Contemporary Mathematics

Remarks on the Additivity Conjectures for Quantum Channels

Christopher King

Abstract. In this article we present the statements of the additivity conjectures for quantum channels, together with some brief history and context. The conjectures were recently shown to be false in general, and we include a review of the state of knowledge concerning counterexamples. The article concludes with a short list of current open questions and topics of research arising out of the additivity problem.
1. Introduction Quantum information theory (QIT) has emerged in the last two decades as a vibrant and exciting eld of research. As well as providing a novel perspective on quantum theory, the eld has generated new conjectures and results in mathematics. The particular focus of this article is the family of related questions known as the additivity conjectures. These conjectures emerged from the attempt to nd a closed form expression for the information-carrying capacity of a noisy quantum channel. There has been much progress toward this goal, but there are still many interesting questions which are the subject of current research. The reader is referred to the papers [8, 32, 33] for a fuller account of the early history of this topic. The most signicant recent result is Hastings proof of the existence of counterexamples to the additivity conjectures [27]. It might have been expected that this result would kill the problem, but in fact it has stimulated further research toward nding explicit counterexamples, and has generated new questions about the extent and typicality of the additivity violations that can occur. The purpose of this article is to state and explain the additivity conjectures, and to indicate some directions of current research. There is no attempt to provide a completely comprehensive survey of all research in this eld, however it is hoped that enough references are provided to allow easy access to the literature. Progress is rapid, so
1991 Mathematics Subject Classication. Primary 81P45, 94A40. Key words and phrases. Quantum information theory, channel capacity, additivity conjectures. c 2010 by the author. This paper may be reproduced, in its entirety, for non-commercial purposes.
c 0000 (copyright holder)

C. KING

the state of current research presented here is a snapshot, and may be soon out of date. 2. Quantum channels The basic object of study in quantum information theory is the quantum channel. Quantum channels arise naturally in several dierent ways; as a physical description of decoherence, as the quantum analog of Shannons notion of a discrete memoryless channel, and as a natural class of maps on matrix algebras. We shall quickly review these three descriptions below, starting with the mathematical denition. 2.1. Quantum channel as a completely positive map on a matrix algebra. Stinespring dened completely positive maps in the context of C -algebras [48]. Here we restrict attention to maps on nite-dimensional matrix algebras (see [11] for more details). Let Mn denote the algebra of complex n n matrices. Definition 2.1. A linear map : Mn Mm is called completely positive (or CP) if Ik : Mn Mk Mm Mk is positivity preserving for every k 1, where Ik is the identity map on Mk. Various characterizations of CP maps are known, we will describe several. The rst is the existence of a Kraus representation. Namely, a map : Mn Mm is CP if and only if there are m n matrices A1 ,. , AK such that for all M Mn ,

(M ) =

Ai M A i
The matrices {Ai } are known as Kraus operators for the map. The Kraus representation is not unique, however if A1 ,. , AK and B1 ,. , BL are both Kraus representations for the same map with K L then there is a K L matrix W = (wij ) such that (2.2) Ai =

wij Bj ,

W W = IK
Furthermore there is a Kraus representation with K nm. Stinesprings original denition of a CP map was expressed in the Heisenberg representation, namely as an operator on the observables of a quantum system. In the following we will use the Schrdinger representation, and regard a CP map o as acting on the states of a quantum system. We rst review some denitions of quantum systems and states. Let H be the Hilbert space of a nite-dimensional quantum system, thus H = Cn for some n. We denote by S(H) Mn the set of states on H, that is the convex set of positive semidenite operators with trace one. A state is pure if it has rank one, otherwise it is mixed. The pure states are the extreme points of S(H). If is a quantum state and is a CP map acting on S(H) then () is required to also be a state, that is a positive semidenite matrix with trace 1. The CP condition ensures that () 0, but in order to preserve probabilities the map is also required to be trace-preserving (TP). Thus a quantum channel is nally dened to be a CPTP map between matrix algebras.

