Вы находитесь на странице: 1из 8

Reply to the criticism in “Theory of channel simulation and bounds

for private communication”


Mark M. Wilde∗

November 24, 2018

Abstract
In a paper recently published as [PBL+ 18], Pirandola, Braunstein et al. have criticized the
proof of the strong converse theorem from [WTB17]. Their criticism suffers from a central mis-
interpretation of the proof and arguments presented in [WTB17]. That is, they have given the
impression that they have refuted the proof from [WTB17], while they have instead incorrectly
presented the argument of [WTB17]. In addition, their criticism is based on a misinterpre-
tation of the definition of a secret-key-agreement protocol from that given in prior works, as
well as a misinterpretation of convergence issues involved with the continuous-variable bosonic
teleportation protocol.

1 Introduction
Pirandola, Braunstein, et al. [PBL+ 18] have recently criticized the proof of the strong converse
theorem from [WTB17], regarding the secret-key-agreement capacity of pure-loss bosonic Gaussian
channels. In particular, the criticism is regarding the proof of Theorem 24 of [WTB17].
In this comment, I address this criticism and show that the proof of [WTB17, Theorem 24]
stands correct as written there. In short, the criticism of Pirandola, Braunstein, et al. is based on a
central misinterpretation of the proofs and argument presented in [WTB17]. They have incorrectly
presented the proof of [WTB17, Theorem 24], and then they have criticized this modified argument
in order to leave the impression that they have refuted the proof of [WTB17, Theorem 24].

2 Review of the argument of [WTB17, Theorem 24]


Once we accept two key facts, it is clear that the argument for [WTB17, Theorem 24] is sound and
complete, as presented therein. The first regards the convergence of the continuous-variable bosonic
teleportation protocol [BK98] to the ideal channel, and the second regards the basic definition of a
secret-key-agreement protocol from [TGW14b, TGW14a, WTB17].
For the first, recall from [BK98, Eq. (9)] that the continuous-variable bosonic teleportation
protocol realizes an additive-noise Gaussian channel T σ̄ of the following form:
exp(− |α|2 /σ̄)
Z
σ̄
ρ → T (ρ) ≡ d2 α D(α)ρD† (α), (1)
πσ̄

Hearne Institute for Theoretical Physics, Department of Physics and Astronomy, Center for Computation and
Technology, Louisiana State University, Baton Rouge, Louisiana 70803, USA

1
where ρ is an input density operator (positive semi-definite operator with trace equal to one), σ̄ > 0
is a noise parameter, and D(α) is a quantum optical displacement operator. It was argued in the
original paper [BK98] (in particular, just after Eq. (4) and implicitly in Eq. (11)) that for any fixed
density operator ρ, the fidelity F (ρ, T σ̄ (ρ)) approaches one in the limit as σ̄ → 0. That is, for any
fixed density operator ρ, the following limit holds

lim F (ρ, T σ̄ (ρ)) = 1. (2)


σ̄→0

One can also easily deduce from the formula in [BK98, Eq. (11)] that, for any fixed density operator
ρRA , where R is a reference system and A is the one to be teleported, that the following limit holds

lim F (ρRA , (idR ⊗T σ̄ )(ρRA )) = 1, (3)


σ̄→0

where idR denotes the identity channel acting on the reference system R. We can summarize the
convergence argument from [BK98] more compactly as

inf lim F (ρRA , (idR ⊗T σ̄ )(ρRA )) = 1. (4)


ρRA σ̄→0

Since the publication of [BK98], a vast number of papers have been published and have employed
the teleportation protocol and the convergence statement in (4) (e.g., see [GIC02, NFC09] for some
prominent cases). It is important to emphasize, albeit even if it is obvious, that the convergence
in (3) holds for any fixed density operator ρRA ; i.e., the only requirement on ρRA is that it is a
density operator, i.e., positive semi-definite and normalized, with trace equal to one.
The next key fact that we need to recall is the definition of a secret-key-agreement protocol for
a quantum channel [TGW14b, TGW14a, WTB17]. Let NA→B be a quantum channel connecting a
sender Alice to a receiver Bob. An (n, R, ε) secret-key-agreement protocol makes n invocations of
the channel NA→B , and between every channel use, the sender and receiver are allowed to perform
a local operations and classical communication (LOCC) channel [BDSW96, CLM+ 14]. So, for
example, if n = 3, the protocol has the following form:

