Beruflich Dokumente
Kultur Dokumente
I. INTRODUCTION
A’
d−1
X preparing a resource state followed by a single round of
|Φd iAB = d−1/2 |iiA |iiB (2) LOCC [4, 8].
i=0 Due in part to the large experimental interest in
bosonic continuous-variable quantum systems, given
and two classical channels, each of dimension d, can be
their practical applications [12, 13], the teleportation pro-
used to simulate an ideal d-dimensional quantum chan-
tocol was extended to this paradigm [11] (see Figure 1).
nel [1]. The teleportation primitive has been extended in
The standard protocol begins with a sender and receiver
multiple non-trivial ways, including a method to simulate
sharing a two-mode squeezed vacuum state of the follow-
an unideal channel using a noisy, mixed resource state
ing form:
[4, Section V] (see also [5–8]) and as a way to implement
nonlocal quantum gates [9, 10]. The former extension ∞ r !n
has been used to bound the rates at which quantum in- 1 X NS
|Φ(NS )iAB ≡ √ |niA |niB ,
formation can be conveyed over a quantum channel as- NS + 1 n=0 NS + 1
sisted by local operations and classical communication (3)
(LOCC) [4, 8], and more generally, as a way to reduce where NS ∈ [0, ∞) represents the squeezing strength and
a general LOCC-assisted protocol to one that consists of {|ni}n denotes the photon-number basis. Suppose that
2
the goal is to teleport a mode A0 . The sender mixes, on a tocols that make use of infinite-energy states, such as the
50-50 beamsplitter, the mode A0 with the mode A of the Basel states in (20) and (124). Although there should be
state in (3). Afterward, the sender performs homodyne great skepticism concerning whether these infinite-energy
detection on the modes emerging from the beamsplitter Basel states could be generated in practice, this latter
and forwards the measurement results over classical chan- byproduct is nevertheless of theoretical interest.
nels to the receiver, who possesses mode B of the above The rest of the paper proceeds as follows. In the
state. Finally, the receiver performs a unitary displace- next section, I discuss the precise form of convergence
ment operation on mode B. In the limit as NS → ∞ and that occurs in continuous-variable quantum teleporta-
in the limit of ideal homodyne detection, this continuous- tion and then develop various extensions of this notion
variable teleportation protocol is often loosely stated in of convergence. I then prove that teleportation simula-
the literature to simulate an ideal channel on any state of tions of the pure-loss, thermal, pure-amplifier, amplifier,
the mode A0 , such that this state is prepared in mode B and additive-noise channels converge both strongly and
after the protocol is finished. Due to the lack of a precise uniformly to the original channels in the limit of ideal
notion of convergence being given, there is the potential squeezing and detection for the simulations. The uniform
for confusion regarding mathematical proofs that make convergence results are then extended to the teleporta-
use of the continuous-variable teleportation protocol. tion simulations of particular multi-mode bosonic Gaus-
With this in mind, one purpose of the present paper sian channels. Section III gives a physical interpreta-
is to clarify the precise kind of convergence that occurs tion of the aforementioned convergence results, by means
in continuous-variable quantum teleportation, which is of the CV Teleportation Game. After that, Section IV
typically not discussed in the literature on this topic. In briefly reviews what is meant by a secret-key-agreement
particular, I prove that the convergence is in the strong protocol and non-asymptotic secret-key-agreement ca-
topology sense, and not in the uniform topology sense pacity. Finally, in Section V, I review the proof of [15,
(see, e.g., [14, Section 3] for discussions of these notions Theorem 24] and carefully go through some of its steps
of convergence). I then show how to extend this strong therein, confirming its correctness, while showing how
convergence result to the teleportation simulation of n the strong convergence of teleportation simulation ap-
parallel ideal channels, and I also show how this strong plies. In Section VI, I conclude with a brief summary
convergence extends to the teleportation simulation of n and a discussion.
ideal channels that could be used in any context.
Strong convergence and uniform convergence are then
II. NOTIONS OF QUANTUM CHANNEL
discussed for the teleportation simulation of bosonic CONVERGENCE, WITH APPLICATIONS TO
Gaussian channels. For this latter case, and in contrast TELEPORTATION SIMULATION
to the result discussed above for the continuous-variable
teleportation protocol, I prove that the teleportation sim-
ulations of the pure-loss, thermal, pure-amplifier, ampli- One main technical issue discussed in this paper is
fier, and additive-noise channels converge both strongly how the continuous-variable bosonic teleportation pro-
and uniformly to the original channels, in the limit of tocol from [11] converges to an identity channel in the
ideal squeezing and detection for the simulations. Here I limit of infinite squeezing and ideal detection. This issue
give explicit uniform bounds on the accuracy of the tele- is often not explicitly clarified in the literature on the
portation simulations of these channels, and I suspect topic, even though it has been implicit for some time in
that these bounds will be useful in future applications. various works that the convergence is to be understood
After this development, I then extend these uniform con- in the strong sense (topology of strong convergence), and
vergence results to particular multi-mode bosonic Gaus- not necessarily the uniform sense (topology of uniform
sian channels. convergence) (see, e.g., [14, Section 3]). For example, in
the original paper [11], the following statement is given
These convergence results are important, even if they regarding this issue:
might be implicit in prior works, as they provide mean-
ingful clarification of mathematical proofs that make use “Clearly, for r → ∞ the teleported state
of teleportation simulation, such as those given in recent of Eq. (4) reproduces the original unknown
work on bounding non-asymptotic secret-key-agreement state.”
capacities. In particular, one can employ these con-
vergence statements to confirm the correctness of the Although it is clear that this statement implies conver-
proof of such bounds given in my joint work with Berta gence in the strong sense, it could be helpful to clarify
and Tomamichel from [15]. Furthermore, these strong this point, and the purpose of this section is to do so.
convergence statements can be used to conclude that In what follows, I first recall the definitions of strong
the energy-constrained diamond distance is not neces- and uniform convergence from [14, Section 3]. I then
sary to arrive at a proof of the bounds from [15]. An- discuss the precise form of convergence that occurs in
other byproduct of the discussion given in the present continuous-variable bosonic teleportation and show how
paper is that it is clarified that the methods of [15] al- strong convergence and uniform convergence are ex-
low for bounding secret-key rates of rather general pro- tremely different in the setting of continuous-variable
3
teleportation. After that, I prove that strong convergence that for every mixed state ρRA , there exists a pure state
of a channel sequence implies strong convergence of n- ψR0 A with the auxiliary Hilbert space HR0 taken to be
fold tensor powers of these channels and follow this with isomorphic to HA , such that
a proof that strong convergence of a channel sequence im-
plies strong convergence of n uses of these channels in any ε(k, ρRA ) ≤ ε(k, ψR0 A ). (8)
context in which they could be invoked. I also prove that
the teleportation simulations of pure-loss, thermal, pure- Thus, when considering strong convergence, it suffices to
amplifier, amplifier, and additive-noise channels converge consider only pure states ψR0 A , so that
both strongly and uniformly to the original channels, in
the limit of ideal squeezing and detection for the simula- sup lim ε(k, ρRA ) = sup lim ε(k, ψR0 A ). (9)
ρRA k→∞ ψR0 A k→∞
tions. The uniform convergence results are then extended
to the teleportation simulations of particular multi-mode Strong convergence is strictly different from uniform
bosonic Gaussian channels. convergence [14, Section 3], which amounts to a swap of
the supremum and the limit in (7). That is, the channel
k
sequence {NA→B }k converges uniformly to the channel
A. Definitions of strong and uniform convergence NA→B if the following holds
Before discussing the precise statement of convergence lim sup ε(k, ρRA ) = 0. (10)
k→∞ ρRA
in the continuous-variable bosonic teleportation protocol,
let us begin by recalling general definitions of strong and Even though this swap might seem harmless and is of no
uniform convergence from [14, Section 3]. I adopt slightly consequence in finite dimensions, the issue is important
different definitions from those given in [14, Section 3], in to consider in infinite-dimensional contexts, especially for
order to suit the needs of the present paper, but note that bosonic channels. That is, a sequence of channels could
they are equivalent to the original definitions as shown converge in the strong sense, but be as far as possible
in [14] and [16, Lemma 2]. In this context, see also [17]. from converging in the uniform sense, and an example of
k
Let {NA→B }k denote a sequence of quantum channels this behavior is given in the next subsection.
