Contem porary Physics, 1998, volum e 39, num ber 6, pages 431 ± 446

Teleportation, entanglement and thermodynamics in the

quantum world

M ARTIN B. P LENIO and V LATKO V EDR AL

Q uantum mechanics has m any counter -intuitive consequences which contradict our intuition
w hich is based on classical physics. H ere we discuss a special aspect of quantum mechanics,
nam ely the possibility of entangl ement between two or more particles. W e w ill establish the
basic properties of entangl ement using quantum state teleportation. These principles will
the n allow us to formulate quantit ative measures of entangle ment. Finally we will show that
the sam e general principles can also be used to prove seemingly di cult questions regarding
entangle ment dynam ics very easily. This w ill be used to m otivate the hope that we can
construct a thermodynam ics of entangl ement.

1. Introduction
Quantum mechanics is a non-classical theory and therefore
exhibits m any eŒects that are counter-intuitive. This is
because in our everyday life we experience a classical
(m acroscopic) w orld w ith respect to w hich w e de® ne
`com mon sense’. One principle that lies at the heart of
quantum mechanics is the superposition principle. In itself
it m ight still be understood w ithin classical physics, as it
crops up, for exam ple in classical electrodynam ics. H ow -
ever, unlike in classical theory the superposition principle in
quantum mechanics also gives rise to a property called
entan glement between quantum mechanical systems. This
is due to the H ilbert space structure of the quantum
mechanical state space. In classical mechanics particles can
be correlated over long distances simply because one
observer can prepare a system in a particular state and
then instruct a diŒerent observer to prepare the same state.
However, all the correlations generated in this way can be
understood perfectly well using classical probability dis-
tributions and classical intuition. The situation changes
dram atically w hen w e consider correlated systems in
quantum m echanics. In quantum m echanics we can prepare
two particles in such a way that the correlations betw een
them cannot be explained classically. Such states are called
entan gled states. It was the great achievem ent of Bell to
recognize this fact and to cast it into a math ematical form
that, in principle, allows the test of quantum mechanics
Authors’ address: Black ett Lab oratory, Im perial C ollege, Prin ce C onsort
Road, London SW 7 2BZ, U K .
V. V edral’ s present address: C laren don Laboratory, U niversity of O xford,
Parks Ro ad, Oxford O X1 3PU , U K .
0010-7514 /98 $12.00 Ó 1998 Taylor & Francis Ltd
again st local realistic theories [1 ± 4]. Such tests have been
performed, and the quantum mechanical predictions have
been con® rmed [5] although it should be noted that an
experim ent that has no loopholes (these are insu ciencies
in the experim ent that allow the simple construction of a
local hidden variable theory) has not yet been perform ed
[6]. W ith the form ulation of the Bell inequalities and the
experim ental demonstration of their violation, it seemed
that the question of the non-locality of quantum m echanics
had been settled once and for all. How ever, in recent years
it turned out this conclusion w as prematu re. W hile indeed
the entanglement of pure states can be view ed as well
understood, the entanglement of mixed states still has m any
properties that are m ysterious, and in fact new problems
(some of which w e describe here) keep appearing. T he
reason for the problem with m ixed states lies in the fac t that
the quantum content of the correlations is hidden behind
classical correlations in a m ixed state. O ne might expect
that it would be im possible to recover the quantum content
of the correlations but this conclusion w ould be w rong.
Special methods have been developed that allow us to
`distil’ out the quantum content of the correlations in a
m ixed quantum state [7 ± 11]. In fact, these methods show ed
that a m ixed state which does not violate B ell inequalities
can nevertheless reveal quantum mechanical correlations,
as one can distil from it pure m axim ally entan gled states
that violate Bell inequalities. Therefore, B ell inequalities are
not the last w ord in the theory of quantum entanglement.
T his has opened up a lot of interesting fundam ental
questions about the natu re of entan glem ent and w e w ill
discuss some of them here. W e will study the problem of
432 M. B. Plenio and V. Vedral
how to quantify entanglement [12 ± 15], the fundam ental
laws that govern entanglement transformation and the
connection of these laws to thermodynam ics.
On the other hand, the new interest in quantum
entan glement has also been triggered by the discovery
that it allows us to transfer (teleport) an unknown
quantum state of a tw o-level system from one particle to
another distant particle without actually sending the
particle itself [16]. As the particle itself is not sent, this
represents a method of secure transfer of inform ation
from sender to receiver (com monly called A lice and
Bob), and eavesdropping is impossible. T he key ingre-
dient in teleportation is that Alice and Bob share a
publicly know n maxim ally entan gled state between them.
To generate such a state in practice one has to employ
methods of quantum state distillation as mentioned
above which w e review in section 3. The protocol of
quantum teleportation has been recently im plemented
experim entally using single photons in laboratories in
Innsbruck [17] and Rome [18], which only adds to the
enormous excitem ent that the ® eld of quantum informa-
tion is currently generating.
