PHYSICAL REVIEW A 73, 012345 共2006兲 Class of bound entangled states of N + N qubits revealed by nondecomposable maps Marco Piani* Institute of Theoretical Physics and Astrophysics, University of Gdańsk, 80-952 Gdańsk, Poland 共Received 26 August 2005; published 31 January 2006兲 We introduce a family of linear maps which are positive but not completely positive. We exhibit examples of nondecomposable maps and 2N 丢 2N, N 艌 2, bound entangled states—e.g., nondistillable bipartite states of N + N qubits. Such states, as the standard Bell diagonal states, are diagonal in a maximally entangled basis and, apart remaining positive under partial transposition, are not detected by the realignment criterion. DOI: 10.1103/PhysRevA.73.012345 PACS number共s兲: 03.67.Mn, 02.10.Ud I. INTRODUCTION Entanglement appears to be a basic resource in the fields of quantum information and quantum computation 共see 关1,2兴 and references therein兲. Even if there is a sound definition of what an entangled state is 关3兴, it is in general difficult to determine whether a given state is entangled or not. There are different results in the literature regarding the classification of states. One of the more interesting 关4,5兴 is based on the use of linear maps which are positive 共P兲 关6–8兴 but not completely positive 共CP兲 关9–11兴: we shall refer to them as PnCP maps. A map is P if it trasforms any state into another positive operator. In the case of a bipartite system, a state is entangled if and only if there exists a PnCP map such that the operator obtained acting with the map on only one of the two subsystems is not positive any more. The simplest example of a PnCP map is the operation of transposition T 共with respect to a given basis兲. The action of trasposition on one of the subsystems is called partial transposition 共PT兲. Because of the structure of the set of positive maps 关7,12兴, in the 共2 丢 2兲- and 共2 丢 3兲-dimensional cases PT can “detect” all entangled states: states that remain positive under PT 共PPT states兲 are separable; states that develop negative eigenvalues under PT 共NPT states兲 are entangled. Unfortunately in higher dimensions PT is not a “complete” test any more and there are PPT states which are entangled 关13兴. The PnCP map approach to the problem of entanglement characterization can also give information about the distillability of the state 共see 关14兴 for a review兲. A state is said to be distillable if, having at disposal a large number of copies of the state, it is possible to obtain some maximally entangled states under the constraint of performing only local operations and using classical communication. It turns out that a PPT entangled state 共PPTES兲 can not be distilled, so that its entanglement can be considered “bound” 关15兴; however, it can be useful for tasks that would be impossible to perform classically 关16,17兴. It is still unknown whether all bound entangled states are PPT—i.e., whether the PPT property is a necessary condition for nondistillability, besides being a sufficient one. If so, the set of bound entangled states would correspond to the set of PPTES. On the other hand, it is evident that, in order to identify PPT bound entangled states *Electronic address: piani@ts.infn.it 1050-2947/2006/73共1兲/012345共10兲/$23.00 by means of linear maps, it is necessary to use PnCP that are not decomposable—i.e., that cannot be written as the sum of a CP map and a CP map composed with transposition. It is therefore clear that the study of P maps is strictly related to the study of entanglement, the link being provided by the Choi-Jamiołkowsky isomorphism 关18,19兴. In this work we contribute to the phenomenology of positive maps 关20–22兴, giving some general methods to construct classes of PnCP maps. In one instance we test their decomposability by finding at the same time examples of PPT 共and therefore bound兲 entangled states of N + N qubits. Moreover, we prove that such states are not detected by the so-called realignment criterion 关23,24兴, so that the symmetry of the considered states seems to “defeat” permutational separability criteria 关25–27兴. In Sec. II we review some basic notions and results concerning the properties of positivity and complete positivity of maps and their relation to entanglement. In Sec. III we provide a method to construct a class of PnCP maps. We are led quite naturally to consider a set of states to test the decomposability properties of representative maps in such a class. Therefore in Sec. IV we focus on states such that the condition of positivity under PT has a simple form. We exhibit examples of 共2N 丢 2N兲-dimensional PPTES, thus proving at the same time that the representative maps are not decomposable. Moreover, we prove that the only other independent permutational criterion in the bipartite case, realignment, does not detect such bound entangled states, either. II. LINEAR MAPS AND ENTANGLEMENT We start with some basic facts about positive maps and entanglement, presented for finite d-dimensional systems Sd described by the algebra of d ⫻ d matrices with complex entries M d共C兲. We shall denote by Sd the space of the states 共density matrices兲—that is, the convex set of positive 苸 M d共C兲 of unit trace. The action of any Hermiticity-preserving linear map ⌳ : M d共C兲 → M d共C兲 can be written as 关28兴 d2−1 M d共C兲 苹 X 哫 ⌳关X兴 = kiFkXF†i , 兺 k,i=0 共1兲 where Fk’s are d2 matrices d ⫻ d, forming an orthonormal basis in M d共C兲 with respect to the Hilbert-Schmidt scalar 012345-1 ©2006 The American Physical Society PHYSICAL REVIEW A 73, 012345 共2006兲 MARCO PIANI product, Tr共F†i Fk兲 = ␦ki, and C⌳ = 关ki兴 is a generic Hermitian matrix. The map is also trace preserving if and only if d2−1 kiF†i Fk = 1. 