See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/229076138

Holevo's bound from a general quantum fluctuation theorem

Article in Physical review A, Atomic, molecular, and optical physics · July 2012
DOI: 10.1103/PhysRevA.86.044302 · Source: arXiv

CITATIONS READS
91 203

2 authors:

Dvir Kafri Sebastian Deffner

University of Maryland, College Park University of Maryland, Baltimore County
36 PUBLICATIONS 1,743 CITATIONS 161 PUBLICATIONS 6,908 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Sebastian Deffner on 16 May 2014.

The user has requested enhancement of the downloaded file.

 Holevo’s bound from a general quantum fluctuation theorem
Dvir Kafri1 and Sebastian Deffner2
1
Joint Quantum Institute, Department of Physics,
University of Maryland, College Park, MD 20742, USA
2
Department of Chemistry and Biochemistry and Institute for Physical Sciences and Technology,
University of Maryland, College Park, MD 20742, USA
(Dated: October 8, 2012)
We give a novel derivation of Holevo’s bound using an important result from nonequilibrium
statistical physics, the fluctuation theorem. To do so we develop a general formalism of quantum
fluctuation theorems for two-time measurements, which explicitly accounts for the back action of
quantum measurements as well as possibly non-unitary time evolution. For a specific choice of ob-
servables this fluctuation theorem yields a measurement-dependent correction to the Holevo bound,
arXiv:1207.2978v2 [quant-ph] 5 Oct 2012

leading to a tighter inequality. We conclude by analyzing equality conditions for the improved
bound.

PACS numbers: 05.30.-d, 05.40.-a, 89.70.Kn

Thermodynamics and information theory are inti- theorems [23], must therefore be treated with care.
mately connected. The most prominent evidence for this The purpose of the present paper is twofold. In the
relationship is that the Clausius entropy [1] is given by first part we derive a general quantum fluctuation theo-
the Shannon information [2] in systems at thermal equi- rem that accounts for the back action of measurements
librium [3]. In particular, Landauer’s principle [4, 5] illus- on reduced systems (see [24] for a similar approach). To
trates that information is a physical, measurable quan- this end, we consider an experimental point of view; we
tity. Thermodynamic work has to be performed in or- assume that the system of interest is coupled to an en-
der to create or erase information. Landauer’s principle vironment which is experimentally inaccessible. Such
can therefore be understood as a statement of the second measurements on open quantum systems are inherently
law of thermodynamics in an information theoretic con- incomplete, since they ignore environmental degrees of
text. This is also true for Holevo’s bound [6], which limits freedom. Information is lost that in principle could have
the amount of classical information that can be encoded been acquired by concurrent measurement of the reser-
in a generic quantum system. Recently, nano-devices - voir. A general formulation of quantum fluctuation the-
for which these principles are directly applicable - have orems must explicitly account for these effects. The inte-
become experimentally accessible [7]. These controlled gral fluctuation theorem we derive is applicable to arbi-
quantum systems have applications ranging from quan- trary orthogonal measurements, for systems undergoing
tum simulation [8, 9], cryptography [10], computing [11– both unitary and non-unitary dynamics.
13], to metrology [14–17]. The main obstacles towards re- In the second part we focus on an information the-
alization of such devices are control noise and interactions oretic consequence of the general quantum fluctuation
with the environment. Thus, a thermodynamic study theorem - Holevo’s theorem. The derivation and implica-
is necessary to fully understand their information theo- tions of this result have attracted much attention [25–29].
retic properties. However, most nano-devices operate far Like more recent works [30–32], our derivation results in
from thermal equilibrium, so tools from non-equilibrium a sharpened statement of Holevo’s bound, which takes
statistical physics are required. In recent years, formu- into account the choice of measurement used to obtain
lations of the second law have been derived which are the encoded information. Further our novel treatment
valid arbitrarily far from equilibrium. These so-called is based on results weaker than the monotonicity of rela-
fluctuation theorems, in particular the Jarzynski equal- tive entropy, and directly leads to necessary and sufficient
ity [18], enable the calculation of equilibrium quantities equality conditions. This illustrates an interesting con-
from non-equilibrium averages over many realizations of nection between quantum thermodynamics and quantum
a single process. They also encompass non-equilibrium, information theory.
information theoretic generalizations of the second law.
As Landauer’s principle is a direct implication of this ap-
General quantum fluctuation theorem Consider
proach [19–21], one may ask whether Holevo’s bound is
a time-dependent quantum system, S, with Hilbert space
also such a result. One complication in this case is that
HS and initial density matrix ρ0 . Information about the
the approach to fluctuation theorems for quantum sys-
state of the system is obtained by performing measure-
tems is mathematically and conceptually more involved.
ments on S at the beginning and end of a specific pro-
Thermodynamic quantities, which are not given as state
cess. Initially a quantum measurement is made of ob-
functions, cannot be assigned an Hermitian operator [22].
servable Ai , with eigenvalues aim . Letting Πim denote the
The proper formulation of quantum thermodynamics for
orthogonal
P projectors into the eigenspaces of Ai , we have
non-equilibrium systems, especially quantum fluctuation
A = m am Πm . Note that the eigenvalues {aim } can
i i i
 2

