You are on page 1of 20

An introduction to Entanglement Theory [Transfer Essay I]

for supervisor Martin Plenio and academic tutor Terry Rudolph

Oscar Dahlsten1
1

QOLS, Blackett Laboratory, Imperial College London,


Prince Consort Road, London SW7 2BW, UK
(Dated: June 30, 2005)

Entanglement is a quantum effect that is not possible in the classical picture. It is here described
how an understanding of this effect has evolved to the current state of entanglement theory. The EPR
paradox and Bells inequalities are firstly discussed. Then the modern paradigm of entanglement
as a resource in e.g. teleportation protocols is described. Finally efforts to quantify this resource
character of entanglement are reviewed.

Contents

Introduction

Motivation

Outline

How Quantum Theory Predicts Entanglement 3

Classical Version

Quantum version

10

Entanglement Theory

11

Characterising Entanglement

11

Manipulating Entanglement

12

Quantifying Entanglement

13

Superposition principle

Von Neumann entropy of entanglement

14

Tensor Product

Entanglement of formation

15

Entangled state

Entanglement cost, and distillability

15

Relative Entropy of Entanglement

16

Logarithmic Negativity

16

EPR Incompleteness Argument

Completeness

Realism

Locality

A test to settle the issue: Bells Inequalities


Aspects experiment

Open issues in Entanglement Theory

17

Multipartite Entanglement

17

Formal analogy with thermodynamics

18

Conclusion

18

References

19

Uses of Entanglement

Superdense Coding

Quantum Computing

Teleportation

2
INTRODUCTION

outcome of the other.

This is an expose of the topic of entanglement (ii) Computer components are getting increasin quantum information theory. Consider a quan- ingly small. If this trend continues at the current
tum state that can be divided into two parts, e.g. two rate(given by Moores Law) the size of information
particles that can be separated. If measurement out- processing units will be comparable to that of atoms
comes on these two parts are correlated in a certain around 2015. Hence quantum physical effects will
way the state is called entangled. One can have cor- impact on the performance of computers at this
relation between variables in classical physics too. stage. This is one of the reasons for scientists to
However in quantum mechanical correlations go fur- model a computer as a quantum system(a Quantum
ther. In general in quantum mechanics it is typically Computer), with superpositions, entanglement etc.
held the a state does not possess its property until Strikingly such a computer would be significantly
being measured. The surprising effect of quantum more powerful. Other uses of entangled states
correlations(entanglement) is that one can see cor- include teleportation, superdense coding and
relations even between measurement outcomes that quantum key distribution. The latter is currently
are supposedly decided upon measuring.

Motivation

Entanglement is a fascinating phenomenon on

being commercialised.

Outline

The essay is loosely placed in a chronological or-

two accounts:

der according to when ideas were formulated and

(i)its implications about the nature of reality

tested. It will first be described how quantum the-

(ii)its uses in technological progress.

ory predicts entanglement. Then we will discuss


how Einstein, Podolsky and Rosen used this to ar-

(i) Entanglement is an effect that clearly dis- gue that quantum mechanics description of reality
tinguishes quantum mechanics from the classical is incomplete. We then see how Bell indicated an
physics model of the world, so it lends itself as a experimental method that determined -with almost
good phenomenon to discuss the general conceptual certainty- that completing quantum state description
problems in quantum mechanics. The interpretation with some more variable(s) obeying Einstein locality
of entanglement revolves around issues of realism could be ruled out. Having then established the pheand what is known as Einstein Locality, the idea nomenon as real we will discuss what it can be used
that two events can be sufficiently separated so that for. This gives rise to the current quantum informathe outcome of one of them could not affect the tion theoretical paradigm of entanglement as a re-

3
source. Accordingly work to quantify entanglement tem are tensored together as H1 H2 . One way
as a resource has been done which we will describe. of forming new state vectors is then to take a tensor
Finally we list some open questions in Entanglement product of two individual states of the systems. In
Quantification relevant to the thesis.

the above example this would imply

|1 i|2 i = (a1 |0i + b1 |1i)(a2 |0i + b2 |1i)

HOW QUANTUM THEORY PREDICTS


ENTANGLEMENT

= a1 a2 |0i|0i + a1 b2 |0i|1i
+ b1 a2 |1i|0i + b1 b2 |1i|1i

Two axioms/basic rules of quantum mechanics

(2)

seem to imply the existence of entangled states: the


However other states are possible in H1 H2 .

superposition principle and the tensor product.

Superposition principle

Entangled state

In quantum theory all possible measurement outcomes are modelled by a vector basis in a complex

The state below is in H1 H2 .


1
|i = (|01i + |10i)
2

vector space satisfying certain requirements. The


state is however not restricted to one of these basis

(3)

vectors, but can be in a superposition. A simple ex- It cannot be written as a product state | i| i. Fur1
2
ample is a two-level atom (a possible realisation of a thermore we will see there are physical processes
qubit).

that produce such states.

We will call this and

generic states entangled as the outcome of the mea|i = a|0i + b|1i

(1)

surement on one system immediately tells us what


the other outcome is, implying a sort of inter link-

An experimenter would then find |0i with probabil- age.


ity |a|2 and |1i with probability |b|2 respectively.