ADDITIVITY CONJECTURES

Definition 2.2. A linear map : Mn Mm is called a quantum channel if is completely positive and trace preserving (CPTP). For a CPTP map , the matrices in the Kraus representation (2.1) must satisfy the condition

A Ai = In i

The Stinespring Dilation Theorem [48] implies a second way to describe a quantum channel, namely as an isometric embedding followed by a partial trace. Let W : Cn Cd Cm be an isometric embedding, that is a linear map satisfying W W = I, and dene : Mn Mm by (2.4) () = Tr Cd W W Every quantum channel : Mn Mm can be described in this way with some choice of W and d. In particular, if the channel is presented by a Kraus representation as in (2.1), then taking d = K and A1 A2 W =. (2.5) . . AK gives the representation (2.4). 2.2. Quantum channels as models for decoherence. This section describes a physical interpretation of quantum channels (this material is explained in detail in many places, two excellent sources being the text by Nielsen and Chuang [42], and the online notes of J. Preskills course at CalTech). Consider a bipartite quantum model with state space H E, where H is the state space of our system (which we can control and measure, at least to some extent), and where E is the state space of the environment, which is outside our control. In the Schrdinger pico ture, the dynamics of the coupled system is governed by an interaction Hamiltonian H leading to the state evolution (2.6) where U = eiHt/ (we henceforth suppress the time dependence as we are not concerned with dynamics, but rather with the description of the system at a xed time). In quantum theory an observable is represented by a Hermitian operator acting on the state space. A local observable A of the system H acts on the coupled system as A I, where I is the identity operator on E. The expected value of the result of a measurement of A I in the state U U is (2.7) (2.8) Tr [(A I) U U ] = Tr H A = Tr E [U U ] where is the reduced density matrix of the system given by Here Tr E is the partial trace over the environment state space, and Tr H is the partial trace over the system. Thus contains all information about the state of the coupled system which can be accessed by local measurements. eiHt/ eiHt/ = U U

Suppose now that the system is prepared in a state (this assumes that the experimenter can isolate the system from its environment for long enough to prepare the state). Then the initial state of the coupled system will be a product where is some state of the environment. Thus in (2.8) can be viewed as the result of a linear map applied to the initial state : (2.9) which serves as the denition of the map. In general maps pure states into mixed states, which corresponds to our view of decoherence as introducing noise into a system through entanglement with the environment. From the denition (2.9) it follows that is trace-preserving and completely positive, and thus is a quantum channel. In the case where is a pure state it can be seen that (2.9) is equivalent to the formulation (2.4), with the same input and output spaces (if is a mixed state then can still be written in the form (2.4) but with a larger environment). 2.3. Quantum channel as an information device. Shannons model of a discrete memoryless channel [44] is based on the notion that an information source can be viewed as a stochastic process, producing strings of random letters drawn from a source alphabet. Such a string is then transmitted through a channel, and the output of the channel is another stochastic process which is correlated with the input. The simplest assumption to make is that the channel acts independently on each letter, randomly changing it according to a xed transition matrix {pij }. That is, letting X denote the input and Y the output letters, (2.10) P (Y = j|X = i) = pij ,

= Tr E [U ( ) U ] = ()

pij = 1
If X is a random variable with distribution i = P (X = i), then the distribution of Y is given by (2.11) qj = P (Y = j) =
From this point of view a discrete memoryless channel is a linear map T on probability distributions = (1 , 2 ,. ): (2.12) T : q, qj = T ()j =
The channel acts independently on successive letters in the input string: This can be expressed as the action of the m-fold product map T m on input product distributions: letting (X) denote the distribution of the input letter X, (2.14) By linearity the map T m extends uniquely to a map on the set of all probability distributions on m-letter input strings, and it is straightforward to check that this map is also a channel. This viewpoint on a classical channel leads directly to the denition of a quantum channel as a linear map on quantum states. Namely, let Hin and Hout be the input and output state spaces of the channel. Then the set of states S(Hin ) T m ((X1 ) (Xm )) = T ((X1 )) T ((Xm )) (2.13) P (Y1 = j1 ,. , Ym = jm |X1 = i1 ,. , Xm = im ) = pi1 ,j1 pim ,jm