(completely positive, trace-preserving maps), which each It was also stressed in [14] that the topology of uniform
accept as input a trace-class operator acting on a separa- convergence is too strong for physical applications and
ble Hilbert space HA and output a trace-class operator should typically not be considered.
acting on a separable Hilbert space HB . This sequence
converges strongly to a channel NA→B if for all density
operators ρRA acting on HR ⊗ HA , where HR is an ar- B. Strong and uniform convergence considerations
bitrary, auxiliary separable Hilbert space, the following for continuous-variable teleportation
limit holds
lim ε(k, ρRA ) = 0, (4) We now turn our attention to convergence in the
k→∞ continuous-variable bosonic teleportation protocol and
focus on [11, Eq. (9)], which states that an unideal
where the infidelity is defined as continuous-variable bosonic teleportation protocol with
input mode A realizes the following additive-noise quan-
ε(k, ρRA ) ≡ tum Gaussian channel TAσ̄ on an input density opera-
k tor ρA :
1 − F ((idR ⊗NA→B )(ρRA ), (idR ⊗NA→B )(ρRA )), (5)
Z
idR denotes the identity map on the auxiliary space, and ρA → TA (ρA ) ≡ d2 α Gσ̄ (α) D(α)ρA D(−α), (11)
σ̄
√ √ 2
F (τ, ω) ≡ k τ ωk1 is the quantum fidelity [18], defined
for density operators τ and ω. The quantum fidelity
obeys a data processing inequality, which is the state- where D(α) is a displacement operator [13] and
ment that 2
!
1 |α|
Gσ̄ (α) ≡ exp − (12)
F (M(τ ), M(ω)) ≥ F (τ, ω), (6) πσ̄ σ̄
for states τ and ω and a quantum channel M. We can is a zero-mean, circularly symmetric complex Gaussian
summarize strong convergence more compactly as the fol- probability density function with variance σ̄ > 0. To
lowing mathematical statement: be clear, the integral in (11) is over the whole complex
plane α ∈ C. For an explicit proof of (11), one can also
sup lim ε(k, ρRA ) = 0. (7) consult [19, 20]. The variance parameter σ̄ quantifies
ρRA k→∞
unideal squeezing and unideal detection. Thus, for any
Due to purification, the Schmidt decomposition theorem, σ̄ > 0, the teleportation channel TAσ̄ is unideal and intu-
and the data processing inequality for fidelity, we find itively becomes ideal in the limit σ̄ → 0. However, it is
4
istic function of the reduced density operator ψA . By The drastic difference is due to the fact that for all fixed
applying the Hölder inequality, we conclude that χψA (α) k
σ̄ > 0, one can find a sequence of states {ψRA }k such that
is bounded for all α ∈ C because k
supk ε(σ̄, ψRA ) = 1, establishing (21). For example, one
k
could pick each ψRA to be a two-mode squeezed vacuum
|Tr{D(α)ψA }| ≤ kD(α)k∞ kψA k1 = 1. (15)
state with squeezing parameter increasing with increas-
Exploiting the continuity of χψA (α) at α = 0 and the ing k. In the limit of large squeezing, the ideal channel
fact that χψA (0) = 1 [21, Theorem 5.4.1], as well as in- and the additive-noise channel for any σ̄ > 0 become
voking the boundedness of χψA (α) and [22, Theorem 9.8] perfectly distinguishable, having infidelity approaching
regarding the convergence of nascent delta functions, we one, implying (21). One can directly verify this calcula-
then conclude that for a given state ψRA , the following tion by employing the covariance matrix representation
strong convergence holds of the two-mode squeezed vacuum and the additive-noise
channel, as well as the overlap formula in [13, Eq. (4.51)]
lim ε(σ̄, ψRA ) = 0, (16) to calculate entanglement fidelity.
σ̄→0 I now give details of the aforementioned calculation,
which can be written, as before, more compactly as regarding how the continuous-variable bosonic teleporta-
tion protocol from [11] does not converge uniformly to
sup lim ε(σ̄, ψRA ) = 0. (17) an ideal channel. Consider a two-mode squeezed vacuum
ψRA σ̄→0 state Φ(NS ) with mean photon number NS for one of
its reduced modes, as defined in (3). Such a state has a
Note that, as before and due to (9), Eq. (17) implies that Wigner-function covariance matrix [13] as follows:
sup lim ε(σ̄, ρRA ) = 0 (18) p
ρRA σ̄→0 p 2N S + 1 2 N S (NS + 1)
2 NS (NS + 1) 2NS + 1
for any mixed state ρRA where p
p2NS + 1 −2 NS (NS + 1)
⊕ . (22)
ε(σ̄, ρRA ) ≡ 1 − F (ρRA , (idR ⊗TAσ̄ )(ρRA )) (19) −2 NS (NS + 1) 2NS + 1
After sending one mode of this state through an additive-
and HR is an arbitrary auxiliary separable Hilbert space. noise channel with variance σ̄ (corresponding to an
One should note here that the convergence in (17) al- unideal continuous-variable bosonic teleportation), the
ready calls into question any claim regarding the neces- covariance matrix becomes as follows, corresponding to
sity of an energy constraint for the states that are to a state τ (NS , σ̄):
be teleported using the continuous-variable teleportation
protocol. Clearly, the state ψRA to be teleported could
p
be chosen as the following Basel state: p 2NS + 1 2 NS (NS + 1)
2 NS (NS + 1) 2NS + 1 + 2σ̄
∞ r
r p
6 X 1 2N S + 1 −2 NS (NS + 1)
|βiRA = |niR |niA , (20) ⊕ p . (23)
π 2 n=1 n2 −2 NS (NS + 1) 2NS + 1 + 2σ̄
The overlap Tr{ωσ} of two zero-mean, two-mode Gaus-
which has mean photon number equal to ∞, but it also sian states ω and σ is given by [13, Eq. (4.51)]
satisfies (16). Such a state is called a “Basel state,” p
due to its normalization factor being connected with Tr{ωσ} = 4/ det(Vω + Vσ ), (24)
5
where Vω and Vσ are the Wigner-function covariance ma- C. Strong and uniform convergence for
trices of ω and σ, respectively. We can then employ this tensor-power channels
formula to calculate the overlap
A natural consideration to make in the context of quan-
hΦ(NS )|τ (NS , σ̄)|Φ(NS )i = Tr{Φ(NS )τ (NS , σ̄)} (25) tum Shannon theory is the convergence of n uses of a
channel on a general state of n systems, where n is a
positive integer. To this end, suppose that the strong
as convergence in (7) holds for the sequence {NA→B k
}k of
channels. Then it immediately follows that the sequence
1 ⊗n
hΦ(NS )|τ (NS , σ̄)|Φ(NS )i = . (26)
k
{(NA→B )⊗n }k converges strongly to NA→B . Indeed, for
σ̄ + 2σ̄NS + 1 ⊗n
an arbitrary density operator ρRAn acting on HR ⊗ HA ,
we are now interested in bounding the infidelity for the
⊗n
Thus, for a fixed σ̄ > 0 and in the limit as NS → ∞, we
k
tensor-power channels (NA→B )⊗n and NA→B :