But perhaps the m ost spectacular application of en-
tan glement is the quantum com puter, which could allow,
once realized, an exponential increase of com putational
speed for certain problems such as for exam ple the
fac torization of large num bers into primes, for further
explanations see the reviews [19 ± 21]. A gain at the heart of
the idea of a quantum computer lies the principle of
entan glement. This oŒers the possibility of massive
parallelism in quantum system s as in quantum mechanics
n
n quantum systems can represent 2 numbers simulta-
neously [19,20,22]. The disruptive in¯ uence of the environ-
ment makes the realization of quantum computing
extremely di cult [23,24] and many ideas have been
developed to combat the noise in a quantum computer,
incidentally again using entanglement [25 ± 28]. M any other
applications of entanglement are now being developed and
investigate d, e.g. in frequency standards [29], distributed
quantum computation [30,31], multiparticle entan glem ent
swapping [32] and multiparticle entan glem ent puri® cation
[33].
In this article we w ish to explain the basic ideas and
problems behind quantum entan glem ent, address som e
fundam ental questions and present some of its conse-
quences, such as teleportation and its use in (quantum )
communication. Our approach is somewhat unconven-
tional. Entan glem ent is usually introduced through
quantum states which violate the classical locality
requirement (i.e. violate Bell’s inequalities) as w e have
done above. H ere w e abandon this approach altogether
and show that there is much more to entan glem ent than
the issue of locality. In fact, concentrating on other
aspects of entanglement helps us to view the natu re of
quantum mechanics from a diŒerent angle. W e hope that
the reader w ill, after studying this article, share our
enthusiasm for the problems of the new and rapidly
expanding ® eld of quantum inform ation theory, at the
heart of which lies the phenomenon of quantum correla-
tions and entanglement.
2. Q uantum teleportation
W e ® rst present an exam ple that crucially depends on the
existence of quantum mechanical correlations, i.e. entan -
glem ent. The procedure w e will analyse is called quantum
teleportation and can be understood as follows. The naive
idea of teleportation involves a protocol whereby an object
positioned at a place A and time t ® rst `dem ate rializes’ and
then reappears at a distant place B at som e later time t+ T.
Quantum teleportation im plies that w e wish to apply this
procedure to a quantum object. H owever, a genuine
quantum teleportation diŒers from this idea, because we
are not teleporting the w hole object but just its state from
particle A to particle B. As quantum particles are
indistinguishable anyway, this am ounts to `real’ teleporta-
tion. O ne way of performing teleportation (an d certainly
the way portrayed in var ious science ® ction movies, e.g.
The Fly) is ® rst to learn all the properties of that object
(thereby possibly destroying it). W e then send this
information as a classical string of data to B where another
object with the same properties is re-created. One problem
with this picture is that, if we have a single quantum system
in an unknown state, we cannot determine its state
completely because of the uncertainty principle. M ore
precisely, we need an in® nite ensemble of identically
prepared quantum systems to be able completely to
determ ine its quantum state. So it would seem that the
laws of quantum mechanics prohibit teleportation of single
quantum system s. H ow ever, the very featu re of quantum
mechanics that leads to the uncertainty principle (the
superposition principle) also allows the existence of
entan gled states. These entangled states will provide a
form of quantum channel to conduct a teleportation
protocol. It w ill turn out that there is no need to learn
the state of the system in order to teleport it. O n the other
hand, there is a need to send some classical inform ation
from A to B, but part of the inform ation also travels down
an entangled channel. This then provides a w ay of
distinguishing quantum and classical correlations, which
we said w as at the heart of quantifying entanglement. After
the teleportation is com pleted, the original state of the
particle at A is destroyed (although the particle itself
remains intact) and so is the entan glem ent in the quantum
channel. T hese tw o features are direct consequences of
fundam ental law s that are central for understanding
entan glement as we explain in more detail in the next
subsection.
 Teleportation, entanglement and thermodynamics in the quantum world
2.1. A basic description of teleportation
Let us begin by describing quantum teleportation in the
form originally proposed by Bennett et al. [16]. Suppose
that Alice and B ob, w ho are distant from each other, wish
to im plement a teleportation procedure. Initially they need
to share a maxim ally entan gled pair of quantum mechan-
ical two level systems. A two level system in quantum
mechanics is also called a quantum bit, or qubit [34], in
direct analogy with the classical bit of information (which is
just tw o distinguishable states of som e system ). U nlike the
classical bit, a qubit can be in a superposition of its basis
states, like | ñ 5 a | 0ñ 1 b | 1ñ . This m eans that if A lice and
Bob both have one qubit each then the joint state m ay for
exam ple be
| AB ñ 5 ( | 0A ñ | 0B ñ 1 | 1A ñ | 1B ñ ) 22 1 /2 , ( 1)
where the ® rst ket (with subscript A) belongs to Alice
and second (with subscript B) to Bob. This state is
entan gled, m eaning that it cannot be written as a product
of the individual states (like e.g. | 00ñ ). N ote that this state is
diŒerent from a statistical m ixture ( 00ñ á 00| 1 | 11ñ á 11| ) /2
which is the most correlated state allowed by classical
physics.
