## Pesquisar no resumo do documento

Annals of Physics 329 (2013) 1–27

Contents lists available at SciVerse ScienceDirect

Annals of Physics

journal homepage: www.elsevier.com/locate/aop

Deterministic LOCC transformation of three-qubit pure states and entanglement transfer Hiroyasu Tajima ∗

Department of Physics, The University of Tokyo, 4-6-1 Komaba, Meguro, Tokyo, 153-8505, Japan

a r t i c l e i n f o

Article history: Received 19 June 2012 Accepted 6 November 2012 Available online 23 November 2012

Keywords: Quantum information LOCC transformation Entanglement

a b s t r a c t

A necessary and sufficient condition of the possibility of a deter- ministic local operations and classical communication (LOCC) transformation of three-qubit pure states is given. The condition shows that the three-qubit pure states are a partially ordered set parametrized by five well-known entanglement parameters and a novel parameter; the five are the concurrences CAB, CAC , CBC , the tangle τABC and the fifth parameter J5 of Acín et al. (2000) Ref. [19], while the other new one is the entanglement charge Qe. The order of the partially ordered set is defined by the possibility of a deterministic LOCC transformation from a state to another state. In this sense, the present condition is an extension of Nielsen’s work (Nielsen (1999) [14]) to three-qubit pure states. We also clarify the rules of transfer and dissipation of the entanglement which is caused by deterministic LOCC transformations. Moreover, the minimum number of times of measurements to reproduce an arbitrary deterministic LOCC transformation between three-qubit pure states is given.

© 2012 Elsevier Inc.

1. Introduction

Entanglement is known to be a promising resource which enables us to execute various quantum tasks such as quantum computing, teleportation, superdense coding, etc. [1–4]. Quantification of the entanglement is a very important subject.

The quantification has been successful for bipartite pure states. Many indices such as the concurrence [5–7] and the negativity [8] have been proposed. Bennett et al. [9] have proven that all

∗ Tel.: +81 3 5452 6156; fax: +81 3 5452 6155. E-mail address: h-tajima@iis.u-tokyo.ac.jp.

0003-4916 © 2012 Elsevier Inc. doi:10.1016/j.aop.2012.11.001

Open access under CC BY-NC-ND license.

Open access under CC BY-NC-ND license.

2 H. Tajima / Annals of Physics 329 (2013) 1–27

of them can be expressed by the set of the coefficients of the Schmidt decomposition [10–13]. Based on the properties of this set, the following have been given:

(i) an explicit necessary and sufficient condition to determine whether an arbitrary bipartite pure state is an entangled state or a separable state;

(ii) an explicit necessary and sufficient condition to determine whether we can transform an arbitrary bipartite pure state into another arbitrary bipartite pure state with local unitary transformation (LU-equivalence);

(iii) an explicit necessary and sufficient condition to determine whether a deterministic LOCC transformation from an arbitrary bipartite pure state to another arbitrary bipartite pure state is executable or not [14];

(iv) an explicit necessary and sufficient condition to determine whether a stochastic LOCC transformation from an arbitrary bipartite pure state to an arbitrary set of bipartite pure states with arbitrary probability is executable or not [15];

(v) the fact that copies of an arbitrary partially entangled pure state can be distilled to the Bell states by an LOCC transformation, where the ratio between the copies and the Bell states is proportional to the entanglement entropy of the partially entangled state [9];

(vi) the fact that copies of an arbitrary partially entangled pure state can be reduced from the Bell states by an LOCC transformation, where the ratio between the copies and the Bell states is inversely proportional to the entanglement entropy of the partially entangled state [9].

Extension of the above to multipartite states has been vigorously sought, but this albeit it is a hard problem. Still, the multipartite results corresponding to (i) and (ii) have been given: The tangle has been defined [16], which together with the concurrences gives a solution to (i) for three-qubit pure states. The tangle τABC for three qubits A, B and C has an important property that C2A(BC) = C2AB + C

2 AC + τABC , where CAB is the concurrence between the qubits A and B, CAC is the concurrence

between the qubits A and C , and CA(BC) is the concurrence between the qubit A and the set of the qubits B and C .

It has been shown that the entanglement of three-qubit pure states is expressed by five parameters [17,18]. With the coefficients of the generalized Schmidt decomposition, we can determine whether two three-qubit pure states are LU-equivalent or not [19]. The latter gives the result corresponding to (ii) for three-qubit pure states. The result corresponding to (ii) for multipartite pure states, namely a necessary and sufficient condition for LU-equivalence, is given in Ref. [20].

There are many other important researches to clarify the features of entanglement in multipartite systems. In Ref. [21], a stochastic LOCC classification of three-qubit pure states has been done. This clarifies that there are six classes in the space of three-qubit pure states and gives a necessary and sufficient condition whether we can perform an LOCC transformation from a state to another state with finite probability or not. The tangle can be expressed by the hyper-determinant [22]. A generalization of the Schmidt decomposition [23] gives a sufficient condition for LU-equivalence of arbitrary multipartite pure states. A set of operational entanglement measures which characterize the LU-equivalence classes of three-qubit pure states (up to complex conjugation) was recently given, and it was shown thatwe can determine operationally whether a state is LU-equivalent to its complex conjugate or not [24]. However, the results corresponding to (iii)–(vi) have not been provided yet.

In the present paper, we obtain the following four results. First, a complete solution corresponding to (iii) for three-qubit pure states is given. To be precise, we give an explicit necessary and sufficient condition to determine whether a deterministic LOCC transformation from an arbitrary three-qubit pure state |ψ⟩ to another arbitrary state

ψ ′ is executable or not. We express the present condition in terms of the tangle, the concurrence between A and B, the concurrence between A and C , the concurrence between B and C , along with J5, which is a kind of phase, and a new parameter Qe, which means a kind of charge. We thereby clarify the rules of conversion of the entanglement by arbitrary deterministic LOCC transformations. Thus, defining the order between two states

ψ ′ 4 |ψ⟩ by the existence of an executable deterministic LOCC transformation from |ψ⟩ to

ψ ′, we can make the whole set of three-qubit pure states a partially ordered set. To summarize the above, we find that three-qubit pure states are a partially ordered set parametrized by the six entanglement parameters. This is an extension of Nielsen’s work [14] to three-qubit pure states.

H. Tajima / Annals of Physics 329 (2013) 1–27 3

Second, as we alreadymentioned above, we introduce a new entanglement parameterQe. The new parameter has the following three features:

1. Arbitrary three-qubit pure states are LU-equivalent if and only if their entanglement parameters (CAB, CAC , CBC , τABC , J5,Qe) are the same.

2. The parameter Qe has a discrete value: −1, 0 or 1. 3. The complex conjugate transformation on |ψ⟩ reverses the sign of Qe.

Third, we clarify the rules of conversion of the entanglement by deterministic LOCC transfor- mations. We also find that we can interpret the conversion as the transfer and dissipation of the entanglement.

Fourth, we obtain the minimum number of times of measurements to reproduce an arbitrary deterministic LOCC transformation. The minimum number of times depends on the set of the initial and the final states of the deterministic LOCC transformation;wewill list up the dependence in Table 1 in Section 2. We also show that the order of measurements are commutable; we can choose which qubit is measured first, second and third.

After completing the present work, we noticed other important results [25–33] which give partial solutions to (iii) for three qubits. In particular, two recent studies [28,33] are remarkable. The former [28] gives a necessary and sufficient condition of the possibility of a deterministic LOCC transformation of truly multipartite states whose tensor rank is two; the latter [33] gives a necessary and sufficient condition of the possibility of a deterministic LOCC transformation of W- type states, both using approaches different from ours. These studies [28,33] also give the results which correspond to the first column of Table 1. However, these studies have not achieved the complete solution to (iii) for three-qubit pure states. Specifically, they cannot determine whether a deterministic LOCC transformation from an arbitrary GHZ-type truly tripartite state to an arbitrary bipartite state is possible or not. Rules of conversion of entanglement have been provided only in implicit forms, and explicit forms of the rules have been yet to be given. Nevertheless, we decided to employ these results [28,33] partially in the proof of ourmain theory primarily in order to shorten the proof. Our original proof is given in the supplementary materials. We believe that our original proof still contains novel techniques and is valuable in its own right.

Let us overview the structure of the present paper. In Section 2, we present Theorems 1–3 of the present paper, and overview their physical meanings. Theorem 1 gives a necessary and sufficient condition of the LU-equivalence, Theorem 2 gives a necessary and sufficient condition of a deterministic LOCC transformation, and Theorem 3 gives the minimum number of necessary times of measurements to reproduce an arbitrary deterministic LOCC transformation. In Section 3, we prove Theorem 1. In Section 4, we prove Theorem 2. In supplementary materials, we prove Theorem 3.

2. Theorems and their physical meanings

In this section, we list up theorems in this article and describe their physical meaning. We consider only three-qubit pure states throughout the present paper. Before listing up theorems,

we introduce a useful expression of three-qubit pure states. An arbitrary pure state |ψ⟩ of the three qubits A, B and C is expressed in the form of the generalized Schmidt decomposition

|ψ⟩ = λ0 |000⟩ + λ1eiϕ |100⟩ + λ2 |101⟩ + λ3 |110⟩ + λ4 |111⟩ (1)

with a proper basis set [19,34]. (There are two kinds of decompositionswhich are called generalization of the Schmidt decomposition. One was given in Ref. [19] and the other was given in Ref. [23]. We use the former in the present paper.) The coefficients {λi| i = 0, . . . , 4} in (1) are nonnegative real numbers and satisfy that

4 i=0 λ

2 i = 1. Note that the phase ϕ can take any real values if one of the

coefficients {λi| i = 0, . . . , 4} is zero, in which case we define the phase ϕ to be zero in order to remove the ambiguity.

Two different decompositions of the form (1) are possible for the same state |ψ⟩, one with 0 ≤ ϕ ≤ π and the other with π ≤ ϕ ≤ 2π . These two decompositions are LU-equivalent; in other words, they can be transformed into each other by local unitary (LU) transformations. Hereafter,

4 H. Tajima / Annals of Physics 329 (2013) 1–27

Table 1 The minimum number of times of measurements to reproduce an arbitrary deterministic LOCC transformation.