find that 1 − hΦ(NS )|τ (NS , σ̄)|Φ(NS )i → 1, so that the
continuous-variable bosonic teleportation protocol from ε(n) (k, ρRAn ) ≡
[11] does not converge uniformly to an ideal channel. ⊗n
k
1−F ((idR ⊗(NA→B )⊗n )(ρRAn ), (idR ⊗NA→B )(ρRAn )),
To summarize, the kind of convergence considered in (27)
(17) is the strong sense (topology of strong convergence),
whereas the kind of convergence considered in (21) is the in the limit as k → ∞. By employing the fact that
uniform sense (topology of uniform convergence) (see, p
e.g., [14, Section 3]). That is, (17) demonstrates that P (τ, ω) ≡ 1 − F (τ, ω) (28)
unideal continuous-variable bosonic teleportation con-
verges strongly to an ideal quantum channel in the limit obeys the triangle inequality [24–27], we conclude that,
of ideal squeezing and detection, whereas (21) demon- for an arbitrary density operator ρRAn , the following in-
strates that it does not converge uniformly. equality holds
⊗n
k
)⊗n )(ρRAn ), (idR ⊗NA→B
P (idR ⊗(NA→B )(ρRAn )
Xn
⊗n−i ⊗n−i+1
k
)⊗i ⊗ NA→B k
)⊗i−1 ⊗ NA→B
≤ P (idR ⊗(NA→B )(ρRAn ), (idR ⊗(NA→B )(ρRAn ) (29)
i=1
Xn
P (idR ⊗ id⊗i−1 k ⊗n−i
)(ρRAn ), (idR ⊗ id⊗i−1 ⊗n−i
≤ A ⊗NA→B ⊗ NA→B A ⊗NA→B ⊗ NA→B )(ρRAn ) . (30)
i=1
The first inequality follows from the triangle inequality This result can be summarized more compactly as
for P (τ, ω), and the second follows from data processing
for the fidelity under the channel sup lim ε(k, ρRA ) = 0
ρRA k→∞
k
idR ⊗(NA→B )⊗i−1 ⊗ id⊗n−i+1
A (31)
=⇒ sup lim ε(n) (k, ρRAn ) = 0. (35)
acting on the states ρRAn k→∞
⊗i−1 k ⊗n−i
(idR ⊗ idA ⊗NA→B ⊗ NA→B )(ρRAn ), (32)
That is, the strong convergence of the channel sequence
⊗i−1 ⊗n−i k
(idR ⊗ idA ⊗NA→B ⊗ NA→B )(ρRAn ). (33) {NA→B }k implies the strong convergence of the tensor-
power channel sequence {(NA→Bk
)⊗n }k for any finite n.
The method used in (29)–(30) is related to the telescop-
ing approach of [28], employed in the context of continu- For the case of continuous-variable teleportation, we
ity of quantum channel capacities (see the proof of [28, have that n unideal teleportations with the same perfor-
Theorem 11] in particular). Now employing strong con- mance corresponds to the tensor-power channel (TAσ̄ )⊗n .
vergence of the channel sequence {NA→B k
}k and the fact By appealing to (17) and (35), or alternatively employ-
⊗n−i ing Wigner characteristic functions, [21, Theorem 5.4.1],
that NA→B (ρRAn ) is a fixed state independent of k for
and [22, Theorem 9.8], the following convergence holds:
each i ∈ {1, . . . , n}, we conclude that for all density op-
for a pure state ψRAn , we have that
erators ρRAn
lim ε(n) (k, ρRAn ) = 0. (34) lim ε(n) (σ̄, ψRAn ) = 0, (36)
k→∞ σ̄→0
6
Again, more compactly, this is the same as Proposition 1 Let {NAk1 →B1 }k be a channel sequence
that converges strongly to a channel NA1 →B1 , and
let {MkA2 →B2 }k be a channel sequence that converges
sup lim ε(n) (σ̄, ψRAn ) = 0. (38) strongly to a channel MA2 →B2 . Then the channel se-
ψRAn σ̄→0
quence {NAk1 →B1 ⊗ MkA2 →B2 }k converges strongly to
NA1 →B1 ⊗ MA2 →B2 .
and drastically different from
Proof. A proof is similar to what is given above, and
I give it for completeness. Let ρRA1 A2 be an arbitrary
lim sup ε(n) (σ̄, ψRAn ) = 1. (39) state. Consider that
σ̄→0 ψ
RAn
P ((idR ⊗NAk1 →B1 ⊗ MkA2 →B2 )(ρRA1 A2 ), (idR ⊗NA1 →B1 ⊗ MA2 →B2 )(ρRA1 A2 ))
≤ P ((idR ⊗NAk1 →B1 ⊗ MkA2 →B2 )(ρRA1 A2 ), (idR ⊗NAk1 →B1 ⊗ MA2 →B2 )(ρRA1 A2 ))
+ P ((idR ⊗NAk1 →B1 ⊗ MA2 →B2 )(ρRA1 A2 ), (idR ⊗NA1 →B1 ⊗ MA2 →B2 )(ρRA1 A2 ))
≤ P ((idRA1 ⊗MkA2 →B2 )(ρRA1 A2 ), MA2 →B2 (ρRA1 A2 ))
+ P ((idR ⊗NAk1 →B1 ⊗ idA2 )(ρRA1 A2 ), (idR ⊗NA1 →B1 ⊗ idA2 )(ρRA1 A2 )). (40)
The first inequality follows from the triangle inequality and the second from data processing. Using the strong
convergence of {NAk1 →B1 }k and {MkA2 →B2 }k , applying the inequality in (40), and taking the limit k → ∞, we find
that
lim P ((NAk1 →B1 ⊗ MkA2 →B2 )(ρRA1 A2 ), (NA1 →B1 ⊗ MA2 →B2 )(ρRA1 A2 )) = 0. (41)
k→∞
Since the state ρRA1 A2 was arbitrary, the proof is com- in the rather general sense described below.
plete. To clarify what is meant by an adaptive protocol for
channel discrimination, suppose that the task is to dis-
k
tinguish n uses of the channel NA→B from n uses of the
D. Strong and uniform convergence in arbitrary channel NA→B . The most general protocol for doing
contexts so begins with the preparation of a state ρR1 A1 , where
system A1 is isomorphic to the channel input system A
The most general way to distinguish n uses of two dif- and R1 corresponds to an arbitrary auxiliary separable
ferent quantum channels is by means of an adaptive pro- Hilbert space. The system A1 is then fed in to the first
k
tocol. Such adaptive channel discrimination protocols channel use of NA→B or NA→B , depending on which of
have been considered extensively in the literature in the these channels is chosen from the start. The resulting
context of finite-dimensional quantum channel discrimi- state is then either
nation (see, e.g., [29–34]). However, to the best of my k
NA→B (ρR1 A1 ) or NA→B (ρR1 A1 ), (42)
knowledge, the issues of strong and uniform convergence
have not yet been considered explicitly in the literature depending on which channel is selected, and where I
in the context of infinite-dimensional channel discrimi- have omitted the identity map on R1 for simplicity. Af-
nation using adaptive strategies. The purpose of this ter this, the discriminator applies a quantum channel
(1)
section is to clarify these issues by defining strong and AR1 B1 →R2 A2 , where R2 corresponds to another arbitrary
uniform convergence in this general context and then to separable Hilbert space, which need not be isomorphic to
prove explicitly that strong convergence of a channel se- R1 , and A2 corresponds to a separable Hilbert space iso-
k
quence {NA→B }k to NA→B implies strong convergence morphic to the channel input A. The discriminator then
k k
of n uses of each channel in {NA→B }k to n uses of NA→B calls the second use of NA→B or NA→B , such that the
7
state is now either Figure 2 depicts these channels and states, which are used
in a general adaptive strategy to discriminate three uses
k
(1)
(NAk2 →B2 ◦ AR1 B1 →R2 A2 ◦ NAk1 →B1 )(ρR1 A1 ), (43) of NA→B from three uses of NA→B .
