(MT) 0408094 (Measures of Entanglement in Quantum Mechanics) Geir Ove Myhr

Norges teknisk-naturvitenskapelige universitet - NTNU
Fakultet for naturvitenskap og teknologi

Institutt for fysikk
arXiv:quant-ph/0408094v1 14 Aug 2004
HOVEDOPPGAVE
FOR
STUD. TECHN. GEIR OVE MYHR
Oppgaven gitt: 01.02.2004

Besvarelsen levert: 27.06.2004
FAGOMRÅDE: TEORETISK FYSIKK
Norsk tittel: “Mål på sammenfiltring i kvantemekanikk”

Engelsk tittel: “Measures of entanglement in quantum mechanics”
Hovedoppgaven er utført ved Institutt for fysikk, Fakultet for naturvitenskap og teknologi,
under veiledning av professor Jan Myrheim og stipendiat Stein Olav Skrøvseth.
Trondheim, 27. juni 2004,
Jan Myrheim
ansvarlig faglærer
professor ved Institutt for fysikk
iv
Abstract
I give an overview of some of the most used measures of entanglement. To make the
presentation self-contained, a number of concepts from quantum information theory are
first explained. Then the structure of bipartite entanglement is studied qualitatively, before
a number of bipartite entanglement measures are described, both for pure and mixed states.
Results from the study of multipartite systems and continuous variable systems are briefly
discussed.
v
vi
Preface
“What’s the name again of this quantum effect when one particle on one side of the universe
magically affects a particle on the other side of the universe?”, I asked while walking out
the door to the corridor. I was on my way to a group of teenagers who were waiting to hear
how it is to study physics at the university. In case someone asked why I found physics
so exciting I could always mention this quantum, non-locality, spooky action-at-a-distance
thing. If I only remembered the name of it. . . “That’d be entanglement”, a graduate student
next-door suggested, using the English term. Yes, that was the word lost in the back of my
mind. I had actually never heard a Norwegian name for it.
This happened one of the last days of October 2003. About a month later it was
decided that my master’s thesis would be about this very action-at-a-distance thing, or
more precisely, about how to quantify entanglement. From hardly remembering the name
of the game to understanding how an appropriate measure needs to behave has been a long
and interesting journey. In writing this text I have tried to keep in mind my own state of
knowledge as of October 2003. With that I hope that any 5th year physics student will
find it sufficiently understandable and self-contained as to be able to grasp the main points.
Whether I have succeeded or not is for others to decide.
This thesis is the final work leading up to the degree of Master of Science in Applied
Physics and Mathematics at the Norwegian University of Science and Technology (NTNU).
It was written during the 2004 spring term.
I would like to thank Jan Myrheim and Stein Olav Skrøvseth for taking the extra effort
to find a topic in quantum information theory for my thesis, and for supervising it. I would
also like to thank Viktor H. Havik and Henrik Tollefsen for interesting discussions about
entanglement.
vii
viii
Contents
Abstract v
Preface vii
Contents ix
1 Introduction 1
2 Preliminaries 3
2.1 The qubit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Entangled states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Density operators and mixed states . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 The Schmidt decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Generalised measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.6 Transformation of states and LOCC . . . . . . . . . . . . . . . . . . . . . . . 13
2.6.1 Local Operations and Classical Communication . . . . . . . . . . . . . 13
2.6.2 Transformation of a Bell state . . . . . . . . . . . . . . . . . . . . . . . 14
2.6.3 Concentrating partial pure state entanglement . . . . . . . . . . . . . 14
2.6.4 Distilling entanglement from mixed states . . . . . . . . . . . . . . . . 16
3 Characterisation of bipartite entanglement 21

3.1 Pure states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Mixed states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.1 Separable states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.2 Free and bound entanglement . . . . . . . . . . . . . . . . . . . . . . . 23
4 Measures of bipartite entanglement 25

4.1 Pure states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1.1 von Neumann entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.2 Reduced von Neumann entropy – entropy of entanglement . . . . . . . 26
4.1.3 Entanglement cost and distillable entanglement . . . . . . . . . . . . . 27
4.1.4 The uniqueness theorem for measures of entanglement . . . . . . . . . 28
4.2 Mixed states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2.1 Entanglement cost and distillable entanglement for mixed states . . . 29
4.2.2 Axioms for abstract measures . . . . . . . . . . . . . . . . . . . . . . . 30
ix
4.2.3 ED and EC as extreme measures . . . . . . . . . . . . . . . . . . . . . 32
4.2.4 Relative entropy of entanglement and other distance based measures . 33
4.2.5 Entanglement of formation . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2.6 Negativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2.7 Squashed entanglement . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2.8 Other entanglement measures . . . . . . . . . . . . . . . . . . . . . . . 40
5 Beyond bipartite entanglement and finite dimension 43

5.1 Multipartite systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.1.1 Separability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.2 Measures of entanglement . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2 Continuous variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
A Expressions for the quantum relative entropy 47
References 49
Index 57
x
Chapter 1
Introduction
Quantum entanglement was first viewed as a curiosity when it was pointed out in the
attempt by Einstein, Podolsky and Rosen [EPR35] to show that quantum mechanics could
not be a complete theory. When John Bell many years later published his now well-known
theorem [Bel64], it was hardly noticed in the scientific community. Bell had showed that for
a theory based on local hidden variables, certain correlations were upper bounded, a result
today known as the Bell inequalities. Quantum mechanics predicted that the inequalities
would be violated, thus giving rise to a way of testing whether the predictions of quantum
mechanics or the assumptions of Einstein et al. were correct. Experiments have later agreed
with quantum mechanics [ADR82], although there are still critics arguing that there are
loopholes in the experimental assumptions [TH02, San04].
Entanglement has since been regarded as a real – albeit strange – phenomenon of quan-
tum mechanics. It was widely regarded as being the same as violation of some Bell inequality.
However, Werner [Wer89] showed that there exist mixed quantum states that allow for a
local hidden-variable theory, but nevertheless are entangled. This led to new criteria to
decide if a state was separable [Per96, HHH96].
Parallel to this, the view of entanglement had gradually changed. From being regarded
as a curiosity, it was now seen as an information processing and communication resource
that could be used for performing tasks that would be impossible without it. Among the
possibilities introduced were quantum cryptography [Eke91], dense coding [BW92], telepor-
tation of a quantum state [BBC+ 93] and exponential speed-up of certain computational
tasks [Sho94, Sho97, Gro96, Gro97]. Because entanglement was now viewed as a resource,
it was natural to try to quantify the entanglement in quantum states.
After the first papers on the quantification of entanglement [SM95, BBP+ 96, BBPS96,
BDSW96], the subject has grown into a whole field of research. The aim of this work
is to describe the entanglement measures that have come out of this. The main focus is
on entanglement between two parties with finite dimensional Hilbert spaces, as this is the
situation where most of the theory is known. Extensions to entanglement between more
than two parties and entanglement with continuous variables are also briefly discussed.
In the course of writing this, several reviews of entanglement and quantum information
theory have been of great help. The textbook by Nielsen and Chuang [NC00] gives a nice
introduction to the prerequisites to understand entanglement theory, but contains little on
1
2 Chapter 1. Introduction
entanglement itself. John Preskill’s lecture notes [Pre98] explain much of the same, but
has a whole chapter devoted to entanglement. The tutorial by Bruß [Bru02] gives a short
introduction to most aspects of entanglement, including some entanglement measures. Both
the lecture notes by Eisert [Eis03] and the review article by Keyl [Key02] include a chapter
on entanglement measures, but the latter is rather technical in nature. The review article
by M. Horodecki about entanglement measures [Hor01] summarises the most of the field in
an elegant way, but it requires much prior knowledge from the reader, and treats almost
exclusively two-party entanglement.
The rest of this work is organised as follows. Chapter 2 gives an introduction to the
concepts needed in the following chapters. A reader with a basic knowledge in quantum
information theory may skip large portions of this chapter. Chapter 3 describes two-party
entanglement qualitatively, preparing the ground for chapter 4 which describes the ways
to quantify two-party entanglement. Finally, in chapter 5 it is briefly indicated how the
notions from the previous chapters may be extended to systems of continuous variables and
to multi-party systems.
Unless stated otherwise all logarithms are taken base-2.
Chapter 2
Preliminaries
Before we can understand how to quantify entanglement, we need to have some concepts
and procedures clear. In this chapter we will have a first look at qubits and entangled states,
get used to density operators and see how quantum states can be transformed into others.
2.1 The qubit

The simplest nontrivial quantum mechanical system is a two-state system which can be
described by a vector in two-dimensional complex Hilbert space. Such a system is called a
qubit. The favourite qubit of many physicists is the spin of a spin-1/2 particle, for instance
an electron. We will use spins of such particles to illustrate entanglement.
If we measure the z-component of the spin, the result is either up or down. This mea-
surement corresponds to the observable operator which we call Z. After the measurement
the state of the particle is an eigenstate of the observable. We denote these states by |↑z i
and |↓z i, for spin parallel to and antiparallel to the z-axis. They are mutually orthogonal,
and we take them as the unit basis vectors of the spin Hilbert space. In addition to the
eigenstates, the particle may be in any superposition of the eigenstates,
|ψi = α |↑z i + β |↓z i (2.1)
such that |α|2 + |β|2 = 1. When the spin along the z-axis is measured, we obtain +1 with
a probability of |α|2 and −1 with probability |β|2 (where the spin is in units of ~/2). After
the measurement the state is collapses into the corresponding eigenstate.
As the overall phase of a quantum state does not have any physical significance, only
the phase difference between α and β is important, and for convenience we often choose α
to be real. In order to also include the property |α|2 + |β|2 = 1, we can write the spin state
(2.1) of the particle as
θ θ
|ψi = cos |↑z i + eiφ sin |↓z i . (2.2)
2 2
This gives a nice and intuitive picture of the state space of the spin-1/2 particle, and thereby
any qubit. All possible states are covered when the parameters are restricted to 0 ≤ φ ≤ 2π
and 0 ≤ θ ≤ π. Thus, any state corresponds to a point on the unit sphere, with the
3
4 Chapter 2. Preliminaries
azimuthal angle given by φ and the polar angle given by θ. Such a sphere is called a Bloch
sphere, and is a very useful tool for visualising states of a qubit. A Bloch sphere with some
states indicated is shown in figure 2.1. Note that for θ = π, the phase of (2.2) changes
as φ changes, but the state is still the same. This is reflected on the Bloch sphere, as the
azimuthal angle is irrelevant at the poles.
z
|↑z i
|ψi = cos 2θ |↑z i + eiφ sin 2θ |↓z i
|↓x i = √1 (|↑z i − |↓z i)

2
y
φ
|↑y i = √1 (|↑z i + i |↓z i)
2
|↑x i = √1 (|↑z i + |↓z i)

√1 (|↑z i
x 2
|↓y i = 2
− i |↓z i)
|↓z i
Figure 2.1: The Bloch sphere of the Hilbert space spanned by |↑z i and |↓z i.
We may choose another set of orthonormal basis vectors. For instance

1 1
|↑x i = √ (|↑z i + |↓z i) |↓x i = √ (|↑z i − |↓z i) . (2.3)
2 2
These are the eigenvectors of the X-operator, and they are the resulting states if we measure
the spin along the x-axis. Likewise, we may measure the spin along any axis, and find a
corresponding set of eigenvectors. If we denote the basis vectors

1 0
|↑z i = |↑z i = , (2.4)
0 1
then the operators for the spin along the x, y and z axes are nothing more than the Pauli
matrices
0 1 0 −i 1 0
X≡ Y ≡ Z≡ . (2.5)
1 0 i 0 0 −1
For the spin along any other unit vector, say a~x + b~y + c~z where a2 + b2 + c2 = 1, the spin
operator is aX + bY + cZ. All operators of this form have eigenvalues ±1 corresponding to
spin parallel and antiparallel to the vector.
2.2. Entangled states 5
As already mentioned, other systems are equally good qubits as a spin-1/2 particle.
Among the more popular ones are the polarisation of a photon and the energy levels of a
two-level system. They can be represented by a Bloch sphere just as well as the spin of an
electron can. The advantage of thinking about a spin-1/2 particle is that the vectors on the
Bloch sphere corresponds to actual spin directions in real space. Thus, a qubit in a state
close to the y-axis on the Bloch sphere, has, when the qubit is a spin-1/2 particle, a spin
pointing in a direction close to the y-axis.
In general the particular implementation of the qubit is not important for our purpose,
so we simply denote the basis vectors |0i and |1i, which in the case of a spin-1/2 particle
are typically |↑z i and |↓z i, respectively. The superpositions of these states corresponding
to the spin along the x-axis, are often denoted |+i and |−i, according to the sign in the
superposition.
1 1
|+i = √ (|0i + |1i) |−i = √ (|0i − |1i) (2.6)
2 2
Three qubit operators are of special importance, namely the unitary Pauli operators of
(2.5). In the context of general qubits, they can be written as
X ≡ |1i h0| + |0i h1| , (2.7a)

Y ≡ i |1i h0| − i |0i h1| , (2.7b)
Z ≡ |0i h0| − |1i h1| . (2.7c)
X is also called the bit flip operator as it turns a |0i into a |1i and vice versa. Z is called
the phase flip operator as it switches between the states α |0i + β |1i and α |0i − β |1i. The
Pauli operators are defined with respect to a particular basis. For instance, in the {|+i , |−i}
basis, Z takes the role of the bit flip operator.
The name qubit comes from quantum bit and it plays the same role in quantum infor-
mation theory as the bit does in classical information theory. The bit is a system which
can be in one of two states, usually denoted 0 and 1. Thus, each bit is capable of storing
enough information to answer exactly one yes/no question. As we have just seen, the qubit
is a much richer structure. In addition to being capable of being in the states |0i and |1i,
just as a bit, it can be in a continuum of superpositions between those states, given by two
parameters. Still, only a single bit of information can be read from a single qubit. When
the spin of a spin-1/2 particle is measured along an axis the only possible results are parallel
and antiparallel to the chosen axis. After the measurement the state is changed, and from
the result of the measurement we know what it is. No more information about the original
state can be retrieved.
2.2 Entangled states

From the simplest quantum systems whatsoever, we now turn to look at the simplest com-
posite quantum systems. A composite system is a system which consists of two or more
parts, and the simplest one is a system consisting of two qubits. We call the two systems
A and B. In communication and entanglement theory the convention is to think of the
subsystems as being in the possession of two well separated observers, called Alice and Bob.
Any state of each of the two systems can be written as
|ψiA = α |0iA + β |1iA |φiB = γ |0iB + δ |1iB (2.8)
with |α|2 + |β|2 = 1 and |γ|2 + |δ|2 = 1 The composite state of the two systems is then
simply the tensor product (or direct product) of the two states.
|Ψprod i = |ψiA ⊗ |φiB (2.9)
Such a state is called a product state, but product states are not the only physically realisable
states. If we let the two systems interact with each other, any superposition of product states
is realisable. Hence, a general composite state can be written as
X
|Ψi = αij |ψi iA ⊗ |φj iB (2.10)
ij
|αij |2 = 1 and the sets {|ψi i} and {|φj i} are orthonormal bases for the two sub-
P
where
systems. Any composite state that is not a product state is called an entangled state.
A composite quantum state consisting of two parts only, is called a bipartite state, as
opposed to multipartite states which consist of more than two parts. For bipartite qubit
states, four entangled states play a major role, namely the singlet state
1
|Ψ− i ≡ √ (|01i − |10i) (2.11a)
2
and the three triplet states
1
|Ψ+ i ≡ √ (|01i + |10i) (2.11b)
2
1
|Φ− i ≡ √ (|00i − |11i) (2.11c)
2
+ 1
|Φ i ≡ √ (|00i + |11i) (2.11d)
2
where we have used |iji as a shorthand notation for |ii ⊗ |ji. They are called Bell states
or EPR pairs. Together they form an orthonormal basis for the state space of two qubits,
called the Bell basis. The Bell states are maximally entangled and one can be converted
into another by applying a unitary transform locally on any one of the subsystems.
Note that if we measure the state of one qubit in a Bell state (that is, measure the Z
operator which has eigenvalues ±1), we immediately know the state of the other particle. In
the singlet Bell state, a measurement of qubit A will yield one of the eigenstates |0i and |1i,
each with a probability of 1/2. These results leave qubit B in state |1i or |0i, respectively.
For a single qubit we could always change to another basis where the outcome of a Z
measurement would be given. For a spin-1/2 particle this means that the spin is always
pointing in some direction, even though the state will show up as a superposition in a
basis where the state is not one of the basis states. If the particle is entangled with another
particle, though, the direction of the spin of that particle alone is not well defined. Actually,
for particles in one of the Bell states, the probability for measuring the spin of the particle
to “up” (while ignoring the other particle) is 1/2 for any direction.
2.3. Density operators and mixed states 7
2.3 Density operators and mixed states