The main difference between |i and a classical


probability distribution is that the probabilities inter-

EPR INCOMPLETENESS ARGUMENT

fere in general.
In a famous 1935 paper [1] Einstein Podolsky
Tensor Product

and Rosen pointed out the existence of similar states


in an argument to question whether the quantum me-

Two systems 1 and 2 are combined by using a chanical description of reality could be considered
tensor product. The Hilbert state spaces of each sys- complete.

4
Completeness

By completeness EPR refer to the issue of

Locality

Locality is essentially the idea that systems can

whether our model of reality has a good mapping be sufficiently separated that doing something to
with the physical reality it aspires to model. They one does not affect the other simultaneously. We
feel a necessary criteria for completeness is[1]:

can more specifically consider two formulations of


locality, that used by Bell and that by Einstein

Every element of the physical reality must have respectively[4]:


a counterpart in the physical theory.

Einstein Locality(L):
The real, physical state of one system is not im-

In their epic paper they set out to show the mediately influenced by the kinds of measurements
quantum theory is incomplete. To argue this re- directly made on a second system, which is suffiquired specifying what they mean by elements of ciently separated from the first.
Bell Locality:

reality

The probabilities of outcomes of the measureRealism

ments of certain quantum-mechanical observables


on one system are not immediately influenced by the

The superposition principle mentioned above kinds of measurements directly made on a second
leads to the question of whether the quantum state system, which is sufficiently spatially separated from
is in state |0i or |1i. The orthodox interpretation of from the first.
quantum mechanics is that we can only say it is in ei-

Mathematically[2], letting A be the result of the

ther if the probability of that outcome is 1, i.e. when first measurement, and B of the second, Bell locality
there is no superposition.

is used in demanding

EPR use what they call a sufficient criterion for

P (
a, b) =

reality:

d()A(
a, )B(b, )

(4)

where is a hypothetical hidden variable(s) distribIf, without in any way disturbing a system, we uted as (). Note the outcome A does not depend
can predict with certainty(i.e. with probability equal on b.
to unity) the value of a physical quantity, then there

EPR reasoned essentially that measuring one par-

exists an element of physical reality corresponding ticle of the entangled pair, regardless of how far apart
to this physical quantity.

the particles are, will change the probability distrib-

EPR furthermore formulate and use a principle ution of the other and thereby violate locality.
called locality.

Their argument can be summarised as [3]

5
they are separated. So how, if at all, are these correlations different from the correlations we would
QM L = Incompleteness

(5)

get from assuming realism and locality, as earlier de-

Where QM refers to the orthodox interpretation of fined? For example, if we considered two balls each
quantum mechanics and L is the Einstein locality de- possessing angular momentum J and -J respectively
fined above. However the argument could also be then a measurement of their angular momenta would
taken to say that either L is violated or QM is incom- create correlated outputs. Is there a difference empirically between such a local, deterministic model

plete.

This looked like an esoteric argument until oth- of the electrons spins and the quantum one?
ers, foremost Bell, began looking for observational
Bell showed there is a difference. The key is that

differences between the two positions.

the quantum state above is in a superposition which


A TEST TO SETTLE THE ISSUE: BELLS

can lead to interference. To see this interference

INEQUALITIES

we need to look at several choices of basis rather


than just the one. In the example of the electron

It was considered that if Quantum mechanics is spins, a choice of basis corresponds to measuring
incomplete, it could be completed by some extra the spin along a different basis. This can be done
variables, who are currently hidden to us for what- by passing the electrons through inhomogeneous
ever reason. Without explicitly defining these vari- magnetic

fields(using

Stern-Gerlach

magnets),

ables, Bell managed to show that if they exist, and sending spin-up or down electrons in different
obey the Bell locality defined above, it would lead to directions. This field can be applied in different
observational predictions.

directions according to our choice of basis. Bell

He considered a formulation of the EPR problem uses four such directions: a and a for Alice and b
by Bohm and Aharonov where spin half particles are and b for Bob respectively. For each experimental
formed in the maximally entangled state:
1
|i = (| i + | i)
2

run Alice picks one axis and Bob one as in figure


(6)

1. The experimental data will then look like in the


table 1.

More succinct examples have been formulated


since the original paper and we will here deal with
one of the more pedagogical ones[3]. Consider the

Bell then considers the type of correlations

above maximally entangled state. One notes that present in this data both for the quantum model and
there will be a correlation between the spin measure- for the local deterministic model. He uses a standard
ment outputs on each particle, regardless of how far correlation coefficient

6
Where ab is the angle between a and b. Similarly

b
b
a

C(a, b) = cosab

C(a, b) = cosa b

Bob

C(a, b) = cosa b

Alice

(10)
In considering the generic local deterministic

FIG. 1: Alice and Bob each has a choice of two axes. model, we dont have a unique form for C(a,b) but it
Each axis corresponds to the angle at which they each can be shown that
measure the spin of their particle.
a

(11)

This is known as a Bell inequality. To understand


-1

-1

..
.

why it is true consider again the experimental data

1
1
..
.

|C(a, b) + C(a, b) + C(a, b) C(a, b)| 2

in table 1. Treat this now as a local realistic model,

-1
1
..
.