be degenerate, so the projectors {Πim } may have rank the environment. It plays a crucial role in the following
greater than one. Unlike the classical case, as long as ρ0 discussion and is given by
and Ai do not have a common set of eigenvectors - i.e.
γ = tr exp −Af E M i (ρ0 ) exp Ai




they do not commute - performing a measurement on S . (6)
i
alters its statistics. Measuring
 am maps ρ0 to the state
Πim ρ0 Πim /pm , where pm = tr Πim ρ0 Πim is the probabil- Similar fluctuation theorems of the form hexp (−Σ)i = γcl
ity of the measurement outcome aim . Generally account- have been derived in the context of classical feedback pro-
ing for all possible measurement outcomes, the statistics cesses, where Σ is an entropy production [53]. We note
of S after the measurement are given by the weighted that by appropriate choice of initial and final observables
average of all projections, Ai and Af , Eq. (6) reproduces many known quantum fluc-
tuation theorems [22–24, 34–38], which we will discuss in
M i (ρ0 ) =
X
Πim ρ0 Πim . (1) detail elsewhere.
m
A complementary result to the fluctuation theorem
is Jensen’s Inequality, which states that for any convex
If ρ0 commutesPwith Ai , it commutes with each Πim , function φ′′ (x) ≥ 0 and random variable x, hφ(x)i ≥
so M i (ρ0 ) = i i φ(hxi) [39]. Applying this to Eq. (5) yields
m Πm Πm ρ0 = ρ0 and the statistics
of the system are unaltered by the measurement. Af-
ter measuring aim , S undergoes a generic time evolu- h∆ai ≥ − ln (γ) . (7)
tion, after which it is given by E(Πim ρ0 Πim )/pm . Here
For specific choices of thermodynamically relevant ob-
E represents any linear (unitary or non-unitary) quan-
tum transformation, which is trace-preserving and maps servables Ai and Af , this relation can be understood as
a formulation of the Clausius inequality. InRparticular,
non-negative operators to non-negative operators. Fur- τ

ther, we require that this holds whenever E is extended for a unitary time-evolution Uτ = T> exp −i 0 H(t)dt ,
an initial Gibbsian state ρ0 = exp (−βH(0))/Z0 and
to an operation E ⊗ IE on any enlarged Hilbert space
HS ⊗ HE (IE being the identity map on HE ). Such a corresponding energy measurements, Ai = βH(0) and
transformation is called a trace-preserving, completely Af = βH(τ ), Eq. (5) re-produces the quantum Jarzynski
positive (TCP) map [33]. After this evolution, a measure- equality [22, 40, 41]. Accordingly, Eq. (7) reduces to the
ment of maximum work theorem, β hW i ≥ β∆F , where hW i =
Pa second (not necessarily the same) observable, hH(τ )i − hH(0)i = h∆ai /β and β∆F = − ln (Zτ /Z0 ) =
Af = n afn Πfn , is performed on S. The probability of
measuring afn , conditioned on
having first measured aim , − ln(γ).
is pn|m = tr Πfn E Πim ρ0 Πim /pm . Accordingly, the
joint probability distribution pm→n reads Holevo’s bound We now use the fluctuation theo-
rem (5) to derive a sharpened version of Holevo’s bound.
pm→n = pm · pn|m = tr Πfn E Πim ρ0 Πim . (2)