N ow suppose that Alice receives a qubit in a state
which is unknow n to her (let us label it | ñ 5 a | 0ñ 1 b | 1ñ )
and she has to teleport it to Bob. T he state has to be
unknow n to her because otherwise she can just phone Bob
up and tell him all the details of the state, and he can then
recreate it on a particle that he possesses. If Alice does not
know the state, then she cannot m easure it to obtain all
the necessary information to specify it. Therefore she has
to resort to using the state | AB ñ that she shares with
Bob. T o see w hat she has to do, we write out the total
state of all three qubits
| AB ñ :5 | ñ | AB ñ 5 (a | 0ñ 1 b | 1ñ )( | 00ñ 1 | 11ñ ) 22 1 /2. ( 2)
However, the above state can be written in the following
convenient way (here we are only rewriting the above
expression in a diŒerent basis, and there is no physical
process tak ing place in between)
| AB ñ 5 (a | 000ñ 1 a | 011ñ 1 b | 100ñ 1 b | 111ñ ) 22 1 /2
1 ( 3)
5 12 [ | ñ ( a | 0ñ 1 b | 1ñ ) 1 | 2 ñ (a | 0ñ 2 b | 1ñ )
1 | 1 ñ ( a | 1ñ 1 b | 0ñ ) 1 | 2 ñ (a | 1ñ 2 b | 0ñ ) ]
where
1 2 1 /2
| ñ 5 ( | 00ñ 1 | 11ñ ) 2 , ( 4)
2 2 1 /2
| ñ 5 ( | 00ñ 2 | 11ñ ) 2 , ( 5)
1
| ñ 5 ( | 01ñ 1 | 10ñ ) 22 1 /2, ( 6)
2
| ñ 5 ( | 01ñ 2 | 10ñ ) 22 1 /2, ( 7)
form an orthonorm al basis of Alice’s tw o qubits
(rem ember that the ® rst two qubits belong to A lice and
433
the last qubit belongs to Bob). The above basis is
frequently called the Bell basis. This is a very useful way
of writing the state of Alice’ s two qubits and Bob’ s single
qubit because it displays a high degree of correlations
between A lice’s and B ob’ s parts: to every state of Alice’ s
1
two qubits (i.e.| ñ , | 2 , | 1 ñ , | 2 ñ ) corresponds a state
of B ob’ s qubit. In addition the state of Bob’ s qubit in all
four cases looks very much like the original qubit that
A lice has to teleport to Bob. It is now straightforward to
see how to proceed with the teleportation protocol [16]:
(1) Upon receiving the unknown qubit in state | ñ Alice
performs projective measurements on her two qubits
in the Bell basis. This means that she will obtain one
of the four Bell states randomly, and with equal
probability.
(2) Suppose Alice obtains the state | ñ . Then the state of
all three qubits (A lice+ B ob) collapses to the
following state
1
| ñ (a | 1ñ 1 b | 0ñ ) . ( 8)
(the last qubit belongs to Bob as usual). A lice now
has to communicate the result of her measurement to
B ob (over the phone, for exam ple). The point of this
com munication is to inform Bob how the state of his
qubit now diŒers from the state of the qubit Alice
w as holding previously.
(3) Now B ob knows exactly what to do in order to
com plete the teleportation. H e has to apply a unitary
transform ation on his qubit which sim ulates a logical
N OT operation: | 0ñ ® | 1ñ and | 1ñ ® | 0ñ . He thereby
transform s the state of his qubit into the state
a | 0ñ 1 b | 1ñ , which is precisely the state that Alice had
to teleport to him initially. This com pletes the
protocol. It is easy to see that if Alice obtained
som e other Bell state then Bob w ould have to apply
som e other simple operation to complete teleporta-
tion. W e leave it to the reader to work out the other
two operations (note that if Alice obtained | 1 ñ he
w ould not have to do anyth ing). If | 0ñ and | 1ñ are
w ritten in their vector form then the operations that
B ob has to perform can be represented by the Pauli
spin matrices, as depicted in ® gure 1.
An important fac t to observe in the above protocol is
that all the operations (Alice’ s measurements and Bob’ s
unitary transformations) are local in natu re. This means
that there is never any need to perform a (global)
transformation or m easurement on all three qubits
sim ultaneously, which is what allows us to call the above
protocol a genuine teleportation. It is also important that
the operations that Bob perform s are independent of the
state that A lice tries to teleport to Bob. Note also that the
classical communication from Alice to Bob in step 2 above
434 M. B. Plenio and V. Vedral
Figure 1. The basic steps of quantum state teleportation. Alice
and Bob are spatially separated, Alice on the left of the dashed
line, Bob on the right. (a) Alice and Bob share a maximally
entangled pair of particles in the state ( | 00 ñ 1 | 11 ñ ) / 2 / . Alice
1 2
wants to teleport the unknown state | w ñ to Bob. (b) The total
state of the three particles that Alice and Bob are holding is
rewritten in the Bell basis equations (4) ± (7) for the two particles
Alice is holding. Alice performs a measurement that projects the
state of her two particles onto one of the four Bell states. (c) She
transmits the result encoded in the numbers 0, 1, 2, 3 to Bob, who
performs a unitary transformation 1, r z , r x , r z r x that depends
only on the measurement result that Alice obtained but not on the
state |w ñ ! (d) After Bob has applied the appropriate unitary
operation on his particle he can be sure that he is now holding the
state that Alice was holding in (a).
is crucial because otherwise the protocol would be
im possible to execute (there is a deeper reason for this: if
we could perform teleportation w ithout classical comm u-
nication then A lice could send m essages to Bob faster than
the speed of light, see e.g. [35]).
