we just
which connects the matrices AT and AR according to need to use the relations in (22). However, the final result
AT = U (s = 1)AR U T (s = 1). would look rather complicated. Instead, in terms of the
A general parameterization can be obtained by con- normal mode frequencies introduced in (6), which can be
structing U according to the decomposition GL(2, R) = used to express (22) as
R × SL(2, R) where certain coordinates, for instance, la-    
beled by τ, θ and ρ, are chosen on the subgroup SL(2, R). 1 ω̃+ ω̃− 1 ω̃−
A fourth coordinate, call it y, will be responsible for the y1 = log , ρ1 = log , (25)
4 ω02 4 ω̃+
remaining R fibre. In this way, inserting U ≡ U (τ, θ, ρ, y)
into the complexity in (24) takes the form

ds2 = GIJ tr(dU U −1 MIT ) tr(dU U −1 MJT )

s
(20) 1

ω̃+
 
ω̃−

C2 = D2 (U0 ) = log2 + log2 . (26)
yields a right-invariant metric. To obtain the geodesic, 2 ω0 ω0
one first identifies the corresponding Killing vectors. Af-
terwards, these can be used to find all conserved charges This simple expression invites to investigate the optimal
which simplify solving the underlying geodesic equations. circuit in terms of the normal mode coordinates x̃± . Such
an observation can also be anticipated from the ground
state wave function which is a factorized Gaussian state
in the normal mode basis, i.e.
8 Note that by definition the complexity is determined by the
shortest geodesic which yields the desired transformation. In (ω̃+ ω̃− )1/4 − 1 (ω̃+ x̃2+ +ω̃− x̃2− )
ψ0 (x̃+ , x̃− ) = √ e 2 . (27)
general, one may find a family of geodesics with various lengths. π
 6

To work out the explicit form of the optimal circuit U0 act in the physical basis to scale/entangle the coordinates
in normal mode basis, we first need to identify the re- x1,2 , whereas M̃I˜ = R2 MI˜R2T scale/entangle the coordi-
quired transformation matrix which yields the change nates x+,− in the normal mode basis. The corresponding
[x̃+ , x̃− ]T = R2 [x1 , x2 ]T . It turns out that such a trans- matrix is orthogonal and can be constructed according to
formation can be realized via the following orthogonal the prescription [59]
rotation matrix
b ˜ = Rka ⊗ Rlb ,
R k, l ∈ {+, −}, a, b ∈ {1, 2} (31)
  IJ
1 1 1
R2 = √ with RRT = RT R = 1, (28)
2 1 −1 such that MI˜ = Rb ˜ MJ . Utilizing the matrix R
IJ
b ˜ , the
IJ
mentioned basis independence is nothing but the conse-
where quence of the equality
 
ÃT =
ω̃+ 0
, ÃR = ω0 1 (29) (20)|GIJ =δIJ = δI˜J˜tr(dU U −1 MI˜T ) tr(dU U −1 MJT˜ ). (32)
0 ω̃−
Very similar to the discussion above, one can show that
follow due to à = R2 AR2T .
Then, once the matrix R2 such a basis independence also holds for the Fκ=2 cost
is known, the optimal straight line circuit U0 can be set by
transformed to operate in the normal mode space via X
Fκ=2 = GI |Y I |2 (33)
the transformation Ũ0 (s) = R2 U0 (s)R2T . In contrast to
I
U0 , the latter transformation results in a considerable
I
simplification of the normal mode circuit9 Ũ0 with the where we will use G = 1 for the moment. In this case,
following diagonal form the complexity in both bases turns out to be
    

1
 
ω̃+
  1 2 ω̃+ 2 ω̃−
2 log ω0 0 Cκ=2 = log + log (34)
Ũ0 (s) = exp    s . (30) 4 ω0 ω0
1 ω̃−
0 2 log ω0
i.e. Cκ=2 = C22 . However, this basis independence does
generally not occur for costs of the form
It can be seen that Ũ0 does not possess any off-diagonal X
entries. In other words, in normal mode subspace, there Fκ = GI |Y I |κ ∀ κ 6= 2. (35)
is no entanglement introduced. This is in line with the I
mentioned factorized Gaussian shape for the states ψ̃R
and ψ̃T . The factorization turns out to be a substantial We should add that the cost Fκ=2 is of particular rele-
simplification when generalizing the results for the regu- vance for our studies, since it leads to the same leading
lated field theory. Now, if the analogous computation is UV divergence as found in holographic complexity com-
made by using the generators M̃I˜ and the velocity vec- putations. More details regarding this issue are discussed
˜ in section III C 2.
tor Ỹ I operating in the normal mode subspace, it follows
that the complexity remains as in (24), i.e. the complex-
ity C2 for the F2 cost is independent of the basis. This B. Lattice
feature can be comprehended if one uses the transforma-
tion matrix which transforms the generators MJ to MI˜. 1. One dimensional
Let us note that the former one, i.e. MJ = R2T M̃J R2 ,
The previous results can be generalized for the regu-
lated field theory by a line lattice. The corresponding
9 Hamiltonian is of the form
Of course, the simple circuit Ũ0 can be obtained if the correct
basis of generators in normal mode subspace are employed. In- N −1
1 X 2
troducing the corresponding set of indices in that subspace, i.e. H= p + ω 2 x2j + Ω2 (xj − xj+1 )2 (36)
I˜ ∈ {++, +−, −+, −−}, the generators M̃I˜ are formally equal 2 j=0 j
to MI˜, but act in a different space. Using the relation

M̃I˜ = R2 MI˜R2T ⇒ MI˜ = R2T M̃I˜R2

which satisfies the periodic boundary condition xj+N =
xj . It describes a discretized field placed on a circle with
the M̃I˜ generators can be transformed to act on states ex- length
pressed in position basis. Then, transforming (30) according to
R2T Ũ0 (s)R2 , where L = N δ. (37)
 ˜ 
Ũ0 (s) = exp Ỹ I (s)M̃I˜ , Motivated by the simplifications in normal mode basis,
we rewrite H in this representation, i.e.
leads to the optimal circuit from (23), i.e.
 ˜  N −1
U0 (s) = exp Y I (s)MI˜ . 1 X
H= |p̃k |2 + ω̃k2 |x̃k |2 . (38)
2
k=0
 7

The normal mode coordinates can be deduced from a For the details showing that Ũ0 from (42) indeed cor-
discrete Fourier transform, responds to the optimal circuit, we refer the interested
reader to [59]. Bringing all together, the complexity for
N −1
the one dimensional regulated field theory becomes
 
1 X 2πik
x̃k = √ exp j xj , (39)
N j=0 N v
uN −1  
1u X ω̃k
C2 = t log2 . (44)
where k ∈ {0, . . . , N − 1} and x̃†k = x̃N −k . For N = 2, 2 ω0
k=0
we find the coordinates
1 1
x̃0 = √ (x0 + x1 ), x̃1 = √ (x0 − x1 ) (40)
2 2 2. d dimensional
which of course correspond to the one introduced in (6)
if we identify x̃0,1 ↔ x̃+,− and x0,1 ↔ x1,2 . The normal The complexity for the one dimensional lattice can be
mode frequencies are expressed in terms of the physical easily extended to the (d − 1) dimensional lattice con-
frequencies ω and Ω, sisting of N d−1 oscillators. The final expression is of the
  from
2 2 2 2 πk
ω̃k = ω + 4Ω sin . (41) v
u NX −1
N 1u
 
ω̃~k
C2 = t log2 , (45)
Note that, in contrast to the system of two coupled har- 2 ω0
{kj }=0
monic oscillators, cf. equation (6), here we have a factor
4 in front of the second term. This goes back to the pe- where kj denote the momentum vector components. The
riodic boundary condition. This can be also seen when corresponding expression for the normal mode frequen-
we set N = 2 in (36) and demand x2 ≡ x0 such that the cies reads
term being proportional to Ω2 becomes doubled, see also
section V C 1. d−1
X 
πkj

