ON ENTANGLED QUANTUM CAPACITY VIACHESLAV P BELAVKIN Abstract. The pure quantum entanglement is generalized to the case of mixed compound states to include the classical and quantum correspondences as particular cases. We show that the classical-quantum encodings can be treated as classical entanglements. The mutual information of the entangled states leads to two di¤erent types of entropy for a given quantum state: the von Neumann entropy, which is achieved as the supremum of the information over all c-entanglements, and the true quantum entropy, which is achieved at the standard entanglement. The q-capacity, de…ned as the supremum over all entanglements, doubles the c-capacity. The conditional q-entropy is positive, and q-information and q-capacity of a quantum channel are additive. 1. Introduction The entanglements as speci…cally quantum (q-) correlations, are used for quantum information processes in quantum computation, quantum teleportation, and quantum cryptography [2, 3, 4]. In this paper, we develope the operational approach to quantum entanglement and study the mathematical structure of quantum entangled states to provide a comparison of di¤erent quantum information processings. We show that any compound state can be achieved by a generalized entanglement, and the classically (c-) entangled states of c-q encodings and q-c decodings can be achieved by d-entanglements, the diagonal c-entanglements for these disentangled states. The pure orthogonal disentangled compound states are most informative among the c-entangled states in the sense that the maximum of mutual information over all c-entanglements is achieved on the extreme c-entangled states as the von Neumann entropy S (%) of a given normal state %. Thus the maximum of mutual entropy over all classical couplings, described by c-entanglements of a classical probe systems A to the system B, is bounded by the c-capacity C = log rankB, where rankB is the dimensionality of a maximal Abelian subalgebra A B. We prove that the truly entangled states are most informative in the sense that the maximum of mutual entropy over all entanglements to the quantum system B is achieved by an extreme entanglement of the probe system A = B, called standard for a given %. The mutual information for such extreme q-compound state de…nes another type of entropy, the q-entropy H (%) 2S (%). The maximum of mutual entropy over all quantum couplings, described by true quantum entanglements of probe systems A to the system B is bounded by Cq = log dim B. Date : July 2, 2000. 1991 Mathematics Subject Classi…cation. Quantum Information. Key words and phrases. Entanglements, Compound States, Dimensional Entropy and Quantum Information. This work was partially supported by the Royal Society grant for UK–Japan collaboration. Published in: Quantum Communication, Computing, and Measurement 3. Ed Tombesi and Hirota, Kluwer/Plenum, 2001, 325— 333. 1 2 VIACHESLAV P BELAVKIN In this short paper we consider the case of a simple algebra B for which some results are rather obvious and given without proofs. Some proofs are given in the paper [1], others will be published elsewhere for a more general case of decomposable algebra B to include the classical discrete systems as a particular quantum case. 2. Compound States and Entanglements Let H denote a separable Hilbert space of quantum system, and B = L (H) be the algebra of all linear operators B : H ! H with the Hermitian adjoints B y on H. A normal state can be expressed as (1) y % (B) = trG B = trB ; B 2 B. Here G is another separable Hilbert space, is a Hilbert-Schmidt operator from G to H and y is its adjoint H ! G. This is called the amplitude operator (just amplitude if G is one dimensional space C, = 2 H with y = k k2 = 1 in which case y is the functional y from H to C). Moreover the density operator y = in (1) is uniquely de…ned as a probability operator in B, i.e. as a positive trace one operator 2 B. We denote