for example it could be the output from the classical

-1
..
.

balls mentioned above where the angular momentum


is resolved along an axis a etc. The data in the table

TABLE I: An example of experimental data that would be

could then correspond to the sign of the momentum

obtained from an experiment with Bells inequalities.

component, with +1 for + and -1 for -[3]. Locality


then implies that the outcomes of for example b
h(a hai)(b hbi)i
C(a, b) =
(h(a hai)2 (b hbi)2 i)1/2
= habi

measurements are not correlated with whether Alice


chose a or a. Realism implies that the values are
(7) possessed and has a more subtle implication. One

Where the second line follows in this case as

way to interpret it is to say that we can measure all


combinations of axes in one experimental run, so

hai = 0 and (a hai)2 = 1

that the expression = (ab) + (ab ) + (a b) (a b )

Quantum theory then predicts

can be associated with one experimental run, so that


2
C(a, b) = h|a S b S|i
~

(8) e.g. a has the same value both times it appears. One

where S is the spin operator. Choosing a to be along


the z-axis and b in the x-z plane then gives
C(a, b) = cosab

can then write out the 16 possible s, and find they


are never more than 2 or less than -2.

(9) Now make suitable choices regarding the ori-

7
entation of the measurement axes in order to arrive

Aspects experiment

at a case where the quantum predictions clearly


violate the Bell inequality.

Aspect et al found results in agreement with

(i) a,a,b,b are coplanar

quantum mechanical predictions and in violation of

(ii) a,b parallel

Bells inequalities by 5 standard deviations. They

(iii) ab = a b = say

used an optical version of the EPR Gedankenex-

This together with equations 9-11 implies

periment, which uses photons polarisations. Photon pairs were obtained from a J=0 J=1 J=0

|C(a, b) + C(a, b) + C(a, b) C(a, b)|


= |1 + 2cos cos2| (12)

cascade in calcium, shown in figure 3. In order to


make sure there is no local interaction between Alice and Bobs axes they set up an experiment such
that the choice of a,a,b,b is done randomly in such
a way that a switch is done every 10ns. The light

Quantum prediction violates inequality would take 20ns to reach the analysers after emis2.5

sion. Hence the photons are space-like separated by


the time the decision of the axes of measurement are

f(angle)

made. This has later been done with much larger

1.5

margins using optical fibres in Innsbruck[18]


1

0.5

0.5

1.5

angle (pi)
FIG. 2: |1 + 2cos cos2| (y-axis) is more than two for
certain s (x-axis)

Figure 2 shows that equation 12 implies a violation of the Bell inequality of equation 11 for certain s. Hence there is an experimental way to tell
generic local realistic quantum theories from the or- FIG. 3: A figure from [13] showing the cascade in Calthodox quantum theory. In a series of famous exper- cium used to create entangled photons
iments Aspect et al found nature ruling in favour of
the orthodox theory[5].

8
of the states in the Bell Basis.

USES OF ENTANGLEMENT

1
|+ i = (|00i + |11i)
2
1
| i = (|00i |11i)
2
1
|+ i = (|01i + |10i)
2
1
| i = (|01i |10i)
2

These experiments have served to prove that entanglement is a genuine phenomenon.

We now

consider how it can be harnessed to achieve tasks.


The simplest example is perhaps superdense coding
which is described firstly. The role of entanglement

(14)

in quantum computing is briefly pointed out. We Note these states form an orthonormal basis. Alice
then show a related but slightly more involved proto- and Bob have agreed on which bit string is associated
col for teleportation i.e. the transfer of a quantum with what state in the Bell Basis.
state from Alice to Bob, without sending the actual

3. Alice sends her qubit to Bob.

particle.
4. Bob performs a measurement on both qubits
in the Bell Basis. From its outcome he knows what
Alice did to her state, and hence which two-bit
Superdense Coding

string she wants to communicate.

Using a method called superdense coding one So the above protocol allows Alice to transmit
can transfer two bits of classical information using two classical bits to Bob through sending one qubit.
one qubit, i.e.

a two-level quantum system.

A It has been at least partially verified experimentally

simple example of the protocol[28] is as follows:

in Innsbruck[17]. Some interesting points to note


are that:

1. Alice and Bob share one qubit each of the


maximally entangled state:

this does not imply superluminal communication as Alice must send her qubit to Bob in a
physical manner.

1
|i = (|00i + |11i)
2

(13)

The maximally entangled state is both necessary and used up in the process. Hence it can

2. Alice decides on one out of four classical bit

be viewed as a resource that is consumed.

strings she wishes to communicate: 00,01,10 or 11.