Initial state Final state Times

Truly tripartite state Truly tripartite state 3 Truly tripartite state Biseparable state or full-separable state 2 Biseparable state Biseparable state or full-separable state 1 Full-separable state Full-separable state 0

we refer to the decomposition (1) with 0 ≤ ϕ ≤ π as the positive decomposition and the one (1) with π < ϕ < 2π as the negative decomposition. We also refer to the coefficients of the positive and negative decompositions as the positive-decomposition coefficients and the negative-decomposition coefficients, respectively. Therefore, a set of coefficients gives a unique set of states that are LU- equivalent to each other, whereas such a set of states may give two possible sets of coefficients: for ϕ ≠ 0, a set of positive-decomposition coefficients and a set of negative-decomposition coefficients are possible, while for ϕ = 0, two sets of positive-decomposition coefficients are possible. When sinϕ ≠ 0 holds, a set of LU-equivalent states and a set of positive-decomposition coefficients have a one-to-one correspondence.

We can express the entanglement parameters in the coefficients of the generalized Schmidt decomposition:

CAB = 2λ0λ3, CAC = 2λ0λ2, CBC = 2|λ1λ4eiϕ − λ2λ3|, (2)

τABC = 4λ20λ 2 4, (3)

J5 = 4λ20(|λ1λ4e iϕ

− λ2λ3| 2 + λ22λ

2 3 − λ

2 1λ

2 4), (4)

where τ is the tangle [16],CAB, CAC andCAC are the concurrences [6], and J5 is four times of J5 in Ref. [19]. Finally, we define the names of types of states. We refer to a state whose τABC is nonzero or whose

CAB, CAC and CBC are all nonzero as a truly tripartite state. We refer to a state which has only a single kind of the bipartite entanglement as a biseparable state (Fig. 1(a)). Note that there is no state which has only two kinds of the bipartite entanglement (Fig. 1(b)). If there were such a state as in Fig. 1(b), the coefficients {λi, ϕ| i = 0, . . . , 4} of the state would satisfy

λ0λ2 ≠ 0, λ0λ3 ≠ 0, λ0λ4 = 0, |λ1λ4eiϕ − λ2λ3| = 0, (5)

but (5) is impossible. The preparation has been now completed. Let us give theorems and see their meaning.

Theorem 1 (Condition for the LU-Equivalence). Arbitrary three-qubit pure states are LU-equivalent if and only if their entanglement parameters (CAB, CAC , CBC , τABC , J5,Qe) are equal to each other. Here Qe is a new parameter which is given by

Qe = sgn sinϕ

λ20 −

τABC + J5 2(C2BC + τABC )

, (6)

where sgn[x] is the sign function,

sgn[x] = x/|x| (x ≠ 0) 0 (x = 0)

, (7)

and the parameters ϕ and λ0 are the coefficients of the generalized Schmidt decomposition.

Physical meaning of Theorem 1: Theorem 1 gives a necessary and sufficient condition for the LU- equivalence of three-qubit pure states. The necessary and sufficient conditions can be given in other ways. However, the parameters used in Theorem1have very clear physicalmeaning of themagnitude, phase and charge of the entanglement, in the space of the LU-equivalence class of three-qubit pure states.

H. Tajima / Annals of Physics 329 (2013) 1–27 5

Fig. 1. The concept of (a) a biseparable state. There is no such state as (b).

The concurrences CAB, CAC and CBC express the amount of the entanglement between the qubits A and B, A and C and B and C , respectively. The tangle τABC expresses the amount of the entanglement among three qubits.

What about J5? The parameter J5 and the concurrences let us derive a phase of the entanglement as follows:

cosϕ5 = J5

CABCACCBC , 0 ≤ ϕ5 ≤ π. (8)

Let us refer to the phase ϕ5 as the entanglement phase (EP). The entanglement phase ϕ5 is invariant with respect to local unitary operations, because all of the parameters J5 and CAB CAC CBC are. When CABCACCBC = 0, the phase becomes indefinite. Hereafter,we refer to a statewhose entanglement phase ϕ5 is definite as an EP-definite state and to a state whose entanglement phase ϕ5 is indefinite as an EP- indefinite state. An EP-indefinite statewith τABC ≠ 0 and an EP-definite state are truly tripartite states. A truly tripartite state is an EP-indefinite state with τABC ≠ 0 or an EP-definite state. A biseparable state is EP indefinite with τABC = 0. An EP-indefinite state with τABC = 0 is a biseparable state.

Finally, we interpret the new parameterQe. As wewill prove in Section 3, the parameterQe is equal for the possible sets of coefficients of a state. Therefore, the parameter Qe is invariant with respect to local unitary transformations. The parameter Qe is a tripartite parameter, because Qe is invariant with respect to the permutation of the qubits A, B and C . This fact is shown in Appendix A. The complex- conjugate transformation of a state does not change the parameters (CAB, CAC , CBC , τABC , J5) nor λ0, but reverses the sign of sinϕ. Thus, the complex-conjugate transformation reverses the sign of Qe. As we have seen, the parameter Qe has characters that the electric charge has; hence, we refer to Qe as the entanglement charge.

The next Theorem 2 gives a necessary and sufficient condition for the possibility of a deterministic LOCC. In order to express Theorem 2 in simpler forms, we define three nonnegative real-valued parameters KAB, KAC and KBC as follows:

KAB = C2AB + τABC , KAC = C 2 AC + τABC , KBC = C

2 BC + τABC . (9)

Then, the five parameters KAB, KAC , KBC , τABC and J5 are independent of each other and are invariant with respect to local unitary operations.We can substitute these five parameters for the entanglement parameters (CAB, CAC , CBC , τABC , J5). Let us refer to the old parameters (CAB, CAC , CBC , τABC , J5) as the C-parameters and to the new parameters (KAB, KAC , KBC , τABC , J5) as the K -parameters. Note that (CAB, CAC , CBC , τABC , J5) and (KAB, KAC , KBC , τABC , J5) have a one-to-one correspondence.

We also define three parameters in order to simplify expressionswhich often appear in the present paper:

Jap ≡ C2ABC 2 ACC

2 BC , Kap ≡ KABKACKBC , K5 ≡ τABC + J5, (10)

∆J ≡ K 25 − Kap ≥ 0, (11)

where the subscript ap is abbreviation of all pairs, and where ∆J is sixteen times of ∆J in Ref. [19]. Note that these parameters Jap, Kap, K5 and ∆J are not included in the K -parameters; we introduce them only for simplicity. By definition, Jap, Kap, K5 and ∆J are invariant with respect to local unitary transformations as well as permutations of A, B and C .

6 H. Tajima / Annals of Physics 329 (2013) 1–27

Theorem 2 (Deterministic LOCC). A deterministic LOCC transformation from an arbitrary state |ψ⟩ to another arbitrary state

ψ ′ is executable if and only if the K-parameters of |ψ⟩ and ψ ′ satisfy the following conditions:

Condition 1: There are real numbers 0 ≤ ζA ≤ 1, 0 ≤ ζB ≤ 1, 0 ≤ ζC ≤ 1 and ζlower ≤ ζ ≤ 1 which satisfy the following equation:

K ′AB K ′AC K ′BC τ ′ABC J ′5

= ζ

ζAζB ζAζC

ζBζC ζAζBζC

ζAζBζC

KAB KAC KBC τABC J5

, (12) where

ζlower = Jap

(KAB − ζCτABC )(KAC − ζBτABC )(KBC − ζAτABC ) . (13)

Condition 2: If the state ψ ′ is EP definite, we check whether

(∆J = 0) ∧ (Jap − J25 = 0) (14)

holds or not. When (14) does not hold, the condition is

Qe = Q ′e and ζ = ζ̃ , (15)

where

ζ̃ ≡ Kap(Jap − J25 ) + ∆J Jap

Kap(Jap − J25 ) + ∆J(KAB − ζCτABC )(KAC − ζBτABC )(KBC − ζCτABC ) . (16)

When (14) holds, the condition is

|Q ′e| = sgn[(1 − ζ )(ζ − ζlower)], (17)

or in other expressions,

Q ′e

= 0 (ζ = 1 or ζ = ζlower), ≠ 0 (otherwise). (18)

Physicalmeaning of Theorem2: Theorem2gives a necessary and sufficient condition for the possibility of a deterministic LOCC (d-LOCC) transformation. Note that the conditions are given as the condition for a diagonal matrix which relates two vectors of the K -parameters. This shows that there exists a vector structure in three-qubit pure states.

The difficulty of seeking necessary and sufficient conditions for the possibility of multipartite d- LOCC transformation lies in the fact that we cannot apply the majorization theory, which played an important part in clarifying bipartite-pure d-LOCC transformation, to multipartite pure states. The majorization theory was applied to a vector structure which exists in the coefficients of the Schmidt decomposition. Unfortunately, there is not a vector structure in the coefficients of the generalized Schmidt decomposition, but a tensor structure. Thus, the majorization theory is not applicable to three-qubit pure states directly; this was the reason of the difficulty of clarification of three-qubit d-LOCC transformation.

However, Theorem 2 implies that there is a vector structure in three-qubit states in view of the entanglement measures. Note that there is a possibility that other multipartite systems have similar structures; it is plausible that the nonlocal features ofN-qubit pure states can be expressed completely in the magnitudes, phases and charges of the entanglement. Then, the approach of the present paper may be applicable to such systems.

H. Tajima / Annals of Physics 329 (2013) 1–27 7

Fig. 2. Entanglement transfer.

There are two important interpretations of Theorem 2. One is the rule of the flow of the entanglement and the other is the preservation of the charge. Let us see the rule of the flow of the entanglement. Consider a d-LOCC transformation whose measurement is performed only on the qubit A. In such a case, the condition 1 reduces to the following condition 1′: there are real numbers 0 ≤ ζA ≤ 1 and ζlower ≤ ζ ≤ 1 which satisfy the following equation:

K ′AB K ′AC K ′BC τ ′ABC J ′5

= ζ

ζA ζA

1 ζA

ζA

KAB KAC KBC τABC J5

, (19) where

ζlower = C2BC

(KBC − ζAτABC ) . (20)

Substituting the C-parameters (CAB, CAC , CBC , τABC , J5) for the K -parameters in (19), we obtain the following condition 1′′: 0 ≤ αA ≤ 1, 0 ≤ βA ≤ 1 which satisfy the following equation:

C ′2AB C ′2AC C ′2BC τ ′ABC J ′5

=

α2A α2A

1 βA(1 − α2A) α2A

α2A

C2AB C2AC C2BC τABC J5

. (21) We can interpret the above as the rule how a deterministic measurement, which is a measurement whose results can be transformed into a unique state by local unitary operations without exception, changes the entanglement. We can express this change as in Fig. 2. After performing a deterministic measurement on the qubit A, the four entanglement parameters, C2AB, C

2 AC , τABC and J5, the last of which

does not appear in Fig. 2, are multiplied by α2A . Note that these four entanglement parameters are related to the qubit A, which is the measured qubit. The quantity βA(1 − α2A)τABC , which is a part of the entanglement lost from τABC , is added to C2BC , which is the only entanglement parameter that is not related to the measured qubit A. The quantity (1 − βA)(1 − α2A)τABC , which is the rest of the entanglement lost from τABC disappear. We call this phenomenon the entanglement transfer.