This bound sets a limit on how much classical informa-
tion can be sent through a (noisy) quantum channel.
We are interested in the probability distribution of possi- Let us consider a message composed of code words wj
ble measurement outcomes, P (∆a) = hδ (∆a − ∆an,m )i, that appear with probability πj . A messenger (Alice) at-
where ∆an,m = afn − aim is a random variable determined tempts to transfer this message to a receiver (Bob) by
in a single measurement run. Its probability distribution encoding each word wj in a quantum state and transmit-
is given by averaging over all possible realizations, ting that state to Bob. We assume that Bob receives the
X state ρj , which may have come through a lossy medium
P (∆a) = δ (∆a − ∆an,m ) pm→n . (3) and therefore be different from the original state pre-
m,n pared by Alice. Bob attempts to infer the word wj from
the encoding by making a generalized measurement of
To derive the integral fluctuation theorem we follow the the state ρj . This corresponds to introducing a probe,
standard approach and compute its characteristic func- initially in a pure state |0i, and making an orthogonal
tion, G(s), which is the Fourier transform of P(∆a) [23] measurement on the compound state ρj ⊗ |0ih0| [42]. If
Z {Πk } represents the set of orthogonal projectors corre-
G(s) = d(∆a) P(∆a) exp (is ∆a) sponding to Bob’s measurement, the probability of mea-
(4) suring Πk , given message wj , is given by
= tr exp isAf E M i (ρ0 ) exp −isAi




.
πk|j = tr {ρj ⊗ |0ih0|Πk } = tr {ρj Mk } , (8)
Choosing s = i, we obtain the general quantum fluctua-
tion theorem where Mk = h0|Πk |0i are operators acting only on the en-
coding degree of freedom.
P Although the operators {Mk }
hexp (−∆a)i = γ . (5) are non-negative and k Mk = I, they are generally
not projectors, Mk2 6= Mk . Such a collection {Mk } is
Since it is explicitly dependent on the map E, the quan- called a positive operator valued measure (POVM), and
tity γ accounts for the information lost by not measuring describes the most general measurement on a quantum
 3

system. The classical message distribution {πj }, output space. Note that HM is not a real, physically acces-
quantum encoding {ρj }, and POVM elements {Mk } de- sible subspace, but rather a mathematical construction
fine a classical-quantum channel [43]. denoting the memory for the classical information of the
A proper measure of how well Bob decodes Alice’s message [43]. We have
message is the mutual information between the en-
coded message and measurement distributions, I = X
P
P ρ0 = πj ρj ⊗ |0ih0| ⊗ |jihj| , (11)
π π
jk j k|j ln πk|j /π k , where πk = j j k|j is
π π
j
the overall P probability of measuring Πk . Note
that I = π
j
j D(π k|j ||πk ), where D(π k|j ||πk ) =
P where the states |ji, each corresponding to word wj , form
π
k k|j ln π k|j /πk is the (classical) relative entropy
an orthonormal basis for HM . We see that, with prob-
[44]. Hence I is a sum of non-negative terms and is 0
ability πj , ρ0 corresponds to the message state |ji. The
if and only if πk|j = πk for all k, j. That is, I vanishes
measured initial and final observables are
only if all outcomes of the measurement are independent
of the encoded word, so that Bob always learns nothing X
Ai = ln ρ̂−1


about the message. j ⊗ |0ih0| ⊗ |jihj|
The probability of the message being wj , conditioned j
on Bob measuring Πk is, πj|k = πk|j πj /πk . We have f
A = − ln (ρ̄ ⊗ |0ih0|) ⊗ IM (12)
X
X X − Ik,j Πk ⊗ |jihj| ,
I = S({πj }) + πk πj|k ln πj|k , (9)
k,j
k j

where Ik,j = ln πk|j /πk . Here ρ̂−1

denotes the inverse
P
where S({πj }) = − j πj ln πj is the Shannon informa- j
−1
tion of distribution {πj }. Since x ln(x) ≤ 0 for |x| ≤ 1, within the support of ρj , so that ρ̂j |ψi = 0 whenever
with equality only for x = 0 and x = 1, we observe that ρj |ψi = 0. This form for Eq. (12) ensures that exp(Ai )
I ≤ S({πj }), with equality if and only if πj|k is 0 or 1 and exp(−Af ) are bounded operators.
for all j, k. In other words, the mutual information I is Note that the states ρj do not represent the original
at most S({πj }), with equality if and only if Bob cor- encoding set up by Alice, but rather its time evolved state
rectly decodes the message in every instance. If Alice’s after undergoing dynamics in a quantum channel. To ap-
encoded states are not perfectly distinguishable (that is, ply the fluctuation theorem (5), we start with the output
if the supports of ρj and ρj ′ are not orthogonal for some of this channel and perform the two measurements, Ai
j 6= j ′ ), then I can never equal S({πj }), no matter what and Af , immediately after each other. The TCP map
measurement Bob chooses to make. Holevo’s theorem is crucial for Eq. (5) is thus the identity map E(ρ) = ρ. Ai
then an upper bound for I, namely commutes with ρ0 , so after measurement of Ai , measure-
X ment of Af is carried out i
 onf the isame state, M (ρ0 ) = ρ0 .
χ ≡ S(ρ̄) − πj · S(ρj ) ≥ I, (10) Computing h∆ai = tr (A − A )ρ0 = χ − I, Eq. (7) is
j