Quantum Computing

3. She applies local operations on her qubit. Each

Modelling the computer register using quantum

one will turn the maximally entangled state into one physics implies it can have superpositions, entan-

9
glement etc. These effects are exploited in the few using secretly correlated one-time pads, and then the
known distinct types of quantum algorithms. The quantum version where shared entangled states are
Deutsch algorithm[14] later extended by Josza, is used instead.
considered to be the first one. It has no known applications but nevertheless demonstrated a way of
using the quantum superposition to achieve a computational task. Shors algorithm[15] however has
a clear application in that a quantum computer running it could factorise integers into their primes in a

Classical Version

manageable time. Such a computer could break the


RSA cryptosystem which is based around the idea

It has been argued the secret classical correla-

that prime factoring is virtually impossible on a clas- tions are the classical analogue of entanglement [23],
sical computer. Shors algorithm provides an expo- and it is instructive to highlight the classical analnential speed up compared to a classical algorithm. ogy of teleportation[27]. One can transfer data in
In addition, Grovers quantum search algorithm[16] an eavesdropping-safe way by using the one-time
provides a polynomial speed up in finding a given pad protocol. Say Alice works for a government
element from a random collection.

and wants to send data to Bob in one of the em-

It is often claimed that entanglement is what bassies far away. She could then manufacture two
gives quantum computing its power, although one sequences of perfectly correlated random numbers.
can note that the Deutsch algorithm uses no entan- The sequences are placed on two separate one-time
gled states. However quantum computing methods pads. One is delivered secretly to Bob and one is
for error correction also use entanglement[20], and kept by Alice. Now Alice and Bob share secret corthe consensus seems to be that quantum computing relations. They can use these to transfer data sewould at least be severely restricted if entanglement curely over any public communication, e.g. email.
could not be used.

Alice would scramble her data using the string and


then send it. Bob could then de-scramble it as he
Teleportation

knows how Alice scrambled it from looking at his


pad. They dont want to reuse the same pads how-

Another effect which is good for highlighting the ever, as it increases the chances of anyone interceptresource character of entanglement is teleportation. ing their public communications figuring out what
This is essentially the ability to transfer an unknown the one-time pad is. Hence we see how the resource
quantum state from one physical system onto an- of classical secret correlations is used up in order to
other. We first consider the classical analogy of this: transfer data secretly from one place to another.

10
as:

Quantum version

In the quantum version Alice and Bob would instead share entangled pairs and use this as their resource to achieve that transfer of a quantum state. A
key difference in the quantum setting is that there is
no option of copying the initial state , as that will
in general collapse it and thereby destroy the original state. This is known as the no-cloning theorem.
Therefore the state remains unknown throughout the
protocol.

(1)Alice measures on her two qubits in the Bell


basis.
This collapses her part of the state in equation 16
to one of the Bell basis states and Bobs part of the
state accordingly.
(2) She communicates this result to Bob using
classical communication, e.g. by phone
(3) Bob then knows which local operation to do on
his state to turn it back into a|0i + b|1i

The simplest protocol to teleport a state a|0i +


b|1iis to use three qubits initially in the state
One can note that
1
|AB i = (a|0i + b|1i)(|00i + |11i)
2

(15)

entanglement again acts as a resource: it is


necessary and used up as part of the protocol

where the first two qubits are in Alices possession and the third in Bobs. This means Alice and
Bob must have in a sense been spatially connected
before this stage in the protocol. Typically an entangled pair of photons is created on another site and

Alice communicates her result to Bob classically, so the protocol cannot teleport faster
than the speed of light.

one each sent to Alice and Bob respectively.


The state in equation 15 can be rewritten in a
form that highlights the next stages of the protocol

Even though, as we will see, not all entangled


states can be used for teleportation, it has become

1
|AB i = [|+ i(a|0i + b|1i)
2
+ | i(a|0i b|1i)

in effect the new Bells inequality for the quantum


information science community. The protocol described above has played an important role in estab-

+ |+ i(a|1i + b|0i)
+ | i(a|1i b|0i)]

(16)

lishing the paradigm of entanglement as a resource


used to overcome the restraint of local operations

where |+ i etc. were defined in the superdense and classical communication, and of EPR pairs as
coding protocol above. The protocol now proceeds the gold standard of entanglement.

11
ENTANGLEMENT THEORY

lations are possible and how they are carried out


Quantify: Decide how much entanglement is in a

The physicists approach to understanding a state and how efficient entanglement manipulations
natural phenomenon tends to be to make a quanti- can be.
tative model of it. We now review efforts towards We now discuss the progress made to date on these
creating such a theory of entanglement. These ef- three aims respectively.
forts have been much influenced by the paradigm of
entanglement as a resource, as discussed in exam-

Characterising Entanglement

ples above. They have also been influenced by inIt was mentioned in the section on how quantum

formation theory. One can see why this is the case

by considering that correlations are an information theory predicts entanglement that entangled states
theoretic concept, and protocols like teleportation are vectors in a Hilbert space H1 H2 of the comhave analogues in classical information theory. In- posite systems 1 and 2, and that they cannot be writformation theory was founded in order to quantify ten as a product state |i1 |i2 . In general we must
tasks like storing and communicating information consider mixed states, where we would say a state
in an engineering setting [12]. A paradigm which is non-entangled, or equivalently separable, if
arose from studying information is that information
is physical. This refers to the fact that informa-

tion processing is done on physical systems, so the

X
i

pi i1 i2

(17)

laws of computation are consequences of the laws where p are the mixed state probabilities and each
i
of physics. A well known cornerstone of this para- i i can be taken to be a pure product state. How1

digm is Landauers erasure theorem which states that ever, finding such a decomposition for some state
erasing information requires work and creates heat. is in general a non-trivial task, and is known as the
Built with the paradigms of information is physical separability problem. The set of separable states is
and entanglement is a resource for quantum infor- convex, as mixing separable states produces a sepamation processing, some significant advances have rable state, so the separability problem can be dealt
been made towards a theory of entanglement.
Three key aims for a theory on entanglement as a
resource have been formulated [26]:

with using convex analysis.