兺k,i=0 Remark 1. Expression 共1兲 does not depend on the choice of the orthonormal basis of the matrices Fk’s. In fact, let 兵Gl其 be another orthonormal basis; there exists a d2 ⫻ d2 unitary d2−1 UlkGl. Thus the matrix U, Ulk = Tr共G†l Fk兲, such that Fk = 兺l=0 action of ⌳ can be written as d2−1 ⌳关X兴 = 兺 lm⬘ GlXGm† , with lm ⬘ = 共C⌳⬘ 兲lm = 共UC⌳U†兲lm. Therefore it is always possible to find an orthonormal basis 兵Gl其 such that C⌳ is diag2 onal and Eq. 共1兲 reads ⌳关X兴 = 兺dj=0−1 jG jXG†j , where the j’s are the eigenvalues of C⌳. We will refer to such an orthonormal basis as a diagonal basis for ⌳. Any linear map ⌳ that is used to describe a physical-state transformation must preserve the positivity of all states ; otherwise, the appearance of negative eigevalues in ⌳关兴 would spoil its statistical interpretation which is based on the use of the state eigenvalues as probabilities. If a map preserves the positivity of the spectrum of all , then we say it is positive; however, it is not sufficient to make ⌳ fully physically consistent. Indeed, the system Sd may always be thought to be statistically coupled to an ancilla n-level system Sn. One is thus forced to consider the action idn 丢 ⌳ over the compound system Sn + Sd, where by idn we will denote in the following the identity action on M n共C兲. It is not only ⌳ that should be positive, but also idn 丢 ⌳ for all n; such a property is called complete positivity 关6,10,11兴. Complete positivity is necessary because of the existence of entangled states of the compound system Sn + Sd: namely, of states that cannot be written as factorized linear convex combinations— that is, as ci 艌 0, i 兺 ci = 1. 1 for all 共normalized兲 兩典 , 兩典 苸 Cd, with 兩*典 denoting the conjugate of 兩典 with respect to the fixed orthonormal basis in C d; 共ii兲 completely positive if and only if 关6,30兴 共idd 丢 ⌳兲关P̂+d 兴 艌 0; ⌳关兴 = d2−1 ⌳关兴 = where 兩j典, j = 1,2, . . . , d is any fixed orthonormal basis in Cd, and its unnormalized version 兩ˆ +d 典 = 冑d兩+d 典; let us denote by P+d ⬅ 兩+d 典具+d 兩 and P̂+d ⬅ dP+d 苸 Sd⫻d the corresponding 共normalized and unnormalized兲 projections onto them. Theorem 1. A linear map ⌳ : M d共C兲 → M d共C兲 is 共i兲 positive if and only if 关19,29兴 共6兲 兺 j=0 d2−1 jG jG†j = 共冑 jG j兲共冑 jG j兲† , 兺 j=0 with 冑 jG j identified as the Kraus operator K j. Theorem 2. The following statements are equivalent 关5兴: 共i兲 苸 Sd⫻d is entangled; 共ii兲 for some positive map ⌳ on M d共C兲 D⌳共兲 ª Tr共共idd 丢 ⌳兲关P̂+d 兴兲 ⬍ 0; 共7兲 共iii兲 for some positive map ⌳ on M d共C兲 ⬘ = 共idd 丢 ⌳兲关兴 i 共3兲 K jK†j , 兺 j=0 with 兺 jK†j K j = 1 if ⌳ is trace-preserving. Remark 2. The properties of positivity and complete positivity depend on the coefficient matrix C⌳ of Eq. 共1兲. In particular a linear map is CP if and only if C⌳ is positive semidefinite 共we will simply say “positive” from now on兲. In fact in this case it is possible to obtain the Kraus-Stinespring of point 共iii兲 of theorem 1 diagonalizing the coefficient matrix and using the fact that the eingenvalues of C⌳ are positive: is not a positive operator any more. d ij兩i典 丢 兩j典 in Cd⫻d there is Remark 3. For any 兩典 = 兺i,j=1 A in M d共C兲 such that 具j兩A兩i典 = ij and therefore 兩典 = 共1d ˆ d 典. Then, considering the spectral decomposition of 丢 A 兲兩 + any state 苸 Sd⫻d, it is clear that there is a CP map ⌳, characterized by a coefficient matrix C⌳, such that = 共idd 丢 ⌳兲关P̂+兴. Thus for any positive map ⌳ characterized by a coefficient matrix C⌳, we have d D⌳共兲 = Tr共C⌳C⌳兲, d 兩j典 丢 兩j典, 冑d 兺 j=1 共5兲 d2−1 共2兲 In fact, any PnCP map ⌳, when acting partially as id 丢 ⌳, moves some entangled states out of the space of states; however, exactly for this reason, it may be used to detect entanglement 关4,5兴. In the first of the following two theorems we collect some useful results concerning positivity and complete positivity; the second one links entanglement detection to positive maps. In the space of states Sd⫻d of the bipartite system Sd + Sd, let us introduce the symmetric state 兩+d 典 = 共4兲 共iii兲 completely positive if and only if it can be expressed in the Kraus-Stinespring form 关10,11兴 k,i=0 Ssep = 兺 ciS共i兲 丢 S共i兲, n d n+Sd 具 丢 兩共idd 丢 ⌳兲关P̂+d 兴兩 丢 典 = 具兩⌳关兩*典具*兩兴兩典 艌 0 with the two coefficient matrices expressed in the same orthonormal basis. Remark 4. We