is the quantum analog of the set of input probability distributions for the classical channel T , and similarly S(Hout ) is the analog of the output distributions. Thus a quantum channel is a linear map (2.15) Physical considerations imply that should be trace-preserving and positivity preserving. Furthermore, as in the case of a classical channel, the quantum channel acts independently on successive states in a string of inputs; letting 1 ,. , m denote the input states, the output string is (2.16) As in the classical case, by linearity the map m extends to a map on the states m of the full tensor product of input spaces Hin. However, unlike in the classical m case, it does not follow automatically that is itself a quantum channel. This requires the additional assumption that is completely positive. 3. The capacity of a quantum channel 3.1. Classical channel. Shannon dened the capacity of a discrete memoryless channel as the maximum rate for transmission of information through the channel [44]. This maximum rate is approached by encoding the information in input strings which are suciently dierent that the resulting output strings can be reliably distinguished. By using longer and longer strings to encode the information, the maximum rate can be asymptotically approached. Furthermore, Shannon provided an explicit formula for this maximum rate. Suppose that the input letters X have distribution {i } and the channel matrix is {pij }, then the mutual information of the input and output is dened as pij (3.1) i pij log I(X, Y ) = qj i,j where again {qj } is the distribution of the channel output Y. Shannon proved that the maximum rate for information transmission using the channel with input source X is I(X, Y ). Following this, the classical channel capacity Cclass is dened to be the maximum of I(X, Y ) over all possible distributions of X, that is (3.2) Cclass (T ) = sup I(X, Y )

: S(Hin ) S(Hout )

m (1 m ) = (1 ) (m )
3.2. Shannon capacity of quantum channel. The capacity of a quantum channel is dened by viewing it as a particular realization of a classical channel. That is, one considers the use of the channel for transmission of a signal composed of a string of letters drawn from a nite alphabet. Transmission is achieved by rst encoding the input signal in a quantum state, then allowing the channel to act on the state, and nally measuring the output state in order to recover the information. In the simplest protocol each letter i of the input alphabet is encoded as a quantum state i in the input space Hin. This input state is mapped by the channel to an output state (i ). At the output a measurement is performed. Recall that in quantum theory a measurement is dened by a POVM, that is a collection of k positive semidenite matrices E1 ,. , Ek satisfying j=1 Ej = I. When applied to the output state (i ), this measurement returns the index j with probability

Tr Ej (i ). The classical channel is thus constructed by choosing a set of states {i } to encode the input letters, and choosing a POVM {Ej } to measure the output. The transition matrix of the channel is (3.3) pij = Tr Ej (i )
Now the formula (3.2) provides the capacity of this channel. The Shannon capacity of is then dened to be the maximum of this capacity taken over all choices of input encoding states and output measurements, that is (3.4) CShan () = sup Cclass (T )
where T is the classical channel with transition matrix (3.3). Note that the number of input states and the number of POVM elements is not xed on the right side of (3.4), and the supremum includes a search over all sizes of these sets (though there are dimension-dependent bounds for the number of states and POVM elements needed). 3.3. Entangled inputs and outputs. However this is not the end of the story. In the formula (3.4) there is an implicit assumption about the way that input strings are encoded, namely as products drawn from a xed set of states. For example, suppose that the alphabet is {0, 1}, and we want to eciently transmit the four strings {00, 01, 10, 11}. Using the above product state protocol we would select two input states 0 , 1 and encode these strings as the product states (3.5) 0 , 1 , 0 , 1
Then at the output we select a POVM {E0 , E1 } which tries to distinguish the states (0 ) and (1 ). The average error probability for these four strings will be (3.6) pe = 1 4
[1 Tr Ei (i ) Tr Ej (j )]
However there may be another way to encode and decode the strings that produces a smaller error probability. For example, the four input strings could be encoded using the four Bell states: these are dened as [42] |00 |01 |10 |11 The encoding would then be (3.7) ||, ||, 10 ||, 11 || = = = = (|00 + |11 ) (|01 + |10 ) (|00 |11 ) (|01 |10 )
Furthermore at the output we may also select a POVM that projects onto states which are entangled across the outputs. So we choose a POVM {E00 , E01 , E10 , E11 }

where (3.8)