2
In normal mode basis, the factorized Gaussian shape ω̃~k2 2
= ω + 4Ω 2
sin . (46)
of both the reference state and target ground state is j=1
N
preserved. Therefore, the previous studies for N = 2 can
be generalized to the case of N oscillators. In total, one
gets N 2 generators which give rise to N × N matrices.
C. Comparison with holographic complexity
This extends the initial group GL(2, R) to GL(N, R). For
the F2 cost from (13), an appropriate parameterization
spanned with N 2 coordinates of a general element U ∈ For the comparison with holographic complexity, it
GL(N, R) is needed. Accordingly, the optimal circuit is advantageous to express the complexity from (45) in
will then be determined by the optimal geodesic in this terms of the field theory parameters. Then, the normal
extended geometry characterized by the corresponding mode frequencies introduced in (46) become
right-invariant metric ds2 .
Following the usual procedure can be very challenging d−1
4 X 2 πkj
 
due to the large number of coordinates parameterizing ω̃~k2 =m + 22
sin (47)
δ j=1 N
the extended metric. Instead, one may expect that, sim-
ilar to the previously discussed simplified diagonal form
of the optimal circuit in normal mode basis, a simplifi- according to the definitions in (4). An estimation for the
cation of such type also applies for the present problem. complexity (45) in terms of the volume of the system,
Hence, the most optimal circuit would simply amplify the V , is also very useful. This can be accomplished if we
Gaussian width for each of the normal mode coordinates. first write V = Ld−1 by assuming an equisided lattice.
In particular, the extended optimal circuit would not in- Afterwards, applying the relation (37), the total number
troduce any entanglement between the normal modes. of oscillators can be expressed as
Indeed, it has been shown that the optimal circuit takes
the form of the generalized straight line circuit V
N d−1 = . (48)
  δ d−1
Ũ0 (s) = exp M̃0 s (42)
By using the relation in (48), it can be shown that the
where complexity scales as
    
1 ω̃0 1 ω̃N −1 N
d−1
2

ω̃~k

M̃0 = diag log , . . . , log . (43) C2 ∼ log . (49)
2 ω0 2 ω0 2 ω0
 8

1. UV divergence in QFT For further interesting similarities in the divergence

structure, we can compare with the leading order contri-
The leading order contribution to complexity in QFT bution in the CA proposal,
is determined by the UV modes. We can take δ to be the V

LAdS

UV cutoff where δm  1. In this UV approximation, CA ∼ d−1 log (56)
we may estimate the normal mode frequencies in (47) by δ Λδ
assuming where LAdS is the curvature scale of the AdS bulk space-
time, Λ is an arbitrary dimensionless coefficient fixing
ω̃~k ∼ 1/δ. (50) the null normals on the WDW patch boundary and δ
is the short distance cutoff scale in the boundary CFT
Using the approximation above, as well as the expressions [28]. As suggested in [59], for eliminating LAdS from
in (48) and (49), the leading order contribution to the the expression—since CA is expected to be defined in
complexity depending on the UV cutoff scales as the boundary CFT which therefore should not depend
r on any bulk AdS scale—one may fix the dimensionless
V coefficient as Λ = ω0 LAdS . Of course, ω0 is still an ar-
C2 ∼ . (51)
δ d−1 bitrary frequency. However, with the latter replacement,
the complexity simply reduces to
Remember that for the two coupled oscillators we have
seen that Cκ=2 = C22 . Taking this relation into account,
 
V 1
we may immediately get the leading order UV divergence CA ∼ d−1 log . (57)
δ δω0
on the lattice for the Fκ=2 cost which scales as
It is remarkable that this expression is similar to the QFT
V expression in (55) which also depends on some unspeci-
Cκ=2 ∼ . (52)
δ d−1 fied frequency ω0 of the reference state. The found co-
incidence may therefore suggest an interesting relation
between the notions of complexity defined in holography
2. UV divergence in holography and QFT.

Studies of the UV divergence in holographic complex-

ity based on the CV and CA proposals have shown that IV. PENALIZED GEOMETRY I: COUPLED
the leading contribution scales as [28] OSCILLATORS

V So far, both the scaling and entangling gates have been

CV,A ∼ . (53)
δ d−1 treated equally, i.e. each of them received the same cost
when constructing the optimal circuit. As pointed out
This behavior is similar to the scaling for the Fκ=2 cost
in the introduction, technically, for implementing the no-
from (52). Apart from that, it has been argued that the
tion of locality one would need to introduce some weight-
following depth function for the circuit complexity
ing factors which vary the strength of the acting gates
Z 1 according to the separation between the corresponding
˜
X
D̃κ = ds |Y I (s)|κ (54) points. Such a situation is sketched in figure 3.
0

{ { } }
with κ ≥ 1 would give rise to the same UV divergence
as in (53). On the field theory side we simply get the
mentioned scaling,
  κ
V  1 
Cκ ∼ d−1 log , (55)
δ δω0  FIG. 3. A quantum circuit constructed by weighted gates.
The color scaling for the gates shall indicate the difference in
however, with an additional logarithmic factor which will the gate weighting.
be discussed in detail below. The reason for the same
prefactor for all κ lies in the fact that for the depth func- However, recalling our motivation in the introduction,
tion in normal mode basis, i.e. D̃κ , the previously dis- we may consider a simpler setup via penalizing the gates
cussed straight line circuit still corresponds to the opti- uniformly. In the following, we first briefly discuss these
mal circuit [59]. This basis independence for all κ includ- effects for the system of two coupled harmonic oscillators
ing the F2 cost is the reason why the optimal geodesic subjected to the F2 cost as already explored in [59]. Af-
remains the same. In the original position basis, i.e. Dκ , terwards, we extend our discussion for Fκ=2 and work
this is generally not the case. The only exceptions are out the differences. The results constitute the basis for
the Fκ=2 and F2 cost functions. the field theory computations later in section V.
 9

A. F2 cost to deduce the complexity. The circuit depths on the

right-hand side of (64) can be computed according to the
Introducing weighting factors for the entangling gates integral form introduced in (14). Note that the latter is
simply means modifying the off-diagonal directions in the associated with the F2 cost.
circuit. To do so, we may write the tensor GIJ in (14) as We should add that the given segmented construction
can also be used for the Fκ=2 cost. Differently, for a gen-
GIJ = diag[1, α2 , α2 , 1] (58) eral κ ≥ 1, it does not yield a circuit depth which comes
close to the optimal one. One might argue that chang-
with α2 > 1. Of course, for α = 1 one would arrive at the ing the basis and computing D̃κ instead of Dκ may solve
original situation in which no extra cost is assigned to the this problem. However, it has been discussed that, once
entangling gates. With the extended penalty tensor from the geometry is penalized, the depth function will not be
(58), the corresponding metric would be modified due to invariant under a basis change [59]. Let us continue with
the appearance of additional mixing terms. The final the complexity for the F2 cost. The final result, which
expression may principally be treated as usual. Namely, can be obtained by plugging the mentioned segmented
after identifying the corresponding Killing vectors, one path into the normalization constant,10 takes the form
would need to find all conserved charges in order to solve
the geodesic equations. However, writing down the full
r
p
√  π 2
metric, it turns out that this task is too complicated to be D2 (Up ) ' 2ρ1 + 2y12 + 2α2
4
performed. This problem has been avoided by assuming (65)
√
 
α  y1 , ρ1  1 and neglecting some components of the π 1
' 2ρ1 + √ α + O .
metric to simplify the full expression. In this way, it is 2 2 α2
possible to construct a kind of segmented path which is
Here, the superscript p indicates that we work in penal-
not a geodesic, but is capable to come very close to the
ized geometry. Now, if we go back and compare this
optimal geodesic [59]. The segmented circuit constructed
result with the unpenalized straight line circuit depth in
by exploiting the described approximate path takes the
(24), we find that
form
D2p (Up )  D2 (U0 ) (66)

 Upa (s), 0 ≤ s ≤ 21
Up (s) ' . (59)
which follows due to the mentioned assumption α 
Upb (s), 12 ≤ s ≤ 1

ρ1 , y1  1. Being in agreement with the naive expec-
The circuits for the two segments are determined by tation, the complexity increases when a higher cost is
 −ρ 2s  assigned to the entangling gates.
e 1 0
Upa (s) = (60)
0 eρ1 2s
B. Fκ=2 cost
and
cos(φ)e−ρ1 − sin(φ)eρ1
 