Finally, let us see the behavior of Qe when we perform a d-LOCC transformation. Hereafter, we refer to a state which satisfies (14) as ζ̃ -indefinite and refer to a state which does not satisfy (14) as ζ̃ -definite. The following statements hold:

Statement ζ̃ -1: Any biseparable state is also a ζ̃ -indefinite state. Statement ζ̃ -2: Any ζ̃ -indefinite state satisfies Qe = 0.

8 H. Tajima / Annals of Physics 329 (2013) 1–27

Statement ζ̃ -3: A d-LOCC transformation from an EP-indefinite state to an EP-definite state is executable only if the initial state is ζ̃ -indefinite.

Statement ζ̃ -4: Among truly multipartite states, a d-LOCC transformation from a ζ̃ -indefinite state to a ζ̃ -definite state may be executable, but the contrary cannot be executable.

Statement ζ̃ -5: When the initial state is ζ̃ -definite, the d-LOCC transformation conserves the entanglement charge Qe.

Because of the above five statements, the ζ̃ -definite state can be considered as a ‘‘charge-definite state’’. When we transform a ζ̃ -indefinite state into a ζ̃ -definite state, we can choose the value of the entanglement charge Qe; once the value is determined, we cannot change it anymore with a deterministic LOCC transformation (Fig. 3).

The third theorem gives theminimumnumber of times ofmeasurements to reproduce an arbitrary deterministic LOCC transformation between three-qubit pure states.

Theorem 3 (The Minimum Number of Necessary Times of Measurements). The minimum number of necessary times of measurements to reproduce an arbitrary deterministic LOCC transformation from an arbitrary three-qubit state |ψ⟩ to another arbitrary state

ψ ′ is given as listed in Table 1. The order of measurements are commutable; we can choose which qubit is measured first, second and third.

3. The proof of Theorem 1

In this section, we prove Theorem 1. We perform the proof in the following two steps. First, we see that the C-parameters (CAB, CAC , CBC , τABC , J5) does not specify an LU-equivalence class uniquely. Second, we show that the entanglement charge Qe eliminate the non-uniqueness.

When we specify the set (CAB, CAC , CBC , τABC , J5), there are still two possible positive decomposi- tions [35,36]:

(λ±0 ) 2

= J5 + τABC ±

∆J

2(C2BC + τABC ) =

K5 ±

∆J

2 KBC , (22)

(λ±2 ) 2

= C2AC

4(λ±0 )2 , (λ±3 )

2 =

C2AB 4(λ±0 )2

, (λ±4 ) 2

= τABC

4(λ±0 )2 , (23)

(λ±1 ) 2

= 1 − (λ±0 ) 2 −

C2AB + C 2 AC + τABC

4(λ±0 )2 , (24)

cosϕ± = (λ±1 )

2(λ±4 ) 2 + (λ±2 )

2(λ±3 ) 2 − C2BC/4

2λ±1 λ ±

2 λ ±

3 λ ±

4 , (25)

where

0 ≤ ϕ± ≤ π. (26)

Thus, there are four possible sets of coefficients for one set of (CAB, CAC , CBC , τABC , J5): the positive- decomposition coefficients {λ+i , ϕ

+ | i = 0, . . . , 4} and {λ−i , ϕ

− | i = 0, . . . , 4} as well as other two

sets of coefficients {λ+i , ϕ̃ + | i = 0, . . . , 4} and {λ−i , ϕ̃

− | i = 0, . . . , 4}, where ϕ̃± = 2π −ϕ± withπ ≤

ϕ̃± ≤ 2π . A state with {λ+i , ϕ + | i = 0, . . . , 4} is LU-equivalent to a state with {λ−i , ϕ̃

− | i = 0, . . . , 4},

while a state with {λ−i , ϕ − | i = 0, . . . , 4} is LU-equivalent to a state with {λ+i , ϕ̃

+ | i = 0, . . . , 4} [35].

Therefore we can focus on two possible positive-decomposition coefficients {λ+i , ϕ + | i = 0, . . . , 4}

and {λ−i , ϕ − | i = 0, . . . , 4} for a set of (CAB, CAC , CBC , τABC , J5).

Next, we prove that the entanglement charge Qe eliminates the non-uniqueness and that two states are LU-equivalent if and only if the six parameters (CAB, CAC , CBC , τABC , J5,Qe) of the two states are equal to each other. If Qe ≠ 0, we can determine one positive-decomposition coefficients and

H. Tajima / Annals of Physics 329 (2013) 1–27 9

Fig. 3. The entanglement chargeQe for ζ̃ -definite states and ζ̃ -definite states. The arrows indicate the executable deterministic LOCC transformations among truly multipartite states; transformations not in this figure are not executable as deterministic LOCC transformations. For example, we cannot transform a ζ̃ -definite state whose Qe is 1 into another ζ̃ -definite state whose Qe is 0.

one negative-decomposition coefficients uniquely from the parameters (CAB, CAC , CBC , τABC , J5,Qe) as follows:

λ20 = J5 + τABC+̈Qe

∆J

2(C2BC + τABC ) =

K5+̈Qe

∆J

2 KBC , (27)

λ22 = C2AC 4λ20

, λ23 = C2AB 4λ20

, λ24 = τABC

4λ20 , (28)

λ21 = 1 − λ 2 0 −

C2AB + C 2 AC + τABC

4λ20 , (29)

cosϕ = λ21λ

2 4 + λ

2 2λ

2 3 − C

2 BC/4

2λ1λ2λ3λ4 , (30)

where +̈ is + or −when {λi, ϕ| i = 0, . . . , 4} are positive-decomposition coefficients or negative- decomposition coefficients, respectively. Thus, if Qe ≠ 0, the set of (CAB, CAC , CBC , τABC , J5) together with the entanglement charge Qe gives a unique set of LU-equivalent states. Note that when Qe ≠ 0, the parameter Qe is equal for the possible sets of coefficients of a state, because of (27)–(30).

If Qe = 0, at least one of sinϕ and ∆J is zero because of (6) and (22). If sinϕ is zero, {λ±i , ϕ ± | i =

0, . . . , 4} and {λ±i , ϕ̃ ± | i = 0, . . . , 4} are equal. If ∆J is zero, {λ+i , ϕ

+ | i = 0, . . . , 4} and {λ−i , ϕ

− | i =

0, . . . , 4} are equal as {λ+i , ϕ̃ + | i = 0, . . . , 4} and {λ−i , ϕ̃

− | i = 0, . . . , 4} are, respectively, because

of (22). Thus, if Qe is zero, the four sets of coefficients {λ+i , ϕ + | i = 0, . . . , 4}, {λ−i , ϕ

− | i =

0, . . . , 4}, {λ+i , ϕ̃ − | i = 0, . . . , 4} and {λ−i , ϕ̃

− | i = 0, . . . , 4} are LU-equivalent (Fig. 4). Thus, if

Qe = 0, the set of (CAB, CAC , CBC , τABC , J5) gives a unique set of LU-equivalent states. Note that the above guarantees that when Qe = 0, the parameter Qe is equal for the possible sets of coefficients of a state. Incidentally, a state is LU-equivalent to its complex-conjugate if and only if its entanglement chargeQe is zero. The complex conjugate transformation of the state only changes the sign ofϕ. Thus, a state is LU-equivalent to its complex conjugate if and only if {λ±i , ϕ

± | i = 0, . . . , 4} are LU-equivalent

to {λ±i , ϕ̃ ± | i = 0, . . . , 4}; this LU-equivalence is illustrated in Fig. 4.

For the reasons stated above, the set of (CAB, CAC , CBC , τABC , J5) together with the entanglement charge Qe gives a unique set of LU-equivalent states. In other words, two states are LU-equivalent if and only if the C-parameters (CAB, CAC , CBC , τABC , J5,Qe) of the two states are equal to each other. This is the statement of Theorem 1, and thus the proof is completed.

Note that a state is LU-equivalent to its complex conjugate if and only ifQe = 0. Thus, the condition Qe = 0 is equivalent to E6 = 0 in Ref. [24].

10 H. Tajima / Annals of Physics 329 (2013) 1–27

Fig. 4. The relation among the four sets of coefficients for Qe = 0. The relations indicated by solid lines are always valid, while those indicated by dotted lines are valid if and only if the noted conditions are satisfied.

4. The proof of Theorem 2

In this section, we prove Theorem 2. All the deterministic LOCC transformations are categorized into any of the following cases determined by the initial and final states:

Case A: Both of the initial and final states are EP definite and the initial state is GHZ-type. Case B: The initial state is EP definite and the final state is EP indefinite. Case C: The initial state is EP indefinite and GHZ-type. Case D: The tangle of initial state is zero.

Note that these four Cases exhaust all cases of the initial and final states; (Proof : The case where both the initial and final states are EP-definite is exhausted by Cases A and D. The case where the initial state is EP-definite and the final state is EP-indefinite is Case B. The case where the initial state is EP-indefinite is exhausted by Cases C and D. )

We carry out the proof of each Case in Sections 4.1–4.4, respectively.

4.1. Case A

First, we prove Theorem 2 in Case A, where both of the initial and final states are EP definite and the initial state is GHZ-type. We will argue that we can assume the final state to be GHZ-type too. The final state after an LOCC transformation is EP definite, and hence the final state is a truly multipartite state.We cannot reach aW-type state from a GHZ-type initial state with an LOCC transformation [21]. Thus the final state must be GHZ-type. We can also show that if the initial and final states satisfy the two conditions of Theorem 2, the final state must be GHZ-type. (Proof : If the final state would not be GHZ-type, then τ ′ABC would be zero, and thus at least one of ζ , ζA, ζB and ζC would be zero. Then, at least one of K ′AB, K

′

AC and K ′

BC would be zero. Because of K ′ ap ≥ J

′ ap, the final state would not beW-type.

An EP-definite state is GHZ-type or W-type, and thus this is a contradiction. ) Thus, in the present case, we only have to consider the case in which both initial and final states are GHZ-type. In this case, a necessary and sufficient condition of the possibility of a d-LOCC transformation is already given [28]. Thus, in the present case, we only have to prove the equivalence between the conditions of Theorem 2 and the necessary and sufficient condition.