Note that there is a distinction between classical and quantum correlations; classical correlations

Characterise: Decide which states are entangled and for example would never lead to a violation of the
which not.

Bells inequalities. Hence one cannot just look for

Manipulate: Decide which local operations and clas- correlations in the data of a quantum mechanical exsical communication(LOCC) entanglement manipu- periment unless the state is certain to be pure. We

12
will see this distinction is important in quantifying
the amount of quantum correlations in a mixed state.
One should also try to characterise entanglement physically, which has already been discussed
to some extent in previous sections. There it was argued entanglement was associated with non-locality
in the Bells inequalities, and the ability to perform
new information processing task in e.g. the superdense coding and teleportation protocols. However FIG. 4: The square represents all quantum states. The upit is interesting to consider whether the mathematical characterisation of entanglement entirely coincides with the physical. In fact there are entangled

per half is the separable states, the lower half the entangled. The entangled states are grouped into subsets. Not
all entangled states can be used for teleportation and/or violating bells inequalities (more specifically single copies

states that do not violate Bells inequalities, there are cannot be used). By usable for teleportation is meant a
even entangled states that cannot be used for tele- teleportation fidelity larger than that for separable states.
portation. The latter are known as Bound entangled Bound entangled states (which only exist for qutrits (distates and have the property that taking the transpose mension=3)and higher dimensions) cannot be used for
of the density matrix with respect to either Alice or teleportation and cannot be used for violating Bells inBob preserves the positivity of the eigenvalues of the
density matrix. i.e. TA 0 Figure 4 summarises the
present view on this.

equalities. There are states that are entangled and usable


for teleportation but not Bell violating. The question mark
to the right of the box signifies the possibility of states
that are not use-able for teleportation but violating Bells

One must conclude, therefore, that even though inequalities.


teleportation has proven a very useful protocol for
forming a theory of entanglement, there is currently selection(SS) is also sometimes referred to. One
no quantum information protocol that can be per- of the key axioms of entanglement theory is the
formed with any type of entangled states and not fundamental law of quantum information processwith any separable states. Such a protocol would be ing, that entanglement cannot be increased under
very interesting.

LOCC[21]. This has not been proven from other


axioms of quantum theory, but appears to hold in all

Manipulating Entanglement

cases studied.
An example where this is significant is in so

The most commonly used state manipulations called purification protocols, where Alice and Bob
in entanglement theory involve using local oper- can apply LOCCSS to get EPR pair(s) from several
ations(LO) and classical communication(CC) Sub- weakly entangled pairs. The teleportation protocol

13
above uses maximally entangled pairs. However in shecan
convert this resultant state to approximately
n
practise Alice and Bob will share states |AB ithat
ln maximally entangled pairs[20].
r
are only partially entangled. This can for example
This offers one way of defining the amount of
occur through an interaction by |AB i with an enentanglement in |AB i, namely how many singlets
vironment leading to entanglement between the two.
can be purified from n copies of it in the asymptotic
The new state |AB i is then mixed, as the environlimit of n . It can be shown that here this is
ment is traced out. (This is indeed used as an explan(a2 ln(a2 )b2 ln(b2 )). This quantity corresponds
nation as to why the universe looks classical rather
to the von Neumann Entropy of Alices reduced denthan quantum.[10]) In order to implement the telesity matrix, which is defined in a later section.
portation protocol we then need to employ a protocol
We can then use the amount of entanglement, as
to convert the mixed states to maximally entangled
states. This will need to be done using LOCCSS. measured by the number of EPR pairs we can puThis means we cannot increase the total entangle- rify in this way as a measure of entanglement in that
ment, but if we have many pairs, we can shift it to state.
fewer of those. Several such protocols exist, some
purifying mixed states and some pure states(where

Quantifying Entanglement

purification refers to the entanglement content, and


pure the absence of classical uncertainty).

One can seek to introduce a notion of different

Consider an example of a purification procedure. degrees of entanglement, and look for maps E()
Let Alice and Bob share N copies of:
|AB i = a|00i + b|11i

that will give us a single real number correspond(18)

ing to the entanglement of . One justification for


this is that it appears that some states will not vi-

Where a and b are considered real. The state can be olate Bells inequalities, and some will do it more
written as:
then others. There are now several such accepted
measures of entanglement, all with various moti-

|AB in = an |000...00i +

vations and strengths/weaknesses. For pure bipar-

+ a(n1) b|000...11i + ...

tite states the von Neumann entropy of entanglement

+ ab(n1) |110...00i + ...


n

+ b |111...11i

has been accepted as a canonical measure, for mixed


(19)

states there are several other measures. The choice

Alice can then do projective measurements onto of mixed states measures are then restricted in that
the subspaces with 0,2,4... excitations, each asso- they must correspond to the von Neumann entropy
ciated with a probability given by the binomial sta- of entanglement when applied on pure states.
tistics. Once she has done that, it has been shown

In general, an entanglement measure E() is a

14
map

The reason is the first law of quantum inforE : 7 x such that x+

mation processing mentioned above. A purification


(20)
procedure uses LOCCSS only and if an entangle-

The measure needs to satisfy four conditions (of- ment measure would give an increased value of
ten referred to as requisites to be be an entanglement entanglement after purification there would be a
contradiction with that law.

monotone)[9]:
(1) Separable states have entanglement zero, i.e.

