You are on page 1of 27

2 Basic Concepts of Entangled States

F. Mintert1,3 , C. Viviescas2,3 , and A. Buchleitner3
Department of Physics, Harvard University, 17 Oxford Street, Cambridge MA,
Departamento de F´ısica, Universidad Nacional de Colombia, Carrera 30
No. 45-03 Edif. 404, Bogot´
a D. C., Colombia
Max-Planck-Institut f¨ur Physik komplexer Systeme, N¨othnitzer Str. 38, 01187
Dresden, Germany

2.1 Introduction

Quantum systems display properties that are unknown for classical ones,
such as the superposition of quantum states, interference, or tunneling. These
are all one-particle effects that can be observed in quantum systems, which
are composed of a single particle. But these are not the only distinctions
between classical and quantum objects – there are further differences that
manifest themselves in composite quantum systems, that is, systems that are
comprised of at least two subsystems. It is the correlations between these
subsystems that give rise to an additional distinction from classical systems,
whereas correlations in classical systems can always be described in terms
of classical probabilities; this is not always true in quantum systems. Such
non-classical correlations lead to apparent paradoxes like the famous Einstein
Podolsky Rosen scenario [1] that might suggest, on the first glance, that there
is remote action in quantum mechanics.
States that display such non-classical correlations are referred to as entan-
gled states, and it is the aim of this chapter to introduce the basic tools that
allow to understand the nature of such states, to distinguish them from those
that are classically correlated, and to quantify non-classical correlations.

2.2 Entangled States

Composite quantum systems are systems that naturally decompose into two
or more subsystems, where each subsystem itself is a proper quantum sys-
tem. Referring to a decomposition as “natural” implies that it is given in an
obvious fashion due to the physical situation. Most frequently, the individual
susbsystems are characterized by their mutual distance that is larger than
the size of a subsystem. A typical example is a string of ions, where each ion
is a subsystem, and the entire string is the composite system. Formally, the
Hilbert space H associated with a composite, or multipartite system, is given
by the tensor product H1 ⊗ · · · ⊗ HN of the spaces corresponding to each of
the subsystems.

Mintert, F. et al.: Basic Concepts of Entangled States. Lect. Notes Phys. 768, 61–86 (2009)
DOI 10.1007/978-3-540-88169-8 2 c Springer-Verlag Berlin Heidelberg 2009

62 F. Mintert et al.

In the following, we shall focus on finite-dimensional bipartite quantum
systems, i.e., systems composed of two distinct subsystems, described by the
Hilbert space H = H1 ⊗H2 . Many of the concepts and ideas that we introduce
can, nevertheless, be generalized to multipartite systems.

2.2.1 Pure States

We start out with a bipartite system with each subsystem prepared in a pure
state |ψi  (i = 1, 2). The state of the composite system |Ψs  is the direct
product thereof:
|Ψs  = |ψ1  ⊗ |ψ2  . (2.1)
Suppose that one could perform only local measurements on the system,
i.e., one had access to only one of the subsystems at a time. Then, after a
measurement of any local observable a ⊗ 1 on the first subsystem, where a is
a hermitian operator acting on H1 , and 1 is the identity acting on H2 , the
state of the first subsystem will be projected onto an eigenstate of a, but the
state of the second subsystem remains unchanged. If later on, one performs
a second local measurement, now on the second subsystem, it will yield a
result that is independent of the result of the first measurement. Hence, the
measurement outcomes on different subsystems are uncorrelated with each
other and depend only on the states of each respective subsystem.
A general pure state in H can be given by a superposition of pure states
of the form (2.1), for example,
|Ψe  = √ (|ψ1  ⊗ |ψ2  + |φ1  ⊗ |φ2 ) , (2.2)
where |ψi  = |φi  (i = 1, 2). We may now ask what the state |Ψe  looks like if
one has access to only one of the subsystems? For a local operator a ⊗ 1 on
the first subsystem, the expectation value observed in an experiment reads

a = Ψe |a ⊗ 1|Ψe 
= tr(a ⊗ 1 |Ψe Ψe |)
= tr1 (a tr2 |Ψe Ψe |)
= tr1 (a1 ) ,

where tr1,2 denotes the partial trace over the first/second subsystem, and 
1 = tr2 |Ψe Ψe | is the reduced density matrix of the first subsystem. Since
(2.3) holds for any local operator a, we need to conclude that the state of the
first subsystem alone is given by 1 . An analogous reasoning leads to the con-
clusion that also the state of the second subsystem is described by its reduced
density matrix 2 = tr1 |Ψe Ψe |. The state of the composite system, however,
is not equal to the product of both subsystem states, ρ = |Ψe Ψe | = ρ1 ⊗ ρ2 .
Moreover, if one performs a local measurement on one subsystem, this leads

such that the state of the system |Ψ  can be written as a product thereof:  |ψ1  ∈ H1 . will in general yield correlated measurement results. the state of a quantum system can be mixed. This reduced state is then typically mixed. As elaborated in more detail in Sect. do not exhibit correlations.  = ρ(1) ⊗ ρ(2) (2. More generally.3. 2 Basic Concepts of Entangled States 63 to a state reduction of the entire system state. Thus.2. i. as in (2. are called product or separable states. and non-negative weights pi . and the state of the system is given by the partial trace over the environment.e.6) thus are called separable mixed states. we can define that states that can be written as a product of pure states. These correlations can be described in terms of the classical probabilities pi . are defined by the non-existence of a decomposition into product states [2]: (1) (2) A mixed state  is entangled if there are no local states ρi .1. however. since hardly any quantum system can be isolated completely from its surroundings. however. such that  can be expressed as a convex mixture thereof: . (2. the probabilities for an outcome of a measurement on one subsystem are influenced by prior mea- surements on the other subsystem. 5. measurement results on – possibly distant and non-interacting – subsystems are correlated.1). Mixed states are in fact the most frequently encountered states in real experiments. in turn. Based on these considerations. it is in general not possible to keep track of the many environmental degrees of freedom. Therefore. |ψ2  ∈ H2 such that |Ψ  = |ψ1  ⊗ |ψ2  . (2. not only of the subsystem on which the measurement had been performed. ρi . States of the form (2.. If on the contrary. Similarly to the case of pure states. Mixed entangled states.5) with ρ(1) and ρ(2) for the respective subsystems. and are therefore considered classical. 2. there are local observables a and b such that tr((a ⊗ b)) = tr((a ⊗ 1)) tr((1 ⊗ b)) = tr1 1 a tr2 2 b. there are no local states |ψ1  ∈ H1 and |ψ2  ∈ H2 .4) then |Ψ  is an entangled state.6) i  with pi > 0 and i pi = 1.2 Mixed States So far we considered only pure states. mixed product states.  (1) (2) = pi ρi ⊗ ρi . A convex sum of different product states.