An arbitrary GHZ-type state can be expressed as follows [28]:

|ψ⟩ = 1

√ N

0̃A0̃B0̃C + z 1̃A1̃B1̃C , (31) where {|0̃i⟩| i = A, B, C} and {|1̃i⟩| i = A, B, C} are normalized states of the qubits A, B and C , with their relative phases adjusted such that all of ci ≡

0̃i|1̃i

are real and nonnegative, while z is an

arbitrary complex number and N is the normalization constant. There are some possible expressions of the form (31) for an arbitrary state. These expressions have the same values of {ci} but different

H. Tajima / Annals of Physics 329 (2013) 1–27 11

values of z. If there is no zeros in the set {ci}, the number of the values of z is two. If there is a zero in the set {ci}, the number of the values of z is infinite. The relation between the values of z is as follows:

If there is no zeros in the set {ci} : z1 = 1 z2

, (32)

If there is a zero in the set {ci} : |z1| = 1

|z2| , (33)

where z1 and z2 are the values of z in the possible expressions of the form (31). When the set {ci} of the states |ψ⟩ and the set {c ′i } of

ψ ′ have no zeros, a d-LOCC transformation from the state |ψ⟩ to the state

ψ ′ is executable if and only if the following expressions are satisfied [28]:

c ′i ≥ ci (k = A, B, C), (34)

s′

s =

cAcBcC c ′Ac

′

Bc ′

C , (35)

n′

n =

cAcBcC c ′Ac

′

Bc ′

C , (36)

where

n ≡ 2Re[z] |z|2 + 1

, s ≡ 2Im[z] |z|2 − 1

(37)

with the parameter z of the state |ψ⟩, and s′ and n′ are defined for the state ψ ′with the parameter z ′.

The parameter z can take the two different values, but n and s have the same values for each value of z, except for z = ±1; when (|z| = 1) ∧ (z ≠ ±1), we consider the parameter s to take one value ∞. In the case where z = ±1, the parameter s becomes indefinite; it becomes 0/0. When the parameter s becomes indefinite, we only leave out (35). The parameter s is indefinite if and only if z = ±1. As we will show later, the equation z = ±1 is equivalent to (14) in Theorem 2. There are two other cases where we have to treat (35) and (36) exceptionally; the case where at least one of s and s′ becomes 0 or∞, and the case where at least one of n and n′ becomes 0. In the former case, we consider that (35) holds if and only if s and s′ has the same value; for example, if s takes∞, the Eq. (35) holds if and only if s′ = ∞. In the same manner, in the latter case, we consider that (36) holds if and only if n and n′ has the same value.

Let us prove that (34)–(36) are equivalent to the conditions of Theorem 2. We can express the parameters z and {ci} in terms of the C-parameters:

z = − √ KABKAC

2λ20 √ KBC

e−iϕ̃5 , (38)

cA = CBC

√ KBC

, cB = CAC

√ KAC

, cC = CAB

√ KAB

, (39)

where

e−iϕ̃5 = λ2λ3 − λ1λ4eiϕ

|λ2λ3 − λ1λ4eiϕ | = cosϕ5 − isgn[sinϕ] sinϕ5. (40)

Because there are two possible sets of {λi, ϕ| i = 0, . . . , 4} in (38), the parameter z takes two values. These two values satisfy (32) and (33). The derivation of the above expressions is given in Appendix B. It is easily seen from (39) that the set {ci} of a state has no zeros if and only if the state is EP-definite. Now that we have reduced CaseA into the case in which the initial and final states are EP-definite and GHZ-type, we only have to prove that (34)–(36) are equivalent to the conditions of Theorem 2.

12 H. Tajima / Annals of Physics 329 (2013) 1–27

Using (38) and (39), we prove that (34) and (36) are equivalent to the condition 1 of Theorem 2 except for ζlower ≤ ζ ≤ 1. Substituting (39) into (34), we transform (34) into

τ ′ABC

τABC ≤

K ′AB KAB

, τ ′ABC

τABC ≤

K ′AC KAC

, τ ′ABC

τABC ≤

K ′BC KBC

. (41)

Because of (40),

Re[z] = − √ KABKAC

2λ20 √ KBC

cosϕ5. (42)

Because the possible two sets of {λi, ϕ| i = 0, . . . , 1} are {λ+i , ϕ + | i = 0, . . . , 1} and {λ−i , ϕ̃

− | i =

0, . . . , 1} or {λ−i , ϕ − | i = 0, . . . , 1} and {λ+i , ϕ̃

+ | i = 0, . . . , 1}, we obtain

n = −

Kap

2 K5 cosϕ5. (43)

Thus, we can transform (36) into

K ′5 K5

= J ′5 J5

(for n ≠ 0), J5 = J ′5 = 0 (for n = 0). (44)

Let us then prove that the expressions (41) and (44) are equivalent to the existence of 0 ≤ ζ ≤ 1, 0 ≤ ζA ≤ 1, 0 ≤ ζB ≤ 1 and 0 ≤ ζC ≤ 1, which satisfy (12). To show this, we only have to define ζ , ζA, ζB and ζC as follows:

ζA = KBCτ ′ABC K ′BCτABC

, ζB = KACτ ′ABC K ′ACτABC

, ζC = KABf τ ′ABC K ′ABτABC

, (45)

ζ = τ ′ABC

τABCζAζBζC . (46)

Because of (44), (45) and (46), the parameters ζ , ζA, ζB and ζC satisfy (12). Because of (41), the parameters ζA, ζB and ζC are less than or equal to one. Because of (45) and (46), the parameters ζ , ζA, ζB and ζC are greater than or equal to zero. Thus, the expressions (34) and (36) are equivalent to the condition 1 of Theorem 2, except for ζlower ≤ ζ ≤ 1.

Now we have proven that the Eqs. (34) and (36), which are the ones on the parameters {ci} and n, are equivalent to the condition 1 of Theorem 2 except for ζlower ≤ ζ ≤ 1. Next, let us prove that the condition on the parameter s is equivalent to the inequality ζlower ≤ ζ ≤ 1 and the condition 2 of Theorem 2. Hereafter, we assume the condition 1 of Theorem 2 except for ζlower ≤ ζ ≤ 1 until the end of Case A. The condition on s is (35) for z ≠ ±1 but is left out for z = ±1. First, we prove that the equation z = ±1 is equivalent to (14). Because of (38), we can transform z = ±1 into

z = ±1 ⇔ (ϕ̃5 = 0, π) ∧ √

KABKAC 2λ20

√ KBC

= 1

(47)

⇔ (Jap sin2 ϕ5 = 0) ∧

Kap

K5 ±

∆J = 1

(48)

⇔ (Jap sin2 ϕ5 = 0) ∧ (∆J = 0) (49)

⇔ (Jap − J25 = 0) ∧ (∆J = 0) (50)

where the double sign ± in (48) is the one in (22); note that the possible two sets of {λi, ϕ| i = 0, . . . , 1} are {λ+i , ϕ

+ | i = 0, . . . , 1} and {λ−i , ϕ̃

− | i = 0, . . . , 1} or {λ−i , ϕ

− | i = 0, . . . , 1} and

{λ+i , ϕ̃ + | i = 0, . . . , 1}.

Thus, the equation z = ±1 is equivalent to (14). In other words, Eq. (14) does not hold for z ≠ ±1.

H. Tajima / Annals of Physics 329 (2013) 1–27 13

Let us next prove that the condition on s is equivalent to the condition 2 of Theorem 2 and ζlower ≤ ζ ≤ 1 in the case z ≠ ±1. In this case, what we have to prove is that (35) is equivalent to the condition 2 of Theorem 2 and ζlower ≤ ζ ≤ 1. In this case, (14) does not hold, and thus |ψ⟩ is ζ̃ -definite. Hence, we only have to prove that (35) is equivalent to (15). (Note that ζlower ≤ ζ̃ ≤ 1.) In order to prove the equivalence of (35) and (15), we first express s in terms of the K -parameters. When Qe ≠ 0, by substituting (27) and (40) into (35), we obtain

s = −

Kap

Qe

∆J sinϕ5. (51)

Because of (39) and (51),

Q ′e

∆J

Qe

∆′J

J ′ap sinϕ

′

5 Jap sinϕ5

= 1 (52)

always holds. Substituting ∆J = K 25 − Kap, Jap sin 2 ϕ5 = Jap − J25 and (12) into (52) and reducing it, we

can obtain Qe = Q ′e and ζ = ζ̃ . Thus, when Qe ≠ 0, (15) is equivalent to (35). Now we consider the case where z ≠ ±1, and thus when Qe = 0, only one of the expressions

sinϕ5 = 0 and ∆J = 0 holds. When sinϕ5 = 0 holds, the Eq. (35) is equivalent to s = s′ = 0. When ∆J = 0 holds, the Eq. (35) is equivalent to s = s′ = ∞. Because of (37) and (38), the equations s = s′ = 0 and s = s′ = ∞ are equivalent to sinϕ5 = sinϕ′5 = 0 and ∆J = ∆

′

J = 0, respectively. Because of (Qe = 0) ⇔ ((sinϕ5 = 0)∨(∆J = 0)), ∆J = K 25 −Kap, Jap sin

2 ϕ5 = Jap−J25 and (12), when the equation sinϕ5 = 0 holds, the equation sinϕ5 = sinϕ′5 = 0 is equivalent to (15). In the same manner, when the equation ∆J = 0 holds, ∆J = ∆′J = 0 is equivalent to (15). Thus, when Qe = 0, the expression (15) is equivalent to (35). We have already proven that when Qe ≠ 0, the expression (15) is equivalent to (35). Thus, (15) is equivalent to (35).

Finally, we consider the case z = ±1. In this case, there is no condition of s and the condition (14) holds. Because of (14), the condition 2 of Theorem 2 becomes (17) in this case. Thus, we only have to prove that (14) is equivalent to ζlower ≤ ζ ≤ 1 and (17). Because we have assumed the condition 1 of Theorem 1 except for ζlower ≤ ζ ≤ 1, we obtain the following equations:

J ′25 J ′ap

= ζlower

ζ

J25 Jap

, (53)

∆′J = (ζ ζAζBζC ) 2(K 25 − ζKap) ≥ (ζ ζAζBζC )

2∆J . (54)

The derivation of (53) and (54) is as follows:

J ′25 J ′ap

= (ζ ζAζBζC )

2J25 ζ 3ζ 2A ζ

2 B ζ

2 C (KBC − ζAτABC )(KAC − ζBτABC )(KAB − ζCτABC )

= ζlower

ζ

J25 Jap

, (55)

∆′J = (ζ ζAζBζC ) 2 K 25 − ζ

3ζ 2A ζ 2 B ζ

2 C Kap

= (ζ ζAζBζC ) 2(K 25 − ζKap) ≥ (ζ ζAζBζC )

2∆J . (56)

By using (53) and (54), let us prove that (14) is equivalent to ζlower ≤ ζ ≤ 1 and (17). First, we prove that (14) is a sufficient condition of ζlower ≤ ζ ≤ 1 and (17). When (14) is valid, the equation J25/Jap = 1 holds. Because J

′2 5 /J

′ ap = cos

2 ϕ′5 ≤ 1 is always valid, when (14) is valid, the inequality ζlower ≤ ζ follows (53). In the same manner, when (14) is valid, the inequality ζ ≤ 1 follows (54). Then, note that the equation J ′25 = J

′ ap holds if and only if ζ = ζlower holds, and that the equation∆

′

J = 0 holds if and only if ζ = 1 holds. Because of (6), the equation Q ′e = 0 holds if and only if (∆

′

J = 0) ∨ (J ′25 = J

′ ap) holds. Thus, when (14) is valid, Q

′ e = 0 if and only if ζ = ζlower or ζ = 1. In the other

words, now we have proven that (14) is a sufficient condition of ζlower ≤ ζ ≤ 1 and (17).