Eij = I I. Then the average error probability becomes p e 1 = 4

[1 Tr Eij ( )(|ij ij |)]

If p < pe for any choice of single-letter protocols {i , Ei } then we may increase the e channel capacity beyond CShan () by encoding input strings with this entangled protocol, thus implying that (3.9) CShan ( ) > 2 CShan ()

The existence of channels satisfying the superadditivity property (3.9) was demonstrated in Holevos 1973 paper [31]. In general, it is possible to encode an input string using a state which is entangled across multiple channel inputs, and it is possible to use a POVM which uses operators which are entangled across the channel outputs. When such entangled encodings and measurements are considered over n uses of the channel the resulting capacity is (3.10) 1 CShan (n ) n
(The factor 1/n is needed because we consider information transfer per channel use). By allowing n to increase arbitrarily we reach the ultimate capacity which is given by (3.11) Cult () = lim

1 CShan (n ) n

Note that for a classical channel T superadditivity does not occur, and thus Cult (T ) = Cclass (T ). 3.4. The Holevo capacity. The Holevo capacity of the channel is dened as [31, 32, 33] (3.12) () = sup

p i i )

pi S((i ))
where the sup on the right side runs over all input ensembles for the channel, and where S() is the von Neumann entropy (the function inside the sup is convex and hence the ensemble may be assumed to consist of pure states). Holevo proved the following bound for the Shannon capacity: (3.13) CShan () ()
(this bound had also appeared in earlier work [25], [40]). The Holevo capacity was given an operational meaning through the later work of Hausladen et al [28], Holevo [33], and Schumacher and Westmoreland [43], who proved that () is equal to a restricted version of the capacity Cult (). The restricted version is obtained by allowing entangled measurements at the output for multiple channel uses, but allowing only product input states. It follows that (3.14) () Cult ()
Furthermore the entangled input states may be re-introduced by considering multiple channel uses, thus leading to 1 (3.15) (n ) Cult () = lim n n 4. The additivity conjectures 4.1. The additivity conjecture for Holevo capacity. The original additivity conjecture [7] was to the eect that the regularization is unnecessary in (3.15), meaning that it can be replaced by the simpler one-shot formula (4.1) (4.2) Cult () = () (n ) = n () Equivalently, the function is additive over n-fold tensor products: A slightly generalized version of this soon became the standard additivity conjecture: for any two quantum channels and , (4.3) There is an operational meaning for this conjecture. It says that the channel capacity is achieved using coding on product states only, in other words using entangled input states for the channel does not increase the capacity. It was known that entangled measurements at the output are necessary to achieve the Holevo capacity, and hence also the full channel capacity, but this conjecture implies that the input states can always be chosen from an ensemble consisting only of product states. 4.2. Equivalence to other additivity conjectures. In a quest for new approaches to the additivity problem, the minimal output entropy and minimal output Renyi entropy were studied. These are: (4.4) and for p > log Tr (()p ) 1p Note that limp1 Sp,min () = Smin (). The additivity conjecture is: for all channels and , and all p 1 (4.5) Sp,min () = inf

( ) = () + ()