We will now discuss some important entangleseparable E() = 0

(21)

ment measures:

the von Neumann entropy of

(2)Entanglement is unchanged by local unitary entanglement, the entanglement cost and distillability, the relative entropy of entanglement and the

transformations

logarithmic negativity.
E() = E(UA UB UA UB )

(22)
Von Neumann entropy of entanglement

(3)Local Operations(LO), Classical Communication(CC) and sub-selection(SS) cannot increase the

We saw the purification protocol above has a


maximal efficiency that corresponds to the von Neu-

expected entanglement.

mann entropy of Alices (or Bobs) part of the total


state.
7

for

LOCCSS

p i i

E()

EvN () = SvN (A ) = tr{A lnA }


pi E(i )

(25)

(23)
The von Neumann entropy is the quantum gener-

alisation of the Shannon entropy. It maps a density


matrix to a number, its von Neumann entropy.

(4) Additivity of Entanglement


= 1 2

SvN () = tr{ln} =

i lni

(26)

where the i are the eigenvalues of the density ma-

E() = E(1 ) + E(2 )


Proposition[20]:

(24)

Any such entanglement

measure is an upper bound on the efficiency of a

trix.
Alices reduced state A of the total state AB is
A = trB (AB )

(27)

purification procedure.
where the partial trace with respect to B is defined

15
However can be decomposed in different ways,

by
trB (|a1 iha2 | |b1 ihb2 |) = |a1 iha2 |tr(|b1 ihb2 |)

which may not give the same E(). Note this decomposition is different from choice of basis or per-

(28)
forming LOCC, so it is not to say entanglement can
and likewise for a linear input of states. One can note
be increased under LOCC. One therefore defines the
this is independent of choice of basis, since the trace
is. Defining A in this way is the unique way that

entanglement of formation as the minimum entanglement needed to create , i.e. choosing the decom-

predicts the correct measurement statistics for it, i.e.


position that minimises E.
A

AB

tr(M ) = tr(M IB

(29)

EvN is the same for pure states regardless of

EF () = min

=pi i

pi EvN (i )

(31)

whether we trace out Alice or Bob. This can be


shown by Schmidt decomposing the total state, and

This has been shown to be a measure of the minimum amount of entanglement needed to create a

then tracing out the party in question.

One physical interpretation of the partial trace is mixed state . It has been shown to satisfy conditions
that the experimenter is measuring the environment 1,2,3 and numerical studies indicate it also satisfies
4. A drawback is that the minimisation in equation

and then forgetting the result.

The von Neumann entropy of Entanglement sat- 31 is, in general, very difficult to compute.
isfies all four entanglement conditions. As it corre-

There were, even after this measure, incentives

sponds to both the maximal purification efficiency, to look for more measures, for example ones corand the entanglement cost for pure states, it is gen- responding to how much entanglement one can puerally accepted as a canonical bipartite measure of rify/distill out of a state, such as the so called entanentanglement for pure states. However for mixed glement cost.
states it is more complicated, in particular there can
be classical correlations too.
Entanglement cost, and distillability
Entanglement of formation

The entanglement cost[29] of a state is the numThe first proper measure for mixed states was the ber of maximally entangled pairs needed to create
entanglement of formation. One might first tenta- that state using LOCC. It is taken in the asymptotic
tively suggest the entanglement of of a mixed state limit, unlike the entanglement of formation. In this
P
asymptotic limit the two coincide. The entanglement
i pi i as E? given by
E? () =

X
i

pi EvN (i )

cost is a measure of the best (i.e. most economical)

(30)
rate r of converting k singlets to n copies of the state

16
in question. where r=k/n. It is given by

ing separable. The relative entropy of entanglement


ERE is given by

EC () =




n
rn
inf r : lim inf D( , ((2 ))) = 0
n

ERE () = min D(||)


S

(32)

(34)

where S is the set of separable states and D(||)


Where D is some distance measure like the trace dis- refers to some distance measure between the states
tance. is any trace preserving LOCC operation, and . A suitable distance function is the relative
and (q) is a maximally entangled state in q dimen- entropy
sions. EC is very difficult to calculate as one can see
from equation 32, but it is important as it quantifies
how many singlet pairs are needed in the asymptotic

SRE (||) = tr{ln ln}

(35)

limit to create m copies of a given state, like an ex- This corresponds to a measure of how different/distinguishable the two probability functions

change rate.