cannot be described in terms of only classical probabilities. As we shall see. but the appropriate decomposition could not be identified. though could be rewritten also as |00 + 2|01 + |10 + 2|11 |Ψ  = √ . 2. . (2. But. But the failure to find one can have two different reasons: either the state is entangled and there is no decomposition into product states. It just turns out that separability √ is more √ easily identified if |Ψ  is√ expressed in the√bases {(|0+|1)/ 2. there is a need for potentially simple criteria to distinguish separable from entangled states that do not require an explicit search. (2. Once a decomposition is found one knows that a state is separable.64 F. but for mixed states similar tools are available only for low- dimensional system. or the state is actually separable. i .3. before we discuss mixed states.8) 2 5 One can see that |Ψ  factorizes into local states – it is separable. in order to show that a given state is separable. For this reason. in the sense that there is always a basis that allows to reveal the entanglement properties.9) 10 where separability is less evident. or into a convex sum of tensor products for mixed states. as we will see later on. Separability is defined via the existence of a decomposition of a state into product states in the case of pure states.7) i Entangled states imply quantum correlations of measurements on different subsystems which. (2. there are criteria that discriminate separable and entangled states unambiguously. in contrast to classical correlations (see above). But checking separability of a given state can turn out to be much more involved than one might expect. we will start out with the comparatively simpler case of pure states. That is.3 Separability Criteria The above definitions of separable and entangled states appear simple on a first sight. pi ≥ 0 such that = pi ρi ⊗ ρi . (1) (2)  (1) (2)  i . these tools can provide only partial information. 2. |1}.1 Pure States Let us consider the exemplary case |0 + |1 |0 + 2|1 |Ψ  = √ ⊗ √ . (2|0−|1)/ 5} of H2 than in the basis {|0. The representation of a state in this basis is called the Schmidt decomposition [3]. (|0−|1)/ 2} of H1 and {(|0+2|1)/ 5. For higher dimensional systems. the observation is generic. one has to look for such decompositions. For pure states. Mintert et al.

the state is given by  |Ψ  = [udv]ij |ϕ˜i  ⊗ |φ˜j  . respectively.12) ij In order to obtain the Schmidt decomposition of |Ψ . since the Schmidt basis {|ϕSi  ⊗ |φSj } is given by separable states. Therefore.1). Furthermore. any pure state |Ψ  in H = H1 ⊗ H2 can be expressed in terms of the corre- sponding product basis  |Ψ  = dij |ϕi  ⊗ |φj  . where inthe third line we used the resolution of the identity on each sub- system.10) ij The expansion coefficients dij are given by the overlap of the state with the basis vectors. we use the fact that for every complex matrix d. called singular values. we can conclude that a pure state |Ψ  is separable if and only if it has only one non-vanishing Schmidt coefficient. vqj = φj |V † |φq . with U and V arbitrary. for any state |Ψ  the Schmidt coefficients are unique. it is not possible to express |Ψ  in the form (2.12) reduces to  |Ψ  = λi |ϕSi  ⊗ |φSi  . the dij change accordingly: d˜ij = ϕ˜i | ⊗ φ˜j |Ψ  = ϕi | U † ⊗ φj |V † |Ψ   = ϕi | U † |ϕp φj |V † |φq ϕp | ⊗ φq |Ψ  (2. when at least two Schmidt co- efficients are different from zero. i |ϕi ϕi | = 1 and i |φi φi | = 1. dij = ϕi | ⊗ φj |Ψ .11) pq = [udv]ij . In the new basis. Hence. and the sum is limited by the dimension of the smaller subsystem. Otherwise. Like eigenvalues of a matrix. (2. local unitary transformations on H1 and H2 . then |Ψ  is separable. there always exist unitary transformations u and v such that udv is diagonal. This provides the singular value decomposition of d [4]. and we defined the unitary matrices uip = ϕi | U † |ϕp . (2. for each state |Ψ . all information on the entanglement of a state is en- coded in the Schmidt coefficients: If there is only one non-vanishing Schmidt coefficient. If one now makes a change of bases |ϕ˜i  = U|ϕi  and |φ˜i  = V|φi . with real.13) i where the λi = Si2are known as Schmidt coefficients. Consequently. (2. also the singular values are uniquely defined. one can always find local bases |ϕSi  and |φSi  in terms of which (2. . 2 Basic Concepts of Entangled States 65 Schmidt Decomposition Given two arbitrary local bases {|ϕi } and {|φi } in the spaces H1 and H2 . non-negative diagonal entries Si .

that is 1 and 2 have the same non-vanishing eigenvalues.3. Mintert et al. The reduced density matrices are particularly helpful in this context.2 Mixed States For pure states. the Schmidt decomposition provides a necessary and suffi- cient criterion for separability. both concepts are closely re- lated. . the degree of mixing of its reduced density matrices is not an indicator of entanglement. In particular. The one of the first subsystem reads 1 = tr2 |Ψ Ψ |  = tr2 λi λj |ϕSi ϕSj | ⊗ |φSi φSj | ij (2. We see that the Schmidt coefficients are given by the eigenvalues of the reduced density matrix 1 . we need to find some new criteria to distinguish entangled from separable mixed states. and the basis vectors of the Schmidt basis are given by the eigenstates of 1 and 2 .66 F. we can restate the separability criterion for pure states: tr2r = 1 ⇒ r is pure ⇒ |Ψ  is separable (2. we should focus on how to evaluate them. we also have related the entan- glement of a pure state |Ψ  to the degree of mixing of the reduced density matrices. An equivalent reasoning holds for the reduced density matrix of the second subsystem 2 = tr1 |Ψ Ψ |. That is. i in terms of the Schmidt decomposition (2. We not only found a simple prescription to evaluate the Schmidt co- efficients of any state |Ψ . The most prominent of such tools are entanglement witnesses and positive maps. As we shall see. for mixed states such an elegant decomposition does not exist. if a state is mixed. where we used the orthonor- mality of the Schmidt basis while performing the trace over the second sub- system. Unfortunately. Reduced Density Matrix Since the Schmidt coefficients are so useful for the distinction of separable and entangled states.14)  = λi |ϕSi ϕSi | . 2. Therefore. but since separability requires that exactly one Schmidt coefficient is different from zero.15) tr2r < 1 ⇒ r is mixed ⇒ |Ψ  is entangled with r referring to either one of the two subsystems.13).

(2. (2.16) for all separable pure states |Ψs . The central benefit of witnesses is that there exists a witness for any entangled state that detects it [5]. . B. . but that yields positive expectation values Ψs |W |Ψs  ≥ 0 . Geometrically. (2. intuitive arguments that allow to understand why entanglement witnesses work. 2 Basic Concepts of Entangled States 67 Entanglement Witnesses An entanglement witness W [5. and C on the one hand. . whereas D is not convex. if a given density matrix  leads to a negative expectation value tr(W ) < 0 . and one says that W detects . one can find a straight line that separates this point from the gray-shaded area. 8) are only exemplary ones. but rather give some geometric. where the shapes A. and i pi = 1.18) implies that the expectation value of an entanglement witness with respect to any separable mixed state is also non-negative. B. For a geometric interpretation of this vector space.18) then  is entangled. this is not always possible. any convex sum λs +(1−λ)s (1 ≥ λ ≥ 0) is again separable. That means that.19) Now. Here we do not go into the details of the formal proof.1. (2. and in the present context. There is one crucial difference between cases A. separable states form a convex set. Since any separable mixed state can be expressed as a convex sum of projectors onto pure separable states. given two arbitrary (1) (2) (1) (2) separable states s and s .  tr(s W ) = pi Ψs |W |Ψs  ≥ 0 . as illustrated in Fig. since it has a trough on its right bottom part. one may find many more. and C represent different convex sets. one can find several lines that separate the grey shaded areas from their white surrounding – the depicted lines Wi (i = 1. There is no straight line that separates D from point Z. (2. 6] is a hermitian operator acting on H that is not positive definite. . For D. and C. this is defined as A|B = tr A† B . Geometry of Quantum States Let’s try to understand quantum states in a geometrical setting.17) i Thus. and D on the other hand: For any point outside the convex sets A. 2. B. Density matrices can be conceived as vectors in a vector space that is referred to as Hilbert–Schmidt space [7]. Now.  (i) (i)  s = i pi |Ψs Ψs | with pi > 0. this means that the set of separable states has no trough. . one needs a scalar product.