where S(ρ) χ − I ≥ − ln (γ) , (13)

P= −tr {ρ ln ρ} is the von-Neumann entropy,
and ρ̄ = j πj ρj is the density matrix describing the
statistics of the encoding given no knowledge of the mes- where the corresponding quantum efficacy is given by
sage word. Heuristically, the Holevo quantity χ can be
considered as the uncertainty of the encoding with no γ = tr exp −Af ρ0 exp Ai .



(14)
knowledge of the message, minus the average remaining
uncertainty given knowledge of the message. Note that
Equations (13) and (14) constitute the sharpened
if the encoded states are distinguishable, i.e. ρj ρj ′ = O
Holevo’s bound as a consequence of the general quantum
for all j 6= j ′ , then χ = S({πj }), so that with a proper
fluctuation theorem (5). Indeed, our new bound is tighter
measurement Bob may always correctly decode Alice’s
than the usual inequality (10), in the sense that the cor-
message.
rection term, − ln (γ), is always non-negative. Consider
We now show how Holevo’s theorem (10) follows as a
consequence of the general quantum fluctuation theorem
X
ρ0 exp Ai = πj ρj exp ln(ρ̂−1


(5). To do this we must appropriately choose initial state, j ) ⊗ |0ih0| ⊗ |jihj|
evolution operation, and observables so that the random j
variable h∆ai averages to χ − I. In the language of the X
= πj P̂j ⊗ |0ih0| ⊗ |jihj| ,
general fluctuation theorem, let the initial state ρ0 reside
j
in a composite Hilbert space HE ⊗ HP ⊗ HM . HE rep-
(15)
resents the encoding Hilbert space, which Alice prepares
and Bob then measures, HP is the probe Hilbert space
accessible only to Bob, and HM is a message Hilbert where P̂j is the projector into the support of ρj . We can
 4

rewrite Eq. (14) with Eq. (15) as such that Ai Rmj = aim Rmj and ρ0 Rmj 6= 0. Since the
function exp (x) is strictly convex, the random variable
γ = tr exp −Af ρ0 exp Ai



∆anm obtained from the measurements of Ai and Af has
( ) to satisfy ∆anm = − ln(γ) for all measurements with
nonzero probability. Hence an initial measurement of aim
f

X
= tr exp −A πj P̂j ⊗ |0ih0| ⊗ |jihj|
j implies with certainty a final measurement aim − ln(γ).
( !) Since Rmj is a projector into an eigenspace of ρ0 , any
state satisfying Rmj |ψi = |ψi must therefore also be an
X X
≤ πj tr exp ln (ρ̄ ⊗ |0ih0|) + Ik,j Πk .
j k eigenstate of Af with eigenvalue aim − ln(γ), so
(16)
(Af − Ai )Rmj = − ln(γ)Rmj . (20)
where the inequality P is justified by noting that
exp (ln (ρ̄ ⊗ |0ih0|) + k Ik,j Πk ) is non-negative and Using the definition (12) and Mk = h0|Πk |0i, Eq. (19)
P̂j ⊗ |0ih0| is a projection operator. We now use a
P
follows by summing on m noting that m Rmj = P̂j ⊗
statement of the Golden-Thompson inequality [45, 46], |0ih0| ⊗ |jihj|.
that is for any Hermitian operators A and B, we have Conversely, assume that Eq. (19) holds for all j. Since
tr {exp (A + B)} ≤ tr {exp (A) exp (B)}. Note that in Ai and ρ0 commute, we have
the present case, A and B are both logarithms of bounded
Hermitian operators, and are only bounded from above, χ − I = tr ρ0 (Af − Ai )

though the Golden-Thompson inequality still holds [47]. ( ! )
Accordingly, we have X X
= πj tr ρj ln(ρj ) − ln(ρ̄) − Ik,j Mk P̂j
γ = tr exp −Af ρ0 exp Ai j k