note two interesting properties of the symmetric state 兩+d 典: 共i兲 For all matrices A,B acting on Cd one has A 丢 B兩+d 典 = 1d 丢 BAT兩+d 典 = ABT 丢 1d兩+d 典; 共ii兲 under partial transposition P+d gives rise to the flip operator 012345-2 PHYSICAL REVIEW A 73, 012345 共2006兲 CLASS OF BOUND ENTANGLED STATES OF N + N… F共i兲 = 共F共i兲兲† d Vª 兺 兩i典具j兩 丢 兩j典具i兩 = 共idd 丢 T兲关P̂+d 兴, for all = 0 , . . . , d2i − 1, i = 1,2. If all the coefficients 共i兲 are 共2兲 positive apart from one, let us say 共2兲 k = −兩k 兩, and all the positive coefficients are greater or equal to 兩共2兲 k 兩, then the map ⌳ : M d1⫻d2共C兲 → M d1⫻d2共C兲, i,j=1 which is such that V兩 丢 典 = 兩 丢 典, V共A 丢 B兲V = B 丢 A. Unlike the case of CP maps, there is no general prescription on C⌳ ensuring that ⌳ preserves the positivity of . For instance, if C⌳ is not positive, then, by separating positive and negative eigenvalues, one sees that every ⌳ can be written as the difference of two CP maps ⌳1,2 关31兴: ⌳关兴 = jG jG†j − 兺 兩 j兩G jG†j , 兺 艌0 ⬍0 j 共8兲 j with G j a diagonal basis. However, no general rule is known that may allow us to recognize the positivity of ⌳ by looking at the eigenvalues j and at the matrices G j. Given a set of positive maps P = 兵⌳1 , . . . , ⌳ p其 we can define a larger set of positive maps, 再冏兺 冏 冎 p ⍀共P兲 = ⌫i ⴰ ⌳i ⌫i CP , 共9兲 i=0 with ⌳0 = id. Then, given a set of PnCP maps 兵⌳PnCP 其i, we i 其兲 of P maps, potencan construct a whole class ⍀共兵⌳PnCP i tially PnCP. It is quite evident that no map in ⍀共兵⌳PnCP 其 i兲 i gives a stronger test for entanglement, in the sense of theorem 2, than the ensemble of tests performed with the single ’s. In particular, if a map is in ⍀共兵T其兲, it is said to be ⌳PnCP i decomposable and cannot provide a stronger test than PT. According to a theorem by Woronowicz 关12兴, all P maps M 2共C兲 → M 2共C兲 are decomposable, whence the transposition detects all the entangled states in S2⫻2; in other words, 共id2 丢 T2兲关兴 is nonpositive if and only if is entangled. On the contrary, when d 艌 3, there are PPT states which are entangled 共PPTES兲 关7,9,13,14兴. The entanglement in a PPTES cannot be distilled by means of local operations and classical communication 关15兴; therefore, it is referred to as bound entanglement. The relation between nondecomposability of maps and PPT entangled states is summarized in the following proposition. Proposition 1. If ⌳ is positive on Sd, 苸 Sd⫻d is PPT, and D⌳共兲 ⬍ 0, then ⌳ is not decomposable and is a PPTES. ⌳ = ⌳1 丢 idd2 + idd1 丢 ⌳2 共11兲 is positive. Remark 5. Any map ⌳ satisfying the requirements of the previous theorem is positive. It is, moreover, PnCP as soon as its matrix of coefficients is not positive—i.e., as soon as 共2兲 the negative contribution in idd1 丢 ⌳2 due to 共2兲 k = −兩k 兩 ⬍ 0 is not actually canceled by terms in ⌳1 丢 idd2. Theorem 3 is suggested by a similar result regarding dynamical semigroups 关32,34兴. A dynamical semigroup is a set of Hermiticity and trace-preserving linear maps ␥t, t 艌 0, on Sd which obey a semigroup composition law ␥t ⴰ ␥s = ␥t+s, for any t , s 艌 0. Semigroups are used to describe the dynamics of a system immersed in an environment and weakly coupled to it 关35–38兴. With the further assumption of continuity in t 共time兲 the semigroup has the form ␥t = exp共tL兲, where L is a map called the generator which determines all the properties of the semigroup. The issue of complete positivity in the description of the evolution of dynamical systems is indeed related to the existence of entangled states 关39–44兴. The generator of a factorized semigroup exp共tL兲 = exp共tL1兲 丢 exp共tL2兲 on Sd 丢 Sd is L = L1 丢 idd + idd 丢 L2, which is 1 2 2 1 similar to Eq. 共11兲. IV. CLASS OF „2N ‹ 2N…-DIMENSIONAL BOUND ENTANGLED STATES We will now use the result of the previous section to detect bound entangled states “canonically” related to maps having the structure of theorem 3. We will take advantage of the simple properties of Pauli matrices under product and transposition. We notice that 兵ˆ ª / 冑2其3 =0, with 0 the twodimensional identity matrix and 1 = 冉 冊 0 1 1 0 , 2 = 冉 冊 0 −i i 0 , 3 = 冉 冊 1 0 0 −1 the Pauli matrices, is a Hermitian orthonormal basis in M 2共C兲. Let L共k兲 be the the set III. CLASS OF POSITIVE MAPS L共k兲 ª 兵共1, . . . , k兲兩i = 0,1,2,3,i = 1, . . . ,k其, In the following theorem we will give a sufficient condition for the positivity of a class of maps on M d1共C兲 丢 M d 共C兲. The proof is presented in the Appendix. 2 Theorem 3. Let ⌳i be maps acting on M di共C兲, i = 1,2, in the following way: whose elements are k-dimensional 共integer兲 vectors . Let us take m 艌 n 艌 1 and let L be the lattice d2i −1 ⌳i关X兴 = 共i兲F共i兲XF共i兲; 兺 =0 i.e., they admit Hermitian diagonal bases 共10兲 L ª L共m兲 ⫻ L共n兲 = 兵共␣, 兲兩␣ 苸 L共m兲,  苸 L共n兲其, 共12兲 with 4N, N = m + n elements. In a geometric representation L can be considered in an N-dimensional integer space as a hypercube whose sides contain four points. Every index among ␣1 , . . . , ␣m , 1 , . . . , n is then a coordinate. We will associate to the points of L the tensor products of Pauli matrices 012345-3 PHYSICAL REVIEW A 73, 012345 共2006兲 MARCO PIANI ␣ ª ␣ 丢  = 共␣1 丢 ¯ 丢 ␣m兲 丢 共 1 丢 ¯ 丢 n兲. 