(2. B. i = 1. The situation of entanglement witnesses is analogous: the set of separable states is convex. Given some specific entangled state . 6) that separates this point from the corresponding convex shape.1. and the failure to find a suitable witness for a state  does not necessarily allow to conclude that  is separable. Therefore. For the non-convex shape D. a witness provides a necessary separability criterion: if a state is separable. it will yield a non-negative expectation value for any witness. . there exists a line (like Wi . just like there is a line to any point outside A.1 still allows to understand the basic mechanism of entanglement witnesses. The condition that trσW vanishes. 2. with tr(Oi W ) = 0 . Geometric Interpretation of Entanglement Witnesses A separable state is characterized by the condition tr W ≥ 0. . Since the separable states form a convex set. whereas shape D is not. or C that separates it from the respective grey shaded areas. the condition trσW = 0 defines a hyperplane in the space of opera- tors – analogous to the lines Wi in Fig 2. there is no such line. and all separable states are sit- uated on one side of this hyperplane (tr W ≥ 0). Due to the complicated structure of the set of separable states that has curved borders. 2. . C) are convex. the set of separable states is high dimensional and more com- plicated than the shapes in Fig. To any point outside the convex shapes. three of which (A. requires σ to be a linear combination of operators Oi that are orthogonal to W :  σ= αi Oi . there exists a witness (the analogue of a line Wi ) to any entangled state (the analogue of a point outside the grey shapes) that separates it from the set of separable states (the analogue of one of the convex shapes) Although.1. one needs infinitely many witnesses to characterize it com- pletely. the basic geometric picture of Fig.68 F. but separability of a state cannot deduced from such a non-negative expectation value.20) i That is. . W3 W5 W7 W4 D W1 A B C W8 Z W2 W6 Fig. it can be rather complicated to find a witness that detects it. B. there is a witness to any entangled state that detects it.1. Four different shapes. Mintert et al. 2. . The sign of tr W then indicates on which side of the hyperplane  is situated. the situation is different: for the point Z.

21) A very counterintuitive property of these positive maps is that the extended map ΛE is not necessarily positive. Thus. Only in systems of small dimension the concept of positive maps allows to formulate a constructive criterion that is both necessary and sufficient: for a system of two qubits or a system of one qubit and one qutrit (three-level system). and that is the transposition T () = T . since the characterization of positive maps is an open problem.e. then the state  is entangled. i. there are states  on H = H1 ⊗ H2 . (2.22) i (1) (2) Since Λ is positive. One can extend this map to the product space H = H1 ⊗ H2 . such that the extended map ΛE acts on B(H1 ) like Λ.. That is. any expectation value of ΛE (s ) is positive. and since also pi and ρi are positive. i. let us take a separable state s . let us consider the case of a bipartite system. and apply a positive linear map to it. or just due to the lack of success to find a suitable map. Now. The inverse statement is more involved. Now. for any separable state. such that ΛE (s ) is not a positive operator.  (1) (2) ΛE (s ) = pi Λ(ρi ) ⊗ ρi . The underlying reason for this is that any positive map . the reflection of a matrix  along the diagonal [5. That is. Such a map Λ is considered positive if ˜ = Λ() is a positive operator. for some maps Λ. and therefore ΛE (s ) remains a positive operator.e. one can check separability by considering only a single positive map. such that (1 ⊗ Λ)() is not a positive operator. there is no positive map Λ. (2. for any positive operator . such maps only provide a necessary separability criterion like above in the case of witnesses: if one has found a map Λ.. and ΛE acts trivially on B(H2 ). 2 Basic Concepts of Entangled States 69 Positive Maps An alternative tool to check on separability is the so-called positive linear maps Λ that map the set of operators B(H) acting on a Hilbert space H on ˜ where H the set B(H). then it is necessary to consider maps Λ that map B(H2 ) on B(H1 ) – that is (1 ⊗ Λ)() is an operator acting on H1 ⊗ H1 . Now. ˜ can – though not necessarily needs to – be a different Hilbert space than H. 8].. if one can find a positive map Λ such that ΛE () has at least one negative eigenvalue for a given state . then one knows for sure that  is entangled. one that has a convex decom- position into product states. ΛE = Λ ⊗ 1 . i. a state is separable if and only if (1 ⊗ Λ)() is positive for all positive linear maps of B(H2 ) on B(H1 ). Λ(ρi ) is a positive operator. then one does not necessarily know whether this is due to separability of . such that ΛE () is not a positive operator. But.e. If one wants to prove separability of a state  on H1 ⊗ H2 . But if one fails to find such a map.

we presented entanglement witnesses and positive maps as indepen- dent concepts. Witnesses and Positive Maps So far. The partial transpose pt . 10]. the spectrum of pt allows to unambiguously distinguish separable from entangled states in 2 × 2-dimensional and 2 × 3-dimensional systems..70 F. the extended maps 1⊗ΛCP are positive maps. (2) then also (1 ⊗ ΛCP )(pt ) is non-negative.e. (2. since 1 ⊗ T is not a positive map. Since the ΛCP are (i) completely positive. the so-called ppt- criterion is a frequently used separability criterion: despite being only a nec- essary separability criterion it still detects many entangled states. Therefore. however. we already know that  is entangled if its partial transpose has at least one negative eigenvalue. maps that take a qubit-operator to a qubit. In that case. But also in high-dimensional systems. One may check that pt actually depends on the basis with the help of which it is constructed. and its partial transpose is  obtained by a simple rearrangement of matrix elements. these two con- cepts are more closely related than they seem to be on the first glance. and the spectrum does not depend on this choice of basis. it has no negative eigenvalue.23) does not characterize all positive maps anymore. .24) = (1 ⊗ ΛCP )() + (1 ⊗ ΛCP )(pt ) ≥ 0 . is not necessarily a positive operator. pt = (1 ⊗ T )() = ij. (i) where pt = (1 ⊗ T )() is called the partial transpose of .lk |ϕi ϕj | ⊗ |φk φl |. (1) (1 ⊗ ΛCP )() is non-negative. however. In higher dimensional systems. and there are entangled states with positive par- tial transpose (ppt). 2) are completely positive maps. However. Entanglement witnesses could be understood in a geometric setting.. if  is such that its partial transpose is non-negative. and T is the transposition [9. i.kl |ϕi ϕj | ⊗ |φk φl |. the condition that (1 ⊗ Λ)() be positive for any positive map Λ reduces to (1) (2) (1 ⊗ Λ)() = (1 ⊗ ΛCP )() + (1 ⊗ ΛCP )(1 ⊗ T )() (1) (2) (2. Therefore. and this implies that  is separable. it is only the spectrum of pt that enters the present separability criterion. from B(C2 ) on B(C2 ).kl ij. i.e. or on B(C3 ). However. On the other hand. and pos- itive maps have rather counterintuitive properties. Mintert et al. they do not seem to have too much in common.kl ij.23) where ΛiCP (i = 1. Therefore. and it is rather straightforward to implement: a general state  of a bipartite system can be expanded in some arbitrary product basis  = ij. But. we can conclude that (1 ⊗ Λ)() is non-negative for arbitrary positive maps Λ.or to a qutrit-operator can be written as Λ = Λ1CP + Λ2CP ◦ T . And indeed. (2.