We want to be able to describe the state of a subsystem that is entangled with another
subsystem to which we do not have access. State vectors cannot be used for this purpose,
and we need another representation of quantum states, namely density operators. This is
not only useful for describing “locally” a subsystem which is part of an entangled system for
which we know the state. The formalism is also necessary for describing quantum mechan-
ical experiments – where noise is inevitable – and doing quantum statistical mechanics. In
the laboratory it is impossible to isolate the quantum systems under study completely. The
systems become entangled with the environment through unwanted, but unavoidable, inter-
actions. Nevertheless, it is necessary to describe the system without taking the environment
into account.
In the density operator formalism, we describe quantum states by operators on the
system’s Hilbert space instead of unit vectors on it. For any quantum state vector |ψi,
the corresponding density operator is the projection operator |ψi hψ|. As linear operators
may be – and often is – represented by matrices, density operators are also called density
matrices.
So far we have only represented the state in a new way, using density operators instead
of vectors. But what if we do not know exactly in what state our system is? Imagine,
for example, that we have a machine in our lab which outputs particles in state |ψi (think
of it as some component of a spin). But once in a while events beyond our control and
knowledge (a spike in the electric grid, the turning on of an electromagnet in a neighbouring
lab, etc.) make the machine malfunction slightly producing states which are close to, but
not exactly |ψi. Let us for simplicity say that the state |ψ ′ i is always created when the
machine malfunctions, and that the probability for that to happen to any given particle is
p. Given this uncertainty, the state cannot be described by a state vector. Still, in terms of
density matrices it is described as
ρ = (1 − p) |ψi hψ| + p |ψ ′ i hψ ′ | .
A state that may be represented by a unit vector is called a pure state. For a pure state
we have maximal knowledge of the state. Other states are convex combinations 1 of pure
states, and are called mixed states. To check if a given density operator, ρ, represents a
pure state, it is sufficient to check if ρ = ρ2 , which holds for all pure states and no mixed
states.
The density operator does not tell us from what pure states the state was prepared. For
example, we can have a state |0i or |1i, each with probability 1/2. The density operator
is then 12 (|0i h0| + |1i h1|) = 12 I. The same density operator can be made by mixing the
states |+i = √12 (|0i + |1i) and |−i = √12 (|0i − |1i) in equal amounts, or the three states |0i,
√ √ √ √
|+i and 12 |0i − 23 |1i in amounts p1 = 21 − 63 , p2 = 3+√3 3 and p3 = 1 − 33 , respectively.
We call these different collections of pure states with corresponding probabilities, {pi , |ψi i},
ensembles. Sometimes it is also called a realisation and we say that an ensemble realises a
1
P
A
Pconvex combination of elements x1 , x2 , . . . , xN is an element which can be written i pi xi for pi ≥ 0
and i pi = 1. A convex combination of two points is a point on the straight line connecting them.
mixed state. Hence, when we describe a state from an ensemble by its density operator, we
discard the information about which ensemble the mixed state was made from. The density
operator still describes the mixed state as well as can be done, as states from different
ensembles having the same density operator are experimentally indistinguishable.
This indistinguishability makes the density operator a perhaps more intuitive represen-
tation of a state than the state vector. For example, a state vector has an arbitrary overall
phase factor, so two vectors differing by a phase factor represent the same state. Still,
when the vector is put together with its dual vector to form a pure state density opera-
tor, the phase factor vanishes because of the complex conjugation. Hence, all state vectors
representing the same state are represented by the same density operator.
To complete our discussion about density operators, we should take note of some of
their properties. For a more thorough discussion, see e.g. [NC00]. Density operators are (i)
positive 2 , meaning that for any vector |vi, hv| ρ |vi ≥ 0, or equivalently that it is Hermitian
with nonnegative eigenvalues. Also, (ii) the trace of a density operator is unity, tr(ρ) = 1.
It turns out that any operator satisfying these two criteria can be realised by a pure state
ensemble. Hence, we may take these criteria as the defining properties of a density operator.
We will now see how the density operators are used to describe individual parts of an
entangled system. In fact, we will see that even when the composite system is in a pure
state, if the subsystems are entangled, the individual subsystems are in mixed states.
Consider a bipartite composite system of two qubits, shared between Alice and Bob.
The qubits are entangled, and the system is in the state
|ψi = α |00i + β |11i . (2.12)
Now, imagine that Alice loses contact with Bob, and wants to describe her system without
any reference to the qubit in Bob’s lab. Whatever Bob happens to be doing to his qubit
should not have observable consequences in Alice’s lab. If it had, it could have been used
to perform superluminal communication in some reference frame which would wreak havoc
in physics by allowing consequences to happen before the cause3 . One of the things Bob
might choose to do with his system, is to measure Z in the {|0i , |1i} basis. If he does, he
will get |0i or |1i with probability |α|2 or |β|2 , respectively. This leaves Alice’s qubit in the
same state as the one Bob measured, but as she wouldn’t know the result, the state of her
qubit would be a mixture of |0i and |1i, namely
ρA = |α|2 |0i h0| + |β|2 |1i h1| (2.13)
This mixture correctly predicts the probabilities for all measurements that can be performed
locally by Alice, regardless of what Bob really does to his system.
Formally, ignoring some degrees of freedom in a system (like the ones corresponding to a
particle out of reach) is done by tracing out the relevant degrees of freedom from the density
2
Positive operators here refer to positive semidefinite operators. Positive definite operators, on the other
hand, satisfy hv| ρ |vi > 0 for any |vi and have only positive eigenvalues.
3
This is of course in itself no compelling reason. Even John Bell was afraid that entanglement might do
away with relativity [Whi98]. The real reason is that quantum mechanics tells us that the probabilities for
outcomes of local measurements on one system do not change when an arbitrary operation is performed on
the other system.
2.3. Density operators and mixed states 9
operator. This is also called taking the partial trace with respect to the degrees of freedom
that are to be traced out. Any bipartite state can be written as
X
ρ= cijkl |ai i haj | ⊗ |bk i hbl |
ijkl
and the partial trace with respect to system B of a bipartite system is defined as
X
trB (ρ) ≡ cijkl |ai i haj | ⊗ tr |bk i hbl |
ijkl
X
= cijkl hbl |bk i |ai i haj |
ijkl
X
= Cij |ai i haj | (2.14)
ij
P
with Cij = kl cijkl hbl |bk i. The density operator obtained by tracing out one part of the
system is called the reduced density operator .
We can perform the partial trace on our example state (2.12). The density matrix is
ρ = |α|2 |00i h00| + αβ ∗ |00i h11| + α∗ β |11i h00| + |β|2 |11i h11| , (2.15)
so the reduced density matrix becomes
ρA = trB (ρ)
= |α|2 h0|0i |0i h0| + αβ ∗ h0|1i |0i h1| + α∗ β h1|0i |1i h0| + |β|2 h1|1i |1i h1|
= |α|2 |0i h0| + |β|2 |1i h1| , (2.16)
which is just what we expected.

In section 2.2 we saw what we meant by an entangled state as long as the states were
pure. But this leaves the question of when a mixed state is entangled. Consider for instance
the mixture of two Bell states
1 + 1
ρ= |Φ i hΦ+ | + |Φ− i hΦ− | . (2.17)
2 2
Written out in an orthonormal product basis, this is
1
ρ= (|00i h00| + |00i h11| + |11i h00| + |11i h11|)
4
1
+ (|00i h00| − |00i h11| − |11i h00| + |11i h11|)
4
1 1
= |00i h00| + |11i h11| (2.18)
2 2
which is a mixture of the product states |00i and |11i. So the mixed state can be realised
by both an ensemble of maximally entangled states and an ensemble of product states. We
say that a mixed state is separable if and only if it can be realised as a mixture (convex
combination) of locally prepared states4 . That is, it can be written as5
(A) (B)
X
pi ρi ⊗ ρi (2.19)
i
P
where {pi } forms a probability distribution; pi ≥ 0 and i pi = 1. This also means
([HHH96]) that the state can be written as a mixture of pure product states,
(A) (A) (B) (B)
X
pkl |ψk i hψk | ⊗ |ψl i hψl | . (2.20)
kl
Hence, a mixture of entangled states need not be entangled, but a mixture of separable states
is always separable. Mixture is a process which destroys entanglement. This is because by
discarding information about which of a number of entangled states the system is in, it can
no longer be distinguished from a mixture of separable states.
2.4 The Schmidt decomposition

A useful tool when working with bipartite, pure state entanglement is the Schmidt decom-
position. It is a decomposition into the biorthogonal basis which gives the smallest possible
number of terms for a product basis.
Given a bipartite pure state |ψi, we may write it in terms of some product basis, or-
thonormal in both subsystems
dA X
X dB
|ψi = cij |ai i ⊗ |bj i . (2.21)
i=1 j=1
The coefficients {cij } may be seen as elements in a dA × dB matrix C. Using the singular
value decomposition this may be written as
C = U DV (2.22)
where D is a dA × dB matrix which is zero except for the diagonal elements which are real
and positive, U is a dA × dA unitary matrix and V a dB × dB unitary matrix. This means
Pmin(d ,d )
that cij may be written as k=1 A B uik dkk vkj , so the state vector may be written
dB min(d
dA X
X X A ,dB )
|ψi = uik dkk vkj |ai i ⊗ |bj i

i=1 j=1 k=1
 
min(dA ,dB ) dA
! dB
X X X
= dkk uik |ai i ⊗ vkj |bj i .
k=1 i=1 j=1
4
This definition is due to Werner [Wer89] who called them “classically correlated” states.
5
In fact it was considered sufficient that that such a sum approximated the state arbitrarily well. It was
shown in [HHH96], though, that in the finite dimensional case any separable state could be expressed in d2
terms, where d is the dimension of the Hilbert space of the composite system.
2.5. Generalised measurements 11
We now take as our new basis vectors

 
dA dB
!
X X
|a′k i ≡ uik |ai i |b′k i ≡  vkj |bj i
i=1 j=1
and define λk = dkk to obtain the final form
min(dA ,dB )
X
|ψi = λk |a′k i ⊗ |b′k i (2.23)
k=1
which is the Schmidt decomposition. What we have achieved is to write any state as a
linear combination of maximum min(dA , dB ) product vectors. Also, not only are the product
vectors in the basis orthogonal, the basis vectors from each subsystem are only used once.
These vectors are still orthogonal as we have only performed a unitary transformation on
the set of basis vectors, making the set of new basis vectors biorthogonal. The coefficients
{λk } are called the Schmidt coefficients, although in the literature this name is sometimes
given to {λ2k }
For many purposes the mere existence of the Schmidt decomposition is enough to make
it usable, whereas it sometimes needs to be calculated explicitly. An example of the former
is to show that the eigenvalues of the two reduced density matrices of a pure, bipartite state
are equal. Actually, it can be seen directly from (2.23) that by forming the density matrix
and tracing out any one of the two subsystems, the reduced matrix is already on diagonal
form with {λ2k } as eigenvalues. It is often easier to calculate the Schmidt coefficients this
way than doing the singular value decomposition explicitly.
2.5 Generalised measurements

To facilitate the quantum operations we will generalise the concept of measurements from the
way it is usually treated in courses of introductory quantum mechanics. In that formalism,
which is called projective measurements, the objects representing the measurements are
Hermitian operators. The result of the measurement is an eigenvalue of the observable and
after the measurement, the new state of the system is in the eigenstate corresponding to the
eigenvalue (or projected into the eigenspace if the eigenvalue is degenerate). The probability
for an outcome of the measurement is the absolute squared expansion coefficient when the
state is expanded in the eigenstates of the observable.
In the general measurement scheme, a measurement is represented by a set of linear
operators {Mi } called measurement operators satisfying the completeness relation
X
†
Mm Mm = 1. (2.24)
m
The probability for the outcome to be m is

†
p(m) = hψ| Mm Mm |ψi (2.25)
and after the measurement with result m, the new state of the system is
Mm |ψi
q . (2.26)
†
hψ| Mm Mm |ψi
We may, at least in theory, implement any measurement that can be described by a set
of measurement operators. Thus, we may construct measurements which convert an initial
state to a desired state with a certain probability, discarding the system if that particular
outcome does not occur. As we will see, this is one of the ideas behind distillation of
entanglement.
The projective measurement formalism is a special case of the generalised measurement
scheme. The observable, O, is a Hermitian operator and can be decomposed according to
the spectral theorem. X
O= λm |vm i hvm | (2.27)
m
where {|vm i} are the eigenvectors of the observable, and {λm } the corresponding eigenvalues.
By taking as measurement operators Mm = |vm i hvm |, the completeness relation (2.24) is
satisfied. The probability for a given outcome, m, of a measurement is
†
hψ| Mm Mm |ψi = hψ|vm i hvm |vm i hvm |ψi = | hvm |ψi |2
which is exactly the absolute square of the expansion coefficient of |ψi for that eigenstate.
The state after the measurement is
Mm |ψi hvm |ψi |vm i
= = eiθ |vm i
| hvm |ψi |
q
†
hψ| Mm Mm |ψi
with an arbitrary phase θ. Like |vm i, this is an eigenstate of O.

The generalised measurement scheme can easily be further generalised to density oper-
ators. For pure states, the generalisation is straightforward. The density operator corre-
sponding to the pure state vector |ψi is ρ = |ψi hψ|. The probability for a given outcome is
of course still the same as in (2.25), but we write it in terms of the density operator:
†
p(m) = tr(Mm ρMm ) (2.28)
When the state changes after a measurement with result m, the density operator corre-
sponding to the state vector (2.26) is obviously
† †
Mm |ψi hψ| Mm Mm ρMm
†
= †
(2.29)
hψ| Mm Mm |ψi tr(Mm ρMm )
The probability (2.28) and new state (2.29) is also valid if the density operator represents
a mixed state. To see this, we write the density operator in terms of one of its pure state
realisations: X X
ρ= pi |ψi i hψi | = p i ρi (2.30)
i i
2.6. Transformation of states and LOCC 13
The joint probability for the system to be in the pure state |ψi i and measuring a result m
is pi p(m|i). Summing over all i we get the total probability for getting m as the result of
the measurement.
X
†
p(m) = pi tr(Mm ρi Mm )
i
hX i
†
= tr Mm pi ρi Mm
i
†
= tr(Mm ρMm )
Similarly, the new state after the measurement of the mixed state becomes
X hX i
† † †
ρout = pi Mm ρi Mm = Mm pi ρi Mm = Mm ρMm .
i i
The general measurement scheme covers all operations that can be performed on a quan-
tum system. A unitary transformation is the special case of a one-outcome measurement as
the measurement operators then will have to be unitary to satisfy the completeness relation
(2.24). Performing a different unitary transformation Um conditional on the outcome of
a measurement described by the operators {Mm }, is described by the new measurement
operators {Um Mm }. It is easy to see that these operators satisfy the completeness rela-
tion, that the probabilities of the outcomes are the same and that the post measurement
† †
state is what we expect, namely Um (Mm ρMm )Um . Similarly, consecutive measurements
can be put together to form a single set of measurement operators, even when the mea-
surements to be done depend on the outcome of the previous ones. When doing this, all
possible combinations of measurement outcomes are considered as single outcomes of the
total measurement.
In applications it depends on the situation whether we use the full generalised mea-
surement scheme or only consider projective measurements. Projective measurements are
sometimes simpler to handle and combined with unitary transformations they describe ev-
erything that can be described by generalised measurements. Therefore, both formalisms
will be used in the following.
2.6 Transformation of states and LOCC

We will see later that the set of operations that we allow to be performed on a quantum
state is essential to the most basic entanglement measures. All operations can be described
as generalised measurements. But we will find it useful to describe them as measurements
combined with unitary transformations where the next step in a sequence of operations may
be decided by a previous measurement outcome.
2.6.1 Local Operations and Classical Communication

In the context of entanglement a subclass of quantum operations is particularly important,
namely local quantum operations (LO). When we have a bipartite or multipartite state, the
subsystems may be separated, or at least well isolated from each other. We may therefore
only perform operations that act locally on each subsystem. That is, the operators will
be of the form OA ⊗ OB ⊗ · · · . In addition, the parties sharing a multipartite state are
usually allowed to communicate classically, and through this classical communication (CC)
they may perform different operations depending on the measurement results of the other
parties. This subclass of operations is called Local Operations and Classical Communication,
usually shortened to LOCC . In the literature it is sometimes named otherwise, e.g. QLCC
or LQCC. LOCC is important in entanglement theory because entanglement is exactly those
correlations that cannot be created using LOCC.
2.6.2 Transformation of a Bell state