In this general context, strong convergence corresponds
to the following statement: for a given protocol P (n) , the
or
following limit holds
(1)
(NA2 →B2 ◦ AR1 B1 →R2 A2 ◦ NA1 →B1 )(ρR1 A1 ). (44) (n)
lim εad (k, P (n) ) = 0, (48)
k→∞
This process continues for n channel uses, and then the or more compactly,
final state is either
(n)
sup lim εad (k, P (n) ) = 0. (49)
k P (n) k→∞
ωR n Bn
≡
h i
NAkn →Bn ◦
n−1
(j) k Uniform convergence again corresponds to a swap of the
j=1 ARj Bj →Rj+1 Aj+1 ◦ NAj →Bj (ρR1 A1 ),
supremum and limit
(45)
(n)
lim sup εad (k, P (n) ) = 0, (50)
or k→∞ P (n)
q
(3) k
εad (k, P (3) ) = P ωR 3 B3
, ωR3 B3
h i
= P N k ◦ A(2) ◦ N k ◦ A(1) ◦ N k (ρR1 A1 ), N ◦ A(2) ◦ N ◦ A(1) ◦ N (ρR1 A1 )
h i
≤ P N k ◦ A(2) ◦ N k ◦ A(1) ◦ N k (ρR1 A1 ), N k ◦ A(2) ◦ N k ◦ A(1) ◦ N (ρR1 A1 )
h i
+ P N k ◦ A(2) ◦ N k ◦ A(1) ◦ N (ρR1 A1 ), N k ◦ A(2) ◦ N ◦ A(1) ◦ N (ρR1 A1 )
h i
+ P N k ◦ A(2) ◦ N ◦ A(1) ◦ N (ρR1 A1 ), N ◦ A(2) ◦ N ◦ A(1) ◦ N (ρR1 A1 )
≤ P N k (ρR1 A1 ), N (ρR1 A1 )
h h i h ii
+ P N k (A(1) ◦ N )(ρR1 A1 ) , N (A(1) ◦ N )(ρR1 A1 )
h h i h ii
+ P N k (A(2) ◦ N ◦ A(1) ◦ N )(ρR1 A1 ) , N (A(2) ◦ N ◦ A(1) ◦ N )(ρR1 A1 ) , (51)
where I have omitted some system labels for simplicity. the channels
The first inequality follows from the triangle inequality
N k ◦ A(2) ◦ N k ◦ A(1) , and (52)
and the second from data processing of the fidelity under
k (2)
N ◦A . (53)
The inequality in (51) can be understood as saying that
8
A1 B1 A2 B2 A3 B3
Nk Nk Nk
ρ A1 A2 ωk
R1 R2 R3
(a)
A1 B1 A2 B2 A3 B3
N N N
ρ A1 A2 ω
R1 R2 R3
(b)
k
FIG. 2. Adaptive protocol for distinguishing three uses of the channel NA→B from three uses of NA→B . The protocol is denoted
(3) (1) (2)
by P and consists of state preparation ρR1 A1 , as well as the channels AR1 B1 →R2 A2 and AR2 B2 →R3 A3 . (a) The protocol P (3)
is used with three uses of the channel NA→B , and the final state is ωR3 B3 . (b) The protocol P (3) is used with three uses of the
k k
k
channel NA→B , and the final state is ωR3 B3 . Strong convergence of the channel sequence {NA→B }k to NA→B implies that,
(3) k
given a fixed protocol P , the infidelity of the states ωR3 B3 and ωR3 B3 converges to zero in the limit as k → ∞.
NA→B (ρRA ) is a fixed state independent of k, applying F. Uniform convergence in the teleportation
the inequality in (60), and taking the limit k → ∞, we simulations of pure-loss, thermal, pure-amplifier,
find that amplifier, and additive-noise channels
Note that any thermal channel Lη,NB can be realized Now taking the limit σ̄ → 0 and using the fact that
in three steps: limσ̄→0 e(NB , η, σ̄) = 0 for η ∈ (0, 1) and NB ≥ 0, we
find that
1. prepare an environment mode in a thermal state
θ(NB ) of mean photon number NB ≥ 0, where lim sup P ((idR ⊗Lη,NB )(ρRA ), (idR ⊗Lη,NB0 )(ρRA )) = 0.
σ̄→0 ρRA
∞ n (80)
1 X NB
θ(NB ) = |nihn|, (76) Thus, the teleportation simulation Lη,NB ◦T σ̄ of the ther-
NB + 1 n=0 NB + 1
mal channel Lη,NB of transmissivity η ∈ (0, 1) and ther-
mal photon number NB ≥ 0 converges uniformly to the
2. interact the channel input mode with the environ- thermal channel.
ment mode at a unitary beamsplitter Bη of trans- The above uniform convergence result holds in partic-
missivity η, and ular for a pure-loss channel of transmissivity η ∈ (0, 1),
3. discard the environment mode. because this channel is a thermal channel with NB = 0.
That is, the environment state for the pure-loss channel
This observation and that in (74) are what lead to uni- is a vacuum state, and its teleportation simulation is a
form convergence of the simulating channel Lη,NB0 to the thermal channel with the same transmissivity and envi-
original channel Lη,NB in the limit as σ̄ → 0. Indeed, ronment state given by a thermal state of mean photon
let ρRA be an arbitrary input state, with R a reference number ησ̄/(1−η). In this case, the uniform upper bound
system corresponding to an arbitrary separable Hilbert e(NB = 0, η, σ̄) simplifies to
space and system A the channel input. Then we find that s
1
P ((idR ⊗Lη,NB )(ρRA ), (idR ⊗Lη,NB0 )(ρRA )) e(NB = 0, η, σ̄) = 1 − , (81)
ησ̄/(1 − η) + 1
≤ P ((idR ⊗Bη )[ρRA ⊗ θ(NB )], (idR ⊗Bη )[ρRA ⊗ θ(NB0 )])
= P (ρRA ⊗ θ(NB ), ρRA ⊗ θ(NB0 )) for which we clearly have that limσ̄→0 e(NB = 0, η, σ̄) =
0. Thus, the teleportation simulation of a pure-loss chan-
= P (θ(NB ), θ(NB0 )) nel Lη,NB =0 converges uniformly to Lη,NB =0 .
≡ e(NB , η, σ̄), (77) Similar results hold for the pure-amplifier and amplifier
channels. Indeed, to see this, let us begin by considering
where e(NB , η, σ̄) explicitly evaluates to a general amplifier channel AG,NB of gain G > 1 and
thermal photon number NB ≥ 0. Such a channel has the
e(NB , η, σ̄) = action as in (65) and (66), but with
"
hp √
1− (NB + 1) (NB + ησ̄/(1 − η) + 1) X = GI2 , (82)
#1/2 Y = (G − 1)(2NB + 1)I2 . (83)
p i−2
− NB [NB + ησ̄/(1 − η)] . (78) By similar reasoning as before, the teleportation simula-
tion AG,NB ◦ T σ̄ of the amplifier channel has the action
as in (65) and (66), but with
The explicit evaluation in (78) is a direct consequence
√
of [34, Eqs. (34)–(35)], found by evaluating the fidelity X = GI2 , (84)
between two thermal states of respective mean photon
Y = [G2σ̄ + (G − 1)(2NB + 1)] I2 (85)
numbers NB and NB + ησ̄/(1 − η). See also [43, 44] for
formulas for the fidelity of single-mode Gaussian states. = (G − 1)(2 [NB + Gσ̄/(G − 1)] + 1)I2 . (86)
The first inequality in (77) follows from data process-
Thus, the teleportation simulation AG,NB ◦ T σ̄ is equiv-
ing. The first equality follows from unitary invariance of
alent to an amplifier channel AG,NB +Gσ̄/(G−1) :
the metric P and the second from its invariance under
tensoring in the same state ρRA . To summarize the in- AG,NB ◦ T σ̄ = AG,NB +Gσ̄/(G−1) . (87)
equality in (77), it is stating that the distinguishability
of the channels Lη,NB and Lη,NB0 , when allowing for any An amplifier channel AG,NB can be realized by the fol-
input probe state ρRA , is limited by the distinguishabil- lowing three steps:
ity of the environment states θ(NB ) and θ(NB0 ), and this
is similar to the observations made in [34, 42]. Thus, the 1. prepare an environment mode in a thermal state
bound in (77) is a uniform bound, holding for all input θ(NB ) of mean photon number NB ≥ 0,
states ρRA , and so we conclude that
2. interact the channel input mode with the environ-
sup P ((idR ⊗Lη,NB )(ρRA ), (idR ⊗Lη,NB0 )(ρRA )) ment mode using a unitary two-mode squeezer SG
ρRA of gain G, and
≤ e(NB , η, σ̄). (79) 3. discard the environment mode.