There is also a measure for the reverse of this and are [28]. The relative entropy of entangleprocess, i.e. what is the best rate at which we can ment satisfies the first three conditions according to
transform n into m singlets. This is termed the analytical proofs but it is unclear whether the fourth
(additivity )is satisfied.

entanglement distillation ED () given by:

A physical insight gained from this is that for so


ED () =

 called Werner states the relative entropy of entangle

n
rn
sup r : lim sup D(( ), ((2 ))) = 1
ment is, in general, less than the entanglement of forn

(33) mation. From the first law and the nature of EF this
implies that one cannot distill out as much entanglewhere the relevant quantities were defined in equa-

ment as it took to create this state, i.e. it is an irre-

tion 32. It seems these quantities have important versible process.


physical interpretations, although they appear notoriously difficult to evaluate for a given state.

Relative Entropy of Entanglement

Logarithmic Negativity

Vidal and Werner[22] proposed a measure of entanglement that is very useful by virtue of being easy

The relative entropy of entanglement measure is to compute, even though it has some weaknesses relless based on the paradigm of teleportation and max- ative to other measures. It is based on the necessary
imally entangled pairs than the previous measures. It criterion for separability of positive partial transis instead a measure of how far a state is from be- pose and uses the trace norm of the partial transpose

17

||TA ||1 = tr A A where TiAAjB kA lb = kA jB iA lb most pure states.


defines TA . Partial transpose is a trace-preserving

map, but does not preserve positivity for two classes Other entanglement measures than those discussed above exist, e.g.

of states:

entanglement.

separable states

the so-called squashed

For a review of entanglement

measures see[9].

positive partial transpose(PPT) entangled


states

Open issues in Entanglement Theory

It creates negative eigenvalues for most entangled


states however.

Open issues remain in entanglement theory.


Apart from the general philosophical conundrum we
can point out two more technical issues relevant to

For separable states


s =

X
k

the thesis: Multipartite Entanglement and Analogies


pk |ek , fk ihek , fk |

(36) between thermodynamics and entanglement theory

we see that the positivity is preserved under partial

Multipartite Entanglement

transpose
Ts A =

X
k

Consider multipartite states such as these

pk |ek , fk ihek , fk | 0

For those that obtain negative eigenvalues i under partial transpose however


X


kTA k1 = 1 +
i

(37)

This can be seen by noting that for hermitian matri-

1
|GHZi = (|000i + |111i)
2
|W i

1
= (|100i + |010i + |001i) (39)
3

These, known as GHZ and W-class states respec-

ces the trace norm used here equals the sum of the tively, are incomparable under LOCC, meaning they
absolute eigenvalues.

cannot be transformed into one another using LOCC

The logarithmic negativity EN is then defined as with a non-zero probability. This causes problems
EN () = log2 ||TA ||1

(38)

when trying to create entanglement measures for


these. However if allowing positive partial trace pre-

This is additive due to the logarithm. It exhibits serving operations(PPT), other transforms that are
a type of monotonicity under LOCC, and is 0 for not possible under LOCC can be possible[26]. For
separable states.

However it does not reduce to example the GHZ to W transition can be done with

von Neumann entropy measure of entanglement for 75% probability, and all N-partite pure entangled

18
states are interconvertible under PPT- operations at

Thermodynamics

the single copy level. This may be an approach to-

2nd Law, No perpetual

wards getting around the incomparability problem.

motion machine

It would also be interesting to understand N-partite

heat engine

classical secret correlations, in case the difficulties

reversible heat engine

Entanglement Theory

increase under LOCC


entanglement transformation

irreversible heat

irreversible

engine

tween entanglement and secret classical correlations

reversible
entanglement transformation

with the multi-particle setting are in essence classical. Rohrlich et al fruitfully study the analogy be-

1st Law: no entanglement

entanglement transformation

energy reservoir

access to entangling

in [23], but do not suggest a classical analogy for

(global) operations

multipartite entanglement. Yet another option may

work

be offered by efforts to use possible analogies be-

heat

tween entanglement theory and thermodynamics.

thermodynamic limit

Formal analogy with thermodynamics

There are similarities between thermodynamics


and entanglement theory. Some examples of analogies that have been used are in the table below.
The last two are examples of open questions. The
issue of equilibria has typically been addressed as
a mathematically typical state, see e.g.[7] but we
are now considering more physical settings. The
no-cloning theorem, or more properly no-perfect

entanglement creation
bound entanglement
asymptotic limit

of many particles

of many copies

density function
..
.

wave function
..
.

secret correlations

entanglement

equilibrium

tion became a new paradigm after Bells inequalities.


One reason to be hopeful of this is that [24] show
work can be drawn from bound entanglement(if not
all types of entanglement), and [25] offer an inequality that is satisfied by all separable states(and violated by some entangled states).

cloning theorem has no known analogy in thermodynamics.