Smin () = inf S(())
Sp,min ( ) = Sp,min () + Sp,min ()
The question of additivity of minimal output entropy was posed in the paper [39], where it was conjectured that this would provide an indirect way to attack the additivity problem for Holevo capacity. This approach was conrmed in 2002 by Shor [47], who proved the equivalence of several additivity conjectures, including additivity of Holevo capacity and additivity of minimal output von Neumann entropy (this result involved also the entanglement of formation but we will not consider that quantity here). Following an inuential article by Amosov, Holevo and Werner [4], it was believed that a promising method for proving additivity of minimal output entropy was to prove rst (4.6) for p > 1, and then hope to recover additivity in the limit p 1. For some special classes of channels this turned out to be a fruitful approach, and led to proofs of additivity, as the following list of papers shows: [1],
[2], [3], [10], [16], [17], [18], [19], [20], [23], [24], [34], [35], [36], [37], [38], [41], [45], [46]. 5. The counterexamples As mentioned above the additivity conjectures are now known to be false. Historically the minimal Renyi entropy was rst shown to be non-additive [49] for large values of p, with successive counterexamples lowering the value, until nally it was shown for all p > 1 [30]. The nal breakthrough came when Hastings [27] proved the existence of counterexamples at p = 1, thereby disproving the original conjectures. The rst family of counterexamples was discovered by Werner and Holevo [49]. These channels are highly structured and symmetric, and this suggested that the search for counterexamples should be directed toward similarly special classes. However the breakthrough came with A. Winters 2007 paper [50] where counterexamples for all p > 2 were proven using random channels. The channels were random unitary channels of the form (5.1) () = 1 N
where Ui are randomly selected d d unitary matrices. Winters key observation was that for any choice of random unitaries the product channel when applied to the maximally entangled state has a large eigenvalue, and that this in turn gives a useful upper bound for Smin ( ). When combined with a lower bound for Smin () = Smin () this provides the contradiction to additivity. The hard part of the proof is nding a good lower bound for Smin (). Winters method was non-constructive, and used a randomized argument to imply the existence of such channels. This randomizing argument was extended by Hayden and Winter in the paper [30], where it was used to prove the existence of counterexamples for all p > 1. There was a brief hope that additivity might hold for p 1, but counterexamples to this conjecture were also found [15]. Shortly afterwards Hastings [27] extended the reach of the counterexamples by introducing some new ideas and techniques. He adopted the same general approach of looking at product channels of the form where is a random unitary channel. His main contribution was to nd improved lower bounds for Smin (). Again the argument is based on a randomized technique and is non-constructive. By exploiting the explicit form of the eigenvalue distribution for the reduced density matrix of a random pure bipartite state, combined with a novel idea for estimating the probability of low entropy output states, Hastings was able to prove the existence of channels for which minimal output entropy is non-additive. 6. Current directions of research and open problems 6.1. Explicit counterexamples. As mentioned before, Werner and Holevo found explicit channels which violate additivity of minimal Renyi entropy for all p > 4.79. Recently, Grudka, Horodecki and Pankowski [26] have found explicit channels which violate additivity of Renyi entropy for all p > 2. It is very tempting to believe now that explicit examples at p = 1 may be found soon.

6.2. Random subspaces and channels. Hayden, Leung and Winter [29] investigated the entanglement properties of random subspaces, using concentration of measure arguments and other tools from random matrix analysis. More recently, several authors have developed new approaches to nding bounds for entanglement, some based on the new methods introduced by Hastings [9], [12], [13], [14], [22]. 6.3. Bounds for capacity. Since additivity fails, the convenient one-shot formula (4.1) for channel capacity does not hold, and instead the more awkward regularized formula (3.15) must be used. This raises the question of nding useful bounds for the capacity, which is related to the question of nding bounds for the size of the violation of additivity. The Hayden-Winter examples provide large violations of additivity for all p > 1, however the dimensions of the spaces diverge as p approaches 1. In contrast, the method of proof used by Hastings produces a small violation of additivity at p = 1 [21], and it remains an open question whether larger violations are possible. 6.4. Additivity of capacity over dierent channels. The question is whether Cult ( ) = Cult () + Cult () for two dierent channels and. Based on experience with the additivity conjectures, it seems reasonable to expect that this is false. 6.5. Geometrical approach. Recently, Aubrun, Szarek and Werner [5] have used Dvoretzkys Theorem from convex geometry to give a new proof of existence of counterexamples for all p > 1. Dvoretzkys Theorem concerns the existence of almost spherical cross-sections of high-dimensional convex bodies, and the additivity problem for Renyi entropy can be restated in precisely this form. The dimensions of the counterexamples diverge as p 1, however in another paper [6] the same authors have used related methods to prove existence of counterexamples at p = 1. 6.6. Non-unital qubit channels. Qubit channels are the simplest quantum channels [39]. It is known that additivity holds for unital qubit channels for all p 1 [34], and for non-unital channels at p = 2 and p 4 [37], [20]. However the additivity question for non-unital channels for p < 2 is still open, and the additivity of channel capacity is still open. There is no evidence that qubit channels can violate additivity, however it seems worthwhile to settle this question. References
[1] R. Alicki and M. Fannes, Note on multiple additivity of minimal entropy output of extreme SU (d)-covariant channels, Open Systems and Information Dynamics 11, (2004). [2] G. G. Amosov, On Weyl channels being covariant with respect to the maximum commutative group of unitaries, J. Math. Phys. 48, (2007). [3] G. G. Amosov, The strong superadditivity conjecture holds for the quantum depolarizing channel in any dimension, Phys. Rev. A 75, (2007). [4] G. G. Amosov, A. S. Holevo and R. F. Werner, On some additivity problems in quantum information theory, Problems in Information Transmission 36, (2000). [5] G. Aubrun, S. Szarek and E. Werner, Non-additivity of Renyi entropy and Dvoretzkys Theorem, J. Math. Phys. 51, 022102 (2010). [6] G. Aubrun, S. Szarek and E. Werner, Hastings additivity counterexample via Dvoretzkys theorem, arXiv:1003.4925. [7] C. H. Bennett, C. A. Fuchs and J. A. Smolin, Entanglement-enhanced classical communication on a noisy quantum channel, Quantum Communication, Computing and Measurement, eds. O. Hirota, A. S. Holevo and C. M. Caves (Plenum Press, NY 1997), 79 88.