X n  o X
= πj tr ρj − ln(γ)P̂j = − ln(γ) πj
( !)
X X
≤ πj tr exp (ln (ρ̄ ⊗ |0ih0|)) · exp Ik,j Πk j j
j k = − ln(γ)
( )
X X (21)
= πj tr (ρ̄ ⊗ |0ih0|) πk|j /πk Πk .
j k
We conclude that Eq. (19) is equivalent to equality in
(17)
Eq. (13). Observe that since χ − I ≥ − ln(γ) ≥ 0, the
P equality condition for χ = I, Eq. (19) with ln(γ) = 0,
From the definition πk = j πj πk|j we finally obtain
is obtained as a corollary of our result [48]. The equal-
( ) ity condition (19) may be used to determine the bound
saturating observable Af self-consistently.
X
γ ≤ tr ρ̄ ⊗ |0ih0| Πk = 1, (18)
k

which shows that − ln (γ) ≥ 0, as desired. We note that Concluding remarks We developed a general
our derivation does not invoke the monotonicity of the framework for quantum fluctuation theorems by explic-
relative entropy or equivalent statements [48]. Instead itly accounting for the back action of quantum measure-
we have used only Jensen’s inequality and the Golden- ments. With this new result, we showed that quantum
Thompson inequality, which are weaker results [49, 50]. mechanical formulations of the second law are intimately
Equality conditions Holevo’s bound (12) is obtained tied to quantum information theory by deriving Holevo’s
with the help of Jensen’s inequality. For strictly convex bound as a consequence of a fluctuation theorem. The
functions φ′′ (x) > 0, the Jensen bound hφ(x)i ≥ φ(hxi) new approach not only provides a simple derivation, but
achieves equality if and only if the random variable x is also a sharpened statement of the original bound and a
constant valued. This allows us to derive the equality corresponding equality criterion.
conditions for (13) in a straightforward manner. Specifi-
cally, equality is achieved only if
! Acknowledgments
X
−1
− ln(ρ̂j ) − ln(ρ̄) − Ik,j Mk P̂j = − ln(γ)P̂j
k The authors thank Jacob Taylor and Eric Lutz for in-
(19) teresting discussions. SD acknowledges financial support
for all j. This follows from a few simple observations. by a fellowship within the postdoc-program of the Ger-
First assume that χ − I = − ln(γ). As Ai , ρ0 and the man Academic Exchange Service (DAAD, contract No
projectors I ⊗ |0ih0| ⊗ |jihj| mutually commute, we con- D/11/40955). DK acknowledges financial support by a
(j)
sider a mutual eigenprojector Rmj = Rm ⊗ |0ih0| ⊗ |jihj| fellowship from the Joint Quantum Institute.
 5

[1] R. Clausius, Abhandlungen über die mechanische 56, 131 (1997).