CONCLUSION

It was mentioned before that it would be interesting with a physical characterisation of entanglement

Entanglement is an interesting topic, both for its

through a process which will detect all types of non- philosophical and technological implications. Philoseparable states. Here e.g. efforts in [24] and [25] are sophically it begs us the question: is the universe lointeresting. They investigate ways of associating en- cal and real? This was first put into an experimentanglement with work extractable from a thermody- tally testable form by Bell. Experiments then ruled
namical system, a line of investigation which could in favour of the quantum mechanical description of
lead to a new paradigm, in the same way teleporta- the world being either non-local and non-real or one

19
of the two. Technologically, entanglement can be
used as a resource to perform protocols like quantum
teleportation. This lead to a paradigm of entanglement as a resource in quantum information processing. Efforts to properly quantify entanglement the-

[1] A. Einstein,B. Podolsky and N.Rosen, Can


Quantum-Mechanical Description of Physical Reality be Considered Complete? Physical Review, Vol
47, 1935

ory highlight various complexities, for example, not


all entangled states exhibit non-locality in the sense
prescribed by the Bell inequalities. Introducing measures of entanglement help to clarify such issues,
and have led to a fundamental law of quantum information processing: that entanglement cannot be
increased using local operations and classical com-

[2] J.S.Bell Speakable and Unspeakable in Quantum


Mechanics, Cambridge Univ. Press 1987
[3] M. Redhead Incompleteness, Non-locality and Realism(Chapter 4) Oxford, Clarendon Press, 1987.
[4] A. Fine The Shaky Game- Einstein Realism and
the Quantum Theory Chicago, 1996.
[5] A. Aspect, J. Dalibard, G.Roger Experimental Test
of Bells inequalities Using Time-Varying Analyz-

munications. Open questions in entanglement theory include multipartite measures of entanglement,

ers PRL,volume 49, no 25, 1982.


[6] B. Groisman, S. Popescu, A.Winter On the quan-

and whether there is a formal analogy between en-

tum, classical and total amount of correlations in a

tanglement theory and thermodynamics. Apart from

quantum state quant-ph/0410091, 2004

technological advances I hope efforts in entanglement theory ultimately can lead to an understanding
of what quantum mechanics is telling us about the
nature of reality.

[7] P.Hayden,D.W.Leung,

A.Winter

Aspects

of

generic entanglement quant-ph/0407049, 2004


[8] D.Collins, S.Popescu A classical analogue of entanglement quant-ph/0107082, 2004
[9] M.B.Plenio,S.Virmani An introduction to Entanglement Measures quant-ph/0504163, 2005
[10] J. Bub Interpreting the Quantum World Cambridge Univ. Press, 1997 .
[11] D.Collins, S.Popescu A classical analog of entanglement Phys. Rev. A 65, 032321 (2002).
[12] C. Shannon, A mathematical theory of communication can be downloaded from http://cm.belllabs.com/cm/ms/what/shannonday/paper.html
[13] A.Aspect, P.Grangier, G.Roger, Expt test of real.
local theories via Bells thm. PRL, Vol47, Number7
[14] Deutsch, D., Quantum Theory, the Church-Turing
Principle and the Universal Quantum Computer.
Procedings of the Royal Society of London, Series

20
A 400 (1985), 97-117.
[15] P.W.ShorPOLYNOMIAL-TIME ALGORITHMS

[22] G.Vidal and R.F.Werner A computable measure of


entanglement quant-ph/0102117.

FOR PRIME FACTORIZATION AND DISCRETE [23] S. Popescu, D.Rohrlich Thermodynamics and the
LOGARITHMS ON A QUANTUM COMPUTER

measure of entanglement quant-ph/0102117.

SIAM J. COMPUT. Vol. 26, No. 5, pp. 1484-1509, [24] J.Oppenheim,M.Horodecki,P.Horodecki,


1997
[16] L. Grover, Quantum Computers Can Search Arbitrarily Large Databases by a Single QueryPRL
VOL 79, NUMBER 23
[17] Klaus Mattle, Harald Weinfurter, Paul G. Kwiat,
and Anton Zeilinger. Dense coding in experimental

R.Horodecki

Thermodynamical

approach

to

Quantifying Quantum Correlations PRL, Vol 89,


no. 18
[25] K. Maruyama, F.Morikoshi, V.Vedral Thermodynamical detection of entanglement by Maxwells
Demons quant-ph/0102117.

quantum communication. Physical Review Letters, [26] S. Ishizaka, M. B. Plenio Multi-particle entangle76(25):4656-4659, 1996.
[18] W. Tittel*, J. Brendel, B. Gisin, T. Herzog, H.

ment manipulation under positive partial transpose


preserving operations quant-ph/0412193.

Zbinden, and N. Gisin, Experimental demonstra- [27] M. B. Plenio Notes for IX Escuala de Otono de
tion of quantum correlations over more than 10 km,
Phys. Rev. A 57, 32293232 (1998)
[19] V.Vedral and M.B.Plenio Entanglement Measures
and Purification Procedures Phys. Rev. A, Vol. 57,
1998.

Fisica Teorica
[28] M.A.Nielsen, I.L.Chuang Quantum Computation
and Quantum Information Cambridge Univ. Press,
2000.
[29] The entanglement cost, entanglement distillability

[20] M.B.Plenio and V.Vedral Teleportation, entangle-

and von Neumann entropy of entanglement are all

ment and thermodynamics in the quantum world

examples of concepts in quantum information sci-

Cont. Phys. , Vol. 39, 1998.

ence where Charles Bennett played a leading role.

[21] V. Vedral, M.B. Plenio, M.A. Rippin and P.L.


Knight., Quantifying entanglement, Phys. Rev.
Lett. 78,2275 (1997).

You might also like