共13兲 stitute a natural extension of the standard Bell states to bipartite states of many qubits; they can be considered bipartite tensor products of N Bell states of two qubits: It is clear that 兵ˆ ␣ ª ␣/共冑2兲 兩␣ 苸 L m 共m兲 兩␣典 = 兩␣1典 丢 ¯ 丢 兩n典, 其 with 兩典 ª 12 丢 兩+2典. We call lattice states 共LS’s兲 the states diagonal in the 兵兩␣典其 basis—that is, the mixtures belonging to the convex span of the projectors P␣: and 兵ˆ  ª /共冑2兲n兩 苸 L共n兲其 are orthonormal Hermitian bases, respectively, in M 2m共C兲 and M 2n共C兲, while 兵ˆ ␣ ª ˆ ␣ 丢 ˆ 兩共␣, 兲 苸 L其 共14兲 is an orthonormal Hermitian basis in M 2m共C兲 丢 M 2n共C兲 ⯝ M 2N共C兲. ª 兺 共␣,兲苸L ␣ P␣, ␣ 艌 0, Let us consider the map ⌳0 = ⌳1 丢 id2n + id2m 丢 ⌳2 , They are a possible generalization of the standard Bell diagonal states 关33兴. C. Partial transposition of lattice states 共15兲 T2 ª id2N 丢 T关I兴 = with 冉 兺 苸L共n兲\兵0其 V␣ ª 12N 丢 ␣V12N 丢 ␣ . 冊 ˆ Xˆ  − ˆ 0Xˆ 0 , and 0 ⫽ 0m, denoting with 0k the k-dimensional null vector 共0 , . . . , 0兲 苸 L共k兲. Since m 艌 n, the map ⌳0 satisfies the hypothesis of theorem 3 and is therefore P. Note that id2k关X兴 = 0kX0k. In the basis 共14兲 the coefficient matrix C⌳ is 0 diagonal with eigenvalues ␣ = 冦 共␣, 兲 = 共0m,0n兲, 2, ␣ = 0 m ∧  ⫽ 0 n,  0 , 1, ␣ ⫽ 0m ∧  = 0n , − 1, 共␣, 兲 = 共0m, 0兲, 0, ␣ ⫽ 0m ∧  ⫽ 0n . 1, 冧 共16兲 It is therefore clear that, because of our choice of 0, the map ⌳0 is PnCP. Accordingly to proposition 1, we will show that it is also nondecomposable, exhibiting a PPT state such that D⌳ 共兲 ⬍ 0. 0 B. Lattice states 共19兲 N 兩␣典 ª 12N 丢 ␣兩+2 典, m+n J␣ = 兺 共− 1兲a a=0 such that 冉 兺 ␣ 共␥,␦兲苸L␣ 冦 1 , 共␣, 兲 苸 I, N ␣ = I 0, 共␣, 兲 苸 I, 共21兲 冧 with I 債 L a subset of L and NI ª card共I兲—that is, states 1 兺 P␣ . NI 共␣,兲苸I 共22兲 Such states are completely characterized by a set I 債 L. The condition of positivity under PT 共21兲 becomes N P␣ P␥ = ␦␣␥␦ P␣ . The states 兩␣典, 共␣ , 兲 苸 L, are 4N maximally entangled N N states forming an orthonormal basis in C2 丢 C2 . They con- 冊 ␥␦ 艌 0, a where L␣ 債 L is the set of lattice points 共␥ , ␦兲 such that exactly a indices among ␥1 , . . . , ␥m , ␦1 , . . . , ␦n are equal to the corresponding indices among ␣1 , . . . , ␣m , 1 , . . . , n. The previous condition for positive partial transposition is necessary and sufficient for the whole class of lattice states. For the sake of simplicity we now focus on a subset of these states. We will call equidistributed LS’s 共ELS’s兲 the LS’s such that I = 共17兲 共20兲 Analyzing the spectral decomposition of V␣ 共see lemma 2兲 we obtain the following condition for the PPT property of a lattice state. Theorem 4. A lattice state remains positive under partial transposition if and only if, for any lattice site 共␣ , 兲 苸 L, We construct orthogonal one-dimensional projectors P␣ = 兩␣典具␣兩, 1 兺 ␣V␣ , 2N 共␣,兲苸L with 1 ⌳1关X兴 = n 兺 ˆ ␣Xˆ ␣ , 2 ␣苸L共m兲 1 2m ␣ = 1. 共18兲 We now analyze the problem of which are PPT. We start by operating the partial transposition on P␣ in Eq. 共17兲, obtaining A. Representative class of PnCP maps ⌳2关X兴 = 兺 共␣,兲苸L J␣ = 1 a 艌 0, 兺 共− 1兲aNI␣ NI a=0 a a a 兲 and I␣ ª I 艚 L␣ . where NIa = card共I␣ 012345-4 ␣ PHYSICAL REVIEW A 73, 012345 共2006兲 CLASS OF BOUND ENTANGLED STATES OF N + N… We now notice the following. 共i兲 共 Na 兲 is the number of different ways to choose a conditions among N—that is, the number of different sets A共a兲. 共ii兲 Given two sets A共a兲, A⬘共a兲, if A共a兲 ⫽ A⬘共a兲, then GA共a兲 and GA⬘共a兲 are disjoint, GA 艚 GA⬘ = 쏗. 共iii兲 Two sets of points GA共a兲, GA⬘共a兲 satisfying different conditions A共a兲, A⬘共a兲 are mapped one onto the other by suitable permutations of the indices. Thus, they must contain the same number of points: card共GA兲 = ga for all A共a兲. a In summary, for each a, the set I␣ can be split into 共 Na 兲 disjointed sets GA共a兲 each containing ga points. We argue that ga = 2N−a, a = 0 , . . . , N. This is certainly true for a = N, since there is only one point satisfying N conditions. Let us suppose the statement is true for a = ã , . . . , N. Then gã−1 is given by N−共ã−1兲 gã−1 = 3 N−共ã−1兲 兺 i=1 − N−共ã−1兲 = 3N−共ã−1兲 − 兺 i=0 冉 冉 N − 共ã − 1兲 i N − 共ã − 1兲 i 冊 冊 g共ã−1兲+i 2N−共ã−1兲−i + 2N−共ã−1兲 共25兲 = 2N−共ã−1兲 . FIG. 1. Geometric representation of IC for 共a兲 N = 2 and 共b兲 N a = 3. The different levels of gray mark points of different sets I␣ to calculate J␣, with 共␣ , 兲 苸 IC. The black circle is the element of IC N whose coordinates are exactly 共␣ , 兲; i.e., it constitutes the set I␣ . 0 The open circles correspond to I␣. It is in principle possible to construct all the ELS’s that are PPT. A similar task has been accomplished in 关34兴 for the case m = n = 1, N = 2—i.e., for the ELS’s I 苸 S4⫻4. In the present work, instead, we just show that for any N 艌 2 among the ELS’s there is at least a PPTES. We first need the following lemma. Lemma 1. The ELS described by IC = 兵共␣, 兲 苸 L兩␣i ⫽ 0 ∧  j ⫽ 0;i = 1, . . . ,m; j = 1, . . . ,n其, 共23兲 NIC = 3N, is positive under partial transposition and J␣ 艌 1 / NIC for all 共␣ , 兲 苸 L. Proof. In Fig. 1 we represent graphically the set IC for N = 2 and N = 3 to make the proof easier to understand. Consider first the case 共␣ , 兲 苸 IC. Then, N J␣ = 冉冊 N 1 兺 共− 1兲a a ga , 3N a=0 The relation between the ga’s written in the first line of Eq. 