For such purposes. But the prior definition of entangle- ment in terms of the nonexistence of a decomposition of a state into product states (cf. indeed.(2.4.25) We can now show that the observable W = (1 ⊗ Λ† )(|χχ|) is an entangle- ment witness. before we can introduce entanglement measures. This allows to arrive at a new concept that allows for a quantitative description of entangled states.7)) will not be helpful for finding such a quantification. Then (1 ⊗ Λ)() has an eigenvector |χ with a negative eigenvalue λ. For an arbitrary separable state |Φs . (2. 2. The idea is to classify all operations that one could apply to a composite quantum system. And. 2 Basic Concepts of Entangled States 71 Let us consider a positive map Λ such that the extended map 1⊗Λ applied to some state  yields a non-positive operator. i. such that (1 ⊗ Λ)(|Φs Φs |) is a positive operator. we need to refine our concept of entanglement.27) = χ|(1 ⊗ Λ)()|χ = λ < 0 because of the above eigenvector relation. (1 ⊗ Λ)()|χ = λ|χ . and it will still be in agreement with the previous definitions of entanglement and separability.. Λ is not a completely pos- itive map. where the inequality is due to the positivity of Λ. that is . we have   Φs |W |Φs  = tr (1 ⊗ Λ† )(|χχ|) |Φs Φs |   (2.1 General Considerations Let us forget for a while about the prior formal definition and focus more on the interpretation that entanglement is tantamount to correlations that cannot be described in terms of classical probabilities. does not allow to compare the amount of entanglement of two different states. however. one would need a quantitative description of entanglement. This. Therefore. (2.26) = tr |χχ| ((1 ⊗ Λ)(|Φs Φs |)) ≥ 0 .4 Entanglement Monotones and Measures So far we contented ourselves with a qualitative distinction between separable and entangled states.4). 2.e. and that can increase only classical correlations. this witness detects  to be entangled:    tr(W ) = tr  (1 ⊗ Λ† )(|χχ|)   = tr (1 ⊗ Λ)() |χχ| (2.

one might wonder if one could come up with a slightly less ‘invasive’ measurement with less dramatic effects. On average. The basic ones that are allowed by the laws of quantum mechanics comprise unitary evolutions  → U U † . (2. one can do so. and finally performs the mea- surement on the ancilla only.3). Quantum Operations To do so.31) (i) and a subsequent measurement in the basis {|Ψa } of ancilla states projects this state on Ψa(i) |U |Ψa Ψa |U † |Ψa(i)  = Ai A†i . those that are captured by probabilities pi as in (2.32) (i) with the operators Ai = Ψa |U |Ψa  that act only on the original system. Neumann measurement takes a state to a purely probabilistic mixture of the states |ϕi . Let us denote such a complete set of eigenstates {|ϕi }. on average the state evolves as    → pi |ϕi ϕi | = |ϕi ϕi ||ϕi ϕi | . the corresponding measurement re- sults in the collapse of  on the state |ϕi ϕi |. Thus. (2. (2.30) where |Ψa  is an ancilla state. 5.72 F. before we can come to the promised quan- tification of entanglement. Neumann measurements in which a quantum state  is projected onto an eigenstate of the associated observable (see Sect. and it destroys all coherences between them com- pletely. one can make the decrease of correlations under all such operations a defining property of entanglement.6). we first have to make a significant detour to end up with what is referred to as local operations and classical communication. with UU † = U † U = 1 . An interaction between the original system and the ancilla results in a global unitary evolution  U  ⊗ |Ψa Ψa | U † . a v. with probability pi = ϕi ||ϕi . let us start out with the most general operations. Once this is done.1. Then. (2. Mintert et al.29) i i Thus.28) and v.33) i . And. That is. Though. (2. the state evolves as   → Ai A†i . if one uses an additional quantum system – often referred to as ancilla – lets this ancilla interact with the original system. (2. indeed. The original state ρ of the combined systems including the ancilla reads ρ =  ⊗ |Ψa Ψa | .

we were discussing positive maps and saw that a trivial extension of a map is not necessarily a positive map again.3. this is the case exactly if the extension of the map is positive. Since any trace preserving. However.35) i i and. it cannot be written as U = U1 ⊗ U2 . of probability.1) [11–13]. 5.38) 2 2 followed by a local unitary transformation that is conditioned on the mea- surement outcome. First. (2.3. and. completely positive map can always be expressed in the form of (2.33) is indeed the most general evolution a quantum state can undergo. if the map is completely positive. i.34) i i This property is crucial. Applying the map to√ |00 takes this separable state to the entangled state U|00 = (|00 + |11)/ 2. 2 Basic Concepts of Entangled States 73  (i) (i) If one utilizes the completeness of the ancilla states i |Ψa Ψa | = 1. therefore. In Sect. since it guarantees the conservation of the trace   † tr Ai A†i = tr Ai Ai  = tr .√ Re- peated measurements yield the two different outcomes (|00 + |11)/ 2. such a global operation can indeed create entanglement. (2. obviously the positivity of the state of the entire system has to be ensured. Some Examples Let us look at a few exemplary cases of operations of the form (2. |ϕ4  = √ .36) 2 2 This is an example of a global operation.33). and its implementation requires an interaction between the two individual subsystems.37) 2 2 |01 + |10 |01 − |10 |ϕ3  = √ .2. and subsequently U † U = 1. (2. that is. (2. Thus. |ϕ2  = √ .. A second example is given by a measurement in the Bell-basis |00 + |11 |00 − |11 |ϕ1  = √ .33) is trace preserving and completely positive (see Sect. 2. consider the specific unitary map 1  1  U = √ |00 + |11 00| + √ |00 − |11 11| + |0101| + |1010| . since any map of the form (2. (2. Let us start again with the separable state |00. and . for any map that describes the evolution of a real quantum system. any such extension needs to be positive: if a map acts only on a subcomponent of a system.e. (2. one can convince oneself that the operators Ai satisfy the resolution of the identity  †  Ai Ai = Ψa |U † |Ψa(i) Ψa(i) |U |Ψa  = 1 .33) to see how they can affect entanglement properties.