by B B the predual space identi…ed with the Banach space T (H) of all trace class operators in H, and the probability operators (densities) PA 2 A , PB 2 B of the states &, % on di¤erent algebras A, B will be usually denoted by di¤erent letters , corresponding to their Greek variations &, %. In general, G is not one dimensional, the dimensionality dim G must be not less than rank , the dimensionality of the range ran H of the density operator : We shall always equip G with an isometric involution J = J y , J 2 = I, having the properties of complex conjugation on G, X X J 8 j 2 C; j 2 G j j = jJ j; with respect to which J = J for the positive and so self-adjoint operator = y = y on G. The latter can also be expressed as the symmetricity property ~& = & of the state & (A) = trA given by the real and so symmetric density operator = = ~ on G with respect to the complex conjugation A = JAJ and the tilda operation (G-transponation) A~ = JAy J on the algebra A = L (G). Given the amplitude operator , one can de…ne not only the states % by = y and & by = y on the algebras B = L (H) and A = L (G) but also a pure entanglement state $ on the algebra A B of all bounded operators on the tensor product Hilbert space G H by $ (A B) = trH B A~ y = trG A~ y B : This state is pure as it is given by an amplitude 2 G H de…ned as ( y ) y = J ; and % and & are the marginals of $: (2) $ (I B) = trH B ; 8 2 G; 2 H; $ (A I) = trG A : As follows from the next theorem, any compound state $ (A B) = y (A B) ; A 2 A; B 2 B given on L (G H) by an amplitude 2 G H with a unique entanglement of its marginal states & and %. y = 1, can be achieved by ON ENTANGLED QUANTUM CAPACITY Theorem 2.1. Let $ : A (3) 3 B ! C be a compound state $ (A y B) = trF (A B) ; de…ned by an amplitude operator : F ! G H on a separable Hilbert space F into the tensor product Hilbert space G H with tr y = 1. Then this state is achieved by an entanglement (4) $ (A B) = trF H (I B) A~ y = trG A~ y (I B) of the states (2) with = y and = trF y , where is an operator G ! F H. The entangling operator is uniquely de…ned up to a unitary transformation U of the minimal space by F = ran y ~ U = , where (5) ( y y ) ~ = (J ) J ; 8 2 F; 2 G; 2 H; is given by the isometric involutions J in the spaces G and F. Note that the entangled state (4) is written as $ (A B) = trH B (A) = trG A where the operator (A) = trF space B for any A 2 L (G), and | (6) (B) = J A~ y y I | (B) ; 2 B, bounded by kAk 2 B , is in the predual By J= y ^B) (I is in A as a trace-class operator in G, bounded by kBk 2 A . The linear map | is written in the Steinspring form [5] of the normal completely positive map | (B), while B 7! ^ : A ! B is written in the Kraus form [6] of the normal A~ completely positive map A 7! 2 F ` (N): (A) = X in the canonical orthonormal basis jki of (hkj I) A~ y (jki I) : k Thus the linear maps | and are dual to each other, || = , both are positive, but they are not completely positive but co-positive as the compositions of the transponation A 7! A~ and the completely positive (CP) maps. In terms of the compound density operator ! = y for the entangled state $ (A B) = tr (A B) ! they can be written simply as (7) (A) = trG (A | I) !; (B) = trH (I B) !: De…nition 2.1. The completely co-positive map | : B ! A (or its dual map : A ! B ), normalized as trG | (I) = 1 (or trH (I) = 1) is called the quantum entanglement of the state % on B to = | (I) (or of & on A to = (I) ). The entanglement = q = | by (8) q (B) = 1=2 ~ B 1=2 ; B2B of the state & = % on the algebra A = B is called standard for the system (B; %). The standard entanglement de…nes the standard compound state ~ 1=2 $q (A B) = trH B 1=2 A~ 1=2 = trH A 1=2 B on the algebra B B. This state is pure, given by the amplitude = j y j 1=2 i = ~ with = 1=2 and ~ de…ned in (5) as ( ) ~= y J . 