Suppose Alice and Bob share a Bell state |Φ+ i = √12 |00i + √12 |11i and want to transform
it into the state cos θ |00i + sin θ |11i. One way of doing this is as follows. Alice performs a
measurement described by the measurement operators

cos θ 0 sin θ 0
M1 = M2 = (2.31)
0 sin θ 0 cos θ
where the local state vectors are denoted as usual:

1 0
|0i = |1i = . (2.32)
0 1
The operators clearly satisfies the completeness relation (2.24), as

2 2
cos θ 0 sin θ 0 1 0
+ = .
0 sin2 θ 0 cos2 θ 0 1
After the measurement we may have one of two states, depending on the outcome,
|ψ1 i = cos θ |00i + sin θ |11i |ψ2 i = sin θ |00i + cos θ |11i
where |ψ1 i is the desired output state. If the result of the measurement is 2, leaving the
system in the state |ψ2 i, Alice applies the X operator (also called bit flip operator) to her
system converting |0i into |1i and vice versa. Then she tells Bob to do the same through
the classical communication channel. The resulting state is |ψ1 i and their common goal is
achieved.
The transformation we just did, transforms a maximally entangled state into a less or
equally entangled state. For θ = 0 and θ = π/2 the output state is a product state, which is
not entangled at all, whereas for θ = π/4 and θ = 3π/4 the output state is also a Bell state.
2.6.3 Concentrating partial pure state entanglement

We may also try to do the transformation the other way around. We begin with the state
|ψ1 i, where we for simplicity suppose that 0 < θ < π/4, and want to end up with the Bell
state |Φ+ i. The lower boundary of this interval is a product state for θ = 0 which cannot
possibly be turned into an entangled state by LOCC. The upper boundary θ = π/4 is the
desired output state and will need no transformation.
The problem is now to construct local measurement operators which give |Φ+ i as a post
measurement state. The operator
1
cos θ 0
M1 = const · 1
0 sin θ
will give the desired output, and we want the constant to be as large as possible to maximise
the probability for this output. Still, for the completeness relation (2.24) to be satisfied none
†
of the diagonal elements can be greater than 1. This is because for any Mm , Mm Mm has
nonnegative diagonal elements, and each diagonal element needs to sum up to 1 to get the
identity operator in (2.24). In the valid range of θ, cos θ > sin θ, so the maximal constant is
sin θ. This gives the optimal measurement operator for the successful outcome

tan θ 0
M1 = = tan θ |0i h0| + |1i h1| .
0 1
The remaining set of matrices representing the measurement operators need to have a van-
ishing element to the lower right, and the remaining elements need to be such that the
completeness relation is satisfied. The easiest way to do this is to simply add one more
operator, namely
1 − tan2 θ 0
√ p
M2 = = 1 − tan2 θ |0i h0| .
0 0
The new states after the measurement are proportional to
M1 |ψ1 i = sin θ |00i + sin θ |11i

p
M2 |ψ1 i = cos θ 1 − tan2 θ |00i .
The probability for the two outcomes are
p(1) = hψ1 | M1† M1 |ψ1 i = sin2 (h00|00i + h11|11i) = 2 sin2 θ

p(2) = hψ1 | M2† M2 |ψ1 i = cos2 θ(1 − tan2 θ) = cos2 θ − sin2 θ = 1 − 2 sin2 θ
and the normalised new states are

1
|ψout,1 i = √ (|00i + |11i) = |Φ+ i
2
|ψout,2 i = |00i .
We see that there is a certain probability for converting a partially entangled state into
a maximally entangled state, but the less entangled the initial state is, the smaller is the
probability. This is consistent with the fact that LOCC cannot increase the expected entan-
glement. Still, we may perform a measurement which may give us a more entangled state at
the risk of losing the entanglement we had. If we have a large number of bipartite two-qubit
systems, we may apply the above procedure and discard the systems where the output is a
product state, thereby effectively concentrating entanglement.
2.6.4 Distilling entanglement from mixed states

Also mixed state entanglement may be concentrated, or distilled. This will be important for
practical applications, as the channels used to create entanglement in general will be noisy.
A collection of distillation protocols was introduced in [BBP+ 96] and [BDSW96]. We will
take a closer look at a slightly modified version of a distillation protocol introduced in the
latter, which for certain mixed states can yield pure Bell states as output.
Consider a mixed state, as always shared between Alice and Bob, which can be rep-
resented as a collection of |Φ+ i states contaminated with a portion of |01i states. The
corresponding density operator is then
ρ = (1 − p) |01i h01| + p |Φ+ i hΦ+ |

p p p p
= (1 − p) |01i h01| + |00i h00| + |00i h11| + |11i h00| + |11i h11| .
2 2 2 2
Alice and Bob now pick two pairs from the ensemble. Each of them will perform a joint
operation on the two subsystems on their side. Then they measure one of the pairs, thereby
destroying any entanglement in that pair, to decide if the other pair is in the desired output
state.
The composite system of the two pairs is described by the density operator
ρ⊗2 =
(1 − p)2 |0101i h0101| + p(1−p)
2
|0100i h0100| + p(1−p)
2
|0100i h0111| + p(1−p)
2
|0111i h0100| + p(1−p)
2
|0111i h0111|
p(1−p) p2 p2 p2 p2
+ 2
|0001i h0001| + 4
|0000i h0000| + 4
|0000i h0011| + 4
|0011i h0000| + 4
|0011i h0011|
p(1−p) p2 p2 p2 p2
+ 2
|0001i h1101| + 4
|0000i h1100| + 4
|0000i h1111| + 4
|0011i h1100| + 4
|0011i h1111|
p(1−p) 2 2 2 2
p p p p
+ 2
|1101i h0001| + 4
|1100i h0000| + 4
|1100i h0011| + 4
|1111i h0000| + 4
|1111i h0011|
2 2 2
p(1−p) p p p p2
+ 2
|1101i h1101| + 4
|1100i h1100| + 4
|1100i h1111| + 4
|1111i h1100| + 4
|1111i h1111|
where the first and third qubit are in Alice’s possession and the second and fourth are in
Bob’s possession. The two first qubits are the first pair, and the two last are the second
pair. It may be written symbolically as |A1 B1 A2 B2 i. Both Alice and Bob applies a CNOT
gate to the two qubits in their possession, using the qubit from the first pair as a control
bit and the second one as the target. A CNOT gate is a unitary transformation which does
nothing when the control bit is zero and flips the target bit if the control bit is set to 1. The
operator is written
UCNOT = |00i h00| + |01i h01| + |10i h11| + |11i h10|

= |0i h0| ⊗ I + |1i h1| ⊗ X
where X is the usual bit flip operator, or Pauli X-operator. But as Alice’s operator works
on the first and third qubit while Bob’s operator works on the second and fourth, in our
basis the operators are
UCNOT (1, 3) = |0i h0| ⊗ I ⊗ I ⊗ I + |1i h1| ⊗ I ⊗ X ⊗ I

UCNOT (2, 4) = I ⊗ |0i h0| ⊗ I ⊗ I + I ⊗ |1i h1| ⊗ I ⊗ X
Operationally it only amounts to flipping the third qubit of ρ⊗2 in all terms where the first
qubit is 1 and flipping the fourth qubit where the second qubit is 1. This leaves the new
state
ρout = UCNOT (1, 3)UCNOT (2, 4)ρ⊗2 UCNOT

† †
(2, 4)UCNOT (1, 3) =
(1 − p)2 |0100i h0100| + p(1−p)
2
|0101i h0101| + p(1−p)
2
|0101i h0110| + p(1−p)
2
|0110i h0101| + p(1−p)
2
|0110i h0110|
p(1−p) p2 p2 p2 p2
+ 2
|0001i h0001| + 4
|0000i h0000| + 4
|0000i h0011| + 4
|0011i h0000| + 4
|0011i h0011|
p(1−p) p2 p2 p2 p2
+ 2
|0001i h1110| + 4
|0000i h1111| + 4
|0000i h1100| + 4
|0011i h1111| + 4
|0011i h1100|
p(1−p) p2 p2 p2 p2
+ 2
|1110i h0001| + 4
|1111i h0000| + 4
|1111i h0011| + 4
|1100i h0000| + 4
|1100i h0011|
p(1−p) p2 p2 p2 p2
+ 2
|1110i h1110| + 4
|1111i h1111| + 4
|1111i h1100| + 4
|1100i h1111| + 4
|1100i h1100| .
In this state Alice and Bob both measure the state of their qubits belonging to the target
pair. This can be done formally with the set of measurement operators
M00 = I ⊗ I ⊗ |0i h0| ⊗ |0i h0| (2.33a)

M01 = I ⊗ I ⊗ |0i h0| ⊗ |1i h1| (2.33b)
M10 = I ⊗ I ⊗ |1i h1| ⊗ |0i h0| (2.33c)
M11 = I ⊗ I ⊗ |1i h1| ⊗ |1i h1| , (2.33d)
and is equivalent to measuring the observable I ⊗ I ⊗ Z3 ⊗ Z4 as the measurement operators

are constructed from eigenvectors of this observable. The interesting outcome is when both
qubits are measured to 1. The probability for that is

†
p(11) = tr M11 ρout M11 (2.34)
where only the boxed terms of ρout survive being operated on by the M11 and of them
only the two diagonal ones give a nonzero contribution when the trace is taken, giving a
2
probability of p2 . The new state after the measurement is
†
M11 ρout M11 1
= |00i h00| + |00i h11| + |11i h00| + |11i h11| ⊗ |11i h11|
p(11) 2
= |Φ+ i hΦ+ | ⊗ |11i h11| . (2.35)
So by using two entangled pairs in a mixed state Alice and Bob have obtained a pure
2
state where one of the pairs is in a Bell state. The probability for this to happen is p2 , so
2
on average we get a yield of p4 Bell states per input pair.
Some additional Bell states may be squeezed out of the mixed states after using the
above protocol. The measurement result 00, with a probability 23 p2 − 2p + 1, leaves the pair
of control qubits in the state
2(1 − p)2 p2
ρ00 = |01i h01| + |Φ+ i hΦ+ | (2.36)
3p2 − 4p + 2 3p2 − 4p + 2
which is on the same form as the original input state with a new mixing ratio
p2
p′ = . (2.37)
3p2 − 4p + 2
p′ 2
These pairs may be run through the same procedure once more with a yield of 4 , increasing
1 p′ 2 1
the total yield with an amount of · 2 p(00) 4 ,
where the factor comes from the fact that 2
we use two pairs and only get one out. This procedure may be iterated, and the total yield
converges quickly so only few iterations are necessary in practice. Figure 2.2 shows the total
yield calculated numerically as a function of p.
0.35
0.3
0.25
0.2
yield
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
p
Figure 2.2: The yield of the distillation protocol for 1, 2 and 3 iterations. The dotted line is the
asymptotic limit.
In general, distillation protocols do not yield pure state pairs after a finite number of
iterations. Likewise, in the above protocol this only happens when the state can be written
as a mixture of a Bell state and another special state. A general feature, though, is the
application of the bilateral CNOT unitary transformation on two pairs.
The purity of the output states is measured by the fidelity of the state ρ, defined as6
F ≡ hΦ+ | ρ |Φ+ i (2.38)
where |Φ+ i is the desired output state. Applying a distillation protocol to an ensemble of
states, leaves a subensemble where the fidelity is higher, along with a subensemble where the
fidelity is lower, often zero. In the limit of reapplying the procedure to the higher-fidelity
states an infinite number of times, the fidelity approaches unity. This comes at a cost,
6
Sometimes the fidelity is defined as the square root of this.
however, as the yield decreases as we demand higher fidelity. For the simplest protocols
the yield approaches zero as the fidelity approaches unity. But by combining protocols and
using the already distilled states as a resource, the limit can be made finite.
Chapter 3
Characterisation of bipartite
entanglement
Before moving on to quantify the entanglement in a bipartite state, we will try to characterise
it more qualitatively. There are a number of questions which does not have an immediate
answer. Given a state, how can we tell wheter it is entangled? Is there more than one type
of entanglement? Can all entangled states be distilled to Bell states?
3.1 Pure states

When we are given a bipartite state the first question of interest is whether it is entangled or
not. This is the question of separability. A separable state is a state which is not entangled,
so each of the subsystems can be given a description on its own. When it comes to pure
states, a state is separable if and only if it is a product state. As indicated in 2.2, a pure
state is a product state if it can be written as
|Ψprod i = |ψiA ⊗ |φiB (3.1)
for any pure states |ψiA and |φiB belonging to Alice and Bob, respectively. If the state is
not a product state, it is an entangled state.
In practice, though, one needs to find the right basis vectors to be able to write a product
state as in (3.1). A simpler criterion is desired. The Schmidt decomposition (section 2.4)
is helpful in this respect. The number of terms in the Schmidt decomposition is called the
Schmidt number , sometimes also Schmidt rank . That is, the number n in
n
X
|ψi = λk |a′k i ⊗ |b′k i . (3.2)
k=1
The Schmidt number is at most min(dA , dB ), where dA (dB ) is the dimension of subsystem
A (B). But when some of the Schmidt coefficients are zero, those terms may be skipped
giving a lower Schmidt number. Remember that the Schmidt coefficients are simply the
square root of the eigenvalues of any of the reduced density matrices. Hence, the Schmidt
number is simply the number of nonzero eigenvalues of ρA or ρB .
21
22 Chapter 3. Characterisation of bipartite entanglement
What the Schmidt number tells us physically is basically how many degrees of freedom
that are entangled. And as a product state can be written as (3.1) it is obvious that
its Schmidt number is 1. For pure states this is a necessary and sufficient criterion for
separability.
It turns out that entanglement from any pure entangled state can be concentrated into
Bell states. This may be done by using the procedure similar to the one described in
section 2.6.3, or a more advanced procedure that operates on more partially entangled pairs
at the same time to give a higher yield (see e.g. [BBPS96]). Bell states may again be
used to produce other entangled states by means of LOCC. Because of this, and the easy
identification through the Schmidt number, the only classification we make for bipartite
pure states is in separable and entangled states.
3.2 Mixed states

Some properties of mixed states were discussed in section 2.3. The structure of mixed
state bipartite entanglement is richer than its pure state equivalent. For instance there are
entangled states that cannot be distilled to Bell states, and cannot be put to use in any
of the applications where entanglement is a vital resource. Also, the more difficult task of
finding out whether a given mixed state is separable or not gives rise to classes of states
which can be identified by other criteria.
3.2.1 Separable states

Recall from section 2.3 that a bipartite state is called separable if it can be realised as a
mixture of product states,
(A) (B)
X
pi ρi ⊗ ρi (3.3)
i
P
with pi ≥ 0 and i pi = 1. Given a density matrix in some basis the task of finding local
density matrices to satisfy (3.3) can be enormous. Even for the simplest system which may
be entangled, a pair of qubits, a density operator would typically look like
√ √
3 2 2
ρ= 24 |00i h00| + 24 i |00i h01| + 12 i |00i h11|
√ √
2 1 2
− 24 i |01i h00| + 4 |01i h01| − 12 i |01i h10|
√ √ (3.4)
2 5 2
+ 12 i |10i h01| + 24 |10i h10| − 24 i |10i h11|
√ √
2 2 5
− 12 i |11i h00| + 24 i |11i h10| + 12 |11i h11|
or in matrix representation in the {|00i , |01i , |10i , |11i} basis,

√ √
3 2 2
 
24
√ 24 i 0
√ 12 i
− 242 i 1
− 122 i 0
 
ρ= √4 √  . (3.5)
2 5
− 242 i

 0 12 i 24
√ √
− 122 i 0 2
24 i
5
12
3.2. Mixed states 23
It is not easy to see that this can be written as (3.3). Still, it is constructed as a mixture of
two product states. It can be written as
(A) (B) (A) (B)
ρ = p1 ρ1 ⊗ ρ1 + p2 ρ2 ⊗ ρ2