(4) (3)
LA0 B3 B 0 →KA KB ◦ NA3 →B3 ◦ LA0 B2 B 0 →A0 A3 B 0 ◦ NA2 →B2
3 3 2 2 3 3
(2) (1)
◦ LA0 B1 B 0 →A0 A2 B 0 ◦ NA1 →B1 ◦ L∅→A0 A1 B 0 . (5)
1 1 2 2 1 1

(1)
The first LOCC channel L∅→A0 A1 B 0 simply creates a separable state ωA01 A1 B10 between Alice’s sys-
1 1
tems A01 A1 and Bob’s system B10 . The channel NA1 →B1 is then called. Alice and Bob then perform
(2)
LOCC, represented as the LOCC channel LA0 B1 B 0 →A0 A2 B 0 . The channel NA2 →B2 is called. This
1 1 2 2
(4)
repeats a third time, and then a final LOCC channel LA0 B3 B 0 →KA KB produces an ε-approximate
3 3
secret key [HHHO05, HHHO09, WTB17] in the systems KA and KB . We can alternatively write
the state at the end of the protocol as

(L(4) ◦ N ◦ L(3) ◦ N ◦ L(2) ◦ N )(ω), (6)

where I have omitted system labels for simplicity.


Even if it may be obvious, it is important to stress that, at each time slice of the protocol,
Alice and Bob share a quantum state, i.e., a positive semi-definite operator that is normalized with

2
trace equal to one. That is, ω is a state, N (ω) is a state, (L(2) ◦ N )(ω) is a state, etc. However,
they are not allowed to possess an ideal position eigenstate, an ideal momentum eigenstate, or an
ideal EPR “state” at any point in the protocol, as these are all not quantum states since they are
unnormalizable and unphysical, the latter point emphasized in several prior works of Braunstein,
Pirandola, and others [BvL05, Eq. (80)], [WPGP+ 12, Footnote 1], and [LMGP+ 11]. The fact that
Alice and Bob are only allowed to possess physical, normalizable quantum states in a secret-key-
agreement protocol plays a key role in the application of continuous-variable teleportation to the
analysis of secret-key rates.
With these two notions in place, the proof of [WTB17, Theorem 24] first argues, as had been
done previously in [NFC09], that when NA→B is a bosonic Gaussian channel [Ser17], such as
a thermal, amplifier, or additive-noise channel, it can be replaced by a teleportation simulation
NA→B ◦ TAσ̄ , where TAσ̄ is the teleportation channel in (1). The simulating protocol thus has the
following form:
(L(4) ◦ N ◦ T σ̄ ◦ L(3) ◦ N ◦ T σ̄ ◦ L(2) ◦ N ◦ T σ̄ )(ω). (7)
This replacement inevitably introduces an error, as discussed in the proof of [WTB17, Theorem 24],
but a key feature is that it allows for a rearrangement, due to the structure of teleportation and the
fact that a bosonic Gaussian channel is covariant with respect to displacement operators, as used in
[NFC09]. That is, the simulating protocol can be rearranged as the preparation of a tensor-power
h i⊗n
state NA→B (ϕN S
RA ) followed by a single, final LOCC channel, where ϕN S
RA denotes a two-mode
squeezed vacuum state defined as
ϕNRA ≡ |ϕ
S NS
ihϕNS |RA , (8)
with NS ≥ 0 and s
∞ n
NS 1 X NS
|ϕ iRA ≡ √ |niR |niA . (9)
NS + 1 n=0 NS + 1
This key insight was discussed in [BDSW96, MH12] for discrete-variable channels and in [NFC09] for
continuous-variable bosonic channels. The parameter NS ≥ 0 is directly related to the parameter σ̄
from (1), in the sense that σ̄ → 0 as NS → ∞ [BK98].
A crucial point of the analysis in the proof of [WTB17, Theorem 24] is the convergence of the
simulating protocol to the original one. It was argued therein, based on the convergence result of
Braunstein [BK98] recalled in (4), that for any fixed protocol of the form in (6), the following limit
holds