1=2 i, where 4 VIACHESLAV P BELAVKIN 3. Quantum Entropy via Entanglements As we shall prove in this section, the most informative for a quantum system (B; %) is the standard entanglement 0 = q = |0 of the probe system A0 ; & 0 = (B; %), described in (8). Let us consider the entangled mutual entropy and quantum entropies of states by means of the di¤erent types of compound states. To de…ne the quantum mutual entropy, we need to apply a quantum version of the relative entropy to compound state on the algebra A B, called also the information divergency of the state $ with respect to a reference state '. It can be de…ned by the density operators !; of these states as I ($; ') = tr! (ln ! (9) ln ) : It has a positive value I ($; ') 2 [0; 1] if the states are equally normalized, say (as usually) tr! = 1 = tr , and it can be …nite only if the state $ is absolutely continuous with respect to the reference state ', i.e. i¤ $ (E) = 0 for the maximal null-orthoprojector E = 0. The most important property of the information divergence I is its monotonicity property [7, 8], i. e. nonincrease of the divergency I ($0 ; '0 ) after the application of the pre-dual K of a normal completely positive unital map K : M ! M0 to the states $0 and '0 on a von Neumann algebra M0 : (10) $ = $0 K; ' = '0 K ) I ($; ') I ($0 ; '0 ) : The mutual information IA;B ( ) = IB;A ( | ) in a compound state $ achieved by an entanglement : B ! A , or by : A ! B with the marginals & (A) = $ (A I) = trG A ; % (B) = $ (I B) = trH B is de…ned as the relative entropy (9) of the state $ on M = A the product state ' = & %: (11) IA;B ( ) = tr ! (ln ! ln ( I) ln (I B with respect to )) : The following proposition follows from the monotonicity property (10) of the relative entropy on M = A B with respect to the predual K (! 0 ) = $0 (K I) to the ampliation K I of a normal completely positive unital map K : A ! A0 . Proposition 3.1. Let 0 : A0 ! B be an entanglement of a state & 0 (A) = tr 0 (A), A 2 A0 = L (G0 ), to (B; %) with = 0 (I) on B, where 0 = |0 , and = 0 K be an entanglement de…ned as the composition of 0 with a normal completely positive unital map K : A ! A0 . Then IA;B ( ) IA0 ;B 0 . In particular, for any c-entanglement to (B; %) there existsP a not less informative dP entanglement d (A) = hnjAjni (n) with the same = (n), and the standard entanglement q (A) = 1=2 A~ 1=2 of & 0 = % on A0 = B is the maximal one in this sense. Note that any extreme d-entanglement decomposed into pure normalized states n = (n) =tr (n) is maximal among all c-entanglements in the sense IA;B ( d ) IA;B ( c ). This is because tr n ln n = 0, and therefore the information gain X IA;B ( d ) = (n) tr n (ln n ln ) n ON ENTANGLED QUANTUM CAPACITY 5 with a …xed d (I) = achieves its supremum trH ln at any such extreme d-entanglement 1d . Thus the supremum of the information gain (11) over all centanglements to the system (B; %) is the von Neumann entropy (12) SB (%) = trH ln : It is P achieved on any extreme d , for example given by a Schatten decomposition = n jnihnj (n). The maximal value ln rankB of the von Neumann entropy is 1=2 de…ned by rankB = (dim B) , i.e. by dim H. De…nition 3.1. The maximal mutual entropy (13) HB (%) = sup IA;B ( ) = IB;B ( q) ; (I)= ~ 1=2 for a …xed achieved on A0 = B by the standard q-entanglement q (B) = 1=2 B state % (B) = trH B , is called q-entropy of the state %. The di¤ erence HBjA ($) = HB (%) IA;B ($) is called the q-conditional entropy on B with respect to A. Obviously, HBjA ($) is positive in contrast to SB (%) IA;B ($) which can achieve also the negative value (14) SB (%) HB (%) = tr ln in the case the standard entanglement as the following theorem states. Theorem 3.2. The q-entropy for the simple algebra B = L (H) is given by (15) HB (%) = 2trH ln = 2SB ( ) ; It is positive, HB (%) 2 [0; 1], and if B is …nite dimensional, it is bounded, with the 1 maximal value HB (% ) = ln dim B which is achieved on the tracial = (dim H) I, 2 where dim B = (dim H) . 