[8] C. H. Bennett and P. W. Shor, Quantum Information Theory, IEEE Trans. Info. Theory 44, (1998). [9] F. G. S. L. Brandao and M. Horodecki, On Hastings counterexamples to the minimum output entropy additivity conjecture, arXiv:0907.3210. [10] D. Bruss, L. Faoro, C. Macchiavello and M. Palma, Quantum entanglement and classical communication through a depolarizing channel, J. Mod. Opt. (2000). [11] M. Choi, Completely Positive Linear Maps on Complex matrices, Linear Algebra and Its Applications 12, (1975). [12] B. Collins and I. Nechita, Random quantum channels I: graphical calculus and the Bell state phenomenon, arXiv:0905.2313 [13] B. Collins and I. Nechita, Random quantum channels II: Entanglement of random subspaces, Renyi entropy estimates and additivity problems, arXiv:0906.1877 [14] B. Collins and I. Nechita, Gaussianization and eigenvalue statistics for Random quantum channels (III), arXiv:0910.1768 [15] T. Cubitt, A. W. Harrow, D. Leung, A. Montanaro and A. Winter, Counterexamples to additivity of minimum output p-Renyi entropy for p close to 0, Commun. Math. Phys. 284, (2008). [16] N. Datta and M. B. Ruskai, Maximal output purity and capacity for asymmetric unital qudit channels, J. Phys. A: Math. Gen. 38, (2005). [17] N. Datta, M. Fukuda and A. S. Holevo, Complementarity and additivity for covariant channels, Quant. Info. Proc. 5, (2006). [18] B. Dierckx and M. Fannes, Additivity of the renyi entropy of order 2 for positive-partialtranspose-inducing channels, Phys. Rev. A, 77, art.nr. 060302 (2008). [19] A. Fujiwara and T. Hashizume, Additivity of the capacity of depolarizing channels, Phys. Lett. A 299, (2002). [20] M. Fukuda, An application of decomposable maps in proving multiplicativity of low dimensional maps, J. Math. Phys. 51, 022201 (2010). [21] M. Fukuda, C. King and D. Moser, Comments on Hastings additivity counterexamples, Commun. Math. Phys. 296, no. 1, 111 (2010). [22] M. Fukuda and C. King, Entanglement of random subspaces via the Hastings bound, J. Math. Phys. 51, 042201 (2010). [23] V. Giovannetti and S. Lloyd, Additivity properties of a Gaussian channel, Phys. Rev. A 69 (6): Art. No. 062307, (2004). [24] V. Giovannetti, S. Lloyd and M. B. Ruskai, Conditions for multiplicativity of maximal lp -norms for xed integer p, J. Math. Phys. 46, 042105 (2005). [25] J. P. Gordon, Noise at optical frequencies: Information theory, Proceedings of the International School of Physics Enrico Fermi, ed. P. A. Mills (Academic Press, New York, 1964), 156 181. [26] A. Grudka, M. Horodecki and L. Pankowski, Constructive counterexamples to additivity of minimum output Renyi entropy of quantum channels for all p > 2, arXiv:0911.2515 [quantph]. [27] M. B. Hastings, Superadditivity of communication capacity using entangled inputs, Nature Phys. 5, (2008), arXiv:0809.3972. [28] P. Hausladen, R. Jozsa, B, Schumacher, M. Westmoreland, W. Wootters, Classical information capacity of a quantum channel, Phys Rev A 54, no. 3, (1996). [29] P. Hayden, D. Leung and A. Winter, Aspects of generic entanglement, Commun. Math. Phys. 265, (2007). [30] P. Hayden and A. Winter, Counterexamples to the maximal p-norm multiplicativity conjecture for all p > 1, Commun. Math. Phys. 284, (2008). [31] A. S. Holevo, Bounds for the quantity of information transmitted by a quantum communication channel, Probl. Inf. Transm. (USSR) 9, (1973). [32] A. S. Holevo, The capacity of the quantum channel with general signal states, IEEE Trans. Info. Theory 44, (1998). [33] A. S. Holevo, Coding theorems for quantum channels, Research Reviews of Tamagawa University, N4, (1998) (available at arXiv:quant-ph/9809023 ). [34] C. King, Additivity for unital qubit channels, J. Math. Phys. 43, no.3, (2002). [35] C. King, The capacity of the quantum depolarizing channel, IEEE Trans. Info. Theory 49, no.1, (2003).