33).. as discussed before (2. with equal probability. (2. This provides a second example of a global operation that can create entanglement. .. yields the final state (|00+|11)/ 2. Conditioned on the measurement result that is associated with the collapse on the states (ai ⊗ 1)(a†i ⊗ 1). This can be understood as an interaction with an ancillary system and a subsequent measurement thereon.    (1) † =ρ ⊗ρ (1) (2) → ai ρ ai ⊗ ρ(2) . the local operation associated with the operators bij is . a local operation has been applied to the first subsystem. .74 F. however.41) i i j The situation changes if one allows for a correlated application of such local operations. . (1) (2) (1) (2) = pi ρi ⊗ ρi → pi (2. (2.. Local Operations and Classical Communication The most general local operation that acts non-trivially only on the first subsystem reads   † → (ai ⊗ 1)(a†i ⊗ 1) .42c) ijp  → (1 ⊗ gijp.. (1 ⊗ gijp. where the operation that is applied at a certain instance depends on the outcomes of previous operations:   → (ai ⊗ 1)(a†i ⊗ 1) (2. (2. and of u = |00|−|11| on√ the second subsystem in case of the second. We will see later. Such opera- tions do not induce any correlations: They map product states on product states. that the situation is different if we restrict our- selves to local operations.42d) ijp.q † ).42a) i  → (1 ⊗ bij )(ai ⊗ 1)(a†i ⊗ 1)(1 ⊗ b†ij ) (2. which. Mintert et al. √ (|00 − |11)/ 2.q In the first step. A conditioned local unitary opera- tion that is comprised of the identity operation in case of the first outcome. is entangled. ai ai = 1 .40) i and separable states on separable states     aj ρi a†j ⊗ ρi ..q ) .39) i i and analogously for operations on the second subsystem alone.42b) ij  → (cijp ⊗ 1)(1 ⊗ bij )(ai ⊗ 1)(a†i ⊗ 1)(1 ⊗ b†ij )(c†ijp ⊗ 1) (2.. or. . once again. (ai ⊗ 1)(a†i ⊗ 1) . to local operations and classical communication that we introduce now.

however. it provides only a necessary. and those parties could apply their individual operations to their part of the composite system. necessary criterion thereof: among all LOCC operations. but not a sufficient condition. Thus. therefore is referred to as ‘classical’. conditioned on the outcome of this operation. The idea behind that terminology is that one could imagine two parties that have access to the individual subsystems. And. so is their entanglement. and one necessarily concludes that any entanglement monotone is invariant under local unitaries M() = M(U1 ⊗ U2  U1† ⊗ U2† ) . and so on.e. (2. Invariance of Entanglement Under Local Unitaries Monotonicity under LOCC as the defining property of an entanglement mono- tone is in general difficult to verify. (2. . it is possible to create correlations with LOCC op- erations. formulate a simpler. since these correlations are based on the classical exchange of information. can be performed via a classical channel. they remain correlations of classical nature. But in order to arrive at the above operation. We can. but LOCC suffice to transform arbitrary separable states into each other. then a monotone M cannot increase after either step M() ≥ M(U1 ⊗ U2  U1† ⊗ U2† ) ≥ M() . Note that this requirement is perfectly compatible with the previous def- inition of separable and entangled states. another local operation is applied to the first subsystem. 15] that an entanglement monotone is a quantity that does not increase under local operations and classical communication. and. and its inverse in a second step.44) This invariance is significantly easier to check than monotonicity under LOCC. 2 Basic Concepts of Entangled States 75 applied to the second subsystem in a consecutive step. they would need to communicate with each other.43) However. does not require any quantum nature. the local unitary transformations  → U1 ⊗ U2  U1† ⊗ U2† are special since they have an inverse that is again LOCC. Yet. If one applies some arbitrary local unitary in a first step. since an entangled state cannot be created from a separable one by LOCC alone. However. as mentioned above. because initial and final states are equal. however. Therefore. Such operations are called local operations and classical communication (LOCC). i. we can refine our concept of entangled states by requiring [14. tell the other party their measurement results. LOCC operations can take product states to states no more necessarily of product form.. This communication.

2. Mintert et al. An Example In order to get a bit better idea of how such an LOCC transformation works. and all entanglement properties can be expressed in terms of only those quantities.45) i=1 i=1 . .. Although the exhaustive search for such invariants turns out to be a very intricate task for a general state. There. also ‘λ(Ψ ) is majorized by λ(Φ) ’.e. . satisfy the set of inequalities (Φ) (Ψ ) λ1 ≥ λ1  2 (Φ)  2 (Ψ ) λi ≥ λi i=1 i=1  3 (Φ)  3 (Ψ ) λi ≥ λi (2.] and similarly for λ(Ψ ) . Therefore. the Schmidt coefficients introduced earlier in Sect. This is possible since λ(Φ) actually majorizes λ(Ψ ) . and reads ‘λ(Φ) majorizes λ(Ψ ) ’. . this majorization criterion does not give a prescription on how such a transformation can be achieved. This set of conditions is often expressed in short-hand notation λ(Φ)  λ(Ψ ) (Φ) (Φ) in terms of the Schmidt vectors λ(Φ) = [λ1 . λ1 ≥ λ2 ≥ . This is possible [16.. Schmidt Coefficients and Majorization Invariance under local unitary transformations is not only a simple test to rule out potential candidates for entanglement monotones as non-monotonous under LOCC. .76 F. it has a surprisingly simple answer in the case of pure states of bipartite systems.. one proceeds a big step forward. λ2 . 17] if and only if their Schmidt coefficients. . toward the systematic construction of entangle- ment monotones.2 provide a complete set of invari- ants. ordered decreasingly (i. Majorization One very useful application of the characterization of entanglement in terms of Schmidt coefficients is a simple test that allows to check whether one state |Φ can be prepared by LOCC starting from another state |Ψ . Consequently. . However. . but indeed it has much deeper implications. we will content ourselves with verifying that the LOCC operation that is comprised of the operators . if one can identify these invariants.). . and aim at the preparation of the state |Φ = λ1 |00 + λ2 |11 using only LOCC operations. let us √ √ |Ψ  =√(|00 + look at the exemplary case to start out with the state |11)/ 2. It implies that any entanglement monotone can be expressed as a function only of invariants un- der local unitaries. or.3.

46) a2 = λ1 |01| + λ2 |10| . consider the action of these operators onto the state |Ψ . b11 = |00| + |11| . however. b21 = |01| + |10| . Then. indeed transforms |Ψ  to |Φ. 2 Basic Concepts of Entangled States 77 √ √ a1 = √λ1 |00| + √λ2 |11| . First. First the ai : . one should verify that the res-  olutions to identity i a†i ai = 1 and b†11 b11 = b†21 b21 = 1 are given. (2.