Upb (s) = ey1 (2s−1) (61) Once the exponential forms in (62) and (63) are known,
sin(φ)e−ρ1 cos(φ)eρ1 finding the complexity in penalized geometry for the
Fκ=2 cost becomes straightforward. Similar to the pre-
where we have defined φ := π4 (2s − 1). Instead of con-
vious case, we first fix
tinuing with these expressions, we want to rewrite them
in exponential form. This can be easily done by utiliz- GI = [1, α2 , α2 , 1]T . (67)
ing the relation (19) which yields the desired exponential
forms Then, by using the depth integral
  
−ρ1 0 Z 1
Upa (s) = exp 2s (62) X
0 ρ1 Dκ=2 = ds̃ GI |Y I (s̃)|2 (68)
0 I
and
π
  
y1 4
s
Upb (s) = exp π . (63)
4 y1 2 10 The normalization constant gives the length of the geodesic. It
The expressions above turn out to be much simpler and corresponds to the complexity if the geodesic is the minimal one.
more suitable for later use. Now, using (62) and (63) For F2 , the normalization constant is simply given by the integral
in combination with (18), we can easily find the velocity
Z 1 q
D(U ) = ds gij ẋi (s)ẋj (s)
vector components. Note that the generators are the one 0
which are given in (17). Once the components of Y I (s̃) where x (s) = [τ (s), ρ(s), θ(s), y(s)]. Here, s is the scaled affine
are found, we can use the splitting parameter. The components of x are the mentioned parameteri-
zation variables in section III A 1 for the subgroup SL(2, R) and
C = D(Us ) = D(Us,a ) + D(Us,b ) (64) the R fibre.
 10

and the splitting in (64), we obtain First, we begin by assuming that U0 is the optimal
circuit which implements the following operation
 π 2
p
Dκ=2 (Up ) ' 2ρ21 + 2y12 + 2α2 . (69) |ψT i = U0 |ψR i (71)
4
in unpenalized geometry. The corresponding complexity
As before, we can compare the result (69) with the
shall be denoted by C(ψT , ψR |U0 ). In addition, we as-
complexity in unpenalized geometry, i.e. Dκ=2 (U0 ) =
sume that the analogous optimal circuit U∗ in penalized
D22 (U0 ). The difference between both complexities be-
geometry (i.e. when the entangling gates are weighted as
comes
above) is known as well, which implements
 π 2
p
Dκ=2 (Up ) − Dκ=2 (U0 ) ' 2α2 1 (70) |ψT i = U∗ |ψR i. (72)
4
For the latter, the complexity shall be denoted by
showing again that assigning a higher cost to the entan- C(ψT , ψR |U∗ ). As previously discussed, we know that
gling gates increases the complexity. both complexities now satisfy the following inequality
C(ψT , ψR |U∗ ) > C(ψT , ψR |U0 ). (73)
V. PENALIZED GEOMETRY II: FIELD Next, let us assume that we construct an artificial target
THEORY ∗
state ψT which satisfies the implementation
∗
In this section, we study the effect of introducing |ψT i = U0 |ψR i (74)
weighting factors for the entangling gates in the regulated in unpenalized geometry, but yields a different complex-
field theory placed on the lattice. As we have stressed ity C(ψT , ψR |U∗ ). Notice that the target state |ψT i de-
above, for the two coupled oscillators, a direct attempt pends on the physical parameters ω and δ by construc-
via analytically finding the minimal geodesic in full pe- tion. Now, if we keep the frequency ω unchanged and
nalized geometry is already very challenging. However, demand (74), it is possible to fulfill such an implemen-
certain assumptions allowed to simplify the problem un- tation if we introduce a modified parameter δ∗ which is
der consideration. Even though the exact geodesic could assigned to the mentioned state |ψT ∗
i. Then, we may re-
not be analytically derived for the simplified case as well, place the original cutoff scale δ in the optimal circuit U0
an approximate segmented path has been constructed by δ∗ . This will simply allow deriving the complexity
which comes very close to the optimal geodesic. This in the penalized geometry by using the optimal geodesic
has been explicitly illustrated by comparing the corre- in the unpenalized geometry. The procedure works for
sponding complexities [59]. the states under consideration, since both are taken to
For the generalization on the lattice, the question is, be positive quadratic forms. As we will show soon, such
how to find the the optimal geodesic in the extended ge- a replacement will regulate the desired change in com-
ometry where a higher cost is attributed to the entangling plexity. It is important to note that this replacement is
gates? For instance, one may argue that by changing to rather technical. The cutoff scale in the theory will still
the normal mode basis one could consider a perturba- be determined by δ.
tion around the previous straight line path, since this
particular solution yields the optimal geodesic in unpe-
nalized geometry for all Fκ costs where κ ≥ 1. In general, A. Modified parameter
this strategy will not simplify the problem either, since
introducing a penalty tensor in one basis does not auto- In the following, we explicitly illustrate the described
matically lead to the same penalty tensor when working strategy from above for the system of two coupled os-
in a different basis. However, for certain choices such as cillators subjected to the Fκ=2 cost. For reasons which
the F2 and Fκ=2 costs, as we will see, the straight line will become clear below, we cannot use the expression
circuit may indeed be used in normal mode basis. Let in (69) for the optimal circuit depth in the penalized ge-
us bring to mind that for these the complexity is basis ometry which we again denote by Dκ=2 p
. Recall that the
independent if no extra cost is assigned to the entangling result in (69) is based on the segmented path approach
gates, see section III. which is only valid when the weighting factor introduced
In the following, we use the results already discussed in in (58) satisfies α  y1 , ρ1  1. A more general expres-
the unpenalized case for deriving the complexity without sion where the range of α is not restricted from below has
solving the geodesic equations explicitly. For clarifying been worked out for the F2 cost [59]. According to our
the underlying procedure, we start again with the simple previous findings, we may also use this expression in the
case of two coupled oscillators. Please note, that all the present case by simply squaring it. Proceeding in this
following steps are described for the F2 cost. But we way yields the desired circuit depth which takes the form
know from earlier discussions that the findings can also
α 2
be applied to the Fκ=2 cost which will be our main choice
p 
p
Dκ=2 =2 tan−1 α2 − 1 + ρ1 + 2y12 . (75)
later. 2
 11

Of course, for α → 1 it reduces to the straight line circuit where we have defined
depth D(U0 ) which equals to the square of (24). √
K
(ω0 δ)2 2 + δ2 ω2
 
Having specified this, what remains to be done is find- e 2
R≡ √ , K ≡ log . (82)
ing the solution δ∗ which solves the equation Dκ=2 = 23 K δ 2 ω02
p
Dκ=2 . Note that the right-hand side of the latter
equation has to be written in terms of the functions Suppose that we fix the reference frequency ω0 ∼ e−σ /δ
y1 (δ, ω, ω0 ) and ρ1 (δ, ω) where we recall from (25) that at some scale ∼ 1/δ. The overall numerical parameter
r ! σ ensures that ω0 > 1/δ. Furthermore, we would like
1 ω2 2 to work in the UV approximation described in section
y1 (δ, ω, ω0 ) = log 1+ 2 2 , III C 1. Using the corresponding assumptions, we may
4 ω02 δ ω
r ! (76) estimate K ∼ − log2 (δω0 ). Observe that the dimension-
1 2 less factor R in (82) is positive in this regime. Also, χ is
ρ1 (δ, ω) = log 1+ 2 2 . taken to be sufficiently small as described above. There-
4 δ ω
fore, we find that the expression in the brackets from (81)
Contrary to that, the functions y1 and ρ1 on the left- decreases as soon as the weighting factor α becomes dif-
hand side have to be written in terms of δ∗ . Finally, ferent from one, i.e. δ∗ < δ. This change increases the
using the square of the straight line circuit depth in (24), complexity due to
the explicit equation we need to solve takes the form
y1 (δ∗ , . . .) > y1 (δ, . . .),
(83)
y12 (δ∗ , ω, ω0 )
+ =ρ21 (δ∗ , ω) y12 (δ, ω, ω0 ) ρ1 (δ∗ , . . .) > ρ1 (δ, . . .),
α p  2 (77)
+ tan−1 α2 − 1 + ρ1 (δ, ω) . see (76). This trend is for sure expected according to the
2 nonperturbative expression from (75).
This equation can be analytically solved. An appropriate Up to a numerical factor N multiplied with R from
treatment yields the following result (82), which is not relevant for our discussion later (thus
we set N ≡ 1), we may also write an expansion of the
 s   −1/2 following form
2+δ 2 ω 2
√ χ(α,ω,δ)+log2 2
δ 2 ω0
δ∗ = 2 ω02 e − ω2  (78)
δ∗ ' δ 1 − Rα̃ + O(α̃2 )