[36] C. King, Maximal p-norms of entanglement breaking channels, Quantum Info. and Comp. 3, no.2, (2003). [37] C. King and N. Koldan, New multiplicativity results for qubit maps, J. Math. Phys. 47, 042106 (2006). [38] C. King, M. Nathanson and M. B. Ruskai, Multiplicativity properties of entrywise positive maps, Lin. Alg. and its Appl. 404, (2005). [39] C. King and M. B. Ruskai, Minimal entropy of states emerging from noisy quantum channels, IEEE Trans. Info. Theory 47, (2001). [40] L. B. Levitin, On the quantum measure of the amount of information, Proceedings of the Fourth all-union conference on Information Theory (in Russian), Tashkent, (1969). [41] S. Michalakis, Multiplicativity of the maximal output 2-norm for depolarized Werner-Holevo channels, J. Math. Phys. 48, no. 12, 122102, (2007). [42] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, Cambridge University Press (2000). [43] B. Schumacher and M. Westmoreland, Sending classical information via a noisy quantum channel, Phys. Rev. A 56, (1997). [44] C. E. Shannon, A mathematical theory of communication, Bell System Tech. J. 27, 379 423, (1948). [45] M. E. Shirokov, The Holevo capacity of innite dimensional channels and the additivity problem, Commun. Math. Phys. 262, (2006). [46] P. W. Shor, Additivity of the classical capacity of entanglement-breaking quantum channels, J. Math. Phys. 43, (2002). [47] P. W. Shor, Equivalence of additivity questions in quantum information theory, Comm. Math. Phys. 246, (2004). [48] W. F. Stinespring, Positive Functions on C -algebras, Proceedings of the American Mathematical Society 6, (1955). [49] R. F. Werner and A. S. Holevo, Counterexample to an additivity conjecture for output purity of quantum channels, J. Math. Phys. 43, no.9, (2002). [50] A. Winter, The maximum output p-norm of quantum channels is not multiplicative for any p > 2, ArXiv:0707.0402. Department of Mathematics, Northeastern University, Boston, Massachusetts 02115 E-mail address: c.king@neu.edu