14 H. Tajima / Annals of Physics 329 (2013) 1–27

Let us prove that (14) is also a necessary condition. Because of ζlower ≤ ζ ≤ 1, (53), (54) and (6), the equation |Q ′e| = sgn[(1 − ζ )(ζ − ζlower)] is valid if and only if both of ∆J and J

2 5 − Jap are zero. In

the other words, (14) is also a necessary condition of ζlower ≤ ζ ≤ 1 and (17). Thus, (14) is equivalent to ζlower ≤ ζ ≤ 1 and (17), and thus we have completed the proof in Case A.

4.2. Case B

In this subsection, we prove Theorem 2 in Case B, where the initial state is EP-definite and the final state is EP-indefinite. In the proof, we will use the following two lemmas which describe how a measurement changes entanglement.

Lemma 1. Let us consider the situation where a measurement {M(i)} is performed on the qubit A of an arbitrary three-qubit pure state |ψ⟩. The state

ψ (i) ≡ M(i) |ψ⟩ /√p(i) is obtained as the i-th result with the probability p(i). Then, the following equations are valid:

C (i)AB = α (i)CAB, C

(i) AC = α

(i)CAC ,

τ (i) ABC = α

(i)√τABC , (57)

p(i)C (i) BC = |k(i)

√ τABCei(θ(i)+ϕ̃5) − CBCb(i)|, (58)

p(i) = λ20a(i) + (1 − λ 2 0)b(i) + 2λ0λ1k(i) cos(θ(i) − ϕ) (59)

where C (i)AB , C (i) AC , C

(i) BC and τ

(i) ABC are the concurrences and the tangle of the state

ψ (i) and MĎ(i)M(i) ≡

a(i) k(i)e−iθ(i)

k(i)eiθ(i) b(i)

, (60)

a(i)b(i) − k2(i) ≥ 0, a(i) ≥ 0, b(i) ≥ 0, k(i) ≥ 0, 0 ≤ θ(i) ≤ 2π, (61)

α(i) ≡

a(i)b(i) − k2(i)

p(i) . (62)

Proof. When we perform a transformation which is expressed as

M = M00 M01 M10 M11

, M00,M01,M10,M11 ∈ C, (63)

on the qubit A of a pure state (1), the state |ψ⟩ is transformed into

M |ψ⟩ = (λ0M00 |0⟩ + λ0M10 |1⟩ + λ1eiϕM01 |0⟩ + λ1eiϕM11 |1⟩) |00⟩ + λ2(M01 |0⟩ + M11 |1⟩) |01⟩ + λ3(M01 |0⟩ + M11 |1⟩) |10⟩ + λ4(M01 |0⟩ + M11 |1⟩) |11⟩ . (64)

We can transform (64) into the form of the generalized Schmidt decomposition (1) with straight- forward algebra;

M |ψ⟩ = λ0 det

√ MĎM

|M01|2 + |M11|2 |000⟩

+ λ0(M00M∗01 + M10M

∗

11) + λ1e iϕ(|M01|2 + |M11|2)

|M01|2 + |M11|2 |100⟩

+

|M01|2 + |M11|2(λ2 |101⟩ + λ3 |110⟩ + λ4 |111⟩). (65)

H. Tajima / Annals of Physics 329 (2013) 1–27 15

Note that each coefficient of the generalized Schmidt decomposition (65) ofM |ψ⟩ above is expressed by the components ofMĎM solely:

MĎM =

|M00|2 + |M10|2 M∗00M01 + M

∗

10M11 M∗01M00 + M

∗

11M10 |M01| 2 + |M11|2

. (66)

Then expressing the components ofMĎ(i)M(i) as in (60), we can also expressM(i) |ψ⟩ and p(i) as

M(i) |ψ⟩ = λ0

a(i)b(i) − k2(i)

b(i) |000⟩ +

λ0k(i)eiθ(i) + λ1eiϕb(i) √ b(i)

|100⟩

+ λ2 b(i) |101⟩ + λ3

b(i) |110⟩ + λ4

b(i) |111⟩ . (67)

p(i) = ⟨ψ |M Ď (i)M(i) |ψ⟩

= λ20a(i) + (1 − λ 2 0)b(i) + 2λ0λ1k(i) cos(θ(i) − ϕ). (68)

Because of ψ (i) = M(i) |ψ⟩ /√p(i), we can express the coefficients of the generalized Schmidt

decomposition {λ(i)|k = 0, . . . , 4} as

λ (i) 0 =

λ0

a(i)b(i) − k2(i)

√ p(i)

b(i)

, (69)

λ (i) 1 e

iϕ(i) =

λ0k(i)eiθ(i) + λ1eiϕb(i) √ p(i)

b(i)

, (70)

λ (i) 2 =

λ2 b(i)

√ p(i)

, λ (i) 3 =

λ3 b(i)

√ p(i)

, λ (i) 4 =

λ4 b(i)

√ p(i)

. (71)

Because of (2), (3) and (69)–(71), the Eqs. (57) and (58) are valid.

Lemma 2. Let the notation {M(i)| i = 1, 2} stand for an arbitrary two-choice measurement which is operated on the qubit A of a three-qubit pure state |ψABC ⟩. We refer to each result of the measurements {M(i)| i = 1, 2} as

ψ (i)ABC . Let the notations (CAB, CAC , CBC , τABC , J5,Qe) and (C (i)AB , C (i)AC , C (i)BC , τ (i)ABC , J (i)5 ,Q (i)e ) stand for the sets of the C-parameters of the states |ψABC ⟩ and

ψ (i)ABC , respectively. Then, the following inequalities hold:

CBC ≤ 1

i=0

p(i)C (i) BC ≤

C2BC + 1 − 1

k=0

p(i)α(i) 2 τABC , (72)

where the probability p(i) and the multiplication factor α(i) are defined in (59) and (62), respectively.

Proof. The average 1

i=0 p(i)C (i) BC is equal to the length of the heavy line in Fig. 5, because we can

interpret (58) as the cosine theorem and because b(0) + b(1) = 1 and

i k⃗(i) = 0. The end points of the heavy line have to coincide with the end points of the segment RS because

i k⃗(i) = 0. Then, the

left inequality CBC ≤ 2

i=1 piC (i) BC clearly holds, since a polygonal line is longer than a straight line.

16 H. Tajima / Annals of Physics 329 (2013) 1–27

Fig. 5. A geometric interpretation of the change of CBC .

To prove the right inequality of this lemma, it suffices to show the inequality (bCBC + k cos θ

√ τABC )2 + (k sin θ

√ τABC )2

+

[(1 − b)CBC − k cos θ

√ τABC ]2 + (k sin θ

√ τABC )2

≤

C2BC + {1 − [

ab − k2 +

(1 − a)(1 − b) − k2]2}τABC (73)

under the conditions ab − k2 ≥ 0, (1 − a)(1 − b) − k2 ≥ 0, 0 ≤ θ ≤ 2π, 0 ≤ a ≤ 1 and 0 ≤ b ≤ 1, where we used the substitutions:

a(0) = a, a(1) = 1 − a, b(0) = b, b(1) = 1 − b, θ(0) + ϕ̃5 = π − θ. (74)

The fact that θ can take any value guarantees the last substitution. Let us find maximum value of the left-hand side of (73) with the values of the measurement

parameters a, b and k fixed. We can express the left-hand side of (73) as u2 + w2 + 2uw cos θ +

v2 + w2 − 2vw cos θ, (75)

where u = bCBC , v = (1 − b)CBC and w = k √

τABC . With the values of u, v and w fixed, the quantity (75) is maximized to be (u + v)

1 + w2/(uv) at the point cos θ = w(u − v)/2uv. Substituting

u = bCBC , v = (1 − b)CBC and w = k √

τABC give that

the maximum of the left-hand side of (73) =

C2BC +

k2τABC b(1 − b)

. (76)

Hence, in order to prove Lemma 2, it suffices to show

k2

b(1 − b) ≤ 1 − [

ab − k2 +

(1 − a)(1 − b) − k2]2. (77)

After straightforward algebra, (77) is reduced to (a − b) + k2

(b2 − (1 − b)2) b(1 − b)

2 ≥ 0. (78)

Hence we have proved the right inequality of (72) and thereby completed the proof of Lemma 2.

Note that Lemma 2 has the following corollary:

0 ≤ i

p(i)αi ≤ 1. (79)

H. Tajima / Annals of Physics 329 (2013) 1–27 17

Now we have completed the preparation. Let us start the proof of Theorem 2 in Case B. In the present Case, the final state is EP-indefinite, and thus we can neglect the condition 2 of Theorem 2. Thus, we only have to show that the condition 1 of Theorem 1 is a necessary and sufficient condition of the possibility of a d-LOCC transformation.

Let us prove that a d-LOCC transformation from |ψ⟩ to ψ ′ is executable if |ψ⟩ and ψ ′ satisfy the

condition 1 of Theorem 2. First we prove that if |ψ⟩ and ψ ′ satisfy the condition 1 of Theorem 2, the

state ψ ′ is biseparable or full-separable. Because |ψ⟩ is EP-definite and because of (9), KAB > τABC , KAB > τABC , KBC > τABC . (80)

Because ψ ′ is EP-indefinite, at least one of

K ′AB = τ ′

ABC , K ′

AB = τ ′

ABC , K ′

BC = τ ′

ABC (81) holds. We can assume K ′AB = τ

′

ABC without losing generality. Because of (80), K ′

AB = τ ′

ABC and the condition 1 of Theorem 2, at least one of ζ , ζA, ζB is zero. When only one of ζA and ζB is zero and ζ is not zero, the state

ψ ′ is biseparable; only K ′BC or K ′AC is not zero. When both of ζA and ζB are zero or ζ is zero, the state

ψ ′ is full-separable; all of the K -parameters of ψ ′ are zero. Now we have proven that if |ψ⟩ and

ψ ′ satisfy the condition 1 of Theorem 2, the state ψ ′ is biseparable or full-separable.