1 1 1 1 1 1
= + +
|1i h1| + |−i h−| ⊗ |ψ 2 i hψ 2 | + I/2 + |+i h+| ⊗ |ψ − −
2 i hψ 2 | (3.6)
2 2 2 3 3 2 2 2 3 3
where r
1 2
|ψ ±
2 i = √ |0i ± i |1i
3 3 3
It could of course be realised by a variety of other mixtures as well.
This example shows that we could well use an operational criterion for separability,
like the Schmidt number for pure states. In general, no easily computable necessary and
sufficient criterion is known. Still, we have a necessary condition which is easy to compute.
This is known as the PPT1 -criterion or Peres-Horodecki criterion [Per96, HHH96]. It
states that if the state is separable, then the partial transpose of the density operator with
respect to one subsystem is positive. A transpose of an operator needs to be taken with
respect to a basis, and the resulting operator depends on the basis. Actually, if AT is
the transpose of the operator with respect to one basis, any operator of the form U AT U †
for unitary U satisfying U = U T (in the same basis) is a transpose of the operator with
respect to another basis. However, since the different bases are connected by a unitary
transformation, the eigenvalues are independent of the basis. Therefore, any orthonormal
basis will do for taking the transposition. More formally, if we choose an orthonormal
product basis {|vi vj i} ≡ {|vi i ⊗ |vj i} for the state ρ, the partial transpose ρTB is defined by
its matrix elements
ρTmµ,nν
B
= hvm vµ | ρTB |vn vν i = ρmν,nµ . (3.7)
In the example matrix (3.5) taking the partial transpose of system B corresponds to trans-
posing within each of the four 2 × 2 blocks, whereas switching the upper right block with
the lower left block corresponds to transposing system A.
For systems of dimension 2 × 2 and 2 × 3 the PPT-criterion is also sufficient for the state
to be separable [HHH96]. For larger systems, though, there exist entangled PPT states
[Hor97]. The PPT-criterion is still so useful that the class of PPT states is an important
class of states, strictly larger than the class of separable states. Anything that can be proved
for all PPT states, is automatically true for separable states.
3.2.2 Free and bound entanglement

Unlike pure states, not all entangled mixed states can be distilled into Bell states [HHH98].
This is an important distinction, because for many applications entanglement must be dis-
tilled in order to be useful. Entanglement that can be distilled is called free entanglement
and entanglement that cannot be distilled is called bound entanglement. Bound entangled
states cannot be written as a convex combination of product states (3.3) and thus cannot
be prepared locally by two classically communicating parties, but its entanglement is well
1
Positive Partial Transpose
24 Chapter 3. Characterisation of bipartite entanglement
hidden and difficult to detect. It has been shown that for a pair of qubits all entangled
states can be distilled [HHH97].
A connection between free entangled states and PPT states was found in [HHH98].
It says that no free entangled state is PPT. Hence, all PPT entangled states are bound
entangled states. This result was actually the first evidence of the existence of bound
entangled states.
This raises the question of whether a nonpositive partial transpose (NPPT) state is
always distillable. The answer to this question is not known, but there are indications that
NPPT bound entangled states do exist [DSS+ 00, DCLB00].
It has been proven that bound entanglement is useless for quantum teleportation [HHH99a],
and it can be proved for other applications as well. One has found, though, that bound en-
tanglement is not completely useless after all. For certain tasks it can be activated [HHH99b]
to facilitate distillation of other quantum states. It has even been shown that a small amount
of PPT bound entanglement as a resource makes it possible to distill entanglement from
any NPPT state [VW02b]. This means that if NPPT bound entangled states really ex-
ist, entanglement can be distilled by putting together two states that are not distillable by
themselves.
Chapter 4
Measures of bipartite entanglement
Entanglement can be utilised for performing many tasks which are impossible without it,
or enhance the performance of other tasks. In most of these tasks entanglement is con-
sumed, so we trade entanglement for something else. It is clear from this that we may treat
entanglement as a resource, just as energy is a resource needed for certain tasks. This is
the reason why we want to quantify the entanglement in an entangled state. Given a state
and a task that consumes entanglement, how much can we achieve? How well can we do?
It is not obvious that an entangled state that performs a given task better than another
is will be the better resource for another task. Indeed, this is only true in a very limited
context. Because of this, we have in general many ways to quantify the nonlocal resources
(or entanglement) in a quantum state.
An entanglement measure E(·) is a functional that takes a quantum state of a multi-
partite system to a nonnegative real number. In this chapter we only consider bipartite
systems, hence
E : D(H) → R+ (4.1)
where D(H) is the set of density operators on the Hilbert space H = HA ⊗ HB . There are
in principle two ways to quantify the entanglement in a state. Operational measures are
based on how well a certain task can be performed, usually compared to Bell states. The
other way, which gives rise to abstract measures, is to work from a set of natural axioms
we believe an entanglement measure should satisfy, and look for functionals that satisfy the
axioms.
There are also two regimes to consider. In the finite regime we consider the resources
contained in a single quantum system, whereas in the asymptotic regime we take into account
an infinite number of systems in the same state and consider the the resources per system.
4.1 Pure states

Before treating the general case of mixed states we will consider the simpler case of en-
tanglement measures of pure states. The treatment is facilitated by the fact that a pure
state contains no classical correlations between the subsystems, so any correlation present
must be of quantum nature. As mentioned in section 2.3, even if an entangled state is
25
26 Chapter 4. Measures of bipartite entanglement
pure the states of the subsystems – described locally by the reduced density operator – are
mixed. For pure states, then, the amount of “mixedness” turn out to be a good measure of
entanglement.
4.1.1 von Neumann entropy

Our measure for how mixed a quantum state ρ is, will be the von Neumann entropy, defined
as
S(ρ) ≡ − tr(ρ log ρ) (4.2)
where we take the logarithm base-2 as is the custom in information theory. It is most easily
calculated from the nonzero eigenvalues λi of ρ as
X
S(ρ) = − λi log λi . (4.3)
i
The von Neumann entropy is often seen as a generalisation of the Shannon entropy
from classical information theory. Although it is true that Shannon entropy arises as a
special case in quantum information theory when only orthogonal states are considered, the
historical influence is the other way around. von Neumann introduced his entropy in 1932
[vN32] and Shannon published his mathematical theory of communication in 1948 [Sha48].
In fact, it was von Neumann who suggested that Shannon should name his uncertainty
function “entropy” as that was already used in statistical physics [TM71, p. 180].
What the von Neumann entropy essentially describes, is the uncertainty in a quantum
state. It is zero for pure states, and smaller for a mixture of two nonorthogonal states
than a similar mixture of two orthogonal states. It is not only an analogue to the entropy
known from thermodynamics, it is the same quantity (up to a constant with the definition
(4.2)). Indeed, it was derived using notions such as the ideal gas law and other laws of
thermodynamics [vN32, p. 191-202]1 .
4.1.2 Reduced von Neumann entropy – entropy of entanglement

For the purpose of quantifying entanglement, the important quantity is the reduced von
Neumann entropy. This is the von Neumann entropy of the reduced density matrix. We
saw an example in section (2.3) that the reduced density operator of a pure entangled
state represents a mixed state. The von Neumann entropy of either of the subsystems (the
entropies are equal) is a good measure of entanglement. Because of this, the reduced von
Neumann entropy is also known under the name entropy of entanglement EE .
It is easy to see that the reduced von Neumann entropy is equal for both reduced density
matrices. The von Neumann entropy only depends on the eigenvalues of the density matrix
(4.3) and because of the existence of the Schmidt decomposition, the eigenvalues are equal,
as was mentioned in section 2.4.
We haven’t yet pointed out what constitutes a good measure, but the entropy of entan-
glement has some properties we find natural. (i) It is zero for any product state, (ii) it is
maximal when the reduced density matrix is completely mixed, e.g. when the subsystems
1
Pages 359-379 in the far worse typeset English translation [vN55].
4.1. Pure states 27
have no individual properties (for the degrees of freedom considered) and (iii) it is invariant
under local unitary transformations. The above requirements were introduced in [SM95],
but requirements have later been refined and placed on firmer ground for both pure and
mixed states.
The entropy of entanglement was first introduced as a measure of entanglement in
[BBP+ 96]. It is an abstract measure in the sense that it satisfies some requirements, and
it does not have an immediate operational interpretation. We will now take a look at two
pure state entanglement measures which have a direct operational meaning.
4.1.3 Entanglement cost and distillable entanglement

We saw in section 2.6.3 an example of a concentration of pure state entanglement. We turned
a partly entangled two-qubit state into a Bell state with a yield depending on how entangled
the initial state was. There are other protocols which give a higher yield if we have more
than one pair to operate on (see e.g. [BBPS96]). The distillable entanglement Ed is defined
as the maximum yield of Bell states that can be obtained, optimised over all possible LOCC
protocols. The distillable entanglement is also sometimes called entanglement of distillation
The entanglement cost Ec is the dual to the distillable entanglement. Two separated
parties cannot prepare an entangled state if they can only communicate classically. But if
they have some entangled state in some standard form like Bell states, they can convert those
into the desired entangled state. Thus, the entanglement cost is defined as the minimum
number of Bell states needed to create a given state by means of LOCC.
The above definitions were implicitly given for the finite regime. It turns out that in
this regime Ed and Ec are hard to calculate even though the states are pure. What is clear
intuitively is that for any state
Ed ≤ Ec . (4.4)
If not, one could create entanglement by means of LOCC by converting Bell states to a
state not satisfying (4.4) and converting them back again.
In the asymptotic limit, the results are known. We define the regularised (or asymptotic)
versions of distillable entanglement and entanglement cost as
Ed (|ψi⊗n )
ED (|ψi) ≡ Ed∞ (|ψi) ≡ lim ; (4.5a)
n→∞ n
Ec (|ψi⊗n )
EC (|ψi) ≡ Ec∞ (|ψi) ≡ lim . (4.5b)
n→∞ n
In this case, it was shown in [BBPS96] that both the distillable entanglement and entangle-
ment cost is equal to the entropy of entanglement, ED (|ψi) = EC (|ψi) = EE (|ψi). In fact,
if two states |ψ1 i and |ψ2 i have the same entropy of entanglement, they can be intercon-
verted with efficiency approaching unity as n → ∞. If they do not have the same entropy
of entanglement, |ψ1 i can be converted into |ψ2 i with asymptotic yield EE (|ψ1 i)/EE (|ψ2 i).
It has even been shown that this conversion can be done such that the amount of classical
√
communication required per Bell state produced approaches zero [LP99] (it scales as n).
Despite the fact that the entropy of entanglement did not have an operational interpre-
tation ab initio, we see that it indeed plays a fundamental role. The fact that the entropy
of entanglement can be asymptotically conserved in the conversion of states using LOCC,

places huge constraints on other candidates for pure state asymptotic entanglement mea-
sures. A measure would not live up to our expectations if we could increase it using only
LOCC. If another measure were to give two different values for states with the same en-
tropy of entanglement, it would be possible increase it by LOCC by converting the state
with lower value to the one with higher value. The conversion between states with con-
version rate EE (|ψ1 i)/EE (|ψ2 i) imposes even stronger constraints, so we would expect any
asymptotic pure state measure to coincide with the entropy of entanglement. The exact
conditions for this to happen are discussed in the next section.
4.1.4 The uniqueness theorem for measures of entanglement

The fact that the pure state asymptotic entanglement cost and distillable entanglement
coincide with the entropy of entanglement is a special case of a more general property of
entanglement measures. The uniqueness theorem for entanglement measures states that any
pure state measure of entanglement that satisfies certain natural criteria coincides with the
entropy of entanglement. The criteria are considered to be too strict for the finite regime,
but all natural asymptotic measures satisfy the conditions.
The uniqueness theorem has been developed gradually in the literature, in the beginning
only as an intuitive idea of the same character as given in the previous section [PR97].
Later, uniqueness was proved from a set of conditions that measures should satisfy [Vid00,
HHH00]. In [DHR02] the minimal conditions for the theorem to hold were found. The
conditions come in different versions, some weaker than others. It turns out that some of
the stronger conditions are not necessary for the theorem, but are in fact implied by the
weaker conditions. The strongest versions are listed first.
(P0) If |ψi is separable (i.e. a product state), then E(|ψi) = 0
(P1a) (Normalisation) For the maximally entangled state in d × d dimensions
d
X 1
|Φ+
di = √ |iiA ⊗ |iiB (4.6)
i=1
d
where the two sets {|iiA } and {|iiB } are orthonormal bases for the two subsystems, E(|Φ+ d i) =
log d, where we as usual take the logarithm base-2.
(P1b) For the Bell state |Φ+ i = √12 (|00i + |11i), E(|Φ+ i) = 1. The previous condition
is simply the generalisation of this.
(P2) For any operation Λ that can be implemented by means of LOCC and any |ψi such
that Λ(|ψi) is a pure state, E(Λ(|ψi)) ≤ E(|ψi).
(P3) (Continuity) Let {|ψn i} and {|φn i} be sequences of pure bipartite states living
on the sequence of Hilbert spaces {Hn }. For all such sequences
√ such that k |ψn i hψn | −
†
|φn i hφn | k1 → 0 where k · k1 is the trace norm kAk1 = tr( A A),
E(|ψi) − E(|φi)
→ 0. (4.7)
1 + log(dim Hn )
(P4a) (Weak additivity) For all pure states |ψi and n ≥ 1,
E(|ψi⊗n )
= E(|ψi) (4.8)
n
(P4b) (Asymptotic weak additivity) Given ǫ > 0 and a pure state |ψi, there exists an
integer N > 0 such that for all n ≥ N ,
E(|ψi⊗n ) E(|ψi⊗n )
− ǫ ≤ E(|ψi) ≤ + ǫ. (4.9)
n n
(P4c) (Existence of a regularisation) For all bipartite pure states |ψi, the limit
E(|ψi⊗n )
E ∞ (|ψi) ≡ lim (4.10)
n→∞ n
exists.
With these conditions given, the uniqueness theorem for entanglement measures states
[DHR02] that for a functional E on pure states, the following are equivalent:
(i) E satisfies (P1b), (P2), (P3) and (P4b).
(ii) E satisfies (P0), (P1a), (P2), (P3) and (P4a).
(iii) E coincides with the entropy of entanglement E = EE .
On the other hand, if E satisfies (P0), (P1a), (P2) and (P3), it automatically satisfies (P4c),
and E ∞ = EE .
(i) represents the weakest set of conditions. When they are satisfied, the stronger con-
ditions in (ii) are also satisfied. The connection is that (P4a) and (P4b) turn out to be
equivalent ((P4a) ⇒ (P4b) is obvious) and (P1b), (P2) and (P4a) together implies (P0) and
(P1a). The uniqueness for the asymptotic regime, previously mentioned, is concretised by
the last sentence.
4.2 Mixed states

The measures of pure state bipartite entanglement constitute a very limited subset of the
measures of generally mixed states. For mixed states there is no unique way to quantify the
nonlocal quantum resources even in the asymptotic regime. In the pure state asymptotic
limit, the entropy of entanglement imposed a total order on the set of states. Any state
could be converted into another state with equal or less entanglement. In the mixed state
case there are states where neither can be converted into the other, giving only a partial
order on the set of states. (For a longer discussion on entanglement measures as ordering
of the states, see [MSV04].) This makes the structure of mixed state entanglement richer
than that of pure states.
4.2.1 Entanglement cost and distillable entanglement for mixed states

The entanglement cost and distillable entanglement can easily be defined for mixed states in
the same way as for pure states. The entanglement cost Ec is the minimum number of Bell
states needed to produce the state by means of LOCC and the distillable entanglement Ed
is the maximum number of Bell states that can be distilled by an optimal LOCC distillation
protocol. However, a perfect conversion between mixed states in the finite regime is usually
not possible, so in the following we will only consider the asymptotic versions ED and EC .
There is a tweak, though, compared to the pure state case. As mentioned in the discussion
of entanglement distillation (section 2.6.4) we do not require the conversion to be perfect.
The only requirement is that the fidelity approaches unity as n → ∞.
Note that to produce mixed states from Bell states using perfect LOCC we will need to
discard some information. This can be done by doing a random local unitary transformation
and not recording which one it was, or by doing a measurement and forgetting the outcome.
A third option is to entangle the local system with an ancilla system and discard the ancilla.
It is not surprising then that the distillable entanglement usually is strictly smaller than the
entanglement cost. Actually, for any measure E which satisfies certain natural conditions
for asymptotic measures, the entanglement cost and distillable entanglement provide upper
and lower bounds [HHH00]2 ,
ED ≤ E ≤ EC (4.11)
The bound entangled states, discussed in section 3.2.2, is an example where the difference
between EC and ED is highly visible. ED is obviously zero, but EC is finite. This illustrates
that one single measure is not enough to quantify the entanglement resources in a mixed
quantum state, even in the asymptotic limit.
4.2.2 Axioms for abstract measures