Im portant to observe is also the fact that the initial state
to be teleported is at the end destroyed, i.e it becomes
maxim ally m ixed, of the form ( | 0ñ á 0| 1 | 1ñ á 1| ) /2. T his has
to happen since otherwise we w ould end up with two qubits
in the sam e state at the end of teleportation (one with Alice
and the other one with B ob). So, eŒectively, we w ould clone
an unknown quantum state, which is impossible by the law s
of quantum m echanics (this is the no-cloning theorem of
W ootters and Zurek [36]). W e also see that at the end of the
protocol the quantum entanglement of | AB ñ is completely
destroyed. Does this have to be the case in general or m ight
we save that state at the end (by perhaps perform ing a
diŒerent teleportation protocol)? Could we for exam ple
Figure 2. Again Alice is on the left of the dashed line and Bob
on the right side. Assume that initially Alice and Bob are sharing
two particles in a maximally entangled state |w ñ . Alice also holds
a particle in an unknown state q while Bob holds a particle in the
known state | 0 ñ . The aim is that ® nally Alice and Bob have
exchanged the states of their particles and that they are still
sharing a pair of particles in the maximally entangled state |w ñ .
The question whether this protocol is possible will be answered in
section 5.
have a situation as depicted in ® gure 2, w here A lice
teleports a quantum state from to B ob and afterwards the
quantum channel is still preserved. This would be of great
practical advantag e, because we could use a single
entan gled state over and over again to teleport an unlimited
num ber of quantum states from A lice to Bob (this question
was ® rst suggested to the authors by A . Ekert). U nfortu-
nately the answer to the above question is N O: the
entan glement of the quantum channel has to be destroyed
at the end of the protocol. T he analytical proof of this
seems to be extrem ely hard, because it appears that we have
to check all the possible puri® cation protocols (in® nitely
many). How ever, the rest of this article introduces new
ideas and principles that will allow us to explain more easily
why this needs to be so. T his explanation will be presented
at the end of this article. First, how ever, w e need to
understand why entan glement is necessary for teleportation
in the ® rst place.
2.2. W hy is entangl ement necessary?
Quantum teleportation does not work if Alice and Bob
share a disentan gled state. If w e tak e that | A B ñ 5 | 00ñ and
run the same protocol as the above, then Bob’ s particle
stays the same at the end of the protocol, i.e. there is no
teleportation. In this case the total state of the three qubits
would be
| 1ñ 5 ( a | 0ñ 1 b | 1ñ ) | 00ñ . ( 9)
W e see that w hate ver we do (or, rather, w hate ver A lice
does) on the ® rst tw o qubits and how ever we transform
them, the last qubit (B ob’ s qubit) will always be in the state
| 0ñ ; it is thus completely uncorrelated to Alice’s tw o qubits
and no teleportation is possible.
 Teleportation, entanglement and thermodynamics in the quantum world
Thus one might be tempted to say that teleportation is
unsuccessful because there are no correlations between A
and B, i.e. A and B are statistically independent from each
other. So, let us therefore try a state of the form
q AB 5 1 /2 ( | 00 ñ á 00 | 1 | 11 ñ á 11 | ).
This state is a statistical mixture of the states | 00 ñ and | 11 ñ ,
both of w hich are disentangled. This is equivalent to Alice
and B ob sharing either | 00 ñ or | 11 ñ , but being com pletely
uncertain about which state they have. This state is clearly
correlated, because if Alice has 0 so does Bob, and if Alice
has 1 so does Bob. H ow ever, since both the states are
disentan gled and neither one of them achieves teleportation
then their mixture cannot do it either. T he interested reader
can convince him self of this fact by actually performing the
necessary calculation, which is m essy but straightforward.
It is im portant to stress that Alice is in general allowed to
perform any measurement on her qubits and B ob any state
independent transform ation on his qubit, but the teleporta-
tion w ould still not work with the above state [37]. In fact,
{ }
it follows that if | a Ai ñ is a set of states belonging to Alice
{ }
and | b B ñ a set of states belonging to Bob, then the m ost
i
general state that cannot achieve teleportation is of the
form
p ij | a ñáa | Ä |b
5
i i
r AB A A
ij
where p ij are a set of probabilities such that S ij p ij = 1. This
is therefore the m ost general disentangled state of two
qubits. This state might have a certain am ount of classical
correlations as we have seen above, but any form of
quantum correlations, i.e. entanglement, is com pletely
absent [11]. So w e can now sum marize: both classical and
quantum correlations are global properties of two corre-
lated system s, however, they can be distinguished because
classical correlations alone cannot lead to teleportation.