Wärmetheorie (Vieweg, Braunschweig, Germany, 1864). [30] B. Schumacher, M. Westmoreland, and W. K. Wootters,
[2] C. E. Shannon, Bell Sys. Tech. J. 27, 379 (1948). Phys. Rev. Lett. 76, 3452 (1996).
[3] H. Callen, Thermodynamics and an Introduction to Ther- [31] K. Jacobs, Phys. Rev. A 68, 054302 (2003).
mostastistics (Wiley, New York, USA, 1985). [32] K. Jacobs, J. Math. Phys. 47, 012102 (2006).
[4] R. Landauer, IBM J. Research and Develop. 5, 183 [33] B. Schumacher, Phys. Rev. A 2614, 54 (1996).
(1961). [34] P. Talkner, M. Campisi, and P. Hänggi, J. Stat. Mech. p.
[5] A. Bérut, A. Arakelyan, A. Petrosyan, S. Ciliberto, P02025 (2009).
R. Dillenscheinder, and E. Lutz, Nature 483, 187 (2012). [35] S. Deffner and E. Lutz, Phys. Rev. Lett. 107, 140404
[6] A. S. Holevo, IEEE Trans. Info. Theo. 44, 269 (1998). (2011).
[7] C. Monroe, Nature 416, 238 (2002). [36] M. Campisi, P. Talkner, and P. Hänggi, Phys. Rev. E 83,
[8] R. Feynman, Int. J. Theo. Phys. 21, 476 (1982). 041114 (2011).
[9] S. Lloyd, Science 273, 1073 (1996). [37] C. Jarzynski and D. K. Wójcik, Phys. Rev. Lett. 92,
[10] A. K. Ekert, Phys. Rev. Lett. 67, 661 (1991). 230602 (2004).
[11] L. K. Grover, Annual ACM Symposium of Theory of [38] Y. Morikuni and H. Tasaki, J. Stat. Phys. 143, 1 (2011).
Computing p. 212 (1996). [39] D. Chandler, Introduction to Modern Statistical Mechan-
[12] P. W. Shor, SIAM Review 41, 303 (1999). ics (Oxford University Press, Oxford, UK, 1987).
[13] A. W. Harrow, A. Hassidim, and S. Lloyd, Phys. Rev. [40] J. Kurchan (2000), arXiv:cond-mat/0007360v2.
Lett. 103, 150502 (2009). [41] H. Tasaki (2000), arXiv:cond-mat/0000244v2.
[14] M. Kasevich and S. Chu, Phys. Rev. Lett. 67, 181 (1991). [42] V. B. Braginsky and F. Y. Khalili, Quantum Measure-
[15] H. J. Kimble, Y. Levin, A. B. Matsko, K. S. Thorne, and ment (Cambridge University Press, Cambridge, UK,
S. P. Vyatchanin, Phys. Rev. D 65, 022002 (2001). 1992).
[16] V. Giovannetti, S. Lloyd, and L. Maccone, Phys. Rev. [43] M. A. Nielsen and I. L. Chuang, Quantum Computation
Lett. 96, 010401 (2006). and Quantum Information (Cambridge University Press,
[17] J. Hudson, D. M. Kara, I. J. Smallman, B. E. Sauer, Cambridge, UK, 2000).
M. R. Tarbutt, and E. A. Hinds, Nature 473, 493 (2011). [44] S. Kullback, Information Theory and Statistics (Peter
[18] C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997). Smith, Gloucester, USA, 1978).
[19] R. Kawai, J. M. R. Parrondo, and C. V. den Broeck, [45] S. Golden, Phys. Rev. 137, B1127 (1965).
Phys. Rev. Lett. 98, 080602 (2007). [46] C. J. Thompson, J. Math. Phys. 6, 1812 (1965).
[20] C. Jarzynski, Eur. Phys. J. B 64, 331 (2008). [47] M. B. Ruskai, Comm. Math. Phys. 26, 280 (1972).
[21] M. Esposito and C. van den Broeck, EPL (Europhys. [48] M. B. Ruskai, J. Math. Phys. 9, 4358 (2002).
Lett.) 95, 40004 (2011). [49] D. Petz, Communications in Mathematical Physics 114,
[22] P. Talkner, E. Lutz, and P. Hänggi, Phys. Rev. E 75, 345 (1988).
050102 (R) (2007). [50] D. Petz, in Functional Analysis and Operator Theory
[23] M. Campisi, P. Hänggi, and P. Talkner, Rev. Mod. Phys. (Banach Center Publications, 1994), vol. 30.
83, 771 (2011). [51] T. Sagawa and M. Ueda, Phys. Rev. Lett. 104, 090602
[24] V. Vedral, J. Phys. A: Math. Theor. 45, 272001 (2012). (2010).
[25] H. P. Yuen and M. Ozawa, Phys. Rev. Lett. 70, 363 [52] T. Sagawa and M. Ueda, Phys. Rev. E 85, 021104 (2012).
(1993). [53] For these the right hand side, γcl , is commonly called the
[26] M. J. W. Hall and M. J. O’Rourke, Quantum Opt. 5, 161 (classical) efficacy of the feedback protocol [38, 51]. In the
(1993). present context γ can then be interpreted as a quantum
[27] C. A. Fuchs and C. M. Caves, Phys. Rev. Lett. 73, 3047 efficacy corresponding to the observables Ai and Af , and
(1994). the TCP map E. Note that γ only takes the from of the
[28] P. Hausladen, R. Jozsa, B. Schumacher, M. Westmore- classical efficacy, and usually may not be regarded as its
land, and W. K. Wootters, Phys. Rev. A 54, 1869 (1996). physical quantum analogue [52].
[29] B. Schumacher and M. D. Westmoreland, Phys. Rev. A

View publication stats