共25兲 is readily explained: a set of points satisfying ã − 1 conditions 共and no further兲 is given by the number of points satisfying at least ã − 1 conditions minus all the disjoint sets satisfying exactly i further conditions chosen among the remaining N − 共ã − 1兲. Therefore, N J␣ = 共26兲 For the state to be PPT, the condition J␣ 艌 0 must hold for any choice of 共␣ , 兲 苸 L. We have already considered the case 共␣ , 兲 苸 IC. We now show that this is the worst case, in the sense that J␣ is the smallest possible. In fact, without loss of generality, consider the case where k indices ␣i1 , . . . , ␣ik among the indices 共␣ , 兲 are equal to zero. Because of our choice of IC, no element 共␥ , ␦兲 of IC will satisfy any of the corresponding k conditions—e.g., “␥i1 = ␣i1 = 0 ” , . . . , “ ␥ik = ␣ik = 0 . ” This amounts to consider N − k instead of N as the maximum number of conditions that one element of IC can satisfy in the previous reasoning. We find 共24兲 N Ia = ␣ where the coefficient ga is such that 共 Na 兲ga is the number of a —that is, of points satisfying exactly a condipoints in I␣ tions, as expressed by the ␦’s appearing, for example, in Eq. 共A8兲. To show the validity of Eq. 共24兲, let us denote the following. 共i兲 A共a兲 a set of a conditions of the form “␣i = ␥i”—that is, equivalently, of a numbers chosen between 兵1,2, . . . , N其. 共ii兲 GA共a兲 債 IC the set of points satisfying conditions A共a兲 and no further ones. 冉冊 N 1 1 兺 共− 1兲a2N−a = 3N . 3N a=0 a 冦 冉 冊 N−k a 0, ga , 0 艋 a 艋 N − k, N − k ⬍ a 艋 N, 冧 with ga = 2N−k−a3k. Therefore J␣ = 1 / 3N−k 艌 1 / 3N. 䊏 We are now able to construct bound entangled states in S2N⫻2N. Theorem 5. The state IBE共0兲, with IBE共0兲 = IC 艛 兵共0m , 0兲其, 0 ⫽ 0n, and IC given by Eq. 共23兲, is a PPTES. Proof. The state is PPT because the sufficient and necessary conditon for positivity under PT of theorem 4 is satis- 012345-5 PHYSICAL REVIEW A 73, 012345 共2006兲 MARCO PIANI fied. For this state NIBE共0兲 = 3N + 1. We use the result of lemma 1 with the slight difference that now all the points in IBE共0兲 have a weight 1 / NIBE共0兲 = 1 / 共3N + 1兲. Therefore for any 共␣ , 兲 苸 L the elements in IC contribute at least with 1 / NIBE共0兲 to J␣. On the other hand, the element 共0m , 0兲 苸 IBE共0兲 contributes with ±1 / NIBE共0兲, the sign depending on the number of identical indices between 共0m , 0兲 and 共␣ , 兲. Therefore, J␣ 艌 1 NIBE共0兲 ± 1 NIBE共0兲 艌 0. We check now that IBE共0兲 is also entangled. In fact, Eq. 共14兲 is a diagonal basis for the associated map ⌳ of any LS , as well as for the map ⌳0 of Eq. 共15兲. The eigenvalues of C⌳ are exactly the weights ␣,, 共␣ , 兲 苸 L; in particular, for IBE共0兲 they are 冦 1 , 共␣, 兲 苸 IBE共0兲, 3 +1 N 0, 共␣, 兲 苸 IBE共0兲, 冧 while those of ⌳0 are listed in Eq. 共16兲. Therefore, in the case of IBE共0兲, we have 1 1 D⌳共IBE共0兲兲 = N 兺 ␣ = − 共3N + 1兲 . 3 + 1 共␣,兲苸IBE共0兲 According to proposition 1, IBE共0兲 is entangled and ⌳0 is not decomposable. 䊏 Two examples of sets I describing PPT entangled ELS for N = 2 and N = 3 are shown in Fig. 2. Remark 6. Notice that local unitary operations preserve the properties of any state as regards entanglement and positivity under PT. It is therefore quite evident that our choice of I is just one among many possible ones. Generalizing the reasoning in 关34兴, we note that, given two 2N ⫻ 2N unitary matrices U , W such that W␣U† = ␥␦ up to a phase, we have 共U* 丢 W兲P␣共UT 丢 W†兲 FIG. 2. Geometric representation of two sets IBE共0兲 identifying PPT entangled equidistributed lattice states: 共a兲 m = 1, n = 1, N = 2, 0 = 共2兲 and 共b兲 m = 2, n = 1, N = 3, 0 = 共2兲. In both cases the black circle corresponds to the element 共0m , 0兲. nates. Therefore every set that can be transformed into a IBE共0兲 by means of the “elementary” operations just described is a PPTES. It is not necessary to consider sets as symmetric as IBE共0兲 to find PPT entangled states using the class of PnCP maps previously depicted, as the results in 关32,34兴 show. The problem is that, even if we have found a simple condition to check if a ELS is PPT, this does not mean its application is straightforward. In this framework, the class of states IBE共0兲 appears to be a “canonical” family of PPTES, whose high symmetry allows an easy check of the PPT condition. N = 关12N 丢 共W␣U†兲兴P+2 关12N 丢 共U␣W†兲兴 = P ␥␦ . D. Realignment of lattice states 共27兲 Since the transformation is unitary and thus invertible, it induces a permutation among the elements of L, so that ⬘ = 共U* 丢 W兲共UT 丢 W†兲 is another LS with permuted eigenvalues. Let us indicate with i, i = 1 , . . . , N, the elements of the vector 共␣ , 兲. For example, U and W can be chosen