This establishes an important fact: entan glem ent plays a
key role in the m anipulation of quantum inform ation.
2.3. The non-inc rease of entangle m ent under local ope rations
The above discussion leads us to postulate one of the
central laws of quantum information processing. W e now
wish to encapsulate the fact that if A lice and Bob share no
entan glement they can by no local means and classical
communication achieve teleportation.
The fundam ental law of quantum information processing.
Alice and B ob cannot, w ith no matter how small a
probability, by local operations and com municating
classically turn a disentangled state r A B into an
entan gled state.
435
The gist of the proof relies on reductio ad absurdum .
Suppose they could turn a disentan gled state r A B into an
entangled state by local operations and classical comm u-
nication. If so, then they can use the so obtained entangled
state for teleportation. T hus in the end it would be possible
(10)
to teleport using disentangled states which contradicts the
previous subsection. N ote the last part of the fundam ental
law which says `w ith no matte r how small a probability’ .
T his is, of course, very im portan t to stress as we have seen
that teleportation is not possible at all with disentangled
states.
In this paper we will work w ith a m ore general var iant of
the above law , which is more suitable for our purposes. W e
have seen that non-local featu res (i.e. entanglement) cannot
be created by acting locally. This implies that if Alice and
B ob share a certain am ount of entanglement (the notion of
the am ount of entan glement will be m ade more precise later
on) initially, they cannot increase it by only local actions
aided with the classical comm unication. So we can now
restate the fundam ental law in the following, more general,
w ay.
The fundam ental law of quantum information processing
(2. form ulation).
ñáb |,
j j
B B (11) By local operations and classical com munication alone,
Alice and B ob cannot increase the total am ount of
entanglement which they share.
Note that, contrary to the previous form ulation, the
addition `with no matter how sm all a probability’ is
m issing. This law thus says that the total (or rather,
expected) entan glem ent cannot be increased. T his still
leaves room, that w ith som e probability, Alice and Bob can
obtain a more entangled state. Then, however, w ith some
other probability they will obtain less entangled states so
that on average the m ean entan glem ent w ill not increase.
T he above law, it must be stressed, looks deceptively
sim ple, but we will see that it leads to som e profound
implications in quantum inform ation processing. Although
it is derived from considerations of the teleportation
protocol, it nevertheless has m uch wider consequences.
F or exam ple, we have established that if Alice and Bob
share disentangled states of the form in equation (11) then
no teleportation is possible. But w hat about the converse: if
they share a state not of the form given in equation (11) can
they always perform teleportation? Nam ely, even if the
state contains a small am ount of entanglement, can that
always be used for teleportation? This am ounts to asking
w hether, given any entangled state (i.e. a state not of the
form in equation (11), Alice and Bob can, with some
probability, obtain the state ( | 00ñ 1 | 11ñ ) 22 1 /2 by acting
only locally and communicating classically. Also we stated
436 M. B. Plenio and V. Vedral
that entanglement cannot increase under local operations,
but in order to check w hether it has increased we need som e
measure of entan glem ent. All these questions will be
discussed in the following section. A t the end, we stress
that the above law is a working assumption and it cannot
be proved math em atically. It just so happens that by
assum ing the validity of the fundam ental law we can derive
some very useful results, as will be shown in the rest of the
article.
3. C an we am plify and quantify entangl em ent?
In the previous section we have learnt that entan glement is
a property that is essentially diŒerent from classical
correlations. In particular entan glement allows the trans-
mission of an unknow n quantum state using only local
operations and classical com munication. W ithout Alice
and Bob sharing one maxim ally entan gled state this task
can not be achieved perfectly. T his impossibility is directly
related to the fac t that it is not possible to create quantum
correlations, i.e. entanglement, using only local operations
and classical comm unication. This m eans that if we start
with a completely uncorrelated state, e.g. a product state,
then local operations and classical com munication can only
produce a classically correlated state, which is the essence of
the fundam ental law stated in the previous section. W e w ill
now discuss quantum state teleportation again but now not
under ideal conditions but under circum stances that may
occur in an experiment, in particular under circumstances
where decoherence and dissipation are im portant. This
new, realistic, situation gives rise to a new idea which is
called entan glem ent puri® cation.
3.1. E ntangle ment puri® cation
In the previous section we have learnt that starting from a
product state and using only local operations and classical
communication, the best we can achieve is a classically
correlated state, but w e w ill never obtain a state that
contains any quantum correlations. In particular w e w ill
not be able to teleport an unknow n quantum state if w e
only share a classically correlated quantum state.
The im possibility of creating entanglement locally poses
an important practical problem to A lice and B ob when they
want to do teleportation in a realistic experimental situation.
Imagine A lice wants to teleport a quantum state to B ob.
Furthermore assum e that A lice and B ob are really far apart
from each other and can exchange quantum states only for
exam ple through an optical ® bre. The ® bre, w hich w e w ill
occasionally call a quantum channel, is really long and it is
inevitable that it contains fau lts such as impurities which w ill
disturb the state of a photon that we send through the ® bre.