(2. λ1 λ2 1 a1 |Ψ  = |00 + |11 = √ |Φ.47) 2 2 2 .


this is exactly what we were aiming at. The situation is different in higher dimensional systems as one can see in the exemplary case of the following two states 1 1 |Ψ1  = √ |00 + √ |11 ..45) reduces to its first line.48) 2 2 The first term is already proportional to |Φ. or one majorizes the other. The entanglement of a pure state of two qubits is characterized by a single independent Schmidt coefficient due to the normalization of the reduced den- sity matrix. i. Let us however briefly illustrate why the characterization of entanglement by a simple scalar quantity might reveal problematic.e. Thus.49) ij And. we found a criterion that excludes some quantities from the list of potential quantifiers of entanglement. But. so that in the next step the identity operation b11 is applied. (2. there is an unambiguous order of pure states with respect to their degree of entangle- ment. and any entanglement monotone will respect this order. Inequivalent Entanglement Properties So far. one obtains b11 a1 |Ψ  = b21 a2 |Ψ  = 1/ 2|Φ. λ1 λ2 a2 |Ψ  = |01 + |10 . but does not yet define one unique entanglement measure. Here. the set of majorization conditions (2. the second term does not have the correct form yet. Therefore. one needs to transform |0 of the second subsystem into |1 and vice versa. λ1 1 = λ1 2 . what √ is exactly what b21 does. That is. Whether such a unique measure exists is still a subject of debate. and beyond the scope of the present introduction. the final state reads  ai ⊗ bij |Ψ Ψ |a†i ⊗ b†ij = |ΦΦ| . 2 2 . (2. For two arbitrary pure states |Ψ1  and |Ψ2  either both Schmidt vec- (Ψ ) (Ψ ) tors coincide. So all together.



3 1 1 |Ψ2  = |00 + |11 + |22 . (2.50) 5 5 5 .

respectively. we list some important ones: . . This implies convexity of an entanglement measure. . This implies that the two states have non- equivalent entanglement properties. Assume one is given n copies of a state  on H1 ⊗ H2 .4. Similarly. and one wants to quantify the entanglement between the subsystems associated with the larger Hilbert spaces H1⊗n and H2⊗n . the use of different entanglement monotones may lead to contradictory conclusions on the relative entanglement content of both states. Entanglement Measures So far. 2. (0 ≤ p ≤ 1). and that.e. A monotone M that satisfies the inequality M( ⊗ σ) ≤ M() + M(σ) is called subadditive. nor vice versa (Ψ ) (Ψ ) λ1 1 = 12 < 35 = λ1 2 2 (Ψ1 ) 2 (Ψ ) i=1 λi = 1 > 45 = i=1 λi 3 (Ψ1 )  3 (Ψ ) i=1 λi = 1 = 1 = i=1 λi . Mintert et al. we required only monotonicity under LOCC for a potential entangle- ment quantifier. we discussed very general properties of entanglement quanti- fiers. ⊗ . .78 F. one expects that a probabilistic mixture p + (1 − p)σ. and neither does λ(Ψ1 ) majorize λ(Ψ2 ) . one can consider two different states  and σ on H1 ⊗ H2 and evaluate the entanglement of the joint state  ⊗ σ on H1⊗2 ⊗ H2⊗2 . 1/5. should be no more entangled than the two individual states on average.. The Schmidt vectors are λ(Ψ1 ) = [1/2. Pure States We saw earlier that any entanglement monotone or measure can be expressed in terms of invariants under local unitary transformations. This is equivalent to a single n-fold state ⊗n =  ⊗ . 1/2. nor is there an LOCC operation that takes |Ψ2  to |Ψ1 . i. An entanglement monotone that fulfills M(⊗n ) = nM() is called additive. 0] and λ(Ψ2 ) = [3/5. and it is not obvious that either one can be considered more entangled than the other. In particular.2 Some Specific Monotones and Measures In the above. Now we will discuss some more specific entanglement monotones and measures that are frequently used in the literature. Thus neither can |Ψ1  be prepared by LOCC from |Ψ2 . 1/5]. Mixing two states  and σ probabilistically can increase only classical correlations. . While there is no general agreement on the complete list of axioms. Therefore. M(p + (1 − p)σ) ≤ pM() + (1 − p)M(σ). There are additional axioms that qualify a monotone as an entanglement measure. in the .

Therefore. (2. Entanglement Entropy The entanglement entropy. Equivalently. (2. i. if Ψ | reads Ψ | = ij βij ij|. and H2 of H [19]. But not every such function is also an entanglement monotone.52) i is indeed invariant under permutation of the λi .55) Ψ ∗ | denotes the complex conjugate of Ψ |.e.. withthe conjugation performed in the same  basis. (2.  E(Ψ ) = S(r ) = −trr ln r = − λi ln λi . whereas Ψ | is ∗ ∗ the adjoint of |Ψ .53) ∂λ1 ∂λ2 λ1 and thus is a valid entanglement monotone. 2 Basic Concepts of Entangled States 79 case of bipartite pure states. For bipartite systems. then Ψ ∗ | reads Ψ | = ij βij ij|. Ψ ∗ | is the transpose of |Ψ . i. A possible generalization of the above definition for higher dimensional systems (see e.. and if F is Schur concave.g.e. |1} of the factor spaces H1 . The following criterion allows to verify this property: A function F(λ) is monotonously decreasing under LOCC if F is invariant under any permutation of the Schmidt coefficients λi . which is the von Neumann entropy of the reduced density matrix. [18] ∂F ∂F (λ1 − λ2 ) − ≤0 (2. [20]) reads [21] . c(Ψ ) = |Ψ ∗ |σy ⊗ σy |Ψ | .54) i 0 represented in a given orthonormal basis {|0. satisfies ∂S(ρr ) ∂S(ρr ) λ2 (λ1 − λ2 ) − = (λ1 − λ2 ) ln ≤0. we can restrict our discussion to functions F(λ) of the Schmidt coefficients only.51) ∂λ1 ∂λ2 It suffices to express the condition for Schur concavity in terms of only the first two Schmidt coefficients because of the required permutation invariance. Concurrence Another frequently used monotone is concurrence c. the Schmidt coefficients provide a complete set thereof. We now evaluate this criterion for a few specific monotones and measures. non-increasing under LOCC. That is.. c is often defined in terms of the local Pauli matrices   0 −i σy = (2.

we were able to give constructive definitions for some entan- glement measures.58) concurrence is a valid monotone. This advantage. but takes positive values if pt has one. Mixed States For pure states. or more negative eigenvalues. The most prominent example is negativity [22]. concurrence reads   c(Ψ ) = 2 λi λj . N vanishes. Earlier.80 F. comes at the price that negativity assigns non-vanishing entanglement only to those states that are detected via their negative partial transpose. In comparison to virtually all other mixed state entanglement monotoness. Therefore. 2. for two-level systems. (2.57) i =j This is invariant under permutations of the λi . In the case of mixed states.56) and is equivalent to (2. N can be evaluated easily. In terms of the Schmidt coefficients. much as for the ppt-criterion itself. If pt is positive semi-definite. . Mintert et al. Negativity So far. since it is an algebraic function of the spectrum of pt .55). ∂λ1 ∂λ2 2c 2c i =1 i =2 (2. (2. negativity is fully reliable only for 2 × 2 or 2 × 3 system.59) 2 what was proved to be monotonously decreasing under LOCC [22]. These need to be dis- tinguished from genuine quantum correlations by a mixed state entanglement monotone.  c(Ψ ) = 2(1 − tr2r ) . we saw that the partial transpose pt of a mixed state  can be very helpful to decide on the separability of : if one of the eigenvalues λi of pt is negative. This inspired the definition of negativity as  ( i |λi |) − 1 N () = . however. then  is entangled. only very few constructively defined quantities were proved to be non-increasing under LOCC. in Sect. and since ⎛ ⎞ ∂c ∂c (λ1 − λ2 ) ⎝  (λ1 − λ2 )2 (λ1 −λ2 ) − = λi − λi ⎠ = − ≤0. it turns out to be much more involved to find a quantity that is monotonously decreasing under LOCC. (2.2.3. The basic difference between mixed and pure states in this specific context is that pure states bear no classical correlations. however.