11
For an arbitrary state ρRA , we find the following upper where I have omitted the identity maps idR acting on
bound by the same reasoning as in (77), but replacing the system R for simplicity.
beamsplitter Bη therein by the two-mode squeezer SG , A similar argument establishes that the teleportation
simulation of the additive-noise channel T ξ with variance
P ((idR ⊗AG,NB )(ρRA ), (idR ⊗AG,NB00 )(ρRA )) ξ > 0 converges uniformly to it. To see this, let us be-
gin by noting that any additive-noise channel T ξ can be
≤ P (θ(NB ), θ(NB00 )) ≡ e(NB , G, σ̄) (88)
realized by the following three steps:
where 1. Prepare a continuous classical environment regis-
ter according to the complex Gaussian distribution
NB00 ≡ NB + Gσ̄/(G − 1) (89) Gξ (α), as defined in (12).
and 2. Based on the classical value α in the environment
register, apply a unitary displacement operation
e(NB , G, σ̄) = D(α) to the channel input. This step can be de-
"
hp scribed as an interaction channel C between the
1− (NB + 1) (NB + Gσ̄/(G − 1) + 1) channel input and the environment register.
Remark 3 By the results of [34], the uniform upper denote a row vector of position- and momentum-
bounds in (79), (91), and (97) are all achievable and quadrature operators, satisfying the canonical commu-
are thus equalities. The achievable strategy consists of tation relations:
taking the input states ρRA to be a sequence of two-mode
0 1
h i
squeezed vacuum states with photon number NS tending R̂j , R̂k = iΩj,k , where Ω ≡ ⊗ Im , (100)
to infinity. −1 0
Remark 4 On the one hand, the thermal, amplifier, and Im denotes the m × m identity matrix. We take
and additive-noise channels are the single-mode bosonic the annihilation
√ operator for the jth mode as âj = (q̂j +
Gaussian channels that are of major interest in appli- ip̂j )/ 2. For z a column vector in R2m , we define the uni-
cations, as stressed in [46, Section 3.5] and [47, Sec- tary displacement operator D(z) = D† (−z) ≡ exp(iR̂z).
tion 12.6.3]. On the other hand, one could consider gen- Displacement operators satisfy the following relation:
eralizing the results of this section to arbitrary single-
mode bosonic Gaussian channels. In doing so, one should 0 0 i T 0
D(z)D(z ) = D(z + z ) exp − z Ωz . (101)
consider the Holevo classification of single-mode bosonic 2
Gaussian channels [48]. However, there is little reason
to generalize the contents of this section to other chan- Every state ρ ∈ D(H) has a corresponding Wigner char-
nels in the Holevo classification. The thermal and am- acteristic function, defined as
plifier channels form the class C discussed in [48], and
the additive-noise channels form the class B2 from [48], χρ (z) ≡ Tr{D(z)ρ}, (102)
which I have already considered in this section. The
classes that remain are labeled A, B1 , and D. The chan- and from which we can obtain the state ρ as
nels in classes A and D are entanglement breaking, as
d2m z
Z
proved in [49]. Thus, the channels in classes A and D ρ= †
m χρ (z) D (z). (103)
can be directly realized by the action of an LOCC on the (2π)
input state (without any need for an entangled resource
state), and thus we would never have any reason to be in- A quantum state ρ is Gaussian if its Wigner characteristic
terested in the teleportation simulation of these channels. function has a Gaussian form as
The final remaining class is B1 , but channels in the class
1 T ρ ρ T
B1 do not seem to be interesting in physical applications. χρ (ξ) = exp − z V z + i [µ ] z , (104)
4
Remark 5 The class B1 has been considered in [50],
where µρ is the 2m×1 mean vector of ρ, whose entries are
where it was shown that the teleportation simulation of
defined by µρj ≡ hR̂j iρ and V ρ is the 2m × 2m covariance
a channel in this class does not converge uniformly, sim-
matrix of ρ, whose entries are defined as
ilar to what occurs for the identity channel. Based on
the above, and the fact that the ideal channel and chan- ρ
Vj,k ≡ h{R̂j − µρj , R̂k − µρk }iρ . (105)
nels in the class B1 have unconstrained quantum capacity
equal to infinity [48], as well as the fact that channels in The following condition holds for a valid covariance ma-
the classes C and B2 have finite unconstrained quantum trix: V ≥ iΩ, which is a manifestation of the uncertainty
capacity [42], we can conclude that, among the single- principle.
mode bosonic Gaussian channels that are not entangle- A 2m × 2m matrix S is symplectic if it preserves the
ment breaking, their teleportation simulations converge symplectic form: SΩS T = Ω. According to Williamson’s
uniformly if and only if their unconstrained quantum ca- theorem [53], there is a diagonalization of the covariance
pacity is finite. This establishes a non-trivial link between matrix V ρ of the form,
teleportation simulation and unconstrained quantum ca-
pacity. T
V ρ = S ρ (Dρ ⊕ Dρ ) (S ρ ) , (106)
where X is a real 2m × 2m matrix, Y is a real 2m × 2m the Gaussian channel NX,Y +σ̄I , where σ̄ > 0 is a pa-
positive semi-definite matrix, and d ∈ R2m , such that rameter characterizing the squeezing strength and the
they satisfy unideal detections involved in the teleportation simula-
tion. Thus, as before, the teleportation simulation of a
Y ≥ i Ω − X T ΩX .
(108) Gaussian channel simply acts as an additive-noise chan-
nel concatenated with the original channel, and the ef-
The effect of the channel on the mean vector µρ and the
fect is that the noise matrix for the channel realized from
covariance matrix V ρ is thus as follows:
the teleportation simulation is Y + σ̄I, while the X ma-
µρ 7−→ X T µρ + d, (109) trix is unaffected. Thus, invoking [52, Theorem 1], the
ρ T ρ
teleportation simulation NX,Y +σ̄I can be realized as the
V 7−→ X V X + Y. (110) following transformation:
All Gaussian channels are covariant with respect to dis- †
ρA → NX,Y (ρA ) = TrE {UAE (ρA ⊗ γE (Y + σ̄I))UAE }.
placement operators, and this is the main reason why (114)
they are teleportation simulable, as noted in [6, 7]. That Then we are led to the following theorem:
is, the following relation holds
Theorem 6 Let NX,Y be a multi-mode quantum Gaus-
NX,Y (D(z)ρD† (z)) = D(X T z)NX,Y (ρ)D† (X T z). sian channel of the form in (107)–(110), such that Ω −
(111) X T ΩX is full rank. Then its teleportation simulation
Just as every quantum channel can be implemented as converges uniformly, in the sense that
a unitary transformation on a larger space followed by
a partial trace [55], so can Gaussian channels be imple- sup P [(idR ⊗NX,Y )(ρRA ), (idR ⊗NX,Y +σ̄I )(ρRA )]
mented as a Gaussian unitary on a larger space with some ρRA
extra modes prepared in the vacuum state, followed by a ≤ P (γE (Y ), γE (Y + σ̄I)), (115)
partial trace [52]. Given a Gaussian channel NX,Y with
Z such that Y = ZZ T we can find two other matrices where γE (Y ) is defined in (113),
XE and ZE such that there is a symplectic matrix
lim P (γE (Y ), γE (Y + σ̄I)) = 0, (116)
σ̄→0
T
X Z
S= T , (112)
XE ZE and one can use the explicit formula [56, Section IV] for
the fidelity of multi-mode, zero-mean Gaussian states to
which corresponds to the Gaussian unitary transforma-
find an analytical expression for
tion on a larger space.