Next, let us prove that if |ψ⟩ and ψ ′ satisfy the condition 1 of Theorem 2, there is an executable

d-LOCC transformation from |ψ⟩ to ψ ′. Now the state ψ ′ is biseparable or full-separable. Without

losing generality, we can assume that C ′BC is the only nonzero parameter in (C ′

AB, C ′

AC , C ′

BC , τ ′

ABC , J ′

5,Q ′ e).

Because |ψ⟩ and ψ ′ satisfy the condition 1 of Theorem 2, the inequality C ′2BC ≤ KBC holds. The set of

full-separable states and biseparable states which have the same kind of bipartite entanglement is a totally ordered set [14]. In other words, when two states |ψ⟩ and

ψ ′ belong to such a set, there is an executable deterministic LOCC transformation from the EP-definite state |ψ⟩ to the EP-indefinite stateψ ′ if and only if the bipartite entanglement of the state |ψ⟩ is greater than or equal to that of the stateψ ′. Thus, if there is the followingmeasurement {M(i)}, there is an executable d-LOCC transformation from |ψ⟩ to

ψ ′; a measurement whose results can be transformed into a unique state ψ ′′ by local unitary operations without exception, where

ψ ′′ is a biseparable state whose C2BC is equal to KBC of |ψ⟩. The measurement {M(i)} is given as follows:

MĎ(0)M(0) =

a(0) k(0)e−iθ(0) k(0)eiθ(0) b(0)

=

a ke−iθ

keiθ b

, (82)

MĎ(1)M(1) =

a(1) k(1)e−iθ(1) k(1)eiθ(1) b(1)

=

1 − a −ke−iθ

−keiθ 1 − b

, (83)

where the measurement parameters a, b, k and θ are defined as follows:

a = 1 2

− λ1 sinϕ

2

λ21 sin 2 ϕ + λ20

, (84)

b = 1 2

+ λ1 sinϕ

2

λ21 sin 2 ϕ + λ20

, (85)

k = λ0

2

λ21 sin 2 ϕ + λ20

, (86)

θ = π

2 . (87)

With substituting (84)–(87) into (57) and (58) and after straightforward algebra, we can confirm that the measurement {M(i)} is the measurement that we sought. Thus, we have proven that the condition 1 of Theorem 2 is a sufficient condition for the existence of an executable d-LOCC transformation.

18 H. Tajima / Annals of Physics 329 (2013) 1–27

Fig. 6. The method of counting the number N . In this figure,M1,M2 andM3 denote measurements and I denotes the identity transformation. The number N is 3 in this example.

Next, let us prove that the condition 1 of Theorem 2 is also a necessary condition. In other words, we prove that if there is an executable d-LOCC transformation from |ψ⟩ to

ψ ′, the states |ψ⟩ andψ ′ must satisfy the condition 1 of Theorem 2. It is possible to substitute two-choice measurements for any measurements of an LOCC transformation on a three-qubit pure state [21]. Hereafter, unless specified otherwise, measurements of LOCC transformations will be two-choice measurements. First, we prove that a deterministic LOCC transformation from an EP-definite state to an EP-indefinite state is executable only if the final state is biseparable or full-separable.Weprove Lemma3,which generally holds for stochastic LOCC transformation including d-LOCC transformations.

Lemma 3. Let the notation TSL stand for an LOCC transformation from an arbitrary EP-definite state |ψ⟩ to arbitrary EP-indefinite states {

ψ (i)}. The subscript SL stands for stochastic LOCC. Then, if this LOCC transformation TSL is executable, there must be full-separable states or biseparable states in the set {

ψ (i)}. Proof. We prove the present lemma by mathematical induction with respect to N , which is the number of times measurements are performed in the LOCC transformation TSL. Let us define how to count the number of times of the measurement. Let the notation T stands for an arbitrary LOCC transformation. We fix the order of measurements in the LOCC transformation T cyclically: If the first measurement of the LOCC transformation T is performed on the qubit A, the second one is on the qubit B, the third one is on the qubit C , the fourth one returns to the qubit A, and so on. If the first measurement is performed on the qubit B, the second one is on the qubit C , and so on. We can attain such a fixed order by inserting the identity transformation as a measurement. The LOCC transformation T may have branches and the numbers of times the measurements are performed may be different in different branches. We refer to the largest of the numbers as the number N . We canmake the number of each branch equal toN by inserting the identity transformations. An example is given in Fig. 6. We use this counting procedure in the proofs of other theorems, too.

H. Tajima / Annals of Physics 329 (2013) 1–27 19

Let the notations (CAB, CAC , CBC , τABC , J5,Qe) and (C (i) AB , C

(i) AC , C

(i) BC , τ

(i) ABC , J

(i) 5 ,Q

(i) e ) stand for the sets of

the C-parameters of the EP-definite state |ψ⟩ and the EP-indefinite states ψ (i), respectively.

First, we prove the present lemma for N = 1. Because of the arbitrariness of the state |ψ⟩, we can assume that the first measurement {M(i)| i = 0, 1} of the LOCC transformation TSL is performed on the qubit Awithout loss of generality. Thus, the operatorM(i) makes CAB, CAC and

√ τABC evenly multiplied

by a real number α(i). The state |ψ⟩ is EP definite, and hence CAB, CAC and CBC are all positive. Because the state

ψ (i) is EP indefinite, at least one of C (i)AB , C (i)AC and C (i)BC has to be zero for all i. When C (i)AB or C (i)AC is zero, the multiplication factor α(i) must be zero, and therefore all of C (i)AB , C

(i) AC and τ

(i) ABC must be zero.

Then, the parameter J (i)5 also must be zero because of C (i) ABC

(i) ACC

(i) BC = 0. Thus, in the case of C

(i) AB = 0

or C (i)AC = 0, the EP-indefinite state ψ (i) is a full-separable state with C (i)BC = 0 or a biseparable state

with C (i)BC ≠ 0. Hence, if there were neither a full-separable state nor a biseparable state in the set of EP-indefinite states {

ψ (i)}, the expressions C (i)AC ≠ 0, C (i)AB ≠ 0 and C (i)BC = 0 would hold for all i. Because of Lemma 2, however, at least one of C (0)BC and C

(1) BC would be greater than or equal to CBC ,

which is positive. This is a contradiction, and thus the expression C (i)BC ≠ 0 has to hold for at least one of i. We have thereby shown the present lemma for N = 1.

Now, we prove Lemma 3 for N = k + 1, assuming that Lemma 3 holds whenever 1 ≤ N ≤ k. Let us assume that the number of times of measurements in the LOCC transformation TSL from the EP- definite state |ψ⟩ to the EP-indefinite states {

ψ (i)} is k+1. Because of the assumption for 1 ≤ N ≤ k, the situation before the last measurement has to be either of the following two situations:

(i) All states are already EP indefinite, and there are full-separable states or biseparable states among them.

(ii) Some states are EP definite.

In the case of (i), there are full-separable states or biseparable states in the final EP-indefinite states { ψ (i)} because an arbitrary full-separable state or an arbitrary biseparable state can be transformed only into full-separable states or biseparable states by a measurement. In the case of (ii), if therewere neither a full-separable state nor a biseparable state in the EP-indefinite states {

ψ (i)}, there would have to be a measurement which could transform an EP-definite state to EP-indefinite states which are neither full-separable states nor biseparable states. Because of the theorem for N = 1, this is impossible.

Therefore, theremust be either full-separable states or biseparable states in the EP-indefinite states { ψ (i)} in the case (ii) as well as in the case (i). This completes the proof of Lemma 3. Because of Lemma 3, if a d-LOCC from an EP-definite state |ψ⟩ to an EP-indefinite state

ψ ′ is executable, then

ψ ′ is biseparable or full-separable. Without losing generality, we can assume that the only nonzero parameter in the K -parameters of

ψ ′ is K ′BC . To show that the K -parameters of |ψ⟩ and

ψ ′ satisfy the condition 1 of Theorem 2, it is sufficient to prove the inequality K ′BC ≤ KBC . In order to prove K ′BC ≤ KBC , we prove the following Lemma 4.

Lemma 4. Let the notation {M(i)| i = 0, 1} stand for an arbitrary two-choice measurement which is operated on a qubit of a three-qubit pure state |ψABC ⟩. Note that we can operate {M(i)| i = 0, 1} on any

one of the qubits A, B and C of the state |ψABC ⟩. We refer to each result of {M(i)| i = 0, 1} as ψ (i)ABC .

Let the notations (KAB, KAC , KBC , τABC , J5,Qe) and (K (i) AB , K

(i) AC , K

(i) BC , τ

(i) ABC , J

(i) 5 ,Q

(i) e ) stand for the sets of the

K-parameters of the states |ψABC ⟩ and ψ (i)ABC , respectively. Then, the following inequality holds:

1 i=0

p(i) K (i)BC ≤

KBC . (88)

20 H. Tajima / Annals of Physics 329 (2013) 1–27

Proof. First, we prove (88) in the case where the measurement {M(i)| i = 0, 1} is performed on the

qubit B or C . In this case C (i)BC = α (i)CBC and

τ

(i) ABC = α

(i)√τABC , where α(i) is the multiplication factor of the measurement {M(i)| i = 0, 1}. Thus, because of (79), we can obtain (88) as follows:

1 i=0

p(i) K (i)BC =

1 i=0

p(i)

(C (i)BC )2 + τ (i) ABC =

1 i=0

p(i)α(i) C2BC + τABC ≤

KBC . (89)

Now, it suffices to prove (88) in the case where the first measurement is performed on the qubit A. Let the notation f stand for the left-hand side of (88). Because of Lemma 1 we obtain

f = b2C2BC + 2bk cos θCBC

√ τABC + abτABC

+

(1 − b)2C2BC − 2(1 − b)k cos θCBC

√ τABC + (1 − a)(1 − b)τABC , (90)

where we substitute θ for the phase π − θ − ϕ̃5, because the range of the phase θ is from 0 to 2π . When 2k cos θCBC = (b − a)

√ τABC , the function f is maximized to

√ KBC . Thus, (88) holds.

The inequality (88) includes K ′BC ≤ KBC , if the number N of times measurements of a d-LOCC transformation from |ψ⟩ to

ψ ′ is equal to 1. Let us assume that K ′BC ≤ KBC also holds whenever 1 ≤ N ≤ k. Then, when N = k+1, the inequality K ′BC ≤ KBC also holds; (Proof : Let the notion {

ψ ′(i)} stand for results of the first measurement the d-LOCC transformation. We refer to the parameter KBC of

ψ ′(i) as K ′(i)BC . Note that the LOCC transformations from ψ ′(i) to ψ ′ are d-LOCC transformations whose the number of timesmeasurements are less than or equal to k. Thus, the inequalities K ′BC ≤ K

′(i) BC

hold for all of K ′(i)BC , and thus the inequality K ′

BC ≤ KBC also holds. ) Hence, nowwe have completed the proof of Theorem 2 in Case B.