lim F ((L(4) ◦ N ◦ L(3) ◦ N ◦ L(2) ◦ N )(ω), (L(4) ◦ N ◦ T σ̄ ◦ L(3) ◦ N ◦ T σ̄ ◦ L(2) ◦ N ◦ T σ̄ )(ω)) = 1. (10)
σ̄→0

This limit is quitepintuitive and is a natural consequence of (4) and the triangle inequality for the
metric P (ρ, σ) ≡ 1 − F (ρ, σ) [Ras02, Ras03, GLN05, Ras06].
Setting the channel N to be the pure-loss channel of transmissivity η ∈ (0, 1), for example, with
the limit in (10) in place, the proof of [WTB17, Theorem 24] provides further arguments for the
following uniform bound on the secret-key rate of an arbitrary (n, R, ε) protocol of the form in (6):
C(ε)
R ≤ − log2 (1 − η) + , (11)
n
where C(ε) ≡ log2 6 + 2 log2 ([1 + ε] / [1 − ε]). This uniform bound is then used to deduce that
− log2 (1 − η) is a strong converse rate for secret-key agreement over a pure-loss channel.

3
3 The misrepresentation of the argument of [WTB17, Theorem 24],
as given by Pirandola, Braunstein, et al.
I now review the misrepresentation of the argument of [WTB17, Theorem 24], which Pirandola,
Braunstein, et al. have given in [PBL+ 18] as a replacement for the above argument from [WTB17].

1. First, although [WTB17] cited Braunstein’s work in [BK98] for the notion of convergence used
in continuous-variable teleportation and recalled in (4), Pirandola, Braunstein, et al. have
asserted that the convergence of continuous-variable teleportation as employed in the proof
of [WTB17, Theorem 24] should be interpreted in the following way:

lim inf F (ρRA , (idR ⊗T σ̄ )(ρRA )) = 0. (12)


σ̄→0 ρRA

It is true that the limit in (12) holds, but it is most certainly not the notion of convergence
employed in the proof of [WTB17, Theorem 24]. The very fact that Braunstein’s work [BK98]
is cited in the proof of [WTB17, Theorem 24] is all the evidence that is needed to support
the claim that the notion of convergence should be understood as in (4) and (10).
2. Furthermore, Pirandola, Braunstein, et al. have argued that it is necessary for a secret-key-
agreement protocol to allow for an “asymptotic state” of the following form:

lim ϕNS
RA , (13)
NS →∞

where ϕNS +
RA is the state from (8). Indeed, quoting from after [PBL 18, Eq. (188)]:
“It is clear that these states need to be included among all possible inputs, because we
are studying unconstrained quantum and private capacities, for which the input alphabet is
energy-unbounded.”
and quoting after [PBL+ 18, Eq. (194)]:
“Among the key generation protocols to be considered in the derivation of upper bounds
for the (unconstrained) secret key capacity of bosonic channels, one clearly needs to include
protocols based on asymptotic input states.”
By inspecting the definition in (9), it is clear that the limiting object in (13) is most certainly
not a quantum state, as it fails the basic requirement of normalization. Note that it has
previously been stressed [LMGP+ 11], including in the works of Braunstein [BvL05, Eq. (80)],
that the limiting object in (13) is not normalizable and is thus unphysical and not a density
operator. As it was defined and considered in [TGW14b, TGW14a, WTB17], a secret-key-
agreement protocol processes quantum states (i.e., density operators) at each step, and is
thus not allowed to employ the unphysical object in (13) at any place in the protocol. For
this reason, it appears that Pirandola, Braunstein et al. have misunderstood in [PBL+ 18]
the basic definition of what constitutes a secret-key-agreement protocol. See Appendix A for
further remarks about the limiting object in (13).
3. Finally, it appears from the discussion surrounding [PBL+ 18, Eq. (180)] that Pirandola,
Braunstein, et al. have missed a key aspect of the bound in (11) and the way that the proof of
[WTB17, Theorem 24] is constructed. The proof of [WTB17, Theorem 24] first fixes a secret-
key-agreement protocol of the form in (6) and then argues that the bound in (11) holds for