For teleportation Alice and Bob need to share a m axim ally
entan gled state, e.g. a singlet state. H ow ever, w henever Alice
prepares a singlet state on her side and then sends one half of
it to Bob the impurities in the ® bre will disturb the singlet
state. Therefore, after the transm ission A lice and Bob w ill
not share a singlet state but some m ixed state that is no
longer m axim ally entan gled. If Alice attem pts teleportation
with this perturbed state, Bob will not receive the quantum
state A lice tried to send but some perturbed (an d usually
mixed) state. F acing this situation, A lice and Bob becom e
quite desperate , because they have learnt that it is not
possible to create quantum entan glement by local opera-
tions and classical comm unication alone. Because Alice and
Bob are so far apart from each other, these are the only
operations availab le to them . T herefore Alice and Bob
conclude that it w ill be impossible to `repair’ the state they
are sharing in order to obtain a perfect singlet between them.
Luckily Alice and B ob have some friends w ho are physicists
(called say C harles, Gilles, Sandu, Benjam in, John and
W illiam) and they tell them of their predicam ent and ask for
advice. In fact Charles, Gilles, Sandu, Benjamin, John and
W illiam con® rm that it is impossible to create entan glem ent
from nothing (i.e. local operations and classical comm u-
nication starting with a product state). H ow ever, they
inform Alice and Bob that w hile it is im possible to create
quantum entanglement locally w hen you have no initial
entan glement, you can in som e sense am plify or, better,
concentrate entanglement from a source of w eakly en-
tan gled states to obtain som e m axim ally entan gled states
[7,8,10,11,26] (this was the more general formulation of the
fundam ental law). T he purpose of this section is to explain
brie¯ y two particular implementations (there are too m any
to discuss all of them ) of these entan glem ent puri® cation
methods in order to convince A lice, Bob and the reader that
these methods really work.
One m ain diŒerence betw een the existing puri® cation
schemes is their generality, i.e. w hether they can purify an
arbitrary quantum state or just certain subclasses such as
pure states. In fact the ® rst puri® cation schemes [7,10] were
not able to purify any arbitrary state. One schem e could
purify arbitrary pure states [7] (to be described in the
following subsection) while the other could purify certain
special classes of mixed state [10]. Here w e will present a
scheme that can purify arbitrary (pure or mixed) bipartite
states, if these states satisfy one general condition. This
condition is expressed via the ® delity F( q ) of the state q ,
which is de® ned as
F( q ) 5 max á w | q |w ñ . ( 12)
{all max. ent.|w ñ }
In this expression the m axim ization is tak en over all
maxim ally entan gled states, i.e. over all states that one can
obtain from a singlet state by local unitary operations. The
scheme we are presenting here requires that the ® delity of
the quantum state is larger than 0.5 in order for it to be
puri® able.
 Teleportation, entanglement and thermodynamics in the quantum world 437

A lthough one can perform entanglement puri® cation
acting on a single pair of particles only [7,10,35], it can be
show n that there are states that cannot be puri® ed in this
way [38]. Therefore we present a scheme that acts on two
pairs simultaneously. This m eans that Alice and Bob need
to create initially two non-maxim ally entan gled pairs of
states w hich they then store. This and the following
operations are show n in ® gure 3. Now that Alice and Bob
are holding the two pairs, both of them perform two
operations. First Alice perform s a rotation on the two
particles she is holding. T his rotation has the eŒect that
T he last step in the puri® cation procedure consists of a
m easurement that both Alice and Bob perform on their
particle of the second pair. They inform each other about
the measurement result and keep the ® rst pair if their
results coincide. Otherwise they discard both pairs. In
each step they therefore discard at least half of the pairs.