4.3. Case C

In this subsection, we prove main theorems in Case C, where the initial state is EP-indefinite and GHZ-type.

First, we consider the case in which the final state is EP-definite. In this case, the final state is truly multipartite. We cannot achieve a W-type state from a GHZ-type state with an LOCC transformation [21]. Thus, the final statemust be a GHZ-type state too.We have already proven that if the initial state is GHZ-type, and if the initial and final states satisfy the two conditions of Theorem 2, the final state must be GHZ-type. Thus, we only have to consider the case in which both initial and final states are GHZ-type. A d-LOCC transformation from a GHZ-type state whose coefficient set {ci} has a zero to a GHZ-type state whose coefficient set {ci} has no zeros is executable if and only if

c ′i ≥ ci (k = A, B, C), (91)

|z| = 1, (92)

z ′ is purely imaginary, (93)

where {ci} and z are defined in (31) [28]. Because of (39), a GHZ-type state is EP-definite if and only if its coefficient set {ci} has no zeros. Thus, in the present case where the final state is EP-definite, we only have to prove that (91)–(93) are equivalent to the two conditions of Theorem 2.

Let us prove the equivalence between (91)–(93) are the two conditions of Theorem 2. Because of (38) and (39), the expressions (91)–(93) are equivalent to

τ ′ABC

τABC ≤

K ′AB KAB

, τ ′ABC

τABC ≤

K ′AC KAC

, τ ′ABC

τABC ≤

K ′BC KBC

, (94)

√ KABKAC

2λ20 √ KBC

=

Kap

K5 ±

∆J = 1, (95)

H. Tajima / Annals of Physics 329 (2013) 1–27 21

ϕ̃′5 = ± π

2 , (96)

where the double sign ± in (95) is the one in (22); note that the possible two sets of {λi, ϕ| i = 0, . . . , 1} are {λ+i , ϕ

+ | i = 0, . . . , 1} and {λ−i , ϕ̃

− | i = 0, . . . , 1} or {λ−i , ϕ

− | i = 0, . . . , 1} and

{λ+i , ϕ̃ + | i = 0, . . . , 1}. We can define such ζ–ζC as (45) and (46), and thus (12) is equivalent to the

existence of ζ , 0 ≤ ζA ≤ 1, 0 ≤ ζB ≤ 1 and 0 ≤ ζC ≤ 1. Because (95) is valid in either case of the multiple signs, the parameter ∆J is zero. Because of ∆J = K 25 − Kap, when ∆J = 0 holds, (95) also holds. Thus, (95) is equivalent to ∆J = 0. Because of

Jap cos ϕ̃5 = J5 and (12), (96) is equivalent to

J5 = Jap = 0. Thus, (95) and (96) are equivalent to (14).We have already proven that (14) is equivalent to ζlower ≤ ζ ≤ 1 and (17) in the Section 4.1. Thus, in the case where the final state is EP-definite, the conditions of Theorem 2 is a necessary and sufficient condition of d-LOCC.

Second, we consider the case where the final state is EP-indefinite and GHZ-type. Because a GHZ- type state is EP-definite if and only if its coefficient set {ci} has no zeros, in this case, a d-LOCC transformation is executable if and only if

c ′i ≥ ci (k = A, B, C), (97)

|z ′| ≥ |z|, (98)

where {ci} and z are defined in (31), and where we choose z and z ′ such that |z| ≥ 1 and |z ′| ≥ 1 [28]. Let us prove that (97) and (98) are equivalent to the condition 1 of Theorem 2. Because of (22), (38),

|z ′| ≥ 1 and |z| ≥ 1,

|z| =

Kap

K5 −

∆J , |z ′| =

K ′ap

K ′5 −

∆′J

. (99)

Because of (39), (97) is equivalent (94). Because both of the initial and final states are EP-indefinite, J5 = J ′5 = 0 is valid. Thus, we can define ζlower ≤ ζ ≤ 1, 0 ≤ ζA ≤ 1, 0 ≤ ζB ≤ 1 and 0 ≤ ζC ≤ 1, which satisfy (12) as (45) and (46). Thus, in the case where the final state is EP-indefinite and GHZ- type, the conditions of Theorem 2 is a necessary and sufficient condition of d-LOCC.

Third, we consider the case where the final state is EP-indefinite but not GHZ-type. In this case, the final state is biseparable or full-separable. Because the final state is not EP-definite, the condition 2 of Theorem 2 is left out; we only have to consider the condition 1. When the final state is full-separable, a d-LOCC transformation to the final state is clearly executable, and the initial and final states clearly satisfy the condition 1 of Theorem 2: ζA = ζB = ζC = 0 satisfy (12). Thus, we only have to consider the case the final state is biseparable.

Let us prove the condition 1 of Theorem 2 is a necessary condition of the possibility of the d-LOCC transformation. Without loss of generality, we can assume that the only nonzero K -parameter of the final state is K ′BC . Because of Lemma 4, we can prove the inequality K

′

BC ≤ KBC , in the same manner as K ′BC ≤ KBC is shown in Case B. Thus, if a d-LOCC transformation is executable, the initial and final states satisfy the condition 1 of Theorem 2. In other words, the condition 1 of Theorem 2 is a necessary condition of the possibility of the d-LOCC transformation.

Finally, let us prove the condition 1 of Theorem 2 is a sufficient condition of the possibility of the d-LOCC transformation. Without loss of generality, we can assume that the only nonzero K - parameter of the final state is K ′BC . The upper limit of K

′

BC is KBC . We can realize this upper limit by the measurement {M(i)}which is defined as (82)–(87). According to Ref. [14], the set of full-separable states and biseparable states which have the same kind of bipartite entanglement is a totally ordered set. Thus, in this case, a d-LOCC transformation from |ψ⟩ to

ψ ′ is executable if and only if K ′BC ≤ KBC . Thus, in the case where the final state is EP-indefinite and not GHZ-type, the conditions of Theorem 2 is a necessary and sufficient condition of the possibility of d-LOCC transformation.

4.4. Case D

In this subsection, we prove Theorem 1 in Case D, where the tangle of the initial state is zero.

22 H. Tajima / Annals of Physics 329 (2013) 1–27

First, we simplify (12). Let us show that we can leave τABC , J5 and Qe out of the discussion hereafter. First, τ ′ABC = 0 follows from τABC = 0, because an arbitrary measurement makes the tangle τABC only multiplied by a constant. Next, because of (2)–(4) and the equation τABC = 0, the equation J5 = CAB CACCBC holds. (Proof : Because of (3) and τABC = 0, at least one of λ0 and λ4 is zero. When λ0 is zero, the equations J5 = CABCACCBC = 0 hold. Thus λ4 is zero, and then J5 = CABCACCBC = 4λ20λ

2 2λ

2 3 follows

(2) and (4). ) Thus, in order to examine the change of J5, it suffices to examine the change of the concurrences CAB, CAC and CBC .

Next, because of τABC = 0 ⇔ λ0 = 0 ∨ λ4 = 0 and because if there is a zero in {λk|k = 0, . . . , 4} then sinϕ = 0, the equation Qe = 0 follows (6). In the same manner, the entanglement charge Q ′e is also zero because of τ ′ABC = 0. Thus, Qe = Q

′ e = 0 holds. Condition 2 of Theorem 2 satisfies this

equation. Let us show this. The state |ψ⟩ is ζ̃ -indefinite, because τABC = 0:

∆J = K 25 − Kap = J 2 5 − Jap = C

2 ABC

2 ACC

2 BC sin

2 ϕ5

= 4C2ABC 2 ACλ

2 1λ

2 4 sin

2 ϕ = 0, (100)

where we use sinϕ = 0. Note that sinϕ = 0 holds when there is a zero in {λi| i = 0, . . . , 4} and that τABC = λ0λ4 = 0. Because the state |ψ⟩ is ζ̃ -indefinite, Condition 2 is reduced to |Q ′e| = sgn[(1− ζ )(ζ −ζlower)]. Incidentally, because of τABC = 0, ζlower = 1 holds. Thus, ζ = 1 follows ζlower ≤ ζ ≤ 1, and thus condition 2 satisfies the equation Q ′e = 0. In order to prove the present theorem, it suffices to show that a necessary and sufficient condition is that there are real numbers αA, αB and αC which are from zero to one and which satisfy the following equation:C ′2ABC ′2AC

C ′2BC

= α2Aα2B α2Aα2C

α2Bα 2 C

C2ABC2AC C2BC

. (101) Note that (101), J5 = CABCACCBC and J ′5 = C

′

ABC ′

ACC ′

BC give J ′

5 = α 2 Aα

2 Bα

2 C J5, and that τABC = 0 ⇒ (KAB =

C2AB) ∧ (KAC = C 2 AC ) ∧ (KBC = C

2 BC ).

We can classify the sets of the initial and final states as follows:

Case D-1: At least one of the concurrences CAB, CBC and CAC is zero. Case D-2: None of the concurrences CAB, CBC and CAC is zero, and at least one of the concurrences

C ′AB, C ′

BC and C ′

AC is zero. Case D-3: All of the concurrences CAB, CBC , CAC , C ′AB, C

′

BC and C ′

AC are nonzero.

Note that we already have τABC = τ ′ABC in the present Case D. In Case D-1, we first note that only the biseparable states are allowed as the initial states in this

case. The set of full-separable states and biseparable states which have the same kinds of bipartite entanglements is a totally ordered set [14].We cannot transform a full-separable state or a biseparable state into other type states with LOCC transformations[21], and thus we can derive the necessary and sufficient condition, which reduces to the following: there is an executable deterministic LOCC transformation from |ψ⟩ to

ψ ′ if and only if CAC ≥ C ′AC . This condition is equivalent to (101) in Case D-1.

In Case D-2, where none of CAB, CBC and CAC is zero and at least one of C ′AB, C ′

BC and C ′

AC is zero, the initial state is EP definite while the final state is EP indefinite. Thus, Case B includes Case D-2, and thus the existence of αA, αB and αC which satisfy (101) is the necessary and sufficient condition.