4
its rate. Since the bound in (11) is uniform, (i.e., holding for any such secret-key agreement
protocol), and decreasing with n for any fixed ε ∈ (0, 1), it is concluded that − log2 (1 − η) is
a strong converse rate for secret-key-agreement over the pure-loss channel.

Pirandola, Braunstein et al. also leave their readers with the impression that the convergence in
(4) holds only for energy-bounded input states (see [PBL+ 18, Eq. (34)]). However, this assertion is
inaccurate, with a counterexample given by the Basel state of [Wil18]. The Basel state of [Wil18]
is defined as

6 X 1
β≡ 2 |nihn|. (14)
π n2
n=1

It thus has infinite energy (hn̂iβ = ∞) and yet is normalizable (Tr{β} = 1), so that the convergence
in (4) holds for it, as shown in [Wil18].

Going forward from here, we can ask the following key question regarding the criticism of
Pirandola, Braunstein et al. from [PBL+ 18]:

If the convergence of continuous-variable teleportation as recalled in (4) has been


known since Braunstein’s 1998 work [BK98] and has been used in many subsequent
works on the topic (e.g., [GIC02, NFC09]), why have Pirandola, Braunstein et al. as-
serted in [PBL+ 18] that [WTB17] was not employing this notion?

A clue for understanding the answer to this question is given by inspecting the seventh version of
the prior work of Pirandola et al. [PLOB17]. In particular, in [PLOB17], on page 3 therein, on
page 15 before Eq. (94), on page 15 in the caption of Figure 9, and on page 15 before Eq. (98),
Pirandola et al. were employing the diamond norm in their own work, which plays the same role
in this context as the following similarity measure:

inf F (ρRA , (idR ⊗T σ̄ )(ρRA )), (15)


ρRA

Recall that this similarity measure was discussed previously here in (12). In [PLOB17], in contrast
to the equality in (12), it was asserted that a teleportation simulation converges to the ideal channel
with respect to this measure (with the ideal channel being a particular bosonic channel). That is,
it was asserted in [PLOB17] that the expression in (12) is equal to one. However, as stated in
(12) and confirmed in [PBL+ 18], convergence in this sense does not hold, and so we are left with
questions about the completeness of the arguments in the seventh iteration of the work of Pirandola
et al. To employ the word choice from [PBL+ 18], one could suggest that the work [PLOB17] suffers
from a “technical gap,” an “exploding bound,” or a “mathematical issue.” That is, by the very
arguments of Pirandola, Braunstein et al. presented in [PBL+ 18], this seventh iteration of Pirandola
et al.’s work in [PLOB17] could thus be considered incomplete and suffers from a “fundamental
mathematical gap” and “technical mathematical issues,” to quote directly from [PBL+ 18]. It is
unfortunate that Pirandola, Braunstein et al. did not point out this gap in the prior work of
Pirandola et al. [PLOB17] and instead opted to prop it up as a straw man for the arguments in
[WTB17].

5
4 Conclusion
In summary, as argued above, Pirandola, Braunstein et al. have incorrectly presented the argument
of [WTB17, Theorem 24] in a way that is not consistent with the proof presented therein, and the
proof of [WTB17, Theorem 24] is correct as written there. In [PBL+ 18, Section 10.5], Pirandola,
Braunstein et al. have levied the same criticism against the work in [KW17], but the arguments
given in [KW17] are correct as well. One can consult [Wil18] for further clarifications.