F rom now on w e are only interested in those pairs that
are not discarded. In the Bell basis of equations (4) ± (7)
w e de® ne the coe cients
1 1
A 5 á |q | ñ, (19)

2 2
| 0ñ 2 i | 1ñ B 5 á |q | ñ, (20)
| 0ñ ® , ( 13)
21 /2 1 1
C 5 á |q | ñ, (21)
| 1 ñ 2 i| 0ñ 2 2
|1 ñ ® . (14) D 5 á |q | ñ. (22)
2 1 /2
Bob performs the inverse of this operation on his F or the state of those pairs that we keep we ® nd that
particles. Subsequently both A lice and B ob, perform a
~ A2 1 B2
controlled N OT (CN OT) gate between the tw o particles A5 , (23)
N
they are holding. The particle of the ® rst pair serves as the
control bit, while the particle of the second pair serves as ~ 2CD
the target [21]. The eŒect of a CN OT gate is that the B 5 , (24)
N
second bit gets inverted (N O T) w hen the ® rst bit is in the
state 1 while it remains unaŒected when the ® rst bit is in ~ C2 1 D2
C5 , (25)
the state 0, i.e. N

|0 ñ | 0 ñ ® | 0 ñ |0 ñ , (15) ~ 2AB
D 5 . (26)
N
|0 ñ | 1 ñ ® | 0 ñ |1 ñ , (16)
2 2
H ere N = (A+ B) + (C + D) is the probability that Alice
|1 ñ | 0 ñ ® | 1 ñ |1 ñ , (17) and Bob obtain the same results in their respective
m easurements of the second pair, i.e. the probability that
|1 ñ | 1 ñ ® | 1 ñ |0 ñ . (18) they keep the ® rst pair of particles. One can quite easily
check that {A , B, C, D} = {1, 0, 0, 0} is a ® xed point of
the m apping given in equations (23) ± (26) and that for
A > 0.5 one also has AÄ > 0.5. The am bitious reader might
w ant to convince him self num erically that indeed the ® xed
point {A , B, C, D} = {1, 0, 0, 0} is an attractor for all
A > 0.5, because the analytical proof of this is quite tricky
and not of much interest here. T he reader should also
note that the m ap equations (23) ± (26) actually has two
® xed points, nam ely {A, B, C , D } = {1, 0, 0, 0} and {A, B,
C , D} = {0, 0, 1, 0}. This means that if we want to know
Figure 3. The quantum network that implements quantum
tow ards which m axim ally entan gled state the procedure
privacy ampli® cation. Alice and Bob share two pairs of
entangled particles. First Alice performs a one bit rotation R w ill converge, we need to have some more inform ation
(given by the R in a circle) which takes | 0 ñ ® ( | 0 ñ 2 i| 1 ñ ) /2 /
1 2 about the initial state than just the ® delity according to
and | 1 ñ ® ( | 1 ñ 2 i| 0 ñ ) /2
1/ 2
on her particles, while Bob performs equation (12). W e will not go into further technical details
the inverse rotation on his side. Then both parties perform a of this puri® cation procedure and instead we refer the
CNOT gate on their particles where the ® rst pair provides the reader to the literature [8,9,12]
control bits (signi® ed by the full circle) while the second pair
Now let us return to the problem that A lice and Bob
provides the target bits (signi® ed by the encircled cross). Finally
w anted to solve, i.e. to achieve teleportation over a noisy
Alice and Bob measure the second pair in the {0,1} basis. They
communicate their results to each other by classical commu- quantum channel. W e sum marize in ® gure 4 w hat Alice and
nication (telephones). If their results coincide they keep the ® rst B ob have to do to achieve their goal. Initially they are given
pair, otherwise they discard it. a quantum channel (for exam ple an optical ® bre) over
438 M. B. Plenio and V. Vedral

which they can transmit quantum states. As this quantum tion procedures, for which the one outlined here is a
channel is not perfect, A lice and Bob will end up with a particular exam ple. N ow we review another important
partially entangled state afte r a single use of the ® bre. puri® cation protocol.
Therefore they repeat the transmission many times which
gives them many partially entangled pairs of particles. N ow
they apply a puri® cation procedure such as the one 3.2. P uri® cation of pure states
described in this section which will give them a smaller The above title is not the m ost fortunate choice of wording,
number of now m axim ally entan gled pairs of particles. because it might wrongly im ply purifying something that is
W ith these m axim ally entangled particles Alice and Bob already pure. T he reader should remember, however, that
can now teleport an unknown quantum state, e.g. |w ñ from the puri® cation m eans entan glem ent concentration and
Alice to Bob. Therefore A lice and Bob can achieve perfect pure states need not be m axim ally entan gled. For exam ple
transm ission of an unknown quantum state over a noisy a state of the form a | 00ñ 1 b | 11ñ is not maxim ally entangled
quantum channel. unless | a | 5 | b | 5 22 1 /2. In this subsection we consider the
The main idea of the ® rst two sections of this article are following problem ® rst analysed by Bennett and co-
the following. Entan glem ent cannot be increased if we are workers in [7]: Alice and B ob share n entangled qubit
allowed to performed only local operations, classical pairs, where each pair is prepared in the state
communication and subselection as shown in ® gure 5.
| AB ñ 5 a | 00ñ 1 b | 11ñ , ( 27)
Under all these operations the expected entan glem ent is
non-increasing. T his im plies in particular that, starting
from an ensemble in a disentangled state, it is impossible to
obtain entangled states by local operations and classical
communication. However, it does not rule out the
possibility that using only local operations w e are able to
select from an ensemble described by a partially entan gled
state a subensemble of system s that have higher average
entan glement. T his is the essence of entanglement puri® ca-

Figure 5. In quantum state puri® cation procedures three

diŒerent kinds of operations are allowed. In part (a) of this
® gure the ® rst two are depicted. Alice and Bob are allowed to
perform any local operation they like. The most general form is
one where Alice adds additional multi-level systems to her
Figure 4. Summary of the teleportation protocol between Alice particle and then performs a unitary transformation on the joint
and Bob in the presence of decoherence. (a) Alice (on the left system followed by a measurement of the additional multi-level
side) holds an unknown quantum state |w ñ which she wants to system. She can communicate classically with Bob about the
transmit to Bob. Alice creates singlet states and sends one half outcome of her measurement (indicated by the telephones). The
down a noisy channel. (b) She repeats this procedure until Alice third allowed operation is given in part (b) of the ® gure. Using
and Bob share many partially entangled states. (c) Then Alice classical communication Alice and Bob can select, based on their
and Bob apply a local entanglement puri® cation procedure to measurement outcomes, subsensembles e 1 , ..., e n from the
distil a subensemble of pure singlet states. (d) This maximally original ensemble e . The aim is to obtain at least one
entangled state can then be used to teleport the unknown state subensemble that is in a state having more entanglement than
|w ñ to Bob. the original ensemble.