(84)
√
where we have defined where α̃ := α − 1. Writing δ∗ ' δ(. . .) as in (84) will al-
 p  low to make observations more comprehensible in certain
χ(α, ω, δ) := 4 α arctan α2 − 1 cases.
α p   (79)
× arctan α2 − 1 + 2ρ1 (ω, δ)
2 C. On the lattice
due to practical reasons. The expression in (78) is our key
result which will be important in the remainder of this 1. N = 2, unpenalized
section. Of course, once we take the unpenalized limit,
the only α dependent part in (78) vanishes, i.e. χ → 0, For the moment, let us consider a periodic lattice with
and we end up with the original cutoff scale δ, N = 2. The related Hamiltonian describing this system
takes the following form
lim δ∗ = δ. (80)
α→1  
1 2
H= p20 + p21 + ω 2 (x20 + x21 ) + 2 (x0 − x1 )2 (85)
2 δ
B. Small penalty
where we have imposed the boundary condition x2 ≡ x0 .
Next, we write the Hamiltonian from (85) in terms of
To highlight the effect of assigning a higher cost to the
entangling gates, it is advantageous to use the pertur- √
δ̄ := δ/ 2 (86)
bative expansion of δ∗ for small penalties. This can be
achieved for moderate α. More precisely, the prefactor which just gives rise to the same Hamiltonian as intro-
in front of 2ρ1 in (79) which only depends on the weight- duced in (5). Recall that according to our previous dis-
ing factor has to be sufficiently small, i.e. χ  1. This cussion in section V A, the optimal circuit depth Dκ=2
can be controlled with a weighting factor that is α 6= 1, from (34) satisfies the relation
but close enough to one. In this case, we may perturb δ∗
around χ = 0 which yields 1X 2
1 
ω̃k

2y12 (δ̄, ω, ω0 ) + 2ρ21 (δ̄, ω) = log (87)
δ∗ ' δ 1 − Rχ + O(χ3 ) 4 ω0


(81) k=0
 12

where the corresponding normal mode frequencies are Analogous to the expression in (84), we may also write
given as an approximation of the form
 
ω̃02 = ω 2 , ω̃12 = ω 2 + 2/δ̄ 2 . (88) δ̆∗ ' δ 1 − R̆α̃ + O(α̃2 ) . (94)
Now, based on the discussion in section III B 2, it is The leading order contribution to the complexity in the
straightforward to generalize these results for the d di- small α̃ limit then becomes
mensional case. For doing so, we may write !
p V 2 1
V = (N δ)d−1 , (89) Dκ=2 ∼ d−1 log . (95)
δ δω0 (1 − R̆α̃)
although we should note that N has been fixed above.
Then, working in the UV approximation, i.e. ω̃~k → 1/δ̄,
we end up with the following leading order contribution 3. Arbitrary N , penalized
to the complexity
  We begin by noticing that the complexity in the unpe-
V 1 nalized case is basis independent for the F2 cost, i.e.
Dκ=2 ∼ d−1 log2 . (90)
δ δ̄ω0

∗ ˜ ∗
 
˜
ds2 = δIJ dY I dY J = δI˜J˜dY I dY J . (96)
Of course, the expression in (90) has been derived in
unpenalized geometry.
In general, introducing a penalty tensor, i.e. δIJ → GIJ ,
makes the metric depending on the choice of the basis.
2. N = 2, penalized That means, choosing a specific metric GIJ in the origi-
nal position basis would result in a different metric GI˜J˜
in the normal mode basis. For the present setup, where
The previous findings can be easily extended to the pe- the entangling gates are uniformly penalized with the
nalized case when α > 1. We have seen that the periodic weighting factor α2 , it can be shown that
lattice with N = 2 reduces to the system of two cou-
pled oscillators if δ̄ from (86) replaces the original cutoff GI˜J˜ = R bT ˜
b ˜ GIJ R (97)
IJ IJ
scale δ. Thus, we may √use the previous straight line solu-
tion and insert δ̆∗ = 2δ∗ into the corresponding places. yields a matrix which takes the form
In the mentioned UV approximation, this would simply
(1 + α2 )1 + Goff,I˜J˜, (98)
mean that ω̃~k → 1/δ̆∗ . Note that according to (80), we
√
get limα→1 δ̆∗ = 2δ which is in agreement with the defi- see e.g. [59]. The matrix Goff,I˜J˜ is symmetric and only
nition in (86). The complexity in the penalized geometry consists of off-diagonal entries of the form 1 − α2 which
then takes the form do not distinguish between the different classes of gates.
We should add that R b ˜ denotes the generalized matrix
 
p V 1 IJ
Dκ=2 ∼ d−1 log2 . (91) with respect to the generators of GL(N, R), see section
δ δ̆∗ ω0
III A 2. So, the expression in (98) shows that in normal
As in the preceding discussion, this expression has been mode basis there is no extra cost attributed to the entan-
derived by fixing N = 2. Even though (91) provides use- gling gates relative to the scaling gates. Moreover, the
ful information on how (uniformly) weighting a certain dominant contribution to the metric GI˜J˜ is determined
class of gates (here, these are the entangling gates) would by δIJ multiplied with the prefactor 1 + α2 . Note that,
influence the complexity in the regulated field theory, we except the difference due to the appearance of the prefac-
still need to extend the results for an arbitrary number tor, this is similar as in the unpenalized geometry where
of oscillators N . In the subsequent part, it will be argued GIJ = δIJ .
that the relation from (91) also applies in the general case In the unpenalized case both the ground state ψT and
for any N . the reference state ψR turn out to be factorized Gaussian
As in the unpenalized case, let us close this part with states in normal mode basis, i.e.
expanding the α dependent parameter δ̆∗ for small penal-  ω N/4  
0 1
ties which yields ψR (x̃k ) = exp − x̃† ÃR x̃ ,
  π 2
δ̆∗ ' δ 1 − R̆χ + O(χ3 ) (92) N −1
Y ω̃k  1/4   (99)
1
ψT (x̃k ) = exp − x̃† ÃT x̃ ,
π 2
where k=0
√
e K̆
(ω δ)2

4 + δ2 ω2
 with ÃR = ω0 1, ÃT = diag[ω̃0 , . . . , ω̃N −1 ] and x̃ = RN x
R̆ := p0 , K̆ := log2 . (93) where RN denotes the unitary matrix generalizing R2 in-
24 K̆ δ 2 ω02 troduced in section III A 2. Such a factorization simplifies
 13