References
[BDSW96] Charles H. Bennett, David P. DiVincenzo, John A. Smolin, and William K. Woot-
ters. Mixed-state entanglement and quantum error correction. Physical Review A,
54(5):3824–3851, November 1996. arXiv:quant-ph/9604024.

[BK98] Samuel L. Braunstein and H. J. Kimble. Teleportation of continuous quantum vari-


ables. Physical Review Letters, 80(4):869–872, January 1998.

[BvL05] Samuel L. Braunstein and Peter van Loock. Quantum information with continu-
ous variables. Reviews of Modern Physics, 77(2):513–577, June 2005. arXiv:quant-
ph/0410100.

[CLM+ 14] Eric Chitambar, Debbie Leung, Laura Mančinska, Maris Ozols, and Andreas Winter.
Everything you always wanted to know about LOCC (but were afraid to ask). Com-
munications in Mathematical Physics, 328(1):303–326, May 2014. arXiv:1210.4583.

[GIC02] Géza Giedke and J. Ignacio Cirac. Characterization of Gaussian operations and
distillation of Gaussian states. Physical Review A, 66(3):032316, September 2002.
arXiv:quant-ph/0204085.

[GLN05] Alexei Gilchrist, Nathan K. Langford, and Michael A. Nielsen. Distance measures to
compare real and ideal quantum processes. Physical Review A, 71(6):062310, June
2005. arXiv:quant-ph/0408063.

[HHHO05] Karol Horodecki, Michal Horodecki, Pawel Horodecki, and Jonathan Oppenheim.
Secure key from bound entanglement. Physical Review Letters, 94(16):160502, April
2005. arXiv:quant-ph/0309110.

[HHHO09] Karol Horodecki, Michal Horodecki, Pawel Horodecki, and Jonathan Oppenheim.
General paradigm for distilling classical key from quantum states. IEEE Transactions
on Information Theory, 55(4):1898–1929, April 2009. arXiv:quant-ph/0506189.

[KW17] Eneet Kaur and Mark M. Wilde. Upper bounds on secret key agreement over
lossy thermal bosonic channels. Physical Review A, 96(6):062318, December 2017.
arXiv:1706.04590.

[LMGP+ 11] Seth Lloyd, Lorenzo Maccone, Raul Garcia-Patron, Vittorio Giovannetti, and Yu-
taka Shikano. Quantum mechanics of time travel through post-selected teleportation.
Physical Review D, 84(2):025007, July 2011.

6
[MH12] Alexander Müller-Hermes. Transposition in quantum information theory. Master’s
thesis, Technical University of Munich, September 2012.

[NFC09] Julien Niset, Jaromı́r Fiurasek, and Nicolas J. Cerf. No-go theorem for Gaussian
quantum error correction. Physical Review Letters, 102(12):120501, March 2009.
arXiv:0811.3128.

[PBL+ 18] Stefano Pirandola, Samuel L Braunstein, Riccardo Laurenza, Carlo Ottaviani,
Thomas P W Cope, Gaetana Spedalieri, and Leonardo Banchi. Theory of channel
simulation and bounds for private communication. Quantum Science and Technology,
3(3):035009, 2018.

[PLOB17] Stefano Pirandola, Riccardo Laurenza, Carlo Ottaviani, and Leonardo Banchi. Jan-
uary 2017. arXiv:1510.08863v7, posted on January 13, 2017.

[Ras02] Alexey E. Rastegin. Relative error of state-dependent cloning. Physical Review A,


66(4):042304, October 2002.

[Ras03] Alexey E. Rastegin. A lower bound on the relative error of mixed-state cloning
and related operations. Journal of Optics B: Quantum and Semiclassical Optics,
5(6):S647, December 2003. arXiv:quant-ph/0208159.