Alternatively, for certain Gaussian channels, there is
p
P (γE (Y ), γE (Y + σ̄I)) = 1 − F (γE (Y ), γE (Y + σ̄I)),
a realization that is analogous to those discussed in the (117)
previous section for the thermal, amplifier, and additive- for σ̄ > 0.
noise channels. In particular, consider a Gaussian chan-
nel with X and Y matrices as discussed above, and sup- Proof. The proof of this theorem follows the same strat-
pose that the mean vector d = 0 (note that the condition egy given in the previous section, which in turn was used
d = 0 is not particularly restrictive because it just corre- in [34, 42]. In detail, letting ρRA be an arbitrary state,
sponds to a unitary displacement at the input or output we have that
of the channel, and capacities or distinguishability mea-
sures do not change under such unitary actions). When- P [(idR ⊗NX,Y )(ρRA ), (idR ⊗NX,Y +σ̄I )(ρRA )]
ever the matrix Ω − X T ΩX is full rank, implying then ≤ P [UAE (ρRA ⊗ γE (Y ))UAE † †
, UAE (ρRA ⊗ γE (Y σ̄ ))UAE ]
by (108) that Y is full rank, the Gaussian channel can
be realized as follows [52, Theorem 1]: = P [ρRA ⊗ γE (Y ), ρRA ⊗ γE (Y + σ̄I)]
= P [γE (Y ), γE (Y + σ̄I)], (118)
†
ρA → NX,Y (ρA ) = TrE {UAE (ρA ⊗ γE (Y ))UAE }, (113)
where Y σ̄ ≡ Y + σ̄I. The justification of these steps are
where ρA is the m-mode input, γE (Y ) is a zero-mean, the same as before, namely, data processing and unitary
m-mode Gaussian state with covariance matrix given invariance. The above bound is clearly a uniform bound,
by KY K T , where K is the invertible matrix discussed holding for all states ρRA , and so we conclude (115).
around [52, Eqs. (27)–(28)], and UAE is a Gaussian uni-
tary acting on 2m modes.
Given the above result, we can then generalize the III. PHYSICAL INTERPRETATION WITH THE
argument from the previous section to argue that the CV TELEPORTATION GAME
teleportation simulations of these channels converge uni-
formly. Indeed, as discussed in [6] and as a generalization We can interpret the results in the previous sections
of the single-mode case discussed previously, the telepor- of this paper in a game-theoretic way, in order to fur-
tation simulation of a Gaussian channel NX,Y realizes ther elucidate the physical meaning of the two kinds of
14
Gaussian channels converge both uniformly and strongly, which represents a dephased version of the state in (20).
so that Even though it is questionable whether such states are
physically realizable in practice, they are certainly nor-
inf sup
GA (ψRA ) − GA ◦T σ̄A (ψRA )
1
σ̄>0 ψRA
malizable, and thus allowed to be used in principle in an
unconstrained secret-key-agreement protocol.
= sup inf
GA (ψRA ) − GA ◦T σ̄A (ψRA )
1 = 0, (122)
One of the main results of [15] is the following bound
ψRA σ̄>0
on PL↔η (n, ε) when the channel is taken to be a pure-loss
the physically observable consequence is that the Tele- channel Lη of transmissivity η ∈ (0, 1):
porter always has the advantage in this modified (and
physically more realistic) version of the CV Teleporta- PL↔η (n, ε) ≤ − log2 (1 − η) + C(ε)/n, (125)
tion Game.
Note that other variations of the CV Teleportation where
Game are possible. One could allow for the Distin-
guisher to employ entangled strategies among the differ- C(ε) ≡ log2 6 + 2 log2 ([1 + ε] / [1 − ε]). (126)
ent rounds, or even adaptive channels in between rounds.
The results given in Propositions 1 and 2 can be used to This bound was established by proving that P ↔ ≤
analyze these other, richer variations of the CV Telepor- − log2 (1 − η) + C(ε)/n for any fixed (n, P ↔ , ε) uncon-
tation Game, but here I will not go into the details. strained protocol. As a consequence of this uniform
bound, one can then take a supremum over all P ↔ such
that there exists an (n, P ↔ , ε) protocol and conclude
IV. NON-ASYMPTOTIC
SECRET-KEY-AGREEMENT CAPACITIES (125), as was done in the proof of [15, Theorem 24]. Sim-
ilar reasoning was employed in [15] in order to arrive at
bounds on the unconstrained capacities of other bosonic
I now briefly review secret-key-agreement capacities of Gaussian channels.
a quantum channel and some results from [15]. The
A critical tool used to establish (125) is the simulation
secret-key-agreement capacity of a quantum channel is
of a quantum channel via teleportation [4, Section V] (see
equal to the optimal rate at which a sender and receiver
also [8, Theorem 14 & Remark 11]), which, as discussed
can use a quantum channel many times, as well as a free
previously, has been extended to bosonic states and chan-
assisting classical channel, in order to establish a reliable
nels [6, 7], by making use of the well known bosonic tele-
and secure secret key. It is relevant in the context of
portation protocol from [11]. More generally, one can al-
quantum key distribution [62, 63]. More generally, since
low for general local operations and classical communica-
capacity is a limiting notion that can never be reached in
tion (LOCC) when simulating a quantum channel from a
practice, one can consider a fixed (n, P ↔ , ε) secret-key-
resource state [5, Eq. (11)], known as LOCC channel sim-
agreement protocol that uses a channel n times and has
ulation. This tool is used to reduce any arbitrary LOCC-
ε error, while generating a secret key at the rate P ↔ [15].
assisted protocol over a teleportation-simulable channel
Such protocols were explicitly discussed in [15, 39, 40],
to one in which the LOCC assistance occurs after the
as well as the related developments in [41, 64–70]. For
final channel use. Another critical idea is the reinter-
a fixed integer n and a fixed error ε ∈ (0, 1), the non-
pretation of a three-party secret-key-agreement protocol
asymptotic secret-key-agreement capacity of a quantum
↔ as a two-party private-state generation protocol and em-
channel N is written as PN (n, ε) and is equal to the op-
ploying entanglement measures such as the relative en-
timal secret-key rate P subject to these constraints [15].
tropy of entanglement as a bound for the secret-key rate
That is,
[71, 72]. Finally, one can employ the Chen formula for
↔
PN (n, ε) ≡ sup{P ↔ | (n, P ↔ , ε) is achievable using ↔}, the relative entropy of Gaussian states [73], as well as
(123) a formula from [54] for the relative entropy variance of
where ↔ indicates the free use of LOCC between every Gaussian states. These tools were foundational for the
use of the quantum channel. results of [15, Theorem 24] in order to argue for bounds
If an (n, P ↔ , ε) protocol takes place over a bosonic on secret-key-agreement capacities of bosonic Gaussian
channel, the stated definition of a secret-key-agreement channels.
protocol makes no restriction on the photon number of A key point mentioned above, which is critical to and
the channel input states, and as such, it is called an clearly stated in the proof of [15, Theorem 24], is as fol-
unconstrained protocol. The corresponding capacity is lows: the proof begins by considering a fixed (n, P ↔ , ε)
called the unconstrained secret-key-agreement capacity. protocol and then establishes a uniform bound on P ↔ ,
For example, in such a scenario, a sender and receiver independent of the details of the particular protocol. The
could freely make use of the following classically corre- discussion given in this paper clarifies that strong conver-
lated Basel state with mean photon number equal to ∞: gence in teleportation simulation suffices for the proof of
∞ [15, Theorem 24]. Furthermore, there is no need to in-
6 X 1 voke the energy-constrained diamond distance [16, 74] in
β AB ≡ |nihn|A ⊗ |nihn|B , (124)
π 2 n=1 n2 order to establish the correctness of the proof.