4. Quantum Channel and its Q-Capacity A noisy quantum channel sends pure input states %1 on the algebra B 1 = L (H1 ) into mixed ones % = (%1 ) given by the predual = jB1 to a normal com1 pletely positive unital map : B ! B , (B) = trF1 Y y BY; B2B where Y is a linear operator from H1 F+ to H with trF+ Y y Y = I, and F+ is a separable Hilbert space of quantum noise in the channel. Each input mixed state %1 is transmitted into an output state % = %1 given by the density operator ( 1) = Y 1 I+ Y y 2 B for each density operator 1 2 B 1 , where I + is the identity operator in F+ . The input entanglements 1 = A ! B 1 dual to 1 : B 1 ! A will be denoted as 1 = = |1 . They de…ne the quantum-quantum correspondences (q-encodings) of probe systems (A; &) with the density operator = | I 1 , to the input B 1 ; %1 of the channel with = (I). Let Kq1 be the set of all normal completely positive maps : A ! B1 with any probe algebra A, normalized as tr (I) = 1, and Kq (%1 ) be the subset of 2 Kq1 with (I) = %1 . Each 2 Kq1 can be decomposed as 1q K, where 1q = 0 is 6 VIACHESLAV P BELAVKIN the standard entanglement on A0 ; & 0 = B 1 ; %1 , and K is a normal unital CP map A ! B1 . FurtherPlet Kc1 be the set of all c-entanglements described by the combinations (A) = n & n (A) 1 (n)of the primitive maps A 7! & n (A) 1 (n), and Kd1 be thePsubset of the diagonalizing entanglements , i. e. the decompositions (A) = n hnjAjni 1 (n). As in the …rst case Kc (%1 ) and Kd ( 1 ) denote the subsets corresponding to a …xed (I) = %1 , and each Kc (%1 ) can be represented as = 1d K, where 1d = 0 is a pure d-entanglement A0 = B 1 ! B1 normalized as tr 1d (B) = %1 (B) for all B 2 B1 , by a proper choice of the CP map K : A ! B 1 . Furthermore let Ko1 and Ko (%1 ) be the subsets with orthogonal 1 (n) (and …xed P 2 Ko (%1 ) can also be represented as = 1o K, where n 1 (n) = 1 ). Each 1 0 1 o = 0 is a pure o-entanglement on A = B corresponding to the same %1 . Now, let us maximize the entangled mutual entropy for a given quantum channel (and a …xed input state %1 ) by means of the above four types entanglements . The mutual entropy (11) was de…ned in the previous section by the density operators ! of the corresponding compound state $ on A B, and the productstate ' = & % of the marginals &, % for $. In each case $ = $01 (K ); ' = '01 (K ); where K is a CP map A ! A0 = B 1 , $01 is one of the corresponding extreme compound states $q1 , $c1 = $d1 , $o1 on B 1 B 1 , and '01 = 0 %1 . Proposition 4.1. The entangled mutual informations achieve the following maximal values (16) sup 2Kq (%1 ) Ic (%1 ; ) = (17) IA;B ( sup 2Kc (%1 ) sup 2Ko (%1 ) IA;B ( IA;B ( 1 q ) = Iq (%1 ; ) := IB1 ;B 1 d ) = sup IB1 ;B 1 d ; = Id (%1 ; ) ; ) = Io (%1 ; ) := sup IB1 ;B 1 o 1 o ; where 1 are the corresponding extremal input entangled states on B 1 tr 1 (B) = %1 (B) for all B 2 B 1 . They are ordered as (18) Iq (%1 ; ) Ic (%1 ; ) = Id (%1 ; ) B 1 with Io (%1 ; ) : We shall denote the information Ic (%1 ; ) = Id (%1 ; ) simply as I1 (%1 ; ). De…nition 4.1. The suprema Cq ( ) = sup IA;B ( 2Kq1 (19) sup IA;B ( 1 2Kd ) = sup Iq (%1 ; ) ; %1 ) = C1 ( ) := sup I1 (%1 ; ) ; %1 Co ( ) = sup IA;B ( 1 2Ko ) = sup Io (%1 ; ) ; %1 are called the q-, c- or d-, and o-capacities respectively for the quantum channel de…ned by a normal unital CP map : B ! B1 . Obviously the capacities (19) satisfy the inequalities Co ( ) C1 ( ) Cq ( ) : ON ENTANGLED QUANTUM CAPACITY 7 Theorem 4.2. Let (B) = Y y BY be a unital CP map B ! B1 describing a quantum deterministic channel. Then I1 (%1 ; ) = Io (%1 ; ) = S (%1 ) ; Iq (%1 ; ) = H (%1 ) ; and thus in this case C1 ( ) = Co ( ) = ln rankB 1 ; Cq ( ) = ln dim B 1 In the general case d-entanglements can be more informative than o-entanglements as it can be shown on an example of a quantum noisy channel for which I1 (%1 ; ) > Io (%1 ; ) ; C1 ( ) > Co ( ) : 5. Additivity of Entangled Information. In order to consider block entanglements let us introduce the product systems (B n ; % n ) in the tensor product H n = nl=1 Hl