such that 共1, . . . , i, . . . , N兲 哫 关1, . . . ,p共i兲, . . . , N兴, with p : 兵0,1,2,3其 → 兵0,1,2,3其 a permutation, or such that 共1, . . . , i, . . . , j, . . . , N兲 哫 共1, . . . , j, . . . , i, . . . , N兲. In the geometric picture the first operation amounts to exchanging two parallel 共N − 1兲-dimensional hyperplanes, while the second one corresponds to exchanging two coordi- Partial transposition, even if the most known and used method to detect entanglement, it is certainly not the only one. Another test is given, for example, by the realignment criterion 关23,24兴. Essentially it corresponds to a linear map R acting on the total state of both subsystems, such that it does not increase the trace norm of product states; i.e., given the norm 储X储 = Tr兩X兩, we have 储R共A 丢 B兲储 艋 1. In 关25兴 it was shown that both partial transposition and realignment are part of a family of permutation criteria, further studied in 关26,27兴. In the bipartite setting the only independent tests in this family are exactly partial transposition and realignment. As regards the latter, we will consider the map defined by R共兩i典具j兩 丢 兩k典具l兩兲 = 兩j典具l兩 丢 兩i典具k兩. 共28兲 The realignment criterion, as partial transposition, is a necessary separability criterion: if is separable, then the re- 012345-6 PHYSICAL REVIEW A 73, 012345 共2006兲 CLASS OF BOUND ENTANGLED STATES OF N + N… aligned matrix r = R共兲 satisfies 储r储 艋 1, so that if 储r储 ⬎ 1, we know that the state is entangled. Let us consider the action of map 共28兲 on maximally entangled states of the form 1 丢 U兩+d 典具+d 兩: 1 R共1 丢 U兩+d 典具+d 兩1 丢 U†兲 = U† 丢 UT . d 共29兲 Therefore for the states 共17兲 we have Lr共兩␣典具␣兩兲 = 1 1 T N ␣ 丢 ␣ = N ␣␣ 丢 ␣ , 2 2 共30兲 = 共−1兲␦␣,2 m ␣ = 兿i=1 ␣i, with ␣ aligned LS corresponds to r ª R共兲 = 共similarly 兲, so that the re- 1 兺 ␣␣␣ 丢 ␣ . 2N 共␣,兲苸L 共31兲 As regards eigenvalues and eigenvectors of a general realigned LS we find r 兩␥␦典 = 冉 冊 1 兺 ␣␥␦ 兩␥␦典, 2N 共␣,兲苸L 共32兲 m ␣i␥i and ␣␥ defined in Eq. 共40兲 共similarly with ␣␥ = 兿i=1 ␦兲. For a general LS we have 储r 储 = 1 兺 2N 共␥,␦兲苸L 冏 兺 共␣,兲苸L 冏 ␣␣␥␦ , 共33兲 while for ELS the norm of the realigned state assumes the following simple expression: 储Ir储 = 1 兺 2 NI 共␥,␦兲苸L N 冏兺 共␣,兲苸I 冏 ␣␥␦ . 共34兲 We focus on the canonical class of bound entangled states I = IBE共0兲, and for definiteness we consider the case 共0兲 j ⫽ 0 for all j = 1 , . . . , n. As proven in the Appendix, we have, in this case, 储Ir储 = 3N + 共− 1兲m 艋 1, 3N + 1 共35兲 so that the bound entangled state is not revealed by the relignment criterion. V. CONCLUSIONS A general class of positive but not completely positive linear maps was found. The decomposability of a subset of such maps has been studied, exploiting the characterization of entanglement by means of linear maps. We have established the nondecomposability of such a subfamily and states found examples of 共2N 丢 2N兲-dimensional 共N 艌 2兲—e.g., states of a bipartite N + N qubits system— which are PPT but nevertheless entangled. Such states belong to a family of lattice states, which are described by a distribution probability on a lattice and are diagonal in a maximally entangled basis. Moreover, the PPT bound entangled states we found are not revealed by the realignment criterion. The states introduced are potentially interesting to better understand the phenomenon of bound entanglement; in particular, lattice states are a possible generalization of the well-studied Bell diagonal states. Theorem 3 provides a family of PnCP maps in dimensions d = d1d2. In the particular case studied, both factor dimensions are powers of 2—i.e., d1 = 2m and d2 = 2n, with m , n 艌 1. The resulting map is applied to detect PPT bound entangled states among lattice states. As explicitly shown, for lattice states the PPT condition is relatively easy to handle. We exploit the fact that in dimension 2N a complete orthonormal basis for operators may be built out of Pauli matrices. Besides being unitary, the latter are Hermitian, behave simply under transposition, and have easy-to-handle commutation rules. As regards possible generalizations of the results, it is definitely possible to consider PnCP maps for all product dimensions d = d1d2. In particular notice that one may take d1 = 1, so to have PnCP maps in prime dimensions. Unfortunately, in the latter case the structure 共11兲 simplifies too much. It can be proved that the resulting maps are not useful to detect PPT bound entanglement, because they turn out to be decomposable. It is possible to generalize lattice states for d ⫽ 2N, e.g., considering instead of 共normalized兲 Pauli matrices some other 共orthonormal兲 basis in the space of square matrices of dimension greater than two. However, as already recalled, Pauli matrices enjoy nice mathematical properties. Such properties make the evaluation of partial transposition of the corresponding lattice states easy and allow to construct immediately the PnCP map needed to detect bound entanglement by following the prescription of