one can characterize the entanglement proper- ties of  in terms of those of its pure state components. Since any mixed state can be decomposed into a probabilistic mixture of pure states  = pi |Ψi Ψi | .. i. a mixed state does not have a unique pure state decomposition. However. (2. (2. and different decompositions typi- cally yield different average values. However. A valid mixed state generalization that is monotonously decreasing under LOCC is the infimum over all pure state decompositions.60) i with positive prefactors pi . A very sugges- tive generalization  of a pure state monotone for mixed states is the average value i pi M(Ψi ) of the monotone M. the minimal average value  M() = inf pi M(Ψi ) .|Ψi . 2 Basic Concepts of Entangled States 81 Convex Roofs The failure to detect all entangled states finds its remedy with the so-called convex roof measures.61) {pi .e. the solution to this issue comes at the expense of an additional optimization problem that prevents the explicit algebraic evaluation in most cases.

Given the eigenstates |Φj  of .63) = δjk μj μk |Φj Φk | jk  = μj |Φj Φj | =  . j and any pure state decomposition of  can be obtained in this fashion [23]. provided k Vik† Vkj = δjk . To solve the optimization problem implicit in the convex roof definition (2.} i what is called the convex roof.61). any linear combination of the eigenstates √  √ pi |Ψi  = Vij μj |Φj  .e. together with the associated eigenvalues μi . for a left-unitary coefficient matrix V (with adjoint V † ):   √ √ pi |Ψi Ψi | = Vij μj |Φj Φk | μk Vik∗ i ijk  † √ = Vki Vij μj μk |Φj Φk | ijk  √ (2. (2. .. one needs a systematic way to explore all pure state decompositions of .62) j  defines another valid decomposition [23]. i.

we can now focus on the evaluation of concurrence for mixed states. Mintert et al.82 F. (2. for η ≥ 0. So far.2. concurrence is virtually the only quantity for which the convex roof can be evaluated algebraically.64) which allows to rewrite the convex roof expression above as  c() = inf c(ψi ) . however follows from the known solution for concurrence. Later in Sect. Concurrence of Mixed States With this characterization of pure state decompositions at hand. we will see that also the convex roof of the entanglement entropy has an algebraic solution.65) {|ψi . c(η|Ψ Ψ |) = η c(|Ψ ) .4. (2. A crucial property of concurrence in contrast to other monotones is the homogeneity. 2. This solution.

This allows to reformulate (2.} i √ where everything is expressed in terms of subnormalized states |ψi  = pi |Ψi . and the probabilities pi do not enter explicitly any more.61) in the following closed form  c() = inf c(ψi ) {|ψi .

} i  = inf |ψi∗ |σy ⊗ σy |ψi | {|ψi .

(2.62).67) Equation (2.} i   (2. u2 is equal to uT1 . we introduced a short-hand notation. V i where we used (2. enters instead of its adjoint. The differ- ence resides. also holds for the particular case of a symmetric matrix that we are facing here. of course. It stated that any matrix A could be diagonalized with two unitary transformations u1 and u2 as u1 Au2 . and that the transpose of a unitary. Already earlier. however. However. with elements τij = φ∗i |σy ⊗ σy |φj  . in this specific case. This.66) resembles the diagonalization of a hermitean matrix H through a unitary transformation UHU † . we can rephrase the infimum to be evaluated as . In the last line.66) T = inf  Vij φ∗j |σy ⊗ σy |φk Vki  V i jk    = inf [V τ V T ]ii  . in the fact that τ is symmetric and not hermitean. But also a symmetric matrix can be diagonalized in a similar fashion. τ = τ T . where U is unitary.55) and (2. respectively left unitarys.13) we have been invoking the singular value decomposition of a matrix. in (2. Therefore. where τ is a complex symmetric matrix.

Instead of a systematic derivation. S2 . as depicted in Fig. Let us take V˜ equal to V with ⎡ ⎤ 1 eiϕ2 eiϕ3 eiϕ4 1 ⎢ 1 eiϕ2 −eiϕ3 −eiϕ4 ⎥ V= ⎢ ⎥ . we can minimize this expression by proper choices of the free phases. ϕ3 . If S1 > i>1 Si . But it is still not straight- forward to derive an optimal matrix V˜ that achieves the infimum. S1 < i>1 Si as depicted on the right.69) 2 ⎣ 1 −eiϕ2 eiϕ3 −eiϕ4 ⎦ 1 −eiϕ2 −eiϕ3 eiϕ4 where we still have the free phases ϕ2 . S1 < i>1 Si . (2. the  optimal choice to minimize  |S 1 + i>1 iS e2iϕi | of the phases is ϕi = π/2. as depicted on the left. we obtain         Vτd V T  = S1 + e2iϕi Si  . and ϕ4 that we can adjust. Schematic drawing of the singularvalues Si added up with adjustable phases e2iϕi in the complex plane. we will use the convention that S1 is the largest of all diagonal entries. what leads to i |Vτd V T | = S1 − i>1 Si . (2.68) V V˜ i i where V˜ = V U † . But in the following. we have simplified the problem a lot: instead of 20 real parameter that characterize a general complex symmetric matrix. S3 . With the diagonal form τd of τ . In the  latter case. itis optimal to take ϕ2 = ϕ3 = ϕ4 = π/2. With this choice. one can always find a choice of phases such that i |Vτd V T | = 0. on the other hand. S4 ] is the diagonal form of τ .70) i i>1 Now. The order of the diagonal elements is not determined and can be chosen arbitrarily. In the former case. what is most conveniently  done by distinguishing two cases. we are going to take an Ansatz that eventually will turn out to do the job. where S  1 ≥ S i>1 i . we are left with only four real parameters. 2 Basic Concepts of Entangled States 83       c() = inf [V U † U τ U T U ∗ V T ]ii  = inf [V˜ τd V˜ T ]ii  .2.2. and τd = Uτ U T = diag[S1 . 2. That is. 2. we found a pure state decomposition in which the average Fig. If. then one can always find  phases ϕi such that |S1 + i>1 Si e2iϕi | vanishes . (2.