†
the optimal circuit Ũ0 = RN U0 RN in form of a diagonal simplifying the underlying expressions, let us assume11
matrix which amplifies each of the diagonal entries in δω0 . 1 for the remainder of this discussion.
the quadratic forms. As a consequence, the resulting ge- According to our assumptions, we may in addition
ometry of the normal mode subspace remains flat, since drop the leading term in the first line of (100) as well
the generators MI˜ = R b ˜ MJ which construct the opti-
IJ
as the factor 2 in front of the second term which is as-
mal circuit commute with one another. This property sumed to be dominating. By doing so, we may further
explains why the straight line solution yields the optimal estimate
geodesic for the regulated field on the lattice.    
V 1 1
Similarly, we may also expect that the optimal circuit ∆D ' d−1 log log . (101)
δ 1 − R̆α̃ δω0
in the penalized geometry is a diagonal matrix. This goes
back to observation that the metric in normal mode ba- If we now carefully look at the right-hand side of (101),
sis has the same shape as in the unpenalized case, i.e. we notice that this expression is proportional to the com-
GI˜J˜ ∼ δI˜J˜. Thus, we may again use the straight line plexity one would obtain for the Fκ=1 cost. Of course,
solution. The additional change for α > 1 can be incor- the absolute value bars in Dκ=1 can be neglected due to
porated by making the replacement δ → δ̆∗ as illustrated δω0 . 1, see also discussion below. Specifically, defining
in section V C 2. Everything so far has been discussed for Θ := − log(1 − R̆α̃) where 0 < Θ  1, the relation from
the F2 cost. From earlier studies we know that the latter above can equivalently be written as
aspects also apply for the Fκ=2 cost. Proceeding with
the latter, the dominant contribution to the complexity p
Dκ=2 ' ΘDκ=1 + Dκ=2 . (102)
for the regulated field theory is then given by (91). The
corresponding small α̃ expansion of (91) has been intro- Again, we would end up with the unpenalized complex-
duced in (95). ity if Θ = 0. Recall that the Fκ=1 cost yields the
same leading order UV divergence as found via the holo-
graphic complexity proposals [59]. Even more, this par-
ticular cost—which gives rise to the so-called Manhattan
D. Unpenalized vs penalized metric—leads to the same logarithmic factor predicted
on basis of the CA proposal, see section III C 2.
Let us here note that an interesting relation between
Finally, once the complexity for the regulated field in the F2 and Fκ=1 costs has recently been found by explor-
penalized geometry is given, we can compute the differ- ing Nielsen’s geometric approach to circuit complexity for
p
ence ∆D := Dκ=2 − Dκ=2 . From previous findings we two dimensional CFTs.12 It has been shown that the dif-
expect the latter to be positive semidefinite. The corre- ference between distinct metrics on the Virasoro circuits
sponding complexities for both geometries can be found turns out to be vanishing in the holograhic large central
in (91) and (55). Using the small α̃ expansion from (95) charge limit [86].
for the former penalized one in (91), we find that
Coming back to the present case, we can also write
     Dκ=2 = ΦDκ=1 by simply defining Φ := − log(δω0 ).
V 1 1 This is possible, since by assuming δω0 . 1 we can write
∆D ' log + 2 log
δ d−1 1 − R̆α̃ δω0 | log( δω1 0 )| = Φ > 0 as noted before. Following these steps
  (100) and replacing Dκ=2 in (102) appropriately then yields
1
× log .
1 − R̆α̃ p
Dκ=2 ' (Θ + Φ)Dκ=1 . (103)

Of course, in the unpenalized limit, the expression in This relation between complexities for two different costs
(100) vanishes, i.e. limα̃→0 ∆D = 0. As soon as the is interesting. It somewhat resembles the situation de-
gates are penalized, means when α > 1 or α̃ > 0, respec- scribed above in which a relation between two different
tively, the difference in the optimal depth starts to grow, costs has been found as well. Referring to the approx-
similar to the complexity increase in the two coupled os- imate expressions in (102) and (103), we may therefore
cillator case from section V B. It should be noted that ask whether weighting certain classes of gates—in par-
the sign of the approximate ∆D from (100) seems to de- ticular the entangling one, since these also play an im-
pend on O(δω0 ). This basically goes back to the small α̃ portant role in holography related aspects—for a specific
expansion used in (100) for which we need to distinguish cost can compensate the geometry change in the space of
between the different cases. For δω0 . 1, we obviously unitaries for a different cost.
find ∆D & 0 as we would expect. In case of δω0 & 1,
we get a negative second term in the first line, but the
leading positive term will ensure that ∆D is still positive.
However, if R̆α̃ in the argument of the leading logarithm 11 For instance, in cMERA, it is usually assumed that O(δω0 ) ∼ 1.
12
becomes too small, the logarithmic factor in the second The corresponding submanifolds are associated to the underlying
line of course decreases either so that ∆D → 0. For symmetry groups.
 14

Of course, our previous estimations rely on certain sim- engenders that the IR contributions to the complexity
plifications. More detailed studies concerning such as- depend on the UV cutoff scale δ.
pects may be important, particularly, if we presume that To avoid this feature, one may assume ω0  1/δ and
Fκ=1 is the most holographic like cost which might be identify ω0 with some IR frequency [59]. However, in
intrinsic to the CV and CA proposals. For that reason, this case, the additional logarithmic factor will again con-
we are therefore confronted with the question whether tribute to the leading divergence in the limit δ → 0.
weighting gates in a certain way are necessary for a con- As we can see, there are certain ambiguities in the dis-
sistent definition of complexity in QFT. Moreover, what cussions above. An important point is that the scaling
would be the corresponding holographic interpretation? of the complexity clearly depends on the explicit choice
On the other hand, let us bring to mind that notions for the reference frequency ω0 . The dependence on any
such as reference, geometry or gate set are not explicitly reference state parameter is obviously unsatisfying from
set in the holographic CV and CA proposals. Instead, the holographic point of view, since in the AdS/CFT dic-
they appear to be implicit in the duality. Therefore, iden- tionary this choice is not explicit (this is not so for the
tifying possible similarities between QFT and holography target state which in contrast has an holographic inter-
along these lines, might serve valuable information and pretation, see e.g. [57]) and we do not exactly know how
shed light on the role of gate weighting. to get access to it at all.
Let us return to the UV approximation in the penal-
ized geometry for the Fκ=2 cost from (91). Assigning a
E. Divergence higher cost to the entangling gates introduces an addi-
tional degree of freedom which is the weighting factor α.
Resorting to the UV approximation, we have seen that Contrary to the frequency ω0 , whose explicit value plays
there appears an additional logarithmic factor in the a substantial role in the unpenalized case, the factor α
complexity, see expression (55). In the limit δ → 0, this does not characterize the reference state. It specifies the
is the contribution which diverges much faster than the metric in the manifold of unitaries. Hence, instead of fix-
prefactor V δ 1−d , for more details see also [59]. In order ing the reference frequency and thus the reference state
to eliminate the additional divergence, one may, for in- in a certain way, we may simply eliminate the additional
stance, assume that the reference frequency ω0 is fixed at logarithmic contribution by tuning α correspondingly.
some UV scale with ω0 ∼ e−σ /δ. Again, the numerical To derive the explicit form of α, one may solve a similar
factor σ would ensure that ω0 > ω~k . We should note equation as before, i.e.
that these estimations are of course very rough and are
only meant to provide some hint about the approximate δ̆∗ ω0 = e−ν (107)
scaling behavior. A more general approach is presented where ν is again taken to be some overall numerical fac-
in [59] for the unpenalized κ = 1 measure. Nevertheless, tor. The leading contribution to the complexity would
as a first attempt, we would like to continue in the fol- then read14
lowing with the more simple estimations.13 Inserting the
V
explicit choice for ω0 from above into (55) yields CUV,κ=2 ∼ ν 2 , (108)
δ d−1
V
CUV,κ ∼ σ κ , (104) as in (104), i.e. without the additional logarithmic diver-
δ d−1 gence dominating in the limit δ → 0. Thus, avoiding any
so that the additional divergence is eliminated. As we can specific choice for ω0 , a simple power law for the leading
see, the leading order UV divergence becomes a power divergence multiplied by an arbitrary factor can be real-
law. On the other hand, if we suppose to work in the IR ized via choosing α to be the solution of (107). In the
approximation by making the following simple assump- mentioned UV approach with m  1/δ, it can be shown
tion that the weighting factor is determined by
 2
  
ω̃~k ∼ m, (105) p log2 e−2ν
4
+m ω 2 − log2 δ24ω2
α tan−1 ( α2 − 1) ' 0 0
.
log δ22m2


then the IR contribution to the complexity takes the form
(109)
CIR,κ ∼ − logκ (mδ) (106)
Of course, in the unpenalized case, i.e. α = 1, in other
after having assumed mδ  1. From the approximation words if we suppose ω0 to be the only tunable parameter,
above we get that the cancellation of the additional factor we simply get ν = σ according to (109).