The axiomatic approach to quantifying entanglement has been quite successful in finding
measures that quantify certain aspects of entanglement. During the course of finding the
right conditions, some conditions have been discarded as unnecessary and others have been
restricted to a specific regime. The common denominator, though, has always been that
entanglement, irrespective of how we quantify it, should not increase on average by any
LOCC operation, i.e. it is monotonic under LOCC. The first attempt was done by Vedral et
al. in [VPRK97] and was slightly improved by some of the same authors in [VP98]. Vidal
[Vid00] argued that the only absolute requirement was monotonicity under LOCC. This au-
tomatically implies other properties such as nonchange under local unitary transformations,
convexity and constancy on separable states.
The conditions on mixed states discussed here are taken from [DHR02]. The previously
discussed conditions on pure states are specialisations of those.
(E0a) E(ρ) = 0 if and only if ρ is separable. This is a useful property for measures that
satisfy it, but it is too strict in general. Bound entangled states, for instance, have zero
distillable entanglement, but are not separable.
(E0b) E(ρ) = 0 if ρ is separable.
The normalisation conditions are the same as for pure states.
(E1a) (Normalisation) For the maximally entangled state in d×d dimensions, |Φ+ +
d i hΦd |,
2
In [HHH00] the upper bound was EF , the regularised entanglement of formation. This has been shown
to be equal to the asymptotic entanglement cost [HHT01].
where
d
X 1
|Φ+
di= √ |iiA ⊗ |iiB (4.12)
i=1
d
and the two sets {|iiA } and {|iiB } are orthonormal bases, E(|Φ+ +
d i hΦd |) = log d.
+ 1 + +
(E1b) For a Bell state |Φ i = √2 (|00i + |11i), E(|Φ i hΦ |) = 1.
(E2a) (LOCC monotonicity) For any LOCC operation Λ, E(Λ(ρ)) ≤ E(ρ).
(E2b) When Λ is a strictly local operation which is either unitary or adds extra dimen-
sions, then E(Λ(ρ)) = E(ρ).
(E2c) When Λ is a strictly local unitary operation, then E(Λ(ρ)) = E(ρ).
(E3a) (Continuity) Let {ρn } and {σn } be sequences of bipartite states living on the
sequence of Hilbert spaces {Hn }. √ For all such sequences such that kρn − σn k1 → 0 where
k · k1 is the trace norm kAk1 = tr( A† A),
E(ρn ) − E(σn )
→ 0. (4.13)
1 + log(dim Hn )
(E3b) The previous condition (E3a) is weakened by only requiring it to be satisfied for
approximations to pure sates. That is, (E3a) only needs to hold when the ρn are pure.
Many measures satisfy the weak additivity condition, but it is not considered necessary
for a good measure.
(E4a) (Weak additivity) For all states ρ and n ≥ 1,
E(ρ⊗n )
= E(ρ). (4.14)
n
(E4b) (Asymptotic weak additivity) Given ǫ > 0 and a state ρ, there exists an integer
N > 0 such that for all n ≥ N ,
E(ρ⊗n ) E(ρ⊗n )
− ǫ ≤ E(ρ) ≤ + ǫ. (4.15)
n n
(E5a) (Subadditivity) For all states ρ and σ,
E(ρ ⊗ σ) ≤ E(ρ) + E(σ). (4.16)
(E5b) A special case of the above is when ρ and σ are different numbers of the same
state tensored together. For all states ρ, and m, n ≥ 1,
E(ρ⊗(m+n) ) ≤ E(ρ⊗m ) + E(ρ⊗n ). (4.17)
(E5c) (Existence of a regularisation) For all bipartite states ρ, the limit
E(ρ⊗n )
E ∞ (ρ) ≡ lim (4.18)
n→∞ n
exists. E ∞ is called the regularisation of E. This is the weakest of the additivity conditions,
and is usually satisfied.
(E6a) (Convexity) Mixing of states does not increase entanglement. For all bipartite
states ρ and σ,
E(λρ + (1 − λ)σ) ≤ λE(ρ) + (1 − λ)E(σ) (4.19)
for 0 ≤ λ ≤ 1.
Although (E6a) seems very natural as forgetting which of a set of prepared states you
have should not increase entanglement, there are indications that the (asymptotic) distillable
entanglement is not convex [SST01]. This is based on the fact that PPT bound entangled
states together with NPPT bound entangled sates (if they exist) can be distilled. Thus,
even a mixture of two states with zero distillable entanglement can be distilled. A weaker
condition which is shown in [DHR02] to hold also for the distillable entanglement is that
the convexity need only hold on decompositions into pure states. P
(E6b)
P For any bipartite state and any pure state realisation ρ = i pi |ψi i hψi |, pi ≥ 0
and i pi = 1, X
E(ρ) ≤ pi E(|ψi i hψi |). (4.20)
i
Like for pure states, asymptotic weak additivity is sufficient for weak additivity, so (E4a)
and (E4b) are actually equivalent. The same goes for the subadditivity conditions (E5a)
and (E5b).
4.2.3 ED and EC as extreme measures

In [DHR02] the exact conditions for ED and EC to be the lower and upper bound for an
entanglement measure were investigated. Three versions of the theorem were found. With
different conditions on the measures either the nonregularised or the regularised measure
would be bounded by ED and EC .
(i) For an entanglement measure E satisfying (E1a), (E2a), (E3a) and (E4b) (and thereby
(E4a)), for all states ρ,
ED (ρ) ≤ E(ρ) ≤ EC (ρ). (4.21)
These conditions are very strong, and until recently no function was known that satisfied
them. The “squashed entanglement” [CW04], however, satisfies all conditions.
(ii) For an entanglement measure E satisfying (E1a), (E2a), (E3a) and (E5c), then for
all states ρ, the regularised version E ∞ (which exists by (E5c) and always satisfies (E4a),
but not necessarily (E3a)) is bounded by ED and EC .
ED (ρ) ≤ E ∞ (ρ) ≤ EC (ρ) (4.22)
These conditions are easier to satisfy, and they are satisfied by both the entanglement of
formation and the relative entropy of entanglement.
(iii) Let E be an entanglement measure satisfying (E1a), (E2a), (E3b) and (E6b). Then
if weak additivity (E4a) holds,
ED (ρ) ≤ E(ρ) ≤ EC (ρ) (4.23)
and if subadditivity (E5a) holds,
ED (ρ) ≤ E ∞ (ρ) ≤ EC (ρ). (4.24)
4.2.4 Relative entropy of entanglement and other distance based mea-

sures
Along with the first attempts to give realistic conditions for entanglement measures [VPRK97,
VP98], a class of measures that satisfied the conditions was introduced. The conditions con-
sidered was that the measure would be zero if and only if the state is separable (E0a), that
local unitary operations leave it constant (E2c) and that its expectation value does not
increase under LOCC (E2a).
The class of measures is based on some distance function (but not necessarily a metric)
on the set of density matrices. The measure is then the distance from the state in question
to the nearest separable state. That is, for a distance function D(ρ, σ) an entanglement
measure corresponding to the distance function could be defined as
E(ρ) ≡ inf D(ρ, σ) (4.25)

σ∈S(H)
where S(H) is the set of separable states on H. The conditions imposed on E is now
converted into conditions on D.
The quantum relative entropy 3 can be used as a distance function on the set of density
operators. It is defined as
S(ρkσ) ≡ tr[ρ(log ρ − log σ)]. (4.26)
The relative entropy is not a metric and is not even symmetric. It is nonnegative, and
zero only for identical density operators. The same unitary operation on both states leaves
it invariant. The quantum relative entropy can be interpreted as a distinguishability of
quantum states. More precisely, suppose we are given a large (but finite) number n of
quantum systems that are all in the same state, which is either ρ or σ. Our task is to perform
measurements to infer which. The probability for inferring from optimal measurements on
the composite system, that the given state is ρ when it really is σ is [VPJK97]
Pn (σ → ρ) = 2−nS(ρkσ) (4.27)
(for large n). This expression is not symmetric in ρ and σ.

To illustrate this, consider the qubit states
1 1 1
ρ = |0i h0| and σ= |0i h0| + |1i h1| = I2
2 2 2
i.e. one arbitrary pure state (denoted |0i) and the completely mixed state. These states are
optimally distinguished by measurement if we measure each particle in the {|0i , |1i} basis.
In the case that we are performing measurements on σ, the probability for measuring |0i and
|1i are both 12 . If all measurements give |0i, we would wrongly conclude that the state was
ρ. The probability for that to happen is 2−n . Conversely, if we perform the measurements
on systems in state ρ, we will always get the result |0i, so the probability for confusing it
with σ is zero.
3
For reviews of the role of the relative entropy in quantum information theory, see [Ved02, SW02].
The quantum relative entropy can be calculated from the eigenvalues and eigenvectors of
the density operators as shown in Appendix A. In the above example, it can be calculated
from (A.4) giving S(ρkσ) = 1 and S(σkρ) = +∞. This is consistent with the probabilities
above and (4.27). It is a general feature that S(σkρ) = +∞ when ρ is a pure state. This is
because for any pure state there is a complete measurement for which the outcome is 100%
certain, and using this measurement would make it impossible to confuse it with another
state.
This property alone makes the relative entropy with the arguments reversed, D(ρ, σ) =
S(σkρ), unusable in (4.25) as a distance function. It would give infinity for any pure en-
tangled ρ. S(ρkσ) on the other hand, behaves well for separable σ and the entanglement
measure generated by it,
Er (ρ) ≡ inf S(ρkσ), (4.28)
σ∈S(H)
is called the relative entropy of entanglement [VP98]. It could be said that it quantifies
the unlikelihood for a separable state to give measurement outcomes consistent with the
entangled state.
From its definition and the fact that S(ρkσ) = 0 only when ρ = σ, it is obvious that
the relative entropy of entanglement satisfies (E0a) (Er (ρ) = 0 iff ρ ∈ S(H)). It reduces to
the entropy of entanglement on pure states [VP98], and therefore satisfies the normalisation
criterion (E1a) and continuity on pure states (E3b) by the uniqueness theorem. It is nonin-
creasing under LOCC operations (E2a) [VPRK97]. Continuity (E3a) was shown in [DH99].
There have been found counterexamples to weak additivity (E4a) [VW01] and thereby also
to asymptotic weak additivity (E4b). Actually, for very high dimension of a certain class
of states, Er (ρ) ≈ Er (ρ ⊗ ρ). It can also be shown that it is both subadditive (E5a) and
convex (E6a) [DHR02].
The regularisation (E5c)
Er (ρ⊗n )
ER (ρ) ≡ Er∞ = lim (4.29)
n→∞ n
of the relative entropy of entanglement satisfies automatically (E0a), (E1a), (E2a), (E5a)
and (E6a) as those conditions extends to any regularisation of a measure that satisfies them
[DHR02]. In addition, as any regularisation does, it is satisfies weak additivity (E4a).
Any distance based measure, including the relative entropy of entanglement, can be
varied by considering other classes of states than the separable ones. What is important is
that the class is closed under LOCC. For instance, the set can be extended to PPT states.
This will give a measure which is easier to calculate and inferior or equal to the version for
separable states. For 2 × 2 and 2 × 3 dimensional systems the measures coincide as the PPT
is then equivalent to separability.
Even though the quantum relative entropy is by far the most used of the distance based
measures, other distance functions provide usable entanglement measures as well. Another
measure of distinguishability of quantum states is the Bures metric [Bur69, Hüb92, FC95],
in this context defined as [VP98]
p
DB (ρkσ) ≡ 2 − 2 F (ρ, σ) (4.30)
where h n √ √ oi2
1/2
F (ρ, σ) ≡ tr σρ σ (4.31)
is called Uhlmann’s transition probability [Uhl76]. The Bures metric is a true metric and
thus symmetric. It is a generalisation of the pure state Fubini-Study metric to mixed
states [Hüb92]. The entanglement measure generated by it satisfies (E0a), (E1b) and (E2a).
Unlike the relative entropy of entanglement, however, it doesn’t reduce to the entropy of
entanglement on pure states. Actually, it is smaller than the entropy of entanglement
[VP98].
There may be other distance functions that generate entanglement measures with nice
properties, but the two mentioned are the ones used in practice. The Hilbert-Schmidt
distance
DHS (ρ, σ) ≡ kρ − σk2HS = tr (ρ − σ)2

(4.32)
was suggested as a candidate [VP98, WT99]. However, it was shown by Ozawa [Oza00] that
the distance violated one of the criteria it was conjectured to have (nonexpansion under
physical operations), so it did not satisfy the sufficient conditions for it to be nonincreasing
under LOCC. On the other hand, the very similar trace norm distance
hp i
DT (ρ, σ) ≡ kρ − σk1 = tr (ρ − σ)2 (4.33)
has been shown to generate an entanglement measure which is monotonic under LOCC
[EAP03].
4.2.5 Entanglement of formation