[Ras06] Alexey E. Rastegin. Sine distance for quantum states. February 2006. arXiv:quant-
ph/0602112.

[Ser17] Alessio Serafini. Quantum Continuous Variables. CRC Press, 2017.

[TGW14a] Masahiro Takeoka, Saikat Guha, and Mark M. Wilde. Fundamental rate-loss tradeoff
for optical quantum key distribution. Nature Communications, 5:5235, October 2014.
arXiv:1504.06390.

[TGW14b] Masahiro Takeoka, Saikat Guha, and Mark M. Wilde. The squashed entanglement
of a quantum channel. IEEE Transactions on Information Theory, 60(8):4987–4998,
August 2014. arXiv:1310.0129.

[Wil18] Mark M. Wilde. Strong and uniform convergence in the teleportation simula-
tion of bosonic Gaussian channels. Physical Review A, 97(6):062305, June 2018.
arXiv:1712.00145.

[WPGP+ 12] Christian Weedbrook, Stefano Pirandola, Raúl Garcı́a-Patrón, Nicolas J. Cerf, Tim-
othy C. Ralph, Jeffrey H. Shapiro, and Seth Lloyd. Gaussian quantum information.
Reviews of Modern Physics, 84(2):621–669, May 2012. arXiv:1110.3234.

[WTB17] Mark M. Wilde, Marco Tomamichel, and Mario Berta. Converse bounds for private
communication over quantum channels. IEEE Transactions on Information Theory,
63(3):1792–1817, March 2017. arXiv:1602.08898.

7
A Hilbert spaces and EPR states
In this appendix, I provide further remarks about the limiting object in (13), often called the “ideal
EPR state” in the literature. These remarks might be considered elementary, but it is worthwhile
to present them in light of the controversy raised in [PBL+ 18]. Recall that a Hilbert space H is
defined to be a complex inner product space that is complete with respect to the metric induced
by the inner product (this metric is the Euclidean distance). As usual, completeness means that
every Cauchy sequence in H has a limit that is also in H.
Turning to the case of interest, it holds that every individual element of the sequence {|ϕNS iRA }NS ≥0 ,
where |ϕNS iRA is defined in (9), is normalizable and is thus in the Hilbert space corresponding to
the modes R and A. In light of this, one might wonder why the limiting object, often called “ideal
EPR state,”
lim |ϕNS iRA (16)
NS →∞

is not in the Hilbert space.


The issue is that the sequence {|ϕNS iRA }NS ≥0 is not a Cauchy sequence, and thus its limit
need not be contained in the Hilbert space nor be normalizable. The limiting object in (16) is
thus not a state, in spite of the fact that it is typically called “ideal EPR state” in the literature
(see, e.g., [WPGP+ 12, Section II-B-7] for an example of this misleading nomenclature). Indeed,
0
for |ϕNS iRA and |ϕNS iRA with NS , NS0 ≥ 0, the metric induced by the inner product is just the
Euclidean distance:
q
NS NS0 0
|ϕ i − |ϕ i = 2 − 2 Re{hϕNS |ϕNS i} (17)

RA RA
2
q
0
= 2(1 − hϕNS |ϕNS i). (18)

A direct calculation then gives that


0 1
hϕNS |ϕNS i = q . (19)
(NS + 1) NS0 + 1 − NS NS0
 p

For fixed NS0 ≥ 0, we have the following expansion:


   
1 1
q
NS NS0 0 0 0

hϕ |ϕ i = 2NS + 1 + 2 NS NS + 1 + O , (20)
NS NS2

which implies that for fixed NS0 ≥ 0, the following limit holds
0

lim |ϕNS iRA − |ϕNS iRA = 2. (21)

NS →∞ 2

This in turn implies that {|ϕNS iRA }NS ≥0 is not a Cauchy sequence, and so its limit limNS →∞ |ϕNS iRA
need not be contained in the Hilbert space.

Вам также может понравиться