13 Note that, at least, referring to the κ = 1 measure, the simple 14 As in the unpenalized case, although this rough estimation is in
estimation here still leads to the same scaling behavior as present the first instance sufficient to deduce the characteristic leading
in the more general result derived by computing the sum over all order scaling, as well as the subleading IR contribution, a more
normal mode frequencies ω̃~k , cf. [59]. detailed investigation will be subjected elsewhere.
 15

Similarly, we may focus on the IR contribution in the in the same way. In order to do so, we may for instance
penalized case. It should be noted that for the latter we demand (107) to be fulfilled. In the present perturba-
would again end up with (106) as no replacement of the tive approximation (with respect to α̃), we would need
scale parameter by δ̆∗ is necessary. However, referring to solve
to the generalizations derived for the unpenalized Fκ=1
cost by summing over the normal mode frequencies [59], δω0 (1 − R̆α̃) = e−ν (113)
we may expect some subleading UV contributions to the
complexity which would roughly scale as
to find the required α which eliminates the additional log-
CUV-sub,κ ∼ (m∗ δ)2κ (110) arithmic factor. The latter would then yield the leading
order contribution
in the small
 α̃ limit. Let usemphasize that, just formally,
m∗ ≡ m 1 − R̆α̃ + O(α̃2 ) may be seen as some effective V
CA ∼ ν 2 . (114)
mass which reduces to m when the entangling gates are δ d−1
not penalized, i.e. in the limit α̃ → 0.
As mentioned, the dimensionless coefficient Λ is re-
lated to the null normals on the boundary of the WDW
F. Comparison with CA patch in the bulk. To eliminate the curvature scale of
the bulk spacetime LAdS , we have set Λ as in (111). Re-
We can also compare the result in (91) with the pre- ferring to such a relation between Λ and α (note that
diction (56) based on the CA proposal where an extra α̃2 = α − 1), we may ask how weighting a certain class
logarithmic factor arises due to the asymptotic joint con- of gates would be translated via holography. Of course,
tributions in the WDW action [28]. Recall that in the the present studies are clearly too premature to speculate
unpenalized case, the QFT prediction (55) follows if the towards this direction, but the found relations from the
dimensionless coefficient Λ is set equal to ω0 LAdS . This present and the previous sections may motivate further
choice may be motivated by the fact that CA should be investigations.
independent of the AdS curvature scale LAdS when the Note that our findings have been obtained for a reg-
complexity is defined in the CFT. ulated free field. It is clear that this is far from deal-
In the penalized case, the comparison between the log- ing with strongly coupled theories with a large number
arithmic factors becomes even more interesting. As be- of degrees of freedom with existent holographic duals.
fore, we may identify the CFT cutoff scale in (56) with However, choosing the cost appropriately, the resulting
the scale parameter δ appearing in (91). Then, taking, similarities with holographic complexity proposals for the
for instance, the small α̃ expansion from (95), we may leading order UV divergence may provide useful insights.
relate both contributions to each other. The cancellation Namely, as noted above, such an accordance may suggest
of the bulk curvature scale LAdS may be realized if we which cost function is intrinsic to the holographic com-
set plexity conjectures. Indeed, this was actually the main
reason behind utilizing the Fκ costs.
Λ = (1 − R̆α̃)ω0 LAdS . (111)
Despite the mentioned concordance, it will actually be
Inserting Λ from (111) into the general expression (56) necessary to extend the notion of complexity for strongly
then yields coupled theories with holographic duals, or at least,
! for some suitably excited states going beyond the free
V 1 case. As mentioned, in the context of CFTs, interesting
CA ∼ d−1 log . (112) progress has been done in [86]. Note also the recent stud-
δ δω0 (1 − R̆α̃)
ies in [77–79] introducing a notion of complexity which
In contrast to (57), we end up with an additional contri- might shed light on this issue as well. Beyond these, cir-
bution in the denominator which is proportional to R̆α̃ cuit complexity has recently been studied for the TFD
and thus depends on the weighting factor α. Of course, state in a free scalar field theory [69], also see [87] for
the expression (112) still depends on the unspecified ref- similar considerations. The TFD state formed by en-
erence frequency ω0 . At least when gate penalties are not tangling two copies of a CFT is especially important in
implemented, we have encountered that this turns out to holography, since, as already pointed out below figure 1,
be a general feature of the notion of circuit complexity it is dual to an eternal BH in AdS [88].
in QFT. Recall that in the unpenalized situation, elim- Thus, it would be interesting to investigate the role of
inating the additional logarithmic contribution required gate penalties in more holographic like setups in order to
fixing ω0 at an appropriate scale. work out their possible holographic interpretation. On
However, introducing gate penalties, we have shown the other hand, as stressed before, such systems would
that tuning the weighting factor appropriately can elimi- particularly be interesting for making direct comparisons
nate the logarithmic contribution as well. Now, the loga- with recent studies arguing that holographic complexity
rithmic contribution in CA from (112) can be eliminated might be nonlocal [42].
 16

G. Distance dependent penalties graphic proposals due to the found similarities, such ex-
pectations may indeed favor the existence of bilocal gates
To find out what type of conditions are generically in the gate set for holographic states as argued in [42].
needed to be fulfilled by a field theory so that it has a Apart from that, the existence of bilocal gates in recent
bulk dual is important for better understanding the ba- studies of circuit complexity for interacting [65] and time
sic mechanisms behind holography. In previous studies, dependent nonlocal states [70] may explain the found be-
interesting conditions have been worked out by analyzing havior which is in line with the holographic predictions.
the corresponding boundary CFT, see e.g. [89, 90]. It remains interesting to find out whether and how imple-
More recently, it has been discussed that investigting menting distance dependent gate penalties would affect
the complexity growth for certain type of gauge theo- such agreements.
ries may also shed light on such necessity conditions [91].
The motivation here is the conjectured criterion for the
existence of a dual BH, namely, an upper bound in com- VI. CONCLUSION
plexity with the scaling behavior Cmax ∼ eS where the
exponent S denotes the corresponding entropy. Indeed, We have investigated the effects of weighting certain
it has been shown that for both the classical as well as classes of quantum gates constructing a quantum circuit
quantum cases, the complexity first grows and saturates for states in QFT. Our studies are based on the geomet-
afterwards at the maximum Cmax being in accordance ric approach to circuit complexity proposed by Nielsen
with the second law of complexity [24, 30]. More pre- and collaborators. Introducing such gate penalties may
cisely, it has been found that the speed of growth in- be motivated from various perspectives. For instance,
creases when the time evolution is more nonlocal. It has they may incorporate the notion of locality in complex-
been found that only the maximally nonlocal gauge the- ity when the latter is taken to be a physical attribute of a
ories satisfy the mentioned scaling behavior from above QFT. On the other hand, they may have important im-
and thus possess the possibility of a gravity dual. Ac- plications in the holographic dictionary. Closely related,
cordingly, the more nonlocal the interaction is the higher also an application for tensor networks which are believed
is the possibility for the existence of an holographic dual to provide a representation of a time slice of the AdS bulk
of the underlying theory. space may motivate considering a penalized geometry in
On the other hand, as already mentioned, it has been the space of unitaries.
discussed that complexity holographically defined via the In the present work, we have not introduced dis-
CV and CA proposals may be nonlocal as well, but in a tance dependent weighting factors. Instead, we have
slightly different manner. To be more specific, it has been worked with entangling gates which have been uniformly
argued that any gate set for a CFT defining holographic weighted. Of course, this does not introduce the notion
complexity should necessarily contain bilocal gates act- of locality as described above. Nevertheless, an imple-
ing at arbitraryly separated points [42]. Viewing the both mentation of such kind has enabled us obtaining interest-
conjectures as possible entries in the holographic dictio- ing insights which may motivate further investigations, in
nary, the presence of such bilocal gates constructing the particular, by examining similar ideas for strongly cou-
corresponding field theory state may be taken as another pled theories with holographic duals. Our results have
necessity condition, similar to the one described above. generalized earlier findings dealing with the case of a pair
In the present paper, we use such kind of bilocal gates of two coupled harmonic oscillators. More specifically, we
as well. However, we implement gate penalties for the en- have made the extension to the case of a regulated free
tangling gates where those are uniformly weighted, thus, field theory placed on the lattice for which the optimal
distinguishing between the different gate classes. It turns circuit acts in form of a representation of GL(N, R).
out that as long as we insert such distance independent Comparing both the penalized and the unpenalized
penalties, we obtain a similar divergence structure in the results with each other, we have seen that assigning a
complexity as seen in holographic computations, see (91). higher cost to the entangling gates gives rise to a sub-
However, following the discussion in section V, this kind stantial increase in complexity. Furthermore, using the
of agreement might not appear if certain gates are penal- complexities for the Fκ=2 cost function, we have found an
ized according to the distance of points they act on. Even interesting relation between these and the complexity for
more, one may for instance think of entirely preventing the unpenalized Fκ=1 cost which is known to give rise to
such gates from participating in the construction of the the Manhattan metric. From earlier studies we know that
circuit. In either cases, the characteristic power law scal- Fκ=1 behaves as the most holographic like cost, since it
ing of the leading order UV divergence CUV,κ ∼ V δ 1−d , yields the same leading order UV divergence as obtained
but of course the additional logarithmic factor as well, via the holographic complexity proposals, including the
might be modified or even be rescinded completely. In- additional logarithmic factor as appearing in the CA pre-
troducing locality in the way as described above may diction. Referring to earlier results, the found relations
therefore lead to drastic differences between complexity have led us to speculate whether the change in the man-
in field theory and the holographic proposals. Taking into ifold of unitaries determined by a specific cost function
account that certain costs may be intrinsic to the holo- may be compensated by introducing a penalty tensor for
 17