The entanglement of formation was historically the first entanglement measure to appear
[BDSW96]. In the literature it is occasionally called entanglement of creation. It was meant
to be the asymptotic entanglement cost, but this interpretation rests on the weak additivity
which is strongly conjectured, but still not proved. However, the regularisation of it has
been shown to be equal to the asymptotic entanglement cost [HHT01].
The entanglement of formation is a straightforward generalisation of the entropy of
entanglement to mixed states. Remember that in the asymptotic limit, the entanglement
cost in Bell states of preparing a pure state is given by the entropy of entanglement. Thus,
it was natural to define the entanglement of formation for a pure state as the entropy
of entanglement. For a given ensemble of pure states E = {pi , |ψi i} the entanglement of
formation is the average of the entropy of entanglement for the states in the ensemble
X
Ef (E) ≡ pi EE (|ψi i). (4.34)
i
A mixed state can be realised by a multitude of pure state ensembles, with different en-
tanglement of formation. As any of those ensembles realises the mixed state, the natural
definition for the entanglement of formation for a mixed state is the entanglement of forma-
tion for the “most economic” ensemble. That is, the entanglement of formation for a mixed
state is defined as X
Ef (ρ) ≡ inf pi EE (|ψi i) (4.35)
E
i
where the infimum is taken over all ensembles ε = {pi , |ψi i} that realises the state ρ.
How come entanglement of formation is not necessarily equal to the entanglement cost?
It is easy to see that the asymptotic entanglement cost cannot be greater than the entan-
glement of formation. Imagine that we are to produce a large number nout (nout → ∞) of
some bipartite state ρout . We find the pure state realisation {pi , |ψi i} of ρout that has the
lowest average entropy of entanglement (equal to the entanglement of formation). Then we
produce this ensemble by producing the pure states in amounts corresponding to the prob-
ability distribution {pi }. In the asymptotic limit this can be done reversibly. When we mix
the produced states (discard the information saying which is which), we have a collection of
systems in the state ρout . Since the entropy of entanglement is conserved in the first step,
and EE = 1 for Bell states, we have produced systems in state ρout at a cost equal to the
entanglement of formation.
The question is if we can do better than this. Intuitively the above is the most economical
way, as any other realisation of ρout would give a higher or equal cost. But there is another
way of producing a large number of ρout . We can produce nout /N copies of the state ρ⊗N out .
It is not known whether the optimal pure state realisation of this state will have an average
entropy of entanglement N times that of the state ρout . In other words it is not known if
Ef is weakly additive, Ef (ρ) = N1 Ef (ρ⊗N ). If it is, then Ef = EF ≡ Ef∞ which is equal to
the asymptotic entanglement cost EC . Additivity has been shown for some special classes
of states [BN01, VW01, VDC02, Shi02], but whether it holds in general, is one of the big
open questions in quantum information theory.
From the definition of Ef , it is easy to see that it satisfies (E0a) and (E1a). (E0a) by
the fact that the optimal pure state decomposition of a separable states is into a mixture
of product states, (E1a) by the fact that it reduces to the entropy of entanglement on
pure states. LOCC monotonicity (E2a) was shown when it was introduced [BDSW96], and
continuity (E3a) was shown in [Nie00]. Subadditivity (E5a) follows from the discussion of
additivity above, one may use the same type of decomposition for a tensor product, but one
might be able to do better, which would give a lower Ef . Convexity (E6a) follows more or
less directly from the definition.
The way the entropy of entanglement was extended to mixed states in the case of the
entanglement of formation, can also be used for other measures on pure states. Measures
constructed this way are called convex roof measures [Uhl98, Hor01]. The method extends
a measure of some set (here pure states) to the convex hull4 (here mixed states), where it
is the largest function that is convex and compatible with the measure on the original set.
The method was used in [LCOK03] construct an entanglement measure that extends the
entanglement measure called negativity to mixed states in another way than the original
measure.
A concept related to the entanglement of formation is the concurrence [HW97, Woo98].
It is defined for a system of two qubits. For a general state ρ of two qubits, let ρ̃ be the
spin-flipped state
ρ̃ ≡ (Y ⊗ Y )ρ∗ (Y ⊗ Y ) (4.36)
where the Y is the Pauli Y operator and ρ∗ is the complex conjugate of ρ, both taken in
4
The convex hull of a set is the set of all elements that can be written as a convex combination of the
original set. For instance, the convex hull of two points is the line connecting them.
the standard basis {|00i , |01i , |10i , |11i}. Let the Hermitian matrix R be defined as
√ √
q
R≡ ρρ̃ ρ (4.37)
with eigenvalues in decreasing order {λi }. The concurrence is then defined as
C(ρ) ≡ max{0, λ1 − λ2 − λ3 − λ4 }. (4.38)
The concurrence is monotonic under LOCC, and can thus be used as an entanglement
measure for two qubits. The great advantage is that it is easily computable. But more
important is that it is directly related to the entanglement of formation, providing an explicit
formula for the entanglement of formation in the case of two qubits. Let the function E be
given by
√ !
1 + 1 − C2
E(C) ≡ h (4.39)
2
where h is the binary entropy function
h(x) ≡ −x log x − (1 − x) log(1 − x). (4.40)
Then the entanglement of formation is simply given by [HW97, Woo98]
Ef (ρ) = E(C(ρ)). (4.41)
There have been various attempts to generalise the concept of concurrence. Uhlmann
[Uhl00] considered general conjugations instead of the special (4.36). This was further
generalised to concurrence vectors by Auderaert et al. [AVD01]. The previously mentioned
convex roof extended negativity coincides with the concurrence in the case of two qubits
[LCOK03], and thus provides another generalisation. Some of those generalisations – along
with other aspects of concurrence and entanglement of formation – are reviewed in [Woo01].
4.2.6 Negativity
The entanglement measures discussed so far have all had a serious drawback. Their definition
includes some kind of optimisation, which make their evaluation very difficult. In fact,
expressions for the relative entropy of entanglement and entanglement of formation has
only been calculated for highly symmetric states (e.g. [VP98, TV00, VW01, ADVW02])
and low dimensional cases (e.g. [HW97, Woo98]). For general states, evaluating a measure
includes heavy numerical calculations. With the aim of introducing a computable measure
of entanglement, two related quantities were defined by Vidal and Werner [VW02a]. They
can both be seen as a quantification of the PPT criterion for separability, and because of
that they have the disadvantage that they fail to distinguish between separable states and
entangled PPT states. Both quantities are based on the trace norm of the partial transpose
(in some basis) of a state, kρTB k1 , which can easily be calculated using standard linear
algebra packages. The first quantity is the negativity
kρTB k1 − 1
N (ρ) ≡ . (4.42)
2
This quantity is equal to the absolute value of the sum of negative eigenvalues of ρTB . It
was first introduced in [ŻHSL98, Appendix B] (without the factor 21 ) and later shown to
not increase under LOCC [VW02a]. The other quantity is the logarithmic negativity
EN (ρ) ≡ log kρTB k1 , (4.43)
which is not strictly monotonic under LOCC, but does not increase for a subclass of LOCC.
For all separable states (and more generally all PPT states), both N and EN vanish.
This is easily seen from the fact that the partial transpose operation does not change the
trace of the density operator. If ρTB is a positive operator (i.e ρ satisfies the PPT criterion
for separability), then kρTB k1 = 1, since kρTB k1 is the sum of the absolute values of the
eigenvalues of ρTB and all eigenvalues are positive. For states violating the PPT criterion,
kρTB k1 > 1. In this sense, both quantities quantify how much the state violates the PPT
criterion.
A lot of properties of N and EN were deduced in [VW02a]. As already mentioned,
N is zero for all separable states, but is also zero for PPT entangled states. Hence, it
satisfies (E0b) but not (E0a). It is 12 for Bell states, so it doesn’t satisfy the normalisation
criteria either. It could be made to satisfy (E1b) by defining it as the double value, but the
normalisation wouldn’t extend to maximally entangled states in dimension d × d (E1a). It
is monotonic under LOCC (E2a), convex (E6a), and superadditive, satisfying N (ρ ⊗ σ) =
N (ρ) + N (σ) + 2N (ρ)N (σ). Thus, neither additivity of any kind (E4a) nor subadditivity
(E5a) is satisfied.
Like N , logarithmic negativity EN satisfies (E0b) and not (E0a). However, unlike N it
also satisfies the normalisation condition (E1a). But it does not coincide with the entropy
of entanglement for pure states in general. Actually EN ≥ EE on pure states, with equality
only for maximally entangled states. It is not monotonic under LOCC (E2a), but is an
upper bound for the distillable entanglement. Nor is it convex (E6a). The most appealing
property of EN is probably that it satisfies (strong) additivity, EN (ρ⊗σ) = EN (ρ)+EN (σ).
From this, weak additivity (E4a) and subadditivity (E5a) comes for free.
The negativity can be generalised to give rise to other entanglement measures. This is
done by replacing kρTB k1 by another norm kρk. The generalisation comes naturally if we
define the trace norm in another, but equivalent way. Any Hermitian matrix can be written
as a difference of two positive operators,
A = a+ ρ+ − a− ρ− (4.44)
where ρ± are density matrices and a± are nonnegative numbers. There exists a decompo-
sition of the form (4.44) which minimises a+ + a− . For this decomposition the trace norm
is kAk1 = a+ + a− , and a− is the absolute sum of the negative eigenvalues of A [VW02a].
From this we get another expression for the negativity, namely
N (A) = inf{a− | ATB = a+ ρ+ − a− ρ− }. (4.45)
Minimising a− is in this case the same as minimising a+ + a− , since by taking the trace of
the condition, we get a+ = a− + tr(ATB ). If ATB is a density matrix, A is PPT, and we can
take a− to be zero, so the negativity vanishes as it should for PPT states.
The condition ATB = a+ ρ+ − a− ρ− restricts the set of density matrices ρ± over which
we can take the infimum. Another norm and another negativity can be obtained by taking
another set S from which ρ± must be picked. Norms that can be defined in this way
corresponding to a compact set S are called base norms. The norm corresponding to a
general compact set S is then
kAkS = inf{a+ + a− | A = a+ ρ+ − a− ρ− , a± ≥ 0, ρ± ∈ S} (4.46)
and the corresponding negativity
NS (A) = inf{a− | A = a+ ρ+ − a− ρ− , a± ≥ 0, ρ± ∈ S}. (4.47)
We get back the usual trace norm and negativity from this by taking S as the set of all
Hermitian (not necessarily positive) matrices with unit trace such that the partial transpose
is positive. In mathematical terms, S = {A | A = A† , tr(A) = 1, ATB ≥ 0}.
By choosing S as the set of all separable states, another entanglement measure arises.
It was described prior to the introduction of negativity and was introduced in [VT99]. It
is called robustness of entanglement and was defined as the minimum amount of mixing
with locally prepared states needed to make the state separable. Also other measures have
been related to negativities [Hor01], such as the measures introduced by Rudolph based on
the greatest cross norm [Rud01]. It should be noted that only the original negativity has
the advantage of being easily computable. The other negativities need to be evaluated by
finding the infimum over a large parameter space.
4.2.7 Squashed entanglement

Of the entanglement measures considered so far, only few are additive, and when they
are some of the other greatly desired conditions are not fulfilled. Quite recently a new
entanglement measure was proposed, having more nice properties than its predecessors. It
is motivated from the so called intrinsic information of classical cryptography and is called
squashed entanglement [CW04]. It is defined as

1 ABE
Esq (ρ) ≡ inf I(A; B|E) | ρ extension of ρ to HE (4.48)
2
The infimum is taken over all extensions to a third subsystem E, so ρ ≡ ρAB = trE (ρABE ).
I(A; B|E) is the quantum mutual conditional information defined as
I(A; B|E) ≡ S(ρAE ) + S(ρBE ) − S(ρABE ) − S(ρE ). (4.49)
Most of the properties of Esq were shown along with its introduction as an entanglement
measure in [CW04]. The squashed entanglement vanishes on all separable states (E0b), but
it is not known whether it also vanishes for some entangled states (E0a). If there exists
an extension to HE with dim HE < ∞ for which I(A; B|E) vanishes, it implies that ρAB
is separable. But the infimum in (4.48) is taken over all extensions, and thus Esq (ρAB )
may be zero even if no finite dimensional extension makes I(A; B|E) zero. It coincides with
the entropy of entanglement on pure states, and thus satisfies the normalisation condition
(E1a). It is also bounded from below by the distillable entanglement and from above by the
entanglement cost. As a proper entanglement measure it does not increase under LOCC
(E2a) and it is convex (E6a). When it was introduced, it was shown to be continuous on
most of the state space, and indications that it was continuous on the whole state space were
presented. Continuity was finally proved by Alicki and Fannes [AF04] (E3a). The measure
is strongly additive, making it satisfy both weak additivity (E4a) and subadditivity (E5a).
So except from (E0a), which is still unknown, squashed entanglement satisfies the strictest
versions of all conditions presented.
In [VW01] various additivity properties were discussed. The additivities we have pre-
sented so far are for independent pairs, described by tensor products. But if there are
entanglement or classical correlations between the pairs as well, they may be taken advan-
′ ′
tage of. Consider the state ρABA B which lives on the Hilbert space HA ⊗ HB ⊗ HA′ ⊗ HB ′ ,
where HA ⊗ HA′ is controlled by Alice and HB ⊗ HB ′ by Bob. Each of the primed and
unprimed subsystems are individually described by tracing out the other,
′ ′ ′ ′ ′ ′
ρAB = trA′ B ′ (ρABA B ), ρA B = trAB (ρABA B ), (4.50)
′ ′ ′ ′
and if they are independent ρABA B = ρAB ⊗ ρA B . For a strongly additive entanglement
′ ′ ′ ′
measure this implies E(ρABA B ) = E(ρAB )+E(ρA B ). But if there are correlations between
the pairs that can be exploited, we would expect
′ ′ ′ ′
E(ρABA B ) ≥ E(ρAB ) + E(ρA B ). (4.51)
This property is called strong superadditivity [VW01] and is satisfied by the squashed en-
tanglement [CW04].
Like most other entanglement measures, the definition of squashed entanglement involves
an optimalisation which makes the measure hard to compute. Due to its recent introduction,
not much work has been published about its calculation on special sets of states, nor about
efficient numerical algorithms. One of the important open questions is whether it vanishes
on any entangled states, especially on the bound entangled ones.
4.2.8 Other entanglement measures

In addition to the entanglement measures discussed, other measures have been proposed
without gaining the same importance. One of them is the entanglement of assistance [Coh98,
DFM+ 99]. It is based on the following idea. Many copies of an entangled tripartite pure
state, |ΨABC i, is shared between Alice, Bob and Charlie. Now, Alice and Bob want to
use the state to perform some task involving the use of bipartite entanglement. If they
ignore Charlie completely, they share the state ρAB = trC (|ΨABC i hΨABC |). In other words
they share a mixed state, and it might even be separable. But Charlie wants to help,
and Alice and Bob may communicate classically with him. Therefore, Charlie can perform
local operations on his part of the system and tell Alice and Bob about the measurement
outcomes. A measurement by Charlie will then leave the subsystem of Alice and Bob in some
pure state, which depending on the measurement outcome may or may not be entangled.
Alice and Bob can then discard the pairs that are not entangled, and turn the rest reversibly
(in the asymptotic limit) into Bell states. If each measurement by Charlie leaves them with a
state |ΨAB
i i with probability pi , the number of Bell states produced per copy of the tripartite
state will be equal to the average entropy of entanglement of the ensemble E = {pi , |ΨABi i},
namely X
E(E) = pi EE (|ΨAB
i i). (4.52)
i
Now, Charlie wants to do the measurements in a way that leaves Alice and Bob with as
much entanglement as possible on average. It turns out that the optimisation can be taken
over all ensembles E = {pi , |ΨAB
i i} consistent with ρ
AB . Thus, what Charlie can achieve
does not depend on the initial tripartite state |Ψ ABC i, only on ρAB , so all pure tripartite
states for which ρAB is the same, allow Charlie to help Alice and Bob to the same result.
The entanglement of assistance is the average entropy of entanglement Alice and Bob are
left with after Charlie has applied his optimal strategy, namely
X
Ea (ρAB ) ≡ sup pi EE (|ΨAB
i i). (4.53)
E i
We can see that this is dual to the entanglement of formation (4.35), which is the same with
an infimum instead of the supremum. Even though the entanglement of formation is believed
to be additive, there are states for which the entanglement of assistance is superadditive,
Ea (ρAB ⊗ σ AB ) ≥ Ea (ρAB ) + Ea (σ AB ). (4.54)
A recently introduced measure related to the entanglement of assistance is the localizable

entanglement [VPC04]. It is really a multipartite measure for a large number of qubits, but
the goal is, just as for the entanglement of assistance, to optimise (localise) the entanglement
between two subsystems by performing measurements on the others. If we had allowed global
measurements on the other qubits, we could have regarded it as a single “Charlie” system,
and it would have reduced to the entanglement of assistance. Because we can only perform
local measurements on each individual qubit, the localizable entanglement is necessarily
smaller than the entanglement of assistance.
Another entanglement measure recently introduced is the witnessed entropy of entangle-
ment [BV04]. It is based on the concept of entanglement witnesses [HHH96]. An entangle-
ment witness for an entangled state σ is a Hermitian operator W for which tr(W σ) < 0 and
tr(W ρ) ≥ 0 for all separable states ρ. We conventionally normalise entanglement witnesses
to have tr(W ) = 1. We say that an entanglement witness Wσ is optimal for a state σ if
tr(Wσ σ) ≤ tr(W σ) (4.55)
for all entanglement witnesses W . The definition of the witnessed entropy of entanglement
is based on the optimal witness,
Ew (ρ) ≡ log(D/d) max[0, − tr(Wρ ρ)] (4.56)
where D is the dimension of the total Hilbert space and d the smallest of the dimensions
of the Hilbert spaces of the subsystems. Note that the definition applies equally well to
bipartite and multipartite systems. It is not additive, but satisfies the rest of our condi-
tions for entanglement measures, (E0a), (E1a), (E2a), (E3a), (E5a) and (E6a). The main
advantage of this measure is that it can be approximately calculated for all mixed states.
Finding the exact optimal entanglement witness, is a hard optimisation problem, but using
approximation techniques, there are algorithms to find the Ew to a given precision.
Other measures can be defined operationally in the following way. Take your favourite
application of entanglement, study how well or bad it performs for different sets of states,
define an entanglement measure from this and check that it satisfies at least some of the
conditions for entanglement, above all LOCC monotonicity. One of the operational measures
that have had limited impact on the community is an entanglement measure introduced
by Hiroshima [Hir02] based on the capacity of dense coding [BW92]. Another operational
measure by Biham et al. [BNO02] is derived from Grover’s search algorithm [Gro96, Gro97].
It is based on the fact that the algorithm performs worse the more the input state is
entangled.
Chapter 5
Beyond bipartite entanglement and

finite dimension
The systems we have considered until now have all been bipartite systems with finite dimen-
sional Hilbert spaces. Some of the concepts and measures considered are valid also in the
multipartite systems and for entanglement of continuous variables. Others lose their mean-
ing in this more general setting. In this chapter we consider some of the steps that need to
be taken to generalise the concepts from the previous sections to multipartite systems and
systems with continuous variables.
5.1 Multipartite systems

While the basic structure of bipartite entanglement is well understood, multipartite entan-
glement is still an active field of research. In bipartite systems the pure state entanglement
in the asymptotic regime is well understood, since the von Neumann entropy of the reduced
density matrices can be conserved under transformations. In general the Schmidt decompo-
sition cannot be extended to multipartite systems [Per95], so the eigenvalues of the reduced
density matrices of each system need not be the same, nor do the von Neumann entropies.
The questions about entanglement literally acquire new dimensions when more that two
subsystems are considered. Between two parties all relations are between those two parties,
and in terms of quantum communication there is only one single channel. Once we increase
the number of parties to three, we have three pair relations in the system. Still, those three
pair relations do not describe all correlations in the system (neither classical nor quantum).
Consider for instance the pure tripartite entangled state called the GHZ1 state
1
|GHZ i ≡ √ (|000i + |111i). (5.1)
2
If we trace out any one of the systems, we get the density matrix shared by the two remaining
parties
1
ρ = (|00i h00| + |11i h11|) (5.2)
2
1
Daniel M. Greenberger, Michael A. Horne and Anton Zeilinger
43
44 Chapter 5. Beyond bipartite entanglement and finite dimension
which is separable. Thus, the state contains no bipartite entanglement between the systems,
but the tripartite entanglement is maximal.
5.1.1 Separability
Some notions can be taken directly over from bipartite systems, though. The generalisation
of separability from (3.3) is straightforward. A state is separable if and only if it can be
written as a convex combination of product states;
(A) (B) (N )
X
pi ρi ⊗ ρi ⊗ · · · ⊗ ρi (5.3)
i
P
with pi ≥ 0 and i pi = 1.
The classification of states does not end with this. An entangled multipartite state may
or may not still be entangled after one of the subsystems are ignored (i.e. traced out). A
multipartite state that loses its entanglement when any of the subsystems is traced out, is
called multiseparable [Tha99].
The PPT criterion for separability trivially extends to the multipartite case. If a multi-
partite state is separable, then transposing any number of the subsystems must also give a
valid (i.e. positive) density operator.
5.1.2 Measures of entanglement

