arXiv:quant-ph/0201149 v3 14 Oct 2002

Additivity of the Classical Capacity of Entanglement-Breaking Quantum Channels

Peter W. Shor
AT&T Labs, Florham Park, NJ 07922, USA

January 11, 2004

Abstract We show that for the tensor product of an entanglement-breaking quantum channel with an arbitrary quantum channel, both the minimum entropy of an output of the channel and the Holevo-Schumacher-Westmoreland capacity are additive. In addition, for the tensor product of two arbitrary quantum channels, we give a bound involving entanglement of formation for the amount of subadditivity (for minimum entropy output) or superadditivity (for classical capacity) that can occur. One of the more important open questions of quantum information theory is the determination of the capacity of a quantum channel for carrying classical information. This question has been only partially resolved. If entanglement between multiple inputs to the channel is not allowed, a formula for the classical capacity of a quantum channel has indeed been discovered [1, 2]. This capacity formula for a quantum channel is () = max H
pi ,i i

(pi i )
i

pi H (i )

(1)

where H is the von Neumann entropy H () = Tr log , and where the maximization is over probability distributions pi on density matrices i over the input space of the channel. This maximum can be attained because we need at most d2 density matrices i to achieve any attainable value of ({pi , i }) = H
i

(pi i )
i

pi H (i )

(2)

and are thus maximizing over a compact space. The general capacity of a quantum channel , without feedback or prior entanglement between sender and receiver, but possibly using entangled inputs, is 1 C () = lim (n ), (3) n n i.e., the limit for large n of the capacity when we permit the input to be entangled over blocks of n channel uses. This limit can be shown to exist because satises the superadditivity condition ( ) () + (). (4) It is conjectured that equality holds, i.e., that is additive, in which case would give the classical capacity of a quantum channel without feedback. Substantial work has been done

on this conjecture [3, 4], and it has been proven for several special cases. In particular, it has been proven when one of the channels is the identity channel [4, 5], when one of the channels is what A. S. Holevo calls a c-q or q-c channel (these terms will be dened later) [6, 7], and when one of the channels is a unital qubit channel [8]. We will prove additivity for the special case where one of the two channels is entanglement breaking. Entanglement breaking channels are channels which destroy entanglement with other quantum systems. That is, when the input state is entangled between the input space Hin and another quantum system Href , the output of the channel is no longer entangled with the system Href . Both c-q and q-c channels are special cases of entanglement breaking channels. A c-q channel is a channel which can be expressed by the composition of a complete von Neumann measurement on the input space followed by an arbitrary completely positive trace-preserving (CPT) map. A q-c channel can be expressed as the composition of a CPT map followed by a complete von Neumann measurement on the output space. Stated more intuitively, for c-q maps, the input can be treated as being classical, and for q-c maps, the output can be taken to be classical. In either case, the von Neumann measurement eliminates any entanglement between the input space and another system, so c-q and q-c maps are both special cases of entanglement breaking channels. In a conversation with the author, Michal Horodecki [9] gave a simple proof that any entanglement breaking channel can be expressed as a q-c-q channel; that is, the composition of a CPT operator followed by a complete von Neumann measurement followed by another CPT operator. (See also [10] for details of this proof.) As a consequence, the action of an entanglement breaking channel on a state can always be written in the following form introduced by Holevo [6]: () =
i

Tr Xi i

(5)

where {Xi } form a general POVM and {i } are arbitrary states. For a c-q map, Xi = | i i | where | i form an orthonormal basis, and for a q-c map i = | i i |. The additivity problem for capacity is closely related to another additivity problem; that of the minimum entropy output of a channel [11]. For the case of entanglement breaking channels, we rst found the additivity proof for the minimum entropy output, and then discovered a straightforward way to extend this additivity proof to cover the classical capacity. In this paper, we rst give the proof for additivity of minimum entropy output, as this proof contains the important ideas for the capacity proof, but has signicantly fewer technicalities. Theorem 1 For an arbitrary quantum channel , and an entanglement breaking channel min H ( )(AB ) = min H (A ) + min H (B ) .
AB A B

(6)

Proof: The left-hand side is clearly at most the right-hand side, as can be seen by choosing AB = A B . We would like to show that it is at least the right-hand side. We use the strong subadditivity property of von Neumann entropy [12]. Consider the minimum obtainable value of H ( )(AB ) . Because is entanglement breaking, (I )(AB ) =
j

qj | aj aj | | bj bj |

(7)

for some qj , | aj HA and | bj HB . Now, we apply to the state ABC =

j

qj (| aj aj |) | bj bj | | j j |

(8)

the property of strong subadditivity in the form H (AB ) H (ABC ) H (BC ) + H (B ). We have AB =
j

(9)

qj (| aj aj |) | bj bj | ( )(AB ),

(10)

the quantity for the entropy of which we would like a lower bound. Now, note that H (ABC ) H (BC ) = =
j

H (AC ) H (C ) qj H (| aj aj |) .