a different cost. referring to our findings, the insertion of gate penalties

in the space of unitaries makes such connections become
In addition, we have exhibited how the gate weighting even more interesting.
modifies the leading order UV divergence in the com- We have also compared our field theory results with
plexity. In contrast to the unpenalized geometry, penal- the mentioned holographic proposals where we have given
izing the entangling gates has introduced an additional particular consideration to the predictions based on the
degree of freedom, the latter corresponding to the men- CA conjecture. By comparing both predictions with
tioned weighting factor. We have shown that appropri- each other, we have identified a connection between the
ately tuning this factor can eliminate the mentioned loga- weighting factor and quantities which are related to the
rithmic contribution. Importantly, this procedure turned WDW patch in the bulk AdS spacetime. In view of these
out to be independent from the choice for the reference findings, we have finally commented on certain specula-
frequency. Recall that in the penalized case, the latter tive expectations concerning the role of gate penalties in
is the only tunable control parameter. The dependence defining complexity for states in QFT and their possible
on any state parameter of this type is not explicit in the implications in the holographic context.
holographic proposals. Hence, it is of particular inter-
est that the divergence can be modified by tweaking the
weighting factor which specifies the metric in the space VII. ACKNOWLEDGEMENTS
of unitaries instead of characterizing the reference state.
In earlier studies it has been discussed that the geometric I would like to thank Arpan Bhattacharyya, Pawel Ca-
structure in the mentioned space may encode important puta and Tadashi Takayanagi for interesting conversa-
features of the holographic proposals. The main reason tions and helpful comments on this work. I also thank
for such ideas lies in the found similarities in the diver- Mario Flory and Joan Simon for earlier conversations
gence structure for some specific costs such as the Fκ one. and Tadashi Takayangi for the generous hospitality at
Indeed, these similarities may suggest which cost is in- the Yukawa Institute for Theoretical Physics at Kyoto
trinsic to the holographic complexity conjectures. Hence, University.

[1] J. M. Maldacena, Int. J. Theor. Phys. 38, 1113 (1999), [15] L. Susskind, Fortsch. Phys. 64, 44 (2016), [Fortsch.
[Adv. Theor. Math. Phys.2,231(1998)], arXiv:hep- Phys.64,24(2016)], arXiv:1403.5695 [hep-th].
th/9711200 [hep-th]. [16] D. Stanford and L. Susskind, Phys. Rev. D90, 126007
[2] G. ’t Hooft, Conference on Highlights of Particle (2014), arXiv:1406.2678 [hep-th].
and Condensed Matter Physics (SALAMFEST) Trieste, [17] D. A. Roberts, D. Stanford, and L. Susskind, JHEP 03,
Italy, March 8-12, 1993, Conf. Proc. C930308, 284 051 (2015), arXiv:1409.8180 [hep-th].
(1993), arXiv:gr-qc/9310026 [gr-qc]. [18] A. R. Brown, D. A. Roberts, L. Susskind, B. Swingle,
[3] L. Susskind, J. Math. Phys. 36, 6377 (1995), arXiv:hep- and Y. Zhao, Phys. Rev. D93, 086006 (2016),
th/9409089 [hep-th]. arXiv:1512.04993 [hep-th].
[4] M. Rangamani and T. Takayanagi, Lect. Notes Phys. [19] A. R. Brown, D. A. Roberts, L. Susskind, B. Swingle,
931, pp.1 (2017), arXiv:1609.01287 [hep-th]. and Y. Zhao, Phys. Rev. Lett. 116, 191301 (2016),
[5] S. Aaronson (2016) arXiv:1607.05256 [quant-ph]. arXiv:1509.07876 [hep-th].
[6] A. Almheiri, X. Dong, and D. Harlow, JHEP 04, 163 [20] J. Couch, W. Fischler, and P. H. Nguyen, JHEP 03, 119
(2015), arXiv:1411.7041 [hep-th]. (2017), arXiv:1610.02038 [hep-th].
[7] S. Ryu and T. Takayanagi, Phys. Rev. Lett. 96, 181602 [21] M. Alishahiha, Phys. Rev. D92, 126009 (2015),
(2006), arXiv:hep-th/0603001 [hep-th]. arXiv:1509.06614 [hep-th].
[8] S. Ryu and T. Takayanagi, JHEP 08, 045 (2006), [22] J. L. F. Barbon and E. Rabinovici, JHEP 01, 084 (2016),
arXiv:hep-th/0605073 [hep-th]. arXiv:1509.09291 [hep-th].
[9] V. E. Hubeny, M. Rangamani, and T. Takayanagi, JHEP [23] R.-G. Cai, S.-M. Ruan, S.-J. Wang, R.-Q. Yang, and R.-
07, 062 (2007), arXiv:0705.0016 [hep-th]. H. Peng, JHEP 09, 161 (2016), arXiv:1606.08307 [gr-qc].
[10] M. Van Raamsdonk, Gen. Rel. Grav. 42, 2323 (2010), [24] A. R. Brown, L. Susskind, and Y. Zhao, Phys. Rev. D95,
[Int. J. Mod. Phys.D19,2429(2010)], arXiv:1005.3035 045010 (2017), arXiv:1608.02612 [hep-th].
[hep-th]. [25] L. Lehner, R. C. Myers, E. Poisson, and R. D. Sorkin,
[11] N. Lashkari, M. B. McDermott, and M. Van Raamsdonk, Phys. Rev. D94, 084046 (2016), arXiv:1609.00207 [hep-
JHEP 04, 195 (2014), arXiv:1308.3716 [hep-th]. th].
[12] T. Faulkner, M. Guica, T. Hartman, R. C. My- [26] R.-Q. Yang, Phys. Rev. D95, 086017 (2017),
ers, and M. Van Raamsdonk, JHEP 03, 051 (2014), arXiv:1610.05090 [gr-qc].
arXiv:1312.7856 [hep-th]. [27] S. Chapman, H. Marrochio, and R. C. Myers, JHEP 01,
[13] T. Hartman and J. Maldacena, JHEP 05, 014 (2013), 062 (2017), arXiv:1610.08063 [hep-th].
arXiv:1303.1080 [hep-th]. [28] D. Carmi, R. C. Myers, and P. Rath, JHEP 03, 118
[14] L. Susskind, Fortsch. Phys. 64, 49 (2016), (2017), arXiv:1612.00433 [hep-th].
arXiv:1411.0690 [hep-th].
 18