The definition of entanglement cost and the distillable entanglement is based on a standard
bipartite state, namely the Bell state. This is a bipartite state, so it is not obvious how to
extend these measures to the multipartite case. Nielsen [Nie01] argued that this could by
overcome by generalising the definition from Bell states to a definition in terms of qubits
transferred (in the case of creation) or the number of CNOT gates that could be implemented
(in the case of distillation). For instance, to create a Bell state it is necessary to transfer one
qubit. Likewise the transfer of one qubit can be simulated by teleporting a state, spending
one Bell state. If the entanglement cost is defined in terms of how many qubits that need
to be transferred, it will coincide with the usual definition for bipartite states.
Most of the conditions we imposed on measures in section 4.2.2 generalise to the multi-
partite case. The only possible exceptions are of course the normalisation conditions (E1a)
and (E1b) which are based on Bell states. Those would be unnatural to impose on mul-
tipartite entanglement, but it should be possible for a measure to satisfy them for states
where only two of the subsystems are entangled.
The relative entropy of entanglement and other distance based measures do not have
any reference to the bipartite case in their definition. Therefore, they are just as valid in
the multipartite case as in the bipartite case.
The entanglement of formation suffers the same problem as the entanglement cost, and
can be generalised in the same way. For a pure state we define the entanglement of formation
as the number of qubits we need to transfer in order to prepare it (LOCC comes for free).
Then we extend the definition to the mixed states by the convex roof method as usual.
Another more formal generalisation was suggested by Wang [Wan01].
5.2. Continuous variables 45
The meaning of entanglement measures in the multipartite setting is more vague than
in the bipartite. In the bipartite setting it quantifies in some way how much that can be
achieved by the two parties together. In the multipartite case there are more possibilities.
For instance there may be some entanglement between all of the parties, as in the state
1
|W i ≡ √ (|100i + |010i + |001i) (5.4)
3
which on tracing out one of the subsystems gives the density matrix
1 2
ρ= |00i h00| + |Ψ+ i hΨ+ | (5.5)
3 3
which is an entangled state with negativity N = 0.206 and entanglement of formation
Ef = 0.550 (calculated from the expressions in sections 4.2.6 and 4.2.5). Or there may be
no entanglement between the pairs, as in the GHZ state mentioned earlier. Both states can
be prepared by sending two qubits (after one of the parties has created the state locally).
While the different bipartite measures place different orderings on the states [VP00], ordering
is even more ambiguous for multipartite states. The types of entanglement are qualitatively
more different, and a state being very much entangled in one sense (as defined by one
measure), can have very little or no entanglement in another sense. All this says is that,
since a multipartite system is a more complicated structure, we cannot expect one single
parameter to be as good a description of the entanglement in it as it is in the bipartite case.
5.2 Continuous variables

Most of quantum information theory has been formulated in the context of finite dimen-
sional Hilbert spaces. However, from an experimental point of view, entanglement is just
as interesting in the continuous variable regime. This has led to efforts to quantify en-
tanglement also in the case where the Hilbert spaces have infinite dimension. Two papers
by Eisert et al. [ESP02, EP03] summarise many of the new problems that arise and their
possible solutions.
First of all, when the dimension of the Hilbert space is infinite, we have states with
an infinite amount of entanglement. This should come as no surprise, as the entropy of
entanglement for a maximally entangled state in d × d dimensions is log d (c.f. criteria (P1a)
and (E1a)). What is more worrying is that the set of pure states with infinite entropy of
entanglement is actually dense in the trace norm on the set of pure states [ESP02]. This
means that arbitrarily close to any product state is a state which has infinite entanglement.
The solution to this problem is to consider only states where the mean energy is bounded
from above. This is a reasonable physical assumption, and on this subset of states the
entropy of entanglement is continuous. Other measures can also be defined on this subset
without exhibiting strange behaviour.
The definition of separability in a continuous variable bipartite system must be slightly
altered from the one in section 3.2.1. The expression is still the same as (3.3), namely
(A) (B)
X
pi ρi ⊗ ρi (5.6)
i
46 Chapter 5. Beyond bipartite entanglement and finite dimension
with pi is a probability distribution. But in infinite dimension we can only require that the
state can be approximated arbitrarily well by an expression of the form (5.6).
The PPT criterion for separability is also valid in infinite dimension and for the special
class of so-called Gaussian states, it is also sufficient for separability [Sim00].
It is a well known fact in the finite dimensional case that there is a neighbourhood around
the completely mixed state I/d where all states are separable [ŻHSL98, PR00, Myr04].
However, the volume of the separable ball diminishes as the dimension of the system grows.
In the infinite dimensional setting, even when the energy is bounded from above, it has been
shown [ESP02] that arbitrarily close to any state (in the trace norm), there is an entangled
state.
The class of so-called Gaussian states has a special role in entanglement theory of con-
tinuous variables. These are states with a Gaussian Wigner function (a quasi-probability
distribution in phase space which completely describes a quantum state). These states are
often encountered in the laboratory, and they can be completely described by two param-
eters. For Gaussian states, and operators that takes Gaussian states to Gaussian states,
many of the notions from finite dimensional entanglement theory have similar counterparts.
See e.g. [EP03, KGLC03, GWK+ 03, WGK+ 04, RE04].
Appendix A
Expressions for the quantum

relative entropy
The quantum relative entropy is defined as

S(ρkσ) ≡ tr{ρ log ρ − ρ log σ}. (A.1)
We want to calculate it in terms of the eigenvectors and eigenvalues of ρ and σ. We denote
the eigenvalues of ρ (σ) by ri (si ) and its eigenvectors by |ri i (|si i). In the case of degenerate
eigenvalues we assume that the eigenvectors have been properly orthogonalised.
The linearity of the trace gives
S(ρkσ) = tr{ρ log ρ} − tr{ρ log σ}
where the first term can be identified as −S(ρ), where S(·) is the von Neumann entropy.
This depends only on the eigenvalues of ρ:
X
tr{ρ log ρ} = ri log ri .
i
The second term is more complicated as the eigenvectors of the density operators in
general are not the same. We decompose the density operators using the spectral decom-
position.
X X
ρ log σ = ri |ri i hri | log sj |sj i hsj |
i j
X X
= ri |ri i hri | log(sj ) |sj i hsj |
i j
X
= ri log(sj ) hri |sj i |ri i hsj |
ij
P
Then we decompose |ri i in the eigendirections of σ, |ri i = k hsk |ri i |sk i.
X
ρ log σ = ri log(sj ) hri |sj i hsk |ri i |sk i hsj |
ijk
47
48 Appendix A. Expressions for the quantum relative entropy
Now we can take the trace of this to get the second term
ri log sj |hri |sj i|2

X
tr{ρ log σ} =
ij
We can now put the two terms together to get the expression we wanted;
X
log sj |hri |sj i|2 .
X
S(ρkσ) = ri log ri − (A.2)
i j
This may also be written as the single sum

X
S(ρkσ) = ri log ri − hsi | ρ |si i log si . (A.3)
i
In the case when the eigenvectors for the two density matrices are the same (i.e. ρ and
σ commute), we can arrange the eigenvectors so that hri |sj i = δij . (A.2) then reduces to
X
S(ρkσ) = ri log ri − log si . (A.4)
i
References
[ADR82] A. Aspect, J. Dalibard and G. Roger, Experimental Test of Bell’s Inequalities

Using Time-Varying Analyzers, Phys. Rev. Lett. 49(25), 1804–1807 (1982).
[ADVW02] K. Audenaert, B. De Moor, K. G. H. Vollbrecht and R. F. Werner, Asymptotic
relative entropy of entanglement for orthogonally invariant states, Phys. Rev.
A 66, 032310 (2002), quant-ph/0204143.
[AF04] R. Alicki and M. Fannes, Continuity of quantum conditional information, J.
Phys. A: Math. Gen. 37, L55–L57 (2004), quant-ph/0312081.
[AVD01] K. Audenaert, F. Verstraete and B. De Moor, Variational characterizations of
separability and entanglement of formation, Phys. Rev. A 64, 052304 (2001),
quant-ph/0006128.
[BBC+ 93] C. H. Bennett, G. Brassard, C. Crépeau, R. Jozsa, A. Peres and W. K. Woot-
ters, Teleporting an unknown quantum state via dual classical and Einstein-
Podolsky-Rosen channels, Phys. Rev. Lett. 70, 1895–1899 (1993).
[BBP+ 96] C. Bennett, G. Brassard, S. Popescu, B. Schumacher, J. Smolin and W. Woot-
ers, Purification of Noisy Entanglement and Faithful Teleportation via Noisy
Channels, Phys. Rev. Lett. 76(5), 722–725 (1996), quant-ph/9511027.
[BBPS96] C. Bennett, H. Bernstein, S. Popescu and B. Schumacher, Concentrating partial
entanglement by local operations, Phys. Rev. A 53(4), 2046–2052 (1996),
quant-ph/9511030.
[BDSW96] C. Bennett, D. DiVincenzo, J. Smolin and W. Wooters, Mixed-state entangle-
ment and quantum error correction, Phys. Rev. A 54(5), 3824–3851 (1996),
quant-ph/9604024.
[Bel64] J. S. Bell, On the Einstein-Podolsky-Rosen paradox, Physics 1, 195–200 (1964),
Reprinted in J. S. Bell “Speakable and unspeakable in quantum mechanics”,
Cambridge University Press, 1987.
[BN01] F. Benatti and H. Narnhofer, Additivity of the entanglement of formation,
Phys. Rev. A 63(4), 042306 (2001), quant-ph/0005126.
[BNO02] O. Biham, M. A. Nielsen and T. J. Osborne, Entanglement monotone derived
from Grover’s algorithm, Phys. Rev. A 65, 062312 (2002), quant-ph/0112097.
49
50 References
[Bru02] D. Bruß, Characterizing entanglement, J. Math. Phys. 43(9), 4237–4251

(2002), quant-ph/0110078.
[Bur69] D. Bures, An Extension of Kakutani’s Theorem on Infinite Product Measures

to the Tensor Product of Semifinite w∗ -Algebras, Trans. Am. Math. Soc. 135,
199–212 (1969).
[BV04] F. G. S. L. Brandão and R. O. Vianna, Witnessed Entropy of Entanglement,

(May 2004), quant-ph/0405096.
[BW92] C. H. Bennett and S. J. Wiesner, Communication via one- and two-particle

operators on Einstein -Podolsky- Rosen states, Phys. Rev. Lett. 69, 2881–2884
(1992).
[Coh98] O. Cohen, Unlocking Hidden Entanglement with Classical Information, Phys.

Rev. Lett. 80(11), 2493–2496 (1998).
[CW04] M. Christandl and A. Winter, “Squashed Entanglement” – An Additive Entan-

glement Measure, J. Math. Phys. 45(3), 829–840 (2004), quant-ph/0308088.
[DCLB00] W. Dür, J. I. Cirac, M. Lewenstein and D. Bruß, Distillability and partial

transposition in bipartite systems, Phys. Rev. A 61(6), 062313 (2000), quant-
ph/9910022.
[DFM+ 99] D. DiVincenzo, C. Fuchs, H. Mabuchi, J. Smolin, A. Thapliyal and A. Uhlmann,

Entanglement of Assistance, in Quantum Computing and Quantum Comm-
munications: First NASA International Conference, QCQC’98, edited by C. P.
Williams, volume 1509 of Lecture Notes in Computer Science, pages 247–257,
Springer-Verlag, 1999, quant-ph/980333.
[DH99] M. J. Donald and M. Horodecki, Continuity of relative entropy of entanglement,

Phys. Lett. A 264, 257–260 (1999), quant-ph/9910002.
[DHR02] M. J. Donald, M. Horodecki and O. Rudolph, The uniqueness theorem for

entanglement measures, J. Math. Phys. 43(9), 4252–4272 (September 2002),
quant-ph/0105017.
[DSS+ 00] D. P. DiVincenzo, P. W. Shor, J. A. Smolin, B. M. Terhal and A. V. Thapliyal,

Evidence for bound entangled states with negative partial transpose, Phys. Rev.
A 61(6), 062312 (2000), quant-ph/9910026.
[EAP03] J. Eisert, K. Audenaert and M. B. Plenio, Remarks on entanglement measures

and non-local state distinguishability, J. Phys. A: Math. Gen. 36, 5605–5615
(2003), quant-ph/0212007.
[Eis03] J. Eisert, Script of lecture in course: Qauntum Information II, 2003,

http://jense.qip.org/QIT.ps .
References 51
[Eke91] A. K. Ekert, Quantum Cryptography Based on Bell’s Theorem, Phys. Rev.

Lett. 67(6), 661–663 (1991).
[EP03] J. Eisert and M. Plenio, Introduction to the basics of entanglement theory in

continuous-variable systems, Int. J. Quant. Inf. 1(4), 479–506 (2003), quant-
ph/0312071.
[EPR35] A. Einstein, B. Podolsky and N. Rosen, Can Quantum-Mechanical Description

of Physical Reality Be Considered Complete?, Phys. Rev. 47, 777–780 (1935).
[ESP02] J. Eisert, C. Simon and M. B. Plenio, On the quantification of entanglement in

infinite-dimensional quantum systems, J. Phys. A: Math. Gen. 35, 3911–3923
(2002), quant-ph/0112064.
[FC95] C. A. Fuchs and C. M. Caves, Mathematical Techniques for Quantum Com-

munication, Open Systems & Information Dynamics 3, 1 (1995), quant-
ph/9604001.
[Gro96] L. Grover, A Fast Quantum Mechanical Algorithm for Database Search, in

Proc. 28th Annual ACM Symposium on the Theory of Computation, pages
212–219, New York, NY, 1996, ACM Press, New York.
[Gro97] L. K. Grover, Quantum Mechanics Helps in Searching for a Needle in a

Haystack, Phys. Rev. Lett. 79, 325–328 (1997), quant-ph/9706033.
[GWK+ 03] G. Giedke, M. M. Wolf, O. Krueger, R. Werner and J. I. Cirac, Entanglement

of Formation for Symmetric Gaussian States, Phys. Rev. Lett. 91, 107901
(2003), quant-ph/0304042.
[HHH96] M. Horodecki, P. Horodecki and R. Horodecki, Separability of mixed states:

Necessary and sufficient conditions, Phys. Lett. A 223(1-2), 1–8 (November
1996), quant-ph/9605038.
[HHH97] M. Horodecki, P. Horodecki and R. Horodecki, Inseparable Two Spin- 12 Density

Matrices Can Be Distilled to a Singlet Form, Phys. Rev. Lett. 78(4), 574–577
(January 1997).
[HHH98] M. Horodecki, P. Horodecki and R. Horodecki, Mixed-State Entanglement and

Distillation: Is there a “Bound” Entanglement in Nature?, Phys. Rev. Lett.
80(24), 5239–5242 (June 1998), quant-ph/9801069.
[HHH99a] M. Horodecki, P. Horodecki and R. Horodecki, General teleportation channel,

singlet fraction, and quasidistillation, Phys. Rev. A 60(3), 1888–1898 (Septem-
ber 1999), quant-ph/9807091.
[HHH99b] P. Horodecki, M. Horodecki and R. Horodecki, Bound Entanglement Can Be

Activated, Phys. Rev. Lett. 82(5), 1056–1059 (1999), quant-ph/9806058.
52 References
[HHH00] M. Horodecki, P. Horodecki and R. Horodecki, Limits for Entanglement Mea-

sures, Phys. Rev. Lett. 84(9), 2014–2017 (2000), quant-ph/9908065.
[HHT01] P. M. Hayden, M. Horodecki and B. Terhal, The asymptotic entanglement cost

of preparing a quantum state, J. Phys. A: Math. Gen. 34, 6891–6898 (2001),
quant-ph/0008134.
[Hir02] T. Hiroshima, An entanglement measure based on the capacity of dense coding,

[Hor97] P. Horodecki, Separability criterion and inseparable mixed states with posi-
tive partial transposition, Phys. Lett. A 232, 333–339 (August 1997), quant-
ph/9703004.
[Hor01] M. Horodecki, Entanglement Measures, Quantum Inf. Comput. 1(1), 3 (2001),

http://www.rintonpress.com/journals/qic-1-1/miarprz3.pdf.
[Hüb92] M. Hübner, Explicit computation of the Bures distance for density matrices,
Phys. Lett. A 163, 239–242 (1992).
[HW97] S. Hill and W. Wooters, Entanglement of a Pair of Quantum Bits, Phys. Rev.
Lett. 78(26), 5022–5025 (1997), quant-ph/9703041.
[Key02] M. Keyl, Fundamentals of quantum information theory, Physics Reports

369(5), 431–548 (2002), quant-ph/0202122.
[KGLC03] B. Kraus, G. Giedke, M. Lewenstein and J. Cirac, Entanglement properties of

Gaussian States, Fortschr. Phys. 51(4-5), 305–312 (2003).
[LCOK03] S. Lee, D. P. Chi, S. D. Oh and J. Kim, Convex-roof extended negativity as

an entanglement measure for bipartite quantum systems, Phys. Rev. A 68(6),
062304 (2003), quant-ph/0310027.
[LP99] H.-K. Lo and S. Popescu, Classical Communication Cost of Entanglement

Manipulation: Is Entanglement an Inverconvertible Resource?, Phys. Rev. Lett.
83(7), 1459–1462 (1999), quant-ph/9902045.
[MSV04] F. Morikoshi, M. F. Santos and V. Vedral, Accessability of physical states

and non-uniqueness of entanglement measure, J. Phys. A: Math. Gen. 37(22),
5887–5893 (2004), quant-ph/0306032.
[Myr04] J. Myrheim, Separable density matrices, (unpublished), April 2004.
[NC00] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Infor-

mation, Cambridge University Press, Cambridge, UK, 2000.
[Nie00] M. A. Nielsen, Continuity bounds for entanglement, Phys. Rev. A 61(6),

064301 (2000), quant-ph/9908086.
References 53
[Nie01] M. A. Nielsen, On the units of bipartite entanglement: is sixteen ounces of

entanglement always equal to one pound?, J. Phys. A: Math. Gen. 34, 6987–
6995 (2001), quant-ph/0011063.
[Oza00] M. Ozawa, Entanglement measures and the Hilbert-Schmidt distance, Phys.