theorem 3. For general d = d1d2, it is difficult to study the behavior of mixed states under partial transposition to impose the PPT property, even if we confine ourselves to states diagonal in a maximally entangled basis. Anyway, the fact that the maps 共15兲 are nondecomposable suggests that also other maps of the form 共11兲 共and acting on dimensions different than 2N兲 are nondecomposable. Even if, as just explained, it may be difficult to construct analytically examples of PPT bound entangled 共lattice兲 states, it is anyway possible to look numerically for PPT entangled states detected by the depicted maps. Indeed, if some PPT state does not remain positive under the partial action of a given positive map, we know at the same time that the state is 共bound兲 entangled and the map is nondecomposable. We plan to study the properties of the considered bound entangled lattice states with respect to other criteria to detect and possibly quantify entanglement. Indeed, such states constitute a possible test for the power of any 共new兲 separability criterion; it is remarkable that they pass both permutational criteria in the bipartite setting—i.e., partial transposition and entanglement. It would be also interesting to check whether the range criterion 关45兴 is able to detect the non-full-rank states IBE, considering that only recently non-full-rank bound entangled states satisfying such criterion have been found 关46兴. 012345-7 MARCO PIANI PHYSICAL REVIEW A 73, 012345 共2006兲 In 关47兴 bound entangled states of 2N qubits, N 艌 2, were also studied, but those states are considered to be multipartite; i.e., each qubit is held by a different party; the property of multipartite activability 关48,49兴 was studied. In our case there are only two parties, each holding N qubits, so that one may consider only bipartite activation 关16兴. Proof of theorem 4 Lemma 2. The spectral decompostion of V␣ is 兺 V␣ = 共␥,␦兲苸L ⌶␣␥⌶␦ P␥␦ , with m ⌶␣␥ = 兿 ⌶␣i␥i, ACKNOWLEDGMENTS The author thanks F. Benatti and R. Floreanini for fruitful discussions and encouragement. This work was initiated during composition of the author’s Ph.D. thesis at the Department of Theoretical Physics of the University of Trieste. The work is supported by EC grants RESQ, Contract No. IST2001-37559, QUPRODIS, Contract No. IST-2001-38877, and CNR-NATO. i=1 N V␣兩␥␦典 n m = 兿 ␣i␥i 兿  j␦ j12N 丢 关共␣␥␣兲 丢 共␦兲兴兩+2 典 N j=1 i=1 m n i=1 j=1 = 兿 共␣i␥i␣i␥i兲 兿 共 j␦ j j␦ j兲兩␥␦典, Proof of theorem 3 where Proof. We have to check that ␣␥ ª D共, 兲 ª 具兩⌳关兩典具兩兴兩典 艌 0 for all 兩典, 兩典 苸 Cd1⫻d2, which can both be expanded on a basis 兵兩i典 丢 兩j典其: 兩典 = 兺 兺 ⌽ij兩i典 丢 兩j典, i=1 j=1 d1 d2 兩典 = 兺 兺 ⌿kl兩k典 丢 兩l典, 共A1兲 so that they are determined by the coefficient matrices ⌽ and ⌿. It is straighforward to find the following expression for D共 , 兲: D共, 兲 = 兺 共1兲兩Tr共F共1兲⌽⌿†兲兩2 + 兺 共2兲兩Tr共F共2兲共⌿†⌽兲T兲兩2 . 共A2兲 共2兲 d22−1 Since 兵F 其=0 and 兵F 其=0 are two orthonormal bases and Tr共AT兲 = Tr共A兲, we have Tr共⌽⌿†⌽⌿†兲 = 兺 关Tr共F共1兲⌽⌿†兲兴2 = 兺 兵Tr关F共2兲共⌿†⌽兲T兴其2 兺 共␥,␦兲苸L + 兺 兩Tr关F共2兲共⌿†⌽兲T兴兩2 . ⫽k 兺 − 冋兺 兩Tr共F共1兲⌽⌿†兲兩2 + † T 2 兩Tr关F共2兲 k 共⌿ ⌽兲 兴兩 册 艌 0, for all 兩典, 兩典; therefore, ⌳ is P. 冉 兺 共␥,␦兲苸L 冊 ␥␦⌶␣␥⌶␦ P␣ 共A6兲 ␥␦⌶␣␥⌶␦ 艌 0 ∀ 共␣, 兲 苸 L. 共A7兲 m n i=1 j=1 ␥␦ 兿 共− 1兲␦˜␣i␥i 兿 共− 1兲␦˜ j␦ j 艌 0, whence, since it must hold for all 共␣ , 兲, it must be 共A3兲 From the hypothesis of the theorem, the above inequality, and Eq. 共A2兲 we find D共, 兲 艌 兩共2兲 k 兩 共A5兲 ˜ 兲, ˜ , We introduce the bijection L → L given by 共␣ , 兲 哫 共␣ ˜ = 共˜␣1 , . . . , ˜␣m兲 and ˜ ª 共 + 2兲 mod 共4兲, 0 艋 艋 3. It where ␣ then follows that ⌶␣␥ = 共−1兲␦␣,˜␥. The PPT condition can therefore be written as ␥␦苸L 兩Tr共F共1兲⌽⌿†兲兩2 冎 and therefore T1 is positive if and only if and, using the triangle inequality, we have 艋兺 1 兺 2N 共␣,兲苸L T1 = † T 2 兩Tr共F共2兲 k 共⌿ ⌽兲 兲兩 ␣ = 0 ∨ ␥ = 0 ∨ ␣ = ␥, − 1, ␣ ⫽ 0 ∧ ␥ ⫽ 0 ∧ ␣ ⫽ ␥ , 1, and ␣ = 共−1兲␦␣,2. Setting ⌶␥␣ = ␣␥␣␥, the result follows by direct inspection. 䊏 Proof of theorem 4. Using lemma 2, from Eq. 共19兲 we have 共1兲 d21−1 再 k=1 l=1 N Proof. From remark 4 and V兩+2 典 = 兩+2 典, it follows that APPENDIX d1 d2 ⌶␣␥ = 共− 1兲␦兩␣−␥兩,2 . 兺 兩Tr关F共2兲共⌿†⌽兲T兴兩2 J␣ ª ␥␦ 兿 共− 1兲 n ␦␣ ␥ i i i=1 共− 1兲␦ 兿 j=1  j␦ j 艌 0. 共A8兲 We now split the sum over ␥ , ␦ into different sums according to the number of delta conditions that are satisfied; that is, a . Explicitly, we isolate the contributions of different L␣ ⫽k m+n J␣ = 共A4兲 䊏 兺 ␥␦苸L m 兺 兺 a=0 共␥,␦兲苸La ␣ m n i=1 j=1 ␥␦ 兿 共− 1兲␦␣i␥i 兿 共− 1兲␦ j␦ j 艌 0. The theorem follows noticing that 012345-8 PHYSICAL REVIEW A 73, 012345 共2006兲 CLASS OF BOUND ENTANGLED STATES OF N + N… n m 共␥, ␦兲 苸 a L␣ ⇒ 兿 共− 1兲 ␦␣ ␥ i i i=1 共− 1兲␦ 兿 j=1  j␦ j the result of the sum over the remaining nonzero indexes. The latter is equal to = 共− 1兲a . g共k,l兲 = 3m−k兵兩共− 1兲N−共k+l兲3k+l + 1兩f共n − l兲 䊏 + 兩共− 1兲N−共k+l兲3k+l − 1兩关3n−l − f共n − l兲兴其, Evaluation of the norm of the realigned canonical bound entangled state 共A10兲 To evaluate the absolute value in Eq. 