since they sum up to unity. we found a vanishing value for concurrence. which obviously is the infimum. and b = j>1 Wij2 Sj . or if there are decompositions that yield a smaller value. since concurrence cannot be negative. Here we will make use of the fact that for pure states in bipartite two- level systems. We obtained the fifth line using − | aj | ≥ − |aj |. one can determine both Schmidt coefficients in terms of the concurrence: √ 1 ± 1 − c2 λ± = . Thus. 0) (2. i. Therefore. 0). However. Now. we still do not know. (2.  For answering this question.84 F. there is only one independent Schmidt coefficient.. We now show that there is no left-unitary W that could yield a smaller value than what we have found so far.72) i>1 for the concurrence of an arbitrary mixed state of a bipartite two-level system.73) 2 . i |W2ij | = 1. which is the convex roof extension of the entanglement entropy. with aj = Wij Sj .e. and the last line followed again from the left-unitarirty of W . we can restrict ourselves to the case S1 ≥ i>1 Si . let us start out not with τ˜ = Vτd V T with the choice ϕ2 = ϕ3 = ϕ4 = π/2 that we found optimal above. In the other case. Entanglement of Formation of Mixed States With this solution for concurrence. j>1 where going from the second  to the third line we used |a + b| ≥ |a| − |b| with a = Wi1 2 S1 . Mintert et al.  concurrence reads max(S1 − i>1 Si . we used the  2 left-unitarity  condition  of W . if this is optimal.         [W τd W T ]ii  =  Wij2 Sj  i i j     = 2 Wi1 S1 + Wij2 Sj  i j>1     2   ≥ Wi1 S1 −  Wij2 Sj  i j>1    (2. we found the algebraic solution  c() = max(S1 − Si . and in the fourth line. we can now proceed and consider entan- glement of formation.71) = S1 −  Wij2 Sj  i j>1   2  ≥ S1 − Wij  Sj i j>1  = S1 − Sj .

With the help of this particular decomposition. Thus. but actually its exact value: due to its definition as convex roof.. (2. since the entanglement entropy is a function of λ. where we introduced the function E(c). E() is bounded from above by its average value evaluated in any decomposition – in particular {˜ pi .  = i p˜i |Ψ˜i Ψ˜i | in which all pure states do have the same value of concurrence. since the probabilities add up to 1. i.. c(Ψ˜i ) = c() [19]. Here. and convex  ∂ E(c)/∂c  2 2 ≥ 0. we necessarily need to conclude that these two quantities coincide: E() = E(c()) .75) i  = E(inf pi c(Ψi )) i = E(c()) . finally. Now.72) – one can easily also obtain entanglement of formation.74) ≡ E(c) . And.e.e. Since we found above in (2. . in going from the second to the third line. we used the convexity of E. in particular. there is one. E() ≤ i p˜i E(c( ˜i )). we can replace c(Ψ˜i ) by c(). And.76) Therefore. we arrive at the conclusion that entanglement of formation is bounded from above by E(c()). The crucial feature here is the fact that there is not a single optimal decomposition of a mixed state  into pure states that yields the actual value of concurrence. |Ψi }. it can also be ex- pressed in terms of concurrence via √ √ √ √ 1 + 1 − c2 1 + 1 − c2 1 − 1 − c2 1 − 1 − c2 E(Ψ ) = − ln − ln 2 2 2 2 (2. i. we arrive at the following reasoning:  E() = inf pi E(Ψi ) i  = inf pi E(c(Ψi )) i  ≥ inf E( pi c(Ψi )) (2. One easily convinces oneself that E(c) is monotonously increasing ∂E(c)/∂c ≥ 0.75) that it is also bounded from below by the same quantity. but there is actually a continuum  of optimal decompositions. With the help of these properties. once one has evaluated concurrence for a mixed state – what can be done algebraically (2. But. Convexity can equivalently be expressed as i pi E(qi ) ≥ E( i pi qi ). we can now show that E(c()) is not only a lower bound on entanglement of formation. for c ≥ 0. we are close to seeing that this is indeed not only a bound but rather the exact result. 2 Basic Concepts of Entangled States 85 And. so that we end up Ψ  with E() ≤ i p˜i E(c()). we found that entanglement of formation is bounded from below by E(c()). and from the third to the fourth its monotonicity.

Lin. R. Acta. A. R. T. Phys. A. G. Phys. 1 (1996) 6. Soc. 777 (1935) 2. (Elsevier. 2275 (1997) 15. F. 032314 (2002) 23. M. K. 47. Phys. L. Phys. M. Horodecki. 2245 (1998) 20. Phys. Uhlmann: Fidelity and concurrence of conjugated states. A. Reed and B. Inf. 32. Cam- bridge Philos. Rev. E.86 F. 433 (1907) 4. Størmer: Positive linear maps of operator algebras. F. 10. and N. 285 (1975) 14. Wootters: Entanglement of formation of an arbitrary state of two qubits. Am. Rev. A. M. A. Phys. B. A 223. Nielsen and G. F. 77. 4277 (1989) 3. Horodecki. Horn and C. Horodecki: Separability of mixed states: necessary and sufficient conditions. Ando: Majorization. E. Caves. A 62. 1. 1413 (1996) 9. Podolsky. Peres: Separability criterion for density matrices. Appl. 355 (2000) 16. Kraus: States. M. Milburn: Universal state inversion and concurrence in arbitrary dimensions. 76 (2001) 18. Rev. and G. 83. Terhal: Bell inequalities and the separability criterion. Rev. J. 211 (1955) 13. P. M. Linear Algebra Appl. Rev. Man-Duen Choi: Completely positive linear maps on complex matrices. and comparison of eigen- values. Math. A 40. B. Buˇzek. Rev. Rippin. W. Mintert et al. 63. Vedral. E. Lett. P. Lett. 165 (1976) 11. Schmidt: Zur Theorie der linearen und nichtlinearen Integralgleichungen. Opt. Comp. M. Probability relations between separated systems. 233 (1963) 10. Werner: Quantum states with Einstein-Podolsky-Rosen correlations ad- mitting a hidden-variable model. Lett. M. S. Simon: Analysis of Operators. Lett. 042315 (2001) 22. Amsterdam 1978) 8. 80. M. Rev. G. Rev. and R. 118. 032307 (2000) 21. 47. V. Phys. 6. C. Werner: Computable measure of entanglement. Soc. Plenio. A. Mod. Math. W. Quant. Rosen: Can quantum-mechanical description of physical reality be considered complete?. Ann. A 65. L. Lett. References 1. Rungta. Rep. Math. Alg. Entanglement monotones. A 271. Vidal and R. J. Einstein. Hillery. doubly stochastic matrices. Proc. 163 (1989) 19. 436 (1999) 17. Stinespring: Positive functions on c*-algebras. Vidal. 319 (2001) 7. Nielsen: Conditions for a class of entanglement transformations. Effects and Operations: Fundamental Notions of Quantum Theory (Springer. Phys. Vidal: Majorization and the interconversion of bipartite states. P. Schr¨odinger. Lett. Rev. and P. M. A. Johnson: Matrix Analysis (Cambridge University Press. 78. 10. V. Phys. 110. K. Phys. Woronowicz: Positive maps of low dimensional matrix algebras. B. New York 1983) 12. Math. New York 1985) 5. Phys. A 64. 446 (1936) . Phys. Knight: Quantifying entan- glement. Proc. .