Lett. A 268, 158–160 (2000), quant-ph/0002036.
[Per95] A. Peres, Higher order Schmidt decompositions, Phys. Lett. A 202, 16–17
(1995), quant-ph/9504006.
[Per96] A. Peres, Separability criterion for density matrices, Phys. Rev. Lett. 77(8),
1413–1415 (August 1996), quant-ph/9604005.
[PR97] S. Popescu and D. Rohrlich, Thermodynamics and the measure of entangle-

ment, Phys. Rev. A 56(5), R3319–R3321 (1997), quant-ph/9610044.
[PR00] A. O. Pittenger and M. H. Rubin, Separability and Fourier representations of

density matrices, Phys. Rev. A 62, 032313 (2000), quant-ph/0001014.
[Pre98] J. Preskill, Lecture Notes for Physics 229: Quantum Information and Computa-
tion, September 1998, http://www.theory.caltech.edu/people/preskill/ph229/ .
[RE04] G. Rigolin and C. O. Escobar, Lower bounds on the entanglement of forma-

tion for general Gaussian states, Phys. Rev. A 69, 012307 (2004), quant-
ph/0307023.
[Rud01] O. Rudolph, A new class of entanglement measures, J. Math. Phys. 42(11),

5306–5314 (November 2001), math-ph/0005011.
[San04] E. Santos, Optical tests of Bell’s inequalities not resting upon the absurd fair
sampling assumption, (2004), quant-ph/0401003.
[Sha48] C. E. Shannon, A Mathematical Theory of Communication, Bell System

Technical Journal 27, 379–423, 623–656 (July, October 1948), http://cm.bell-
labs.com/cm/ms/what/shannonday/paper.html.
[Shi02] T. Shimono, Towards Additivity of Entanglement of Formation, in Uncon-

ventional Models of Computation: Third International Conference, UMC 2002,
Kobe, Japan, edited by F. P. C. S. Calude, M. J. Dinneen, volume 2509 of
Lecture Notes in Computer Science, pages 252–263, Springer-Verlag, 2002.
[Sho94] P. W. Shor, Algorithms for quantum computation: Discrete logarithms and

factoring, in Proceedings of the 35th Annual Symposium on the Foundations
of Computer Science, edited by S. Goldwasser, pages 124–134, Los Alamitos,
CA, 1994, IEEE Computer Society.
[Sho97] P. W. Shor, Polynomial-time algorithms for prime factorization and discrete

logarithms on a quantum computer, SIAM J. Comput. 26(5), 1484–1509
(October 1997), quant-ph/9508027.
54 References
[Sim00] R. Simon, Peres-Horodecki Separability Criterion for Continuous Variable Sys-

tems, Phys. Rev. Lett. 84(12), 2726–2729 (2000), quant-ph/9909044.
[SM95] J. Schlientz and G. Mahler, Description of entanglement, Phys. Rev. A 52(6),

4396–4404 (December 1995).
[SST01] P. W. Shor, J. A. Smolin and B. M. Terhal, Nonadditivity of Bipartite Dis-

tillable Entanglement Follows from a Conjecture on Bound Entangled Werner
States, Phys. Rev. Lett. 86(12), 2681–2684 (2001), quant-ph/0010054.
[SW02] B. Schumacher and M. D. Westmoreland, Relative entropy in quantum infor-

mation theory, in Quantum Computation and Information, edited by S. J.
Lomonaco Jr. and H. E. Brant, volume 305 of Contemporary Mathematics,
American Mathematical Society, 2002, quant-ph/0004045.
[TH02] C. H. Thompson and H. Holstein, The “Chaotic Ball” model: local realism and
the Bell test “detection loophole”, (2002), quant-ph/0210150.
[Tha99] A. V. Thapliyal, Multipartite pure-state entanglement, Phys. Rev. A 59(5),

3336 (May 1999), quant-ph/9811091.
[TM71] M. Tribus and E. C. McIrvine, Energy and information, Scientific American

225(3), 179–188 (September 1971).
[TV00] B. M. Terhal and K. G. H. Vollbrecht, Entanglement of Formation for Isotropic

States, Phys. Rev. Lett. 85(12), 2625–2628 (2000), quant-ph/0005062.
[Uhl76] A. Uhlmann, The “transition probability” in the state space of a ∗ -algebra,

Rep. Math. Phys. 9(2), 273–279 (1976).
[Uhl98] A. Uhlmann, Entropy and Optimal Decompositions of States Relative to a

Maximal Commutative Subalgebra, Open Systems & Information Dynamics
5, 209–227 (1998), quant-ph/9704017.
[Uhl00] A. Uhlmann, Fidelity and concurrence of conjugated state, Phys. Rev. A 62(3),
032307 (2000), quant-ph/9909060.
[VDC02] G. Vidal, W. Dür and J. I. Cirac, Entanglement Cost of Bipartite Mixed States,
Phys. Rev. Lett. 89(2), 027901 (2002), quant-ph/0112131.
[Ved02] V. Vedral, The role of relative entropy in quantum information theory, Rev.
Mod. Phys 74(1), 197–234 (January 2002), quant-ph/0102094.
[Vid00] G. Vidal, Entanglement monotones, J. Mod. Opt. 47, 355–376 (2000), quant-
ph/9807077.
[vN32] J. von Neumann, Mathematische Grundlagen der Quantenmechanik, Number

XXXVIII in Die Grundlehren der Mathematischen Wissenschaften in Einzel-
darstellungen, Springer, 1932.
References 55
[vN55] J. von Neumann, Mathematical Foundations of Quantum Mechanics, Princeton

University Press, 1955.
[VP98] V. Vedral and M. Plenio, Entanglement measures and purification procedures,

Phys. Rev. A 57(3), 1619–1633 (1998), quant-ph/9707035.
[VP00] S. Virmani and M. B. Plenio, Ordering states with entanglement measures,

[VPC04] F. Verstraete, M. Popp and J. Cirac, Entanglement versus Correlations in Spin

Systems, Phys. Rev. Lett. 92(2), 027901 (2004), quant-ph/0307009.
[VPJK97] V. Vedral, M. B. Plenio, K. Jacobs and P. L. Knight, Statistical inference,

distinguishability of quantum states, and quantum entanglement, Phys. Rev. A
56(6), 4452–4455 (December 1997), quant-ph/9703025.
[VPRK97] V. Vedral, M. Plenio, M. Rippin and P. Knight, Quantifying Entanglement,

Phys. Rev. Lett. 78(12), 2275–2279 (1997), quant-ph/9702027.
[VT99] G. Vidal and R. Tarrach, Robustness of entanglement, Phys. Rev. A 59(1),

141–155 (January 1999), quant-ph/9806094.
[VW01] K. G. H. Vollbrecht and R. F. Werner, Entanglement measures under symmetry,

Phys. Rev. A 64(6), 062307 (December 2001), quant-ph/0010095.
[VW02a] G. Vidal and R. F. Werner, Computable measure of entanglement, Phys. Rev.

A 65, 032314 (2002), quant-ph/0102117.
[VW02b] K. G. H. Vollbrecht and M. M. Wolf, Activating Distillation with an Infinitesi-

mal Amount of Bound Entanglement, Phys. Rev. Lett. 88(24), 247901 (2002),
quant-ph/0201103.
[Wan01] A. M. Wang, Generalization of the entanglement of formation for multi-party

systems, (2001), quant-ph/0011040 v4.
[Wer89] R. F. Werner, Quantum states with Einstein-Podolsky-Rosen correlations ad-

mitting a hidden-variable model, Phys. Rev. A 40(8), 4277–4281 (October
1989).
[WGK+ 04] M. M. Wolf, G. Giedke, O. Krueger, R. Werner and J. Cirac, Gaussian Entan-
glement of Formation, Phys. Rev. A 69, 052320 (2004), quant-ph/0306177.
[Whi98] A. Whitaker, John Bell and the most profound discovery

of science, Physics World 11(12), 29–34 (December 1998),
http://physicsweb.org/article/world/11/12/8.
[Woo98] W. Wooters, Entanglement of Formation of an Arbitrary State of Two Qubits,

Phys. Rev. Lett. 80(10), 2245–2248 (1998), quant-ph/9709029.
56 References
[Woo01] W. K. Wootters, Entanglement of formation and concurrence, Quantum

Inf. Comput. 1(1), 27–44 (2001), http://www.rintonpress.com/journals/qic-
1-1/eof2.pdf.
[WT99] C. Witte and M. Trucks, A new entanglement measure induced by the Hilbert-
Schmidt norm, Phys. Lett. A 257, 14–20 (1999), quant-ph/9811027.
[ŻHSL98] K. Życzkowski, P. Horodecki, A. Sanpera and M. Lewenstein, Volume of the

set of separable states, Phys. Rev. A 58, 883–892 (August 1998), quant-
ph/9804024.
Index
abstract measure, 25 distillation, 16–19

Alice and Bob, 6 localizable, 41
asymptotic regime, 25 measure, 25
uniqueness theorem, 28
base norm, 39 of assistance, 40
Bell basis, 6 of creation, see entanglement, of forma-
Bell states, 6 tion
biorthogonal, 11 of distillation, see entanglement, distill-
bipartite state, 6 able
bit flip operator, 5, 14 of formation, 35
Bloch sphere, 4 squashed, 39
bound entanglement, 23 entropy of entanglement, 26
Bures metric, 34 relative, 34
witnessed, 41
composite system, 5
EPR pair, see Bell state
concurrence, 36
convex roof measure, 36 fidelity, 18
creation, entanglement of, see entanglement, finite regime, 25
of formation formation, entanglement of, see entanglement,
of formation
density matrix, see density operator, 7
free entanglement, 23
density operator, 7
realisation of, 7 GHZ state, 43
reduced, 9
distillable entanglement, see entanglement, Local Operations and Classical Communica-
distillable tion, see LOCC
LOCC, 14
ensemble, 7 LQCC, see LOCC
entangled state, 6
entanglement maximal knowledge, 7
cost measurement
mixed states, 29 generalised, 11–13
pure states, 27 projective, 11
distillable measurement operators, 11
mixed states, 29 mixed state, 7
pure states, 27 multipartite state, 6
57
58 Index
multipartite system, 43
multiseparable, 44
negativity, 37
logarithmic, 38
NPPT, 24
operational measure, 25
partial trace, 9
Pauli operators, 5
Peres-Horodecki criterion, 23
phase flip operator, 5
positive operator, 8
PPT-criterion, 23
product state, 6, 21
pure state, 7
QLCC, see LOCC

qubit, 3
realisation, 7
relative entropy, 33
of entanglement, 34
robustness of entanglement, 39
Schmidt coefficients, 11
Schmidt decomposition, 10
Schmidt number, 21
Schmidt rank, 21
separability, 21
separable state, 10, 21
Shannon entropy, 26
superadditivity, 41
strong, 40
trace norm, 28, 31, 38

trace out, 8
Uhlmann’s transition probability, 35
von Neumann entropy, 26

reduced, 26
X operator, 4, 5
Y operator, 4, 5
Z operator, 4, 5

(MT) 0408094 (Measures of Entanglement in Quantum Mechanics) Geir Ove Myhr

Uploaded by

Copyright:

Available Formats

(MT) 0408094 (Measures of Entanglement in Quantum Mechanics) Geir Ove Myhr

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(MT) 0408094 (Measures of Entanglement in Quantum Mechanics) Geir Ove Myhr

Uploaded by

Copyright:

Available Formats

Norges teknisk-naturvitenskapelige universitet - NTNU

Fakultet for naturvitenskap og teknologi

STUD. TECHN. GEIR OVE MYHR

Oppgaven gitt: 01.02.2004

FAGOMRÅDE: TEORETISK FYSIKK

Norsk tittel: “Mål på sammenfiltring i kvantemekanikk”

Trondheim, 27. juni 2004,

3 Characterisation of bipartite entanglement 21

4 Measures of bipartite entanglement 25

5 Beyond bipartite entanglement and finite dimension 43

A Expressions for the quantum relative entropy 47

2.1 The qubit

|ψi = α |↑z i + β |↓z i (2.1)

|ψi = cos 2θ |↑z i + eiφ sin 2θ |↓z i

|↓x i = √1 (|↑z i − |↓z i)

|↑x i = √1 (|↑z i + |↓z i)

We may choose another set of orthonormal basis vectors. For instance

X ≡ |1i h0| + |0i h1| , (2.7a)

2.2 Entangled states

2.3 Density operators and mixed states

|ψi = α |00i + β |11i . (2.12)

ρA = |α|2 |0i h0| + |β|2 |1i h1| (2.13)

so the reduced density matrix becomes

which is just what we expected.

2.4 The Schmidt decomposition

|ψi = uik dkk vkj |ai i ⊗ |bj i

We now take as our new basis vectors

and define λk = dkk to obtain the final form

2.5 Generalised measurements

The probability for the outcome to be m is

with an arbitrary phase θ. Like |vm i, this is an eigenstate of O.

2.6 Transformation of states and LOCC

2.6.1 Local Operations and Classical Communication

2.6.2 Transformation of a Bell state

where the local state vectors are denoted as usual:

The operators clearly satisfies the completeness relation (2.24), as

2.6.3 Concentrating partial pure state entanglement

M1 |ψ1 i = sin θ |00i + sin θ |11i

The probability for the two outcomes are

p(1) = hψ1 | M1† M1 |ψ1 i = sin2 (h00|00i + h11|11i) = 2 sin2 θ

and the normalised new states are

2.6.4 Distilling entanglement from mixed states

ρ = (1 − p) |01i h01| + p |Φ+ i hΦ+ |

UCNOT = |00i h00| + |01i h01| + |10i h11| + |11i h10|

UCNOT (1, 3) = |0i h0| ⊗ I ⊗ I ⊗ I + |1i h1| ⊗ I ⊗ X ⊗ I

ρout = UCNOT (1, 3)UCNOT (2, 4)ρ⊗2 UCNOT

M00 = I ⊗ I ⊗ |0i h0| ⊗ |0i h0| (2.33a)

and is equivalent to measuring the observable I ⊗ I ⊗ Z3 ⊗ Z4 as the measurement operators

F ≡ hΦ+ | ρ |Φ+ i (2.38)

3.1 Pure states

|Ψprod i = |ψiA ⊗ |φiB (3.1)

3.2 Mixed states

3.2.1 Separable states

or in matrix representation in the {|00i , |01i , |10i , |11i} basis,

3.2.2 Free and bound entanglement

Measures of bipartite entanglement

4.1 Pure states

4.1.1 von Neumann entropy

4.1.2 Reduced von Neumann entropy – entropy of entanglement