共34兲 we observe that for I = IBE共0兲, 共0兲 j ⫽ 0 for all j = 1 , . . . , n, given the definition 共A5兲, 兺 共␣,兲苸I 冋兿 冉 兺 冊册冋兿 冉 兺 冊册 m ␣␥␦ = 3 ␣i=1 i=1 ␣i␥i n 3 j=1  j=1 ␥ j␦ j 兩共− 1兲N−共k+l兲3k+l ± 1兩 = 3k+l ± 共− 1兲N−共k+l兲 . + ␣0␥0␦ = 共− 1兲N−共k+l兲3k+l + 0␦ , where k and l are the number of indexes ␥i and ␦ j equal to zero in ␥ and ␦, respectively. Therefore we can split the outer sum in Eq. 共34兲 according to k and l—i.e., according to the ways to fix k indexes among m and l indexes among n equal to zero: 储Ir储 = 1 兺 兩共− 1兲N−共k+l兲3k+l + 0␦兩 2 NI 共␥,␦兲苸L n 冉 冊冉 冊 m 1 兺 兺 N 2 NI k=0 l=0 k n g共k,l兲, l f共a + 1兲 = 2关3a − f共a兲兴 + f共a兲 = 2 ⫻ 3a − f共a兲, 共A12兲 because for every fixed  ⫽ 0 there are two possible choices of ␦ ⫽ 0 such that ␦ = −1 and one such that ␦ = 1. Therefore, f共a兲 = 关3a + 共− 1兲a兴/2. 共A9兲 共A13兲 Finally we find 储Ir储 = where, for any choice of k + l indexes equal to zero, g共k , l兲 is 关1兴 M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information 共Cambridge University Press, Cambridge, England, 2000兲. 关2兴 Quantum Information: An Introduction to Basic Theoretical Concepts and Experiments, edited by G. Alber et al., Springer Tracts in Modern Physics, Vol. 173 共Springer, Berlin, 2001兲. 关3兴 R. F. Werner, Phys. Rev. A 40, 4277 共1997兲. 关4兴 A. Peres, Phys. Rev. Lett. 77, 1413 共1996兲. 关5兴 M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Lett. A 223, 1 共1996兲. 关6兴 M. D. Choi, Can. J. Math. 24, 520 共1972兲. 关7兴 E. Störmer, Proc. Am. Math. Soc. 86, 402 共1982兲. 关8兴 K. Życzkowski and I. Bengtsson, Open Syst. Inf. Dyn. 11, 3 共2004兲. 关9兴 M. D. Choi, Linear Algebr. Appl. 10, 285 共1975兲. 关10兴 M. Takesaki, Theory of Operator Algebras 共Springer, New York, 1979兲, Vol. 1. 关11兴 K. Kraus, States, Effects and Operations: Fundamental Notions of Quantum Theory, Lecture Notes in Physics, Vol. 190 共Springer, Berlin, 1983兲. 关12兴 S. L. Woronowicz, Rep. Math. Phys. 10, 165 共1976兲. 关13兴 P. Horodecki, Phys. Lett. A 232, 333 共1997兲. 关14兴 M. Horodecki, P. Horodecki, and R. Horodecki, in Quantum Information 关2兴. 关15兴 M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. Lett. 80, 5239 共1998兲. 关16兴 P. Horodecki, M. Horodecki, and R. Horodecki, Phys. Rev. 共A11兲 The coefficient f共a兲 must satisfy f共1兲 = 1 and the recurrence relation N m = where the factor of 3m−k is given by the sum over the m − k nonzero ␥ indexes and f共n − l兲 gives the number of ways to choose the n − l nonzero ␦ indexes so that 0␦ = 1. We note that 3N + 共− 1兲m . 3N + 1 共A14兲 Lett. 82, 1056 共1999兲. 关17兴 K. Horodecki, M. Horodecki, P. Horodecki, and J. Oppenheim, Phys. Rev. Lett. 94, 160502 共2005兲. 关18兴 M. D. Choi, Linear Algebr. Appl. 10, 285 共1975兲. 关19兴 A. Jamiolkowski, Rep. Math. Phys. 3, 275 共1972兲. 关20兴 M. D. Choi, Linear Algebr. Appl. 12, 95 共1975兲. 关21兴 K.-C. Ha, S.-H. Kye, and Y.-S. Park, Phys. Lett. A 313, 163 共2003兲. 关22兴 A. Kossakowski, Open Syst. Inf. Dyn. 10, 1 共2003兲. 关23兴 O. Rudolph, e-print quant-ph/0202121. 关24兴 K. Chen and L.-A. Wu, Quantum Inf. Comput. 3, 193 共2003兲. 关25兴 M. Horodecki, P. Horodecki, and R. Horodecki, e-print quantph/0206008. 关26兴 P. Wocjan and M. Horodecki, e-print quant-ph/0503129. 关27兴 L. Clarisse and P. Wocjan, e-print quant-ph/0504160. 关28兴 V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, J. Math. Phys. 17, 821 共1976兲. 关29兴 A. Kossakowski, Bull. Acad. Pol. Sci., Ser. Sci., Math., Astron. Phys. 12, 1021 共1972兲. 关30兴 M. Horodecki and P. Horodecki, Phys. Rev. A 59, 4206 共1999兲. 关31兴 S. Yu, Phys. Rev. A 62, 024302 共2000兲. 关32兴 F. Benatti, R. Floreanini, and M. Piani, Phys. Lett. A 326, 187 共2004兲. 关33兴 C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, and W. K. Wootters, Phys. Rev. A 54, 3824 共1996兲. 关34兴 F. Benatti, R. Floreanini, and M. Piani, Open Syst. Inf. Dyn. 012345-9 PHYSICAL REVIEW A 73, 012345 共2006兲 MARCO PIANI 11, 325 共2004兲. 关35兴 H. Spohn, Rev. Mod. Phys. 52, 569 共1980兲. 关36兴 R. Alicki and K. Lendi, Quantum Dynamical Semigroups and Applications, Lecture Notes in Physics, Vol. 286 共Springer, Berlin, 1987兲. 关37兴 H.-P. Breuer and F. Petruccione, Theory of Open Quantum Systems 共Oxford University Press, Oxford, 2002兲. 关38兴 Dissipative Quantum Dynamics, edited by F. Benatti and R. Floreanini, Lecture Notes in Physics, Vol. 612 共Springer, Berlin, 2003兲. 关39兴 F. Benatti and R. Floreanini, Banach Cent Publ. 43, 71 共1998兲. 关40兴 F. Benatti, R. Floreanini, and R. Romano, J. Phys. A 35, L551 共2002兲. 关41兴 F. Benatti, R. Floreanini, and R. Romano, J. Phys. A 35, 4955 共2002兲. 关42兴 F. Benatti, R. Floreanini, and M. Piani, Phys. Rev. A 67, 042110 共2003兲. 关43兴 F. Benatti, R. Floreanini, M. Piani, and R. Romano, e-print quant-ph/0310151. 关44兴 M. Piani, Ph.D. thesis, University of Trieste, 2005. 关45兴 P. Horodecki, Phys. Lett. A 232, 333 共1997兲. 关46兴 S. Bandyopadhyay, S. Ghosh, and V. Roychowdhury, Phys. Rev. A 71, 012316 共2005兲. 关47兴 S. Bandyopadhyay, I. Chattopadhyay, V. Roychowdhury, and D. Sarkar, Phys. Rev. A 71, 062317 共2005兲. 关48兴 W. Dur and J. I. Cirac, Phys. Rev. A 62, 022302 共2000兲. 关49兴 J. A. Smolin, Phys. Rev. A 63, 032306 共2001兲. 012345-10