of identical spaces Hl = H, and n : B n 7! B n , the preduals of the normal unital CP the product channels product maps n 1 :B n ! Bn = n 1 l=1 Bl ; where Bl = B for all l. Let us denote Inq (%1 ; ) = Iq %1 n ; n n B n = (B) Cqn ( ) = Cq ; n ; n the suprema of the mutual information IA;B n ( n ) over the set Kq %1 n of all input entanglements : A ! B n with any probe algebra A, normalized as n (I) = 1 , and with any such normalization respectively, 2 Kqn . It is easily seen, by applying the monotonicity property (10) with respect to the normal unital CP map K : A ! A0 = B n in the decomposition = nq K, where n q A0 = n 1=2 1 n 1=2 A~0 1 that the quantities Inq (%1 ; ), Cqn ( ) are additive: Inq (%1 ; ) = nIq (%1 ; ) ; ; A0 2 B n ; Cqn ( ) = nCq ( ) . This additivity cannot be proved but super-additivity for the quantities In1 (%1 ; ) = I1 %1 n ; n C1n ( ) = C1 ; n corresponding to the c-entanglements (or d-entanglements) which are usually used to describe the classical-quantum encodings. This implies that the quantities In = In1 =n; Cn = C1n =n have the limits I (%1 ; ) = lim In ((%1 ; )) n!1 I1 (%1 ; ) ; C ( ) = lim Cn ( ) n!1 C1 ( ) which are usually taken as the bounds of classical information and capacity [9, 10] for a quantum channel . As it has been recently proved in [11] under a certain regularity condition, the upper bound C ( ) is indeed asymptotically achievable by long block classical-quantum encodings. Note that the c-quantities In , Cn are not easy to evaluate for each n, but they all are bounded by the corresponding q-quantities Iq , Cq , and I (%1 ; ) Iq (%1 ; ) ; C( ) Cq ( ) : In the same way one can de…ne the smallest quantities Ino (%1 ; ) = Io %1 n ; n ; Con ( ) = Co n 8 VIACHESLAV P BELAVKIN corresponding to the o-entanglements (orthogonal encodings), and establish the superadditivity for these quantities. One can prove that the information I and the capacity C can be achieved via such the orthogonal block encodings: 1 1 lim Ino (%1 ; ) = I (%1 ; ) ; lim Con ( ) = C ( ) : n!1 n n!1 n In order to measure the ”real”entangled information of quantum channels ”without the classical part”, another quantity, the ”coherent information”was introduced in [3]. It is de…ned in our notations as Is (%1 ; ) = Iq (%1 ; ) S (%1 ) ; Cs1 ( ) = sup Is (%1 ; ) %1 if these quantities are positive, otherwise they are taken to be zero. The supremum Csn ( ) = Cs ( n ) of Is (%n1 ; n ) over all states %n1 2 n1 B 1 in general is not additive but superadditive, and the coherent capacity is de…ned as the limit Cs ( ) = lim Csn ( ) =n: n!1 Obviously this capacity has the bounds Cs1 ( ) Cs ( ) Cq ( ) as Is (%n1 ; n ) Iq (%n1 ; n ) 0 for each input state %1 . Thus the entangled information Iq (%1 ; ) for a single channel corresponding to the standard entanglement 1 and the entangled capacity Cq ( ) = Cq1 ( ) give upper bounds of all other capacities and are good analogs of the corresponding classical quantities. References [1] Belavkin, V.P. and M. Ohya, Los Alamos Archive Quant–Ph/9812082, 1-16, 1998. [2] Bennett, C.H. and G. Brassard, C. Crépeau, R. Jozsa, A. Peres, W.K. Wootters, Phys. Rev. Lett., 70, pp.1895-1899, 1993. [3] Schumacher, B., Phys. Rev. A, 51, pp.2614-2628, 1993; Phys. Rev. A, 51, pp.2738-2747, 1993, Phys. Rev. A, 54, p.2614, 1996. [4] Jozsa, R. and B. Schumacher, J. Mod. Opt., 41, pp.2343-2350, 1994. [5] Stinespring, W. F., Proc. Amer. Math. Soc. 6, p. 211 (1955). [6] Kraus, K., Ann. Phys. 64, p. 311 (1971). [7] Lindblad, G., Comm. in Math. Phys. 33, p305–322 (1973). [8] Uhlmann, A., Commun. Math. Phys., 54, pp.21-32, 1977. [9] Holevo, A.S, Probl. Peredachi Inform., 9, no. 3, pp3-11, 1973. [10] Stratonovich, R.S. and A.G. Vancian, Probl. Control Inform. Theory, 7, no. 3, pp.161-174, 1978. [11] Hausladen P., R. Jozsa, B. Schumacher, M. Westmoreland, W. Wootters, Phys. Rev. A, 54, no. 3, pp. 1869-1876, 1996. Department of Mathematics, University of Nottingham, NG7 2RD Nottingham, UK