In Case D-3, where all of the concurrences CAB, CBC , CAC , C ′AB, C ′

BC and C ′

AC are nonzero, the initial and final states are W-type states. In the present case, we can use the result of Ref. [33]. A d-LOCC transformation from a W-type state |ψ⟩ to another W-type state

ψ ′ is possible if and only if xi ≥ x′i (i = 1, 2, 3), (102)

H. Tajima / Annals of Physics 329 (2013) 1–27 23

where the sets of positive real numbers {xi} and {x′i} are defined by the decompositions of |ψ⟩ andψ ′ [33]: |ψ⟩ = x0 |000⟩ + x1 |100⟩ + x2 |010⟩ + x3 |001⟩ , (103)ψ ′ = x′0 |000⟩ + x′1 |100⟩ + x′2 |010⟩ + x′3 |001⟩ . (104)

Note that we can reduce (103) into a generalized Schmidt decomposition

|ψ⟩ = x1 |000⟩ + x0 |100⟩ + x3 |101⟩ + x2 |110⟩ (105)

with transformation |0A⟩ ↔ |1A⟩. We thereby obtain

2x1x2 = CAB, 2x1x3 = CAC , 2x2x3 = CBC , (106)

x1 =

CABCAC 2CBC

, x2 =

CABCBC 2CAC

, x1 =

CACCBC 2CAB

. (107)

Thus, the existence of αA, αB and αC which are from zero to one which satisfy (101) is equivalent to the existence αA, αB and αC which are from zero to one which satisfy

αA = x′1/x1, αB = x ′

2/x2, αC = x ′

3/x3. (108)

Note that when αA, αB and αC are from 0 to 1, (108) is equivalent to (102). Thus, (101) is a necessary and sufficient condition of d-LOCC in Case D-3.

We thereby have completed the proof of Theorem 2 in all cases.

5. Conclusion

In the present paper, we have given four important results. First, we have introduced the entangle- ment charge Qe. This new entanglement parameter Qe has features which the electric charge has. The set of the six parameters (CAB, CAC , CBC , τABC , J5,Qe) is a complete set for the LU-equivalence; arbitrary three qubit pure states are LU-equivalent if and only if their entanglement parameters (CAB, CAC , CBC , τABC , J5,Qe) are equal to each other. This result means that the nonlocal features of three-qubit pure states can be expressed completely in terms of the magnitudes, phase and charge of the entanglement. The entanglement charge Qe satisfies a conservation law partially. Deterministic LOCC transformations between ζ̃ -definite states conserve the entanglement charge Qe. When we transform a ζ̃ -indefinite state into a ζ̃ -definite state, we can choose the value of the entanglement charge. Once the value is determined, we cannot change it anymore (Fig. 3). In this sense, we can regard ζ̃ -indefinite states as charge-definite states, and a deterministic LOCC transformation between charge-definite states preserves the entanglement charge.

Second, we have given a necessary and sufficient condition of the possibility of deterministic LOCC transformations of three-qubit pure states. The necessary and sufficient condition is given as a condition between the vectors (KAB, KAC , KBC , τABC , J5,Qe) of the initial and final states of deterministic LOCC transformation. In other words, we have revealed that three-qubit pure states are a partially ordered set parametrized by the six entanglement parameters. Note that other multipartite systems may have similar structures; it is plausible that the nonlocal features of N-qubit pure states can be expressed completely in terms of the magnitudes, phases and charges of the entanglement. Then the approach of the present paper may be applicable to such systems.

Third, we have clarified the rule how a deterministic measurement changes the entanglement. We can express this change as in Fig. 2. The rule indicates the transfer of the entanglement. After performing a deterministic measurement on the qubit A, the four entanglement parameters, C2AB, C

2 AC , τ

2 ABC and J5 are multiplied by α

2 A . The quantity βA(1 − α

2 A)τABC , which is a part of the

entanglement lost from τABC , is added to C2BC . The quantity (1 − βA)(1 − α 2 A)τABC , which is the rest

of the entanglement lost from τABC disappear. We call this phenomenon the entanglement transfer. Fourth, we have given the minimum times of measurements to reproduce an arbitrary executable

deterministic LOCC transformation. We can realize the minimum times by performing DMTs. We can

24 H. Tajima / Annals of Physics 329 (2013) 1–27

also determine the order of measurements; we can determine which qubit is measured first, second and third.

Is there entanglement transfer for a stochastic LOCC transformation? For this question, the present paper has given a partial answer. Let us see the inequalities given in Lemma 2:

CBC ≤ 2

i=1

p(i)C (i) BC ≤

C2BC + 1 − 2

k=1

piα(i) 2 τABC . (109)

The left inequality means that the bipartite entanglement CBC between the qubits B and C increases when the qubit A is measured. The right inequality is equivalent to the following inequality:

2 i=1

p(i)C (i) BC

2 +

2

i=1

p(i)

τ (i) ABC

2 ≤ C2BC + τABC , (110)

because

τ (i) ABC = α(i)

√ τABC . We can interpret the left-hand side of (110) as the sum of the bipartite

entanglement CBC between the qubits B and C and the tripartite entanglement τABC among the qubits A, B and C after a measurement. On the other hand, the right-hand side is the sum before a measurement. Thus, (110) means that a measurement decreases the sum. Note that the bipartite entanglement CBC of the qubits B and C increases, whereas the tripartite entanglement τABC among the qubits A, B and C decreases. To summarize the above, a kind of dissipative entanglement transfer also occurs for a two-choice measurement which are not a deterministic measurement. It is expected that the transfer occurs for an n-choicemeasurement too. Indeed, the left inequality of (109) also holds for an n-choicemeasurement. However, the right inequality of (109) for an n-choicemeasurement has not been proven yet.

In the present paper, we have exhaustively analyzed deterministic LOCC transformations of three- qubit pure states. This is the first step of the extension of Nielsen’s work [14] to multipartite entanglements.

Acknowledgments

The present work is supported by CREST of Japan Science and Technology Agency. The author is indebted to Dr. Julio de Vicente for his valuable comments. The author also thanks Prof. Naomichi Hatano and Prof. Satoshi Ishizaka for useful discussions. In particular, Prof. Naomichi Hatano patiently checked this very long thesis and give the author many pieces of important advice.

Appendix A. The proof that Qe is a tripartite parameter

In the present section, we show that Qe defined in (6) is a tripartite parameter; in other words, we show that Qe is invariant with respect to permutations of the qubits A, B and C . First, we perform the proof in the case of Qe = 0. Because of (22) and (6), the equation Qe = 0 holds if and only if ∆J = 0∨ sinϕ = 0. Because of (2) and (4), the expression sinϕ = 0 is equivalent to |J5| = CABCACCBC . Thus, Qe = 0 is equivalent to∆J = 0∨|J5| = CABCACCBC . Therefore, if we can show that the expression ∆J = 0 ∨ |J5| = CABCACCBC is invariant with respect to permutations of A, B and C , we can also show that Qe is invariant with respect to the permutations. The parameters J5 and CABCACCBC are invariant with respect to the permutations of A, B and C [19]. This fact and (11) give that ∆J is also invariant with respect to the permutations of A, B and C . Hence, the quantities∆J , J5 and CABCACCBC are invariant with respect to the permutations of A, B and C , and thus if Qe = 0, then Qe is invariant with respect to permutations of A, B and C . Namely, if Qe = 0, then Qe is a tripartite parameter.

Second, we perform the proof in the case of Qe = ±1. In order to show this, we only have to show that Qe is invariant with respect to the permutation of A and B, because if we can prove the invariance with respect to the permutation of A and B we can also prove the invariance with respect to the permutation of A and C or B and C in the same manner.

H. Tajima / Annals of Physics 329 (2013) 1–27 25

Let us derive the generalized Schmidt decomposition whose order of the qubits is BAC and see the expression of Qe in the new decomposition, which we refer to as Q Be . We can assume that (1) is a positive decomposition and let us permute A and B of (1):

|ψ⟩ = λ0 |0B0A0C ⟩ + λ1eiϕ |0B1A0C ⟩ + λ2 |0B1A1C ⟩ + λ3 |1B1A0C ⟩ + λ4 |1B1A1C ⟩ . (A.1)

We can put (A.1) in the form of the generalized Schmidt decomposition, after straightforward algebra;

|ψ⟩ =

λ23 + λ

2 4

0′B0′A0′C + −eiϕ̃5(λ1λ3eiϕ + λ2λ4) λ23 + λ

2 4

1′B0′A0′C

+ |λ2λ3 − λ1λ4eiϕ |

λ23 + λ 2 4

1′B0A1C + λ0λ3 λ23 + λ

2 4

1′A1′B0C + λ0λ4 λ23 + λ

2 4

1′B1′A1′C (A.2) where

0′A , 1′A , 0′B , 1′B , 0′C and 1′C are new basis of the quits A, B and C . Note that (A.2) is the generalized Schmidt decomposition whose order of the qubits is BAC . The coefficients of (A.2)

correspond to the coefficient of (1);

λ23 + λ 2 4 corresponds to λ0, −e

iϕ̃5(λ1λ3eiϕ + λ2λ4)/

λ23 + λ 2 4

corresponds to λ1eiϕ , and so on. Let us refer to Qe for (A.2) as Q Be . Because of the definition of Qe and (A.2),

Q Be = sgn Im

−eiϕ̃5(λ1λ3eiϕ + λ2λ4)

|(λ1λ3eiϕ + λ2λ4)|

λ23 + λ

2 4 −

K5 2 KAC

(A.3)

= sgn Im

−eiϕ̃5(λ1λ3eiϕ + λ2λ4)

|(λ1λ3eiϕ + λ2λ4)|

sgn

λ23 + λ

2 4 −

K5 2 KAC

(A.4)

holds. Then, we can complete the proof by showing that Qe = Q Be . Because (1) is a positive decomp- osition and because of (6), the following two equations hold:

sgn Im

−eiϕ̃5(λ1λ3eiϕ + λ2λ4)

|(λ1λ3eiϕ + λ2λ4)|

= sgn

Im[−jBCeiϕ̃5(λ1λ3eiϕ + λ2λ4)]

= sgn[λ1λ3 sinϕ(−λ2λ3 + λ1λ4 cosϕ)

− λ1λ4 sinϕ(λ1λ3 cosϕ + λ2λ4)] = sgn[−λ1λ2(λ23 + λ

2 4) sinϕ] = −1. (A.5)

sgn

λ23 + λ 2 4 −

K5 2 KAC

= sgn

KAB 4λ20

− K5

2 KAC

= sgn

KABKBC

2(K5 + Qe

∆J) −

K5 2 KAC

= sgn

Kap − K 25 − QeK5

∆J

2 KAC (K5 + Qe

∆J)

= sgn

−(

∆J + QeK5)

∆J

2 KAC (K5 + Qe

∆J)

= sgn

−Qe

∆J

2 KAC

= −Qe. (A.6)

Because of (A.4)–(A.6), we obtain Qe = Q Be .