[29] A. Reynolds and S. F. Ross, Class. Quant. Grav. 34, [59] R. Jefferson and R. C. Myers, JHEP 10, 107 (2017),
105004 (2017), arXiv:1612.05439 [hep-th]. arXiv:1707.08570 [hep-th].
[30] A. R. Brown and L. Susskind, Phys. Rev. D97, 086015 [60] M. A. Nielsen, arXiv preprint quant-ph/0502070 (2005).
(2018), arXiv:1701.01107 [hep-th]. [61] M. A. Nielsen, M. R. Dowling, M. Gu, and A. C. Do-
[31] Y. Zhao, Phys. Rev. D98, 086011 (2018), herty, Science 311, 1133 (2006).
arXiv:1702.03957 [hep-th]. [62] M. R. Dowling and M. A. Nielsen, Quantum Information
[32] M. Flory, JHEP 06, 131 (2017), arXiv:1702.06386 [hep- & Computation 8, 861 (2008).
th]. [63] R. Khan, C. Krishnan, and S. Sharma, Phys. Rev. D98,
[33] A. Reynolds and S. F. Ross, Class. Quant. Grav. 34, 126001 (2018), arXiv:1801.07620 [hep-th].
175013 (2017), arXiv:1706.03788 [hep-th]. [64] L. Hackl and R. C. Myers, JHEP 07, 139 (2018),
[34] D. Carmi, S. Chapman, H. Marrochio, R. C. Myers, arXiv:1803.10638 [hep-th].
and S. Sugishita, JHEP 11, 188 (2017), arXiv:1709.10184 [65] A. Bhattacharyya, A. Shekar, and A. Sinha, JHEP 10,
[hep-th]. 140 (2018), arXiv:1808.03105 [hep-th].
[35] J. Couch, S. Eccles, W. Fischler, and M.-L. Xiao, JHEP [66] D. W. F. Alves and G. Camilo, JHEP 06, 029 (2018),
03, 108 (2018), arXiv:1710.07833 [hep-th]. arXiv:1804.00107 [hep-th].
[36] R.-Q. Yang, C. Niu, C.-Y. Zhang, and K.-Y. Kim, JHEP [67] J. M. Magan, JHEP 09, 043 (2018), arXiv:1805.05839
02, 082 (2018), arXiv:1710.00600 [hep-th]. [hep-th].
[37] R. Abt, J. Erdmenger, H. Hinrichsen, C. M. Melby- [68] M. Guo, J. Hernandez, R. C. Myers, and S.-M. Ruan,
Thompson, R. Meyer, C. Northe, and I. A. Reyes, JHEP 10, 011 (2018), arXiv:1807.07677 [hep-th].
Fortsch. Phys. 66, 1800034 (2018), arXiv:1710.01327 [69] S. Chapman, J. Eisert, L. Hackl, M. P. Heller, R. Jef-
[hep-th]. ferson, H. Marrochio, and R. C. Myers, (2018),
[38] M. Moosa, JHEP 03, 031 (2018), arXiv:1711.02668 [hep- arXiv:1810.05151 [hep-th].
th]. [70] T. Ali, A. Bhattacharyya, S. Shajidul Haque, E. H. Kim,
[39] M. Moosa, Phys. Rev. D97, 106016 (2018), and N. Moynihan, (2018), arXiv:1810.02734 [hep-th].
arXiv:1712.07137 [hep-th]. [71] T. Ali, A. Bhattacharyya, S. Shajidul Haque, E. H. Kim,
[40] B. Swingle and Y. Wang, JHEP 09, 106 (2018), and N. Moynihan, (2018), arXiv:1811.05985 [hep-th].
arXiv:1712.09826 [hep-th]. [72] J. Jiang and X. Liu, Phys. Rev. D99, 026011 (2019),
[41] A. P. Reynolds and S. F. Ross, Class. Quant. Grav. 35, arXiv:1812.00193 [hep-th].
095006 (2018), arXiv:1712.03732 [hep-th]. [73] M. Sinamuli and R. B. Mann, (2019), arXiv:1902.01912
[42] Z. Fu, A. Maloney, D. Marolf, H. Maxfield, and Z. Wang, [hep-th].
JHEP 02, 072 (2018), arXiv:1801.01137 [hep-th]. [74] F. Liu, R. Lundgren, P. Titum, J. R. Garrison, and A. V.
[43] B. Chen, W.-M. Li, R.-Q. Yang, C.-Y. Zhang, and S.-J. Gorshkov, (2019), arXiv:1902.10720 [quant-ph].
Zhang, JHEP 07, 034 (2018), arXiv:1803.06680 [hep-th]. [75] S. Chapman, M. P. Heller, H. Marrochio, and
[44] S. Chapman, H. Marrochio, and R. C. Myers, JHEP 06, F. Pastawski, Phys. Rev. Lett. 120, 121602 (2018),
046 (2018), arXiv:1804.07410 [hep-th]. arXiv:1707.08582 [hep-th].
[45] R. Abt, J. Erdmenger, M. Gerbershagen, C. M. Melby- [76] H. A. Camargo, P. Caputa, D. Das, M. P. Heller,
Thompson, and C. Northe, JHEP 01, 012 (2019), and R. Jefferson, Phys. Rev. Lett. 122, 081601 (2019),
arXiv:1805.10298 [hep-th]. arXiv:1807.07075 [hep-th].
[46] K. Hashimoto, N. Iizuka, and S. Sugishita, Phys. Rev. [77] P. Caputa, N. Kundu, M. Miyaji, T. Takayanagi, and
D98, 046002 (2018), arXiv:1805.04226 [hep-th]. K. Watanabe, Phys. Rev. Lett. 119, 071602 (2017),
[47] S. Chapman, H. Marrochio, and R. C. Myers, JHEP 06, arXiv:1703.00456 [hep-th].
114 (2018), arXiv:1805.07262 [hep-th]. [78] P. Caputa, N. Kundu, M. Miyaji, T. Takayanagi, and
[48] M. Flory and N. Miekley, (2018), arXiv:1806.08376 [hep- K. Watanabe, JHEP 11, 097 (2017), arXiv:1706.07056
th]. [hep-th].
[49] J. Couch, S. Eccles, T. Jacobson, and P. Nguyen, JHEP [79] B. Czech, Phys. Rev. Lett. 120, 031601 (2018),
11, 044 (2018), arXiv:1807.02186 [hep-th]. arXiv:1706.00965 [hep-th].
[50] K. Goto, H. Marrochio, R. C. Myers, L. Queimada, and [80] T. Takayanagi, JHEP 12, 048 (2018), arXiv:1808.09072
B. Yoshida, (2018), arXiv:1901.00014 [hep-th]. [hep-th].
[51] S. Chapman, D. Ge, and G. Policastro, (2018), [81] G. Vidal, Phys. Rev. Lett. 101, 110501 (2008).
arXiv:1811.12549 [hep-th]. [82] J. Haegeman, T. J. Osborne, H. Verschelde, and
[52] Z.-Y. Fan and M. Guo, (2018), arXiv:1811.01473 [hep- F. Verstraete, Phys. Rev. Lett. 110, 100402 (2013),
th]. arXiv:1102.5524 [hep-th].
[53] S. S. Hashemi, G. Jafari, A. Naseh, and H. Zolfi, (2019), [83] J. Cotler, M. R. Mohammadi Mozaffar, A. Mollabashi,
arXiv:1902.03554 [hep-th]. and A. Naseh, (2018), arXiv:1806.02831 [hep-th].
[54] M. Flory, (2019), arXiv:1902.06499 [hep-th]. [84] B. Swingle, Phys. Rev. D86, 065007 (2012),
[55] R. Yang, H.-S. Jeong, C. Niu, and K.-Y. Kim, (2019), arXiv:0905.1317 [cond-mat.str-el].
arXiv:1902.07586 [hep-th]. [85] B. Swingle, (2012), arXiv:1209.3304 [hep-th].
[56] H. Guo, X.-M. Kuang, and B. Wang, (2019), [86] P. Caputa and J. M. Magan, (2018), arXiv:1807.04422
arXiv:1902.07945 [hep-th]. [hep-th].
[57] A. Bernamonti, F. Galli, J. Hernandez, R. C. Myers, S.- [87] R.-Q. Yang, Phys. Rev. D97, 066004 (2018),
M. Ruan, and J. Simn, (2019), arXiv:1903.04511 [hep- arXiv:1709.00921 [hep-th].
th]. [88] J. M. Maldacena, JHEP 04, 021 (2003), arXiv:hep-
[58] L. Susskind, Fortsch. Phys. 64, 72 (2016), th/0106112 [hep-th].
arXiv:1412.8483 [hep-th].
 19

[89] I. Heemskerk, J. Penedones, J. Polchinski, and J. Sully, [91] K. Hashimoto, N. Iizuka, and S. Sugishita, Phys. Rev.
JHEP 10, 079 (2009), arXiv:0907.0151 [hep-th]. D96, 126001 (2017), arXiv:1707.03840 [hep-th].
[90] S. El-Showk and K. Papadodimas, JHEP 10, 106 (2012),
arXiv:1101.4163 [hep-th].