16
V. DETAILED REVIEW OF THE PROOF OF (47), except the initial state of the protocol is constrained
THEOREM 24 IN WTB17 to be a separable, unentangled state shared between the
sender and receiver, and each channel A(j) in the proto-
We can now step through the relevant parts of the col is constrained to be an LOCC channel between the
proof of [15, Theorem 24] carefully, in order to clarify its sender and receiver. Continuing,
correctness. It is worthwhile to do so, given that Ref. [23] “Due to the fact that continuous-variable teleportation
recently questioned the proof of [15, Theorem 24]. I high- induces a perfect quantum channel when infinite energy
light, in italics, quotations from the proof of [15, Theo- is available [11], the following limit holds for every n:
rem 24] for clarity and follow each quotation with a brief
discussion. The proof begins by stating that: lim sup εTP (n, µ) = 0. ” (128)
µ→∞
“First, consider an arbitrary (n, P ↔ , ε) protocol for
the thermalizing channel Lη,NB . It consists of using the
This is indeed a key step for the proof. In the con-
channel n times and interleaving rounds of LOCC be-
n text of the proof given, the convergence is as it is written
tween every channel use. Let ζ denote the final state
B̂ and is to be understood in the strong sense of (64): for
of Alice and Bob at the end of this protocol.” a fixed protocol used in conjunction with the n telepor-
It is crucial to note here that this is saying, as it is writ- tation simulations, the infidelity converges to zero in the
ten, that we should really start with a particular, fixed limit of ideal squeezing and ideal detection (as µ → ∞ or,
(n, P ↔ , ε) protocol. This does not mean that we should equivalently, as σ̄ → 0). Note that here we are employing
be considering a sequence of such protocols or protocols the notion of strong convergence in (64), given that the
involving unnormalizable states. Thus, proceeding by original protocol has LOCC channels interleaved between
fixing an (n, P ↔ , ε) protocol, the proof continues with every use of Lη,NB .
“By the teleportation reduction procedure [...], such p
Continuing the proof, “By using that 1 − F (ρ, σ) is
a protocol can be simulated by preparing n two-mode
a distance measure for states ρ and σ (and thus obeys
squeezed vacuum (TMSV) states each having energy µ −
a triangle inequality) [...], the simulation leads to an
1/2 (where we think of µ ≥ 1/2 as a very large posi-
(n, P ↔ , ε(n, µ)) protocol for the thermalizing channel,
tive real), sending one mode of each TMSV through each
where
channel use, and then performing continuous-variable
quantum teleportation [11] to delay all of the LOCC op- √
h i2
p
erations until the end of the protocol. Let ρµη,NB denote ε(n, µ) ≡ min 1, ε + εTP (n, µ) . (129)
the state resulting from sending one share of the TMSV
0
through the thermalizing channel, and let ζ B̂
(n, µ) de- Observe that lim supµ→∞ ε(n, µ) = ε, so that the simu-
note the state at the end of the simulation.” lated protocol has equivalent performance to the original
This means, as it states, that the fixed protocol can protocol in the infinite-energy limit.”
be simulated with some error by replacing each chan- This last part that I have recalled is saying how the er-
nel use of Lη,NB with the simulating channel Lµη,NB , for ror of the simulating protocol is essentially equal to the
µ ∈ [0, ∞). Here, the original channel Lη,NB is in cor- sum of the error of the original protocol and the error
respondence with GA from Section II E, Lµη,NB with GA σ̄
, from substituting the original n channels with n unideal
and µ with σ̄, in the sense that the limit µ → ∞ is in teleportations. It is a straightforward √ consequence of
correspondence with the limit σ̄ → 0. Furthermore, since the triangle inequality for the metric 1 − F , as recalled
this is an LOCC simulation and the original protocol con- there, and thus captures a correct propagation of errors.
sists of channel uses of Lη,NB interleaved by LOCC, it is From this point on in the proof, using other techniques,
possible to write the simulating protocol as one that con- the following bound is concluded on the rate P ↔ of the
sists of a single round of LOCC on the state [ρµη,NB ]⊗n , fixed (n, P ↔ , ε) protocol by invoking the meta-converse
as observed in [4, 7, 8]. Continuing, in [15, Theorem 11]:
“Let εTP (n, µ) denote the “infidelity” of the simula- s
tion: 2V (ρµη,NB kση,N
µ
)
P ↔ ≤ D(ρµη,NB kση,N
µ
)+ B
B
n(1 − ε(n, µ))
n
εTP (n, µ) ≡ 1 − F (ζ , ζ 0 (n, µ)).
B̂ ÂB̂
” (127)
+ C(ε(n, µ))/n. (130)
In the context of the proof, this infidelity clearly cor- This bound holds for all µ sufficiently large, ε ∈ (0, 1),
responds to the infidelity of the simulation of the fixed and positive integers n, and as such, it is a uniform
protocol. One might argue that the notation εTP (n, µ) bound. Given the uniform bound, the limit µ → ∞ is
somehow hides this dependence on a fixed protocol, then taken to arrive at
but the dependence of εTP (n, µ) on a fixed protocol is s
clear from the context and the fact that the quantity 2VLη,NB C(ε)
P ↔ ≤ − log (1 − η) η NB − g(NB ) +
+ .
n
1 − F (ζ , ζ 0 (n, µ)) itself clearly depends on a fixed
B̂ ÂB̂
n (1 − ε) n
protocol as stated. This infidelity is similar to that in (131)
17
Since this latter bound is itself a uniform bound, holding ideal quantum channel, and this highlights the role of
for all (n, P ↔ , ε) protocols, it is then concluded that the present paper in providing a precise clarification of
the convergence that occurs in the continuous-variable
PL↔η,N (n, ε) ≤ − log (1 − η) η NB − g(NB )
bosonic teleportation protocol from [11]. I also proved
B
s that the teleportation simulations of the pure-loss, ther-
2VLη,NB C(ε) mal, pure-amplifier, amplifier, and additive-noise chan-
+ + . (132) nels converge both strongly and uniformly to the original
n (1 − ε) n
channels, which is in contrast to what occurs in the tele-
Further arguments are given in the proof of [15, Theo- portation simulation of the ideal channel. I suspect that
rem 24] to establish the bound in (125) for the pure-loss the explicit uniform bounds in (79), (91), and (97), on the
channel. accuracy of the teleportation simulations of these chan-
To summarize, the original proof of [15, Theorem 24] nels, will be useful in future applications. The uniform
as given there is correct as it is written. The proof of [15, convergence results were then generalized to the tele-
Theorem 24] as written there establishes that for a fixed portation simulations of particular multi-mode bosonic
(n, P ↔ , ε) protocol, the bound in (131) holds. Further- Gaussian channels. Finally, I gave a physical interpreta-
more, one might think that a proof may only be given tion of the convergence results discussed in this paper,
by employing the energy-constrained diamond distance, by means of the CV Teleportation Game of Section III.
but this is clearly not the case either, as demonstrated I also reviewed the proof of [15, Theorem 24] and con-
above. firmed its correctness as it is written there. One might
think that it is necessary to use the energy-constrained
diamond distance to arrive at a proof of (125), but this
VI. CONCLUSION is clearly not the case.
It would be interesting in future work to explore phys-
The continuous-variable, bosonic quantum teleporta- ical scenarios of interest in which different topologies of
tion protocol from [11] is often loosely stated to simulate convergence lead to physically distinct outcomes, as was
an ideal quantum channel in the limit of infinite squeezing the case in the CV Teleportation Game. Recent work in
and ideal homodyne detection. The precise form of con- this direction is available in [17] and [74].
vergence is typically not clarified in the literature, and as
a consequence, this has the potential to lead to confusion
in mathematical proofs that employ this protocol.