(11)

The rst equality above follows from the facts that, in Eq. (8), the | j j | form an orthonormal set and the | bj bj | are pure states, so that H (ABC ) = H (AC ) and H (BC ) = H (C ). The second equality follows from the chain rule for entropy, namely H
j

qj j | j j |

H qj | j j |

=
j

qj H (j ).

(12)

Now, note that B =

j

qj | bj bj | TrA (I )(AB ) (TrA AB )

(13)

= =

Putting the above equalities together, we see that H ( )(AB )

j

qj H (| aj aj |) + H (TrA AB ) .

(14)

Since j qj = 1, the right-hand side is clearly at least the sum of the minimum output entropies of and of . We have thus shown that the minimum output entropy is additive for the tensor product of two channels if one of the channels is an entanglement breaking channel. We now prove the corresponding additivity result for the Holevo-Schumacher-Westmoreland capacity ; recall () = max H (
pi ,i i

pi i )
i

pi H (i )

(15)

over probability distributions pi and density matrices i .

Theorem 2 For an arbitrary quantum channel , and an entanglement breaking channel ( ) = () + () (16)

Proof: The capacity is composed of two terms. We will be treating these two terms separately. For the second term, additivity is shown in essentially the same way as in the proof of additivity for minimum entropy, and for the rst term, additivity follows from the subadditivity of von Neumann entropy. Again, we assume that we have an arbitrary quantum channel , and an entanglement breaking channel . We use strong subadditivity. Consider the optimal signal states for , i.e., the pi and i such that ( ) = H ( )() pi H ( )(i ) (17)

where = pi i . Let us consider the state (I )(i ). Because is an entanglement breaking map, this state is separable, and so (I )(i ) =
j

qij | aij aij | | bij bij |

(18)

for some qij , | aij , | bij . Now, we apply strong subadditivity to the state ABC =
j

qij (| aij aij |) | bij bij | | j j | .

(19)

To simplify notation, we let the dependence of on i be implicit. Again, we apply strong subadditivity in the form H (AB ) H (ABC ) H (BC ) + H (B ) As before, H (AB ) = H ( )(i ) . We also have that H (B ) = H (TrA i ) and H (ABC ) H (BC ) =
j

(20) (21) (22) (23)

qij H (| aij aij |) .

We let | aij aij | = ij . Then TrB i = H ( )(i )

j

qij ij . Combining the terms, we observe qij H (ij ) + H (TrA i ) (24)

Now, let us sum over all the states i . We obtain pi H ( )(i )

i i,j

pi qij H (ij ) +
i

pi H (TrA i ) .

(25)

Using subadditivity of von Neumann entropy and the above inequality (25), we get that ( ) = H ( )()
i

pi H ( )(i )

(26)

H (TrB ) + H (TrA )
i,j

pi qij H (ij )
i

pi H (TrA i ) .

However, since pi qij ij =

i,j i

pi TrB i = TrB

and
i

pi TrA i = TrA ,

(27)

we see that ( ) () + ().

(28)

As the opposite inequality is easy, we have additivity of for entanglement breaking channels. We nally give a bound on the amount of superadditivity for general channels. For this, we need to dene the entanglement of formation of a bipartite state. This is another quantity that is conjectured to be additive, but for which additivity has not been proved. Entanglement of formation for a bipartite state AB is dened EF (AB ) =
pi ,i i pi i =AB

min

pi H (TrA i )

(29)

where the minimization is over probability distributions pi on rank-one density matrices i such that i pi i = AB . The theorem is Theorem 3 Suppose we have two quantum channels, i.e., completely positive trace preserving maps, and . Then min H ( )(AB )
AB

min H (A )
A

+ min H (B )
B

(30)

max EF (I )(AB )
AB

and ( ) () + () + max EF (I )(AB ) .

AB

(31)

Note that the formulation of the theorem is asymmetric with respect to and . Thus, to bound the amount of sub- or superadditivity, one can use either the entanglement of formation of (I )(AB ) or of ( I )(AB ), whichever is smaller. Proof: We rst give the proof of the rst part of Theorem 3. Let (I )(AB ) =
i

qi i

(32)

be the decomposition of (I )(AB ) into pure states i that minimizes entanglement of formation, i.e., so that j qj H (TrA j ) is minimum. Now, we consider ABC =
j

qj ( I )(j ) | j j |

(33)

and apply strong subadditivity to this state. We obtain H (AB ) H (ABC ) H (C ) + H (B ) H (BC ) H (C ) . As in (10), we have H (AB ) = H ( )(AB ) . Similar to (13), we get H ( B ) = H
j

(34)

(35)

qj TrA j = H (TrA AB )

(36)

Furthermore, the choice of j and the denition of EF gives H (BC ) H (C ) = EF (I )(AB ) Finally application of the entropy chain rule (12) gives H (ABC ) H (C ) =
j

(37)

qj H ( I )(j )

(38)

The expression (36) is bounded below by min H () . The second expression (37) is bounded above by max EF (I )() . The third expression (38) is bounded below by min H ( I )() , which is known to equal min H () . Combining these three expressions give the rst part of Theorem 3. To prove the second part of the theorem, (38) must be replaced by H (ABC ) H (C )
jk

qj rjk H (| vjk vjk |)

(39)

for states | vjk and probabilities qj rjk such that qj rjk | vjk vjk | = TrB j .
j,k

(40)

We then consider the signal states i and the associated probabilities pi which give the value of ( ) in Equation (1), and let i pi i = . We now use expressions (36), (37), (39) with i in the place of AB . Combining these three expressions yields H ( )(i ) H (TrA i ) +
j,k

qij rijk H (| vijk vijk |)

(41)

EF (I )(i ) . The second part of Theorem 3 then follows in a way entirely analogous to the proof of Theorem 2. We use the equalities TrB =
i

pi TrA i

(42)

and TrA =
i,j,k

pi qij rijk | vijk vijk | ,

(43)

and expand ( ) similarly to Eq. (26) to obtain Eq. (31). We still must prove the inequality (39). The left hand side of (39) is qj H ( I )(j )
j

(44)

Now, j is a purication of TrB j , and H ( I )(j ) = H ( I )( ) for any other quantum state which is a purication of j = TrB j . Let j = k qjk | vjk vjk | be the eigenvector decomposition of j . A different purication is j =
k

qjk | vjk | k | k
k

qjk vjk | k | k |

(45)

It sufces to show that H ( I )(j ) H Tr3 ( I )(j ) H Tr12 ( I )(j ) (46)

as the rst term in the above equation is H ( I )(j ) , the second is H ({qjk }k ) + k qjk H (| vjk vjk |) , and the third is H ({qjk }k ). However, the above equation follows from the inequality H (34 ) H (3 ) H (4 ), which is a consquence (after another purication) of the subadditivity property of entropy [13].

Acknowledgements
I would like to thank Michal and Pawel Horodecki, Chris King, and Beth Ruskai for interesting and informative conversations. In addition, I would like to thank Michal Horodecki for discovering the proof that all entanglement-breaking channels can be expressed in Holevos form (i.e., Eq. (5)), and Chris King and Beth Ruskai for their extremely helpful suggestions for improving the exposition of this paper.

References
[1] A. S. Holevo, The capacity of quantum channel with general signal states, IEEE Trans. Info Thy. 44, 269273 (1998); LANL e-print quant-ph/9611023. [2] B. Schumacher and M. D. Westmoreland, Sending classical information via noisy quantum channels, Phys. Rev. A 56, 131138 (1997). [3] G. G. Amosov and A. S. Holevo, On the multiplicativity conjecture for quantum channels, LANL e-print math-ph/0103015 (2001). [4] G. G. Amosov, A. S. Holevo, and R. F. Werner On some additivity problems in quantum information theory, Probl. Inform. Transm. 36 (4), 2534 (2000); LANL e-print math-ph/0003002. [5] B. Schumacher and M. D. Westmoreland, Relative entropy in quantum information theory, in Quantum Information and Quantum Computation: A Millenium Volume, AMS Contemporary Mathematics Vol. 305, (S. J. Lomomaco, Jr. and H. E. Brandt, eds., AMS Press, Providence, Rhode Island, 2002); LANL e-print quant-ph/0004045. [6] A. S. Holevo, Coding theorems for quantum channels, Russian Math Surveys 53, 12951331 (1998); LANL e-print quant-ph/9809023. [7] C. King, Maximization of capacity and lp norms for some product channels, J. Math. Phys 43, 12471260 (2002); LANL e-print quant-ph/0103086. [8] C. King, Additivity for unital qubit channels, LANL e-print quant-ph/0103156 (2001). [9] M. Horodecki, private communication (2001). [10] M. B. Ruskai, Entanglement breaking channels, Rev. Math. Phys., to appear; LANL e-print quant-ph/0207100. [11] C. King and M. B. Ruskai, Minimal entropy of states emerging from noisy quantum channels, IEEE Trans. Inf. Theory 47, 119 (2001); LANL e-print quant-ph/0103156. [12] E. Lieb and M.B. Ruskai, Proof of the strong subadditivity of quantum mechanical entropy, J. Math. Phys. 14, 19381941 (1973). [13] H. Araki and E. Lieb, Entropy inequalities, Commun. Math. Phys. 18, 160170 (1970).