This paper has clarified various notions of channel con- ACKNOWLEDGMENTS
vergence, with applications to the continuous-variable
bosonic teleportation protocol from [11], and extended I acknowledge discussions with several of my dear col-
these notions to various contexts. This paper provided leagues, as well as support from the Office of Naval Re-
an explicit proof that the continuous-variable bosonic search. I am also indebted to an anonymous referee for
teleportation protocol from [11] converges strongly to suggesting the generalization of the uniform convergence
an ideal quantum channel in the limit of ideal squeez- of the teleportation simulations of thermal, amplifier, and
ing and detection. At the same time, this paper proved additive-noise channels to the case of particular multi-
that this protocol does not converge uniformly to an mode bosonic Gaussian channels.
Theory, 21(5):533–543, September 1975. [62] Charles H. Bennett and Gilles Brassard. Quantum cryp-
[44] Horia Scutaru. Fidelity for displaced squeezed thermal tography: Public key distribution and coin tossing. In
states and the oscillator semigroup. Journal of Physics Proceedings of IEEE International Conference on Com-
A: Mathematical and General, 31(15):3659, April 1998. puters Systems and Signal Processing, pages 175–179,
arXiv:quant-ph/9708013. Bangalore, India, December 1984.
[45] Guy B. Coleman and Harry C. Andrews. Image segmen- [63] Valerio Scarani, Helle Bechmann-Pasquinucci, Nico-
tation by clustering. Proceedings of the IEEE, 67(5):773– las J. Cerf, Miloslav Dušek, Norbert Lütkenhaus, and
785, 1979. Momtchil Peev. The security of practical quantum key
[46] Alexander S. Holevo and Vittorio Giovannetti. Quan- distribution. Reviews of Modern Physics, 81(3):1301,
tum channels and their entropic characteristics. Re- September 2009. arXiv:0802.4155.
ports on Progress in Physics, 75(4):046001, April 2012. [64] Kenneth Goodenough, David Elkouss, and Stephanie
arXiv:1202.6480. Wehner. Assessing the performance of quantum re-
[47] Alexander S. Holevo. Quantum Systems, Channels, In- peaters for all phase-insensitive Gaussian bosonic chan-
formation. de Gruyter Studies in Mathematical Physics nels. New Journal of Physics, 18(6):063005, June 2016.
(Book 16). de Gruyter, November 2012. arXiv:1511.08710.
[48] Alexander S. Holevo. One-mode quantum Gaussian [65] Koji Azuma, Akihiro Mizutani, and Hoi-Kwong Lo.
channels: Structure and quantum capacity. Problems Fundamental rate-loss trade-off for the quantum inter-
of Information Transmission, 43(1):1–11, March 2007. net. Nature Communications, 7:13523, November 2016.
arXiv:quant-ph/0607051. arXiv:1601.02933.
[49] Alexander S. Holevo. Entanglement-breaking channels in [66] Stefan Bäuml and Koji Azuma. Fundamental limitation
infinite dimensions. Problems of Information Transmis- on quantum broadcast networks. Quantum Science and
sion, 44(3):171–184, September 2008. arXiv:0802.0235. Technology, 2(2):024004, June 2017. arXiv:1609.03994.
[50] Pirandola et al. April 2018. arXiv:1712.01615v4. [67] Eneet Kaur and Mark M. Wilde. Amortized en-
[51] I am indebted to an anonymous referee for suggesting tanglement of a quantum channel and approximately
this multi-mode generalization. teleportation-simulable channels. Journal of Physics A,
[52] Filippo Caruso, Jens Eisert, Vittorio Giovannetti, and 51(3):035303, January 2018. arXiv:1707.07721.
Alexander S. Holevo. Multi-mode bosonic Gaussian chan- [68] Eneet Kaur and Mark M. Wilde. Upper bounds on
nels. New Journal of Physics, 10:083030, August 2008. secret key agreement over lossy thermal bosonic chan-
arXiv:0804.0511. nels. Physical Review A, 96(6):062318, December 2017.
[53] John Williamson. On the algebraic problem concerning arXiv:1706.04590.
the normal forms of linear dynamical systems. American [69] Luca Rigovacca, Go Kato, Stefan Baeuml, M. S. Kim,
Journal of Mathematics, 58(1):141–163, January 1936. W. J. Munro, and Koji Azuma. Versatile relative entropy
[54] Mark M. Wilde, Marco Tomamichel, Seth Lloyd, and bounds for quantum networks. New Journal of Physics,
Mario Berta. Gaussian hypothesis testing and quantum 20:013033, January 2018. arXiv:1707.05543.
illumination. Physical Review Letters, 119(12):120501, [70] Masahiro Takeoka, Kaushik P. Seshadreesan, and
September 2017. arXiv:1608.06991. Mark M. Wilde. Unconstrained capacities of quantum
[55] William F. Stinespring. Positive functions on C*- key distribution and entanglement distillation for pure-
algebras. Proceedings of the American Mathematical So- loss bosonic broadcast channels. Physical Review Letters,
ciety, 6:211–216, 1955. 119(15):150501, October 2017. arXiv:1706.06746.
[56] Gh.-S. Paraoanu and Horia Scutaru. Fidelity for mul- [71] Karol Horodecki, Michal Horodecki, Pawel Horodecki,
timode thermal squeezed states. Physical Review A, and Jonathan Oppenheim. Secure key from bound entan-
61(2):022306, January 2000. arXiv:quant-ph/9907068. glement. Physical Review Letters, 94(16):160502, April
[57] Carl W. Helstrom. Quantum detection and estimation 2005. arXiv:quant-ph/0309110.
theory. Journal of Statistical Physics, 1:231–252, 1969. [72] Karol Horodecki, Michal Horodecki, Pawel Horodecki,
[58] Alexander S. Holevo. Statistical decision theory for quan- and Jonathan Oppenheim. General paradigm for dis-
tum systems. Journal of Multivariate Analysis, 3(4):337– tilling classical key from quantum states. IEEE Trans-
394, December 1973. actions on Information Theory, 55(4):1898–1929, April
[59] Carl W. Helstrom. Quantum Detection and Estimation 2009. arXiv:quant-ph/0506189.
Theory. Academic, New York, 1976. [73] Xiao-yu Chen. Gaussian relative entropy of entangle-
[60] Alexander S. Holevo. On quasiequivalence of locally ment. Physical Review A, 71(6):062320, June 2005.
normal states. Theoretical and Mathematical Physics, arXiv:quant-ph/0402109.
13(2):1071–1082, November 1972. [74] Andreas Winter. Energy-constrained diamond norm
[61] Christopher A. Fuchs and Jeroen van de Graaf. Crypto- with applications to the uniform continuity of con-
graphic distinguishability measures for quantum mechan- tinuous variable channel capacities. December 2017.
ical states. IEEE Transactions on Information Theory, arXiv:1712.10267.
45(4):1216–1227, May 1998. arXiv:quant-ph/9712042.