 Teleportation, entanglement and thermodynamics in the quantum world 439

where we tak e a, b Î R, and a + b = 1. How m any

2 2
of them see the initial string of qubits as a classical 0, 1
2 2
maxim ally entangled states can they purify? It turns out string with the corresponding probabilities a and b . This
that the answer is governed by the von Neum ann reduced cannot be com pressed to m ore than its Shannon entropy
entropy S v N ( q A) º tr q A ln q A and is asym ptotically given
2 2 2 2
S S h = Ð a ln a Ð b ln b w hich in this case coincides with
by n ´ S vN ( q A) = n ´ ( Ð a ln a Ð b ln b ). To see why this
2 2 2 2
the von Neum ann entropy) [39]. However, another, less
is so, consider the total state of n pairs given by technical reason, and more in the spirit of this article, w ill
be given in section 5.
Ä n
| AB ñ 5
(a | 00ñ 1 b | 11ñ ) Ä (a | 00ñ 1 b | 11ñ ) Ä .. . Ä (a | 00ñ 1 b | 11ñ )
5 a n | 0000 ... 00ñ 1 a (n 2 1) b( | 0000 ... 11ñ 4. E ntangle m ent measures
1 . .. | 1100 .. . 00ñ ) 1 ... b | 1111 ... 11ñ . n
( 28) In the ® rst two sections we have seen that it is possible to
concentrate entanglement using local operations and
(The convention in the second and the third line is that the classical comm unication. A natu ral question that arises in
states at odd positions in the large joint ket states belong to this context is that of the e ciency w ith which one can
Alice and the even states belong to Bob.) Alice can now perform this concentration. Given N partially entangled
perform projections (locally, of course) onto the subspaces pairs of particles each in the state r , how many maxim ally
which have no states | 1ñ , 2 states | 1ñ , 4 state s | 1ñ , and so on, entangled pairs can one obtain? This question is basically
and comm unicates her results to Bob. The probability of one about the am ount of entanglement in a given quantum
having a successful projection onto a particular subspace state. The more entan glement w e have initially, the m ore
with 2k states | 1ñ can easily be seen for the above equation singlet states we will be able to obtain from our supply of
to be non-m axim ally entan gled states. O f course one could also
n ask a diŒerent question, such as for exam ple: how m uch
p 2k 5 a 2( n2 k) b 2k , ( 29)
k entanglement do we need to create a given quantum state
which follows directly from equation (28). It can be shown by local operations and classical comm unication alone?
that this state can be converted into approxim ate ly 1n ( ( nk )) T his question is somehow the inverse of the question of
singlets [7]. If we assum e that the unit of entan glem ent is how m any singlets we can obtain from a supply of non-
given by the entanglem ent of the singlet state then the total m axim ally entangled states.
expected entan glem ent is seen to be All these questions have been worrying physicists in the
n
n n last tw o to three years and a complete answer is still
E5 a 2( n 2 k) 2k
b ln . ( 30) unknow n. The answer to these questions lies in entangle-
k5 0
k k
m ent measures and in this section w e will discuss these
W e w ish to see how this sum behave s asym ptotically as entanglement m easures a little bit more. F irst we w ill
n ® ` . It can be seen easily that the term w ith the highest explain conditions every `decent’ m easure of entan glem ent
weight is should satisfy. After that we w ill then present some
n n
E ~ (a 2 ) (b 2)
na 2 nb 2 entanglement measures that are known today. Finally w e
ln , ( 31)
b 2n b 2n w ill compare these diŒerent entanglement m easures. This
which can, in turn, be simpli® ed using Stirling’ s approx- com parison will tell us something about the way in which
im ation to obtain the am ount of entanglement changes under local quantum
operations.
E ~ exp ( 2 nS vN ( q A)) exp n ln n 2 a 2 n ln a 2n 2 b 2n ln b 2n
(n ln n 2 a 2n ln a 2n 2 b2 4.1. B asic properties of entangl ement m easures
5 exp ( 2 nS vN ( q A)) exp ( nS vN ( q A) ) 3 nS vN ( q A)
T o determine the basic properties every `decent’ entangle-
5 nS vN ( q A ). ( 32) m ent measure should satisfy w e have to recall what we have
learnt in the ® rst two sections of this article. The ® rst
This now show s that for pure states the singlet yield of a property we realized is that any state of the form equation
puri® cation procedure is determined by the von Neum ann (11), w hich we call separable, does not have any quantum
reduced entropy. It is also im portan t to stress that the correlations and should therefore be called disentan gled.
above procedure is reversible, i.e. starting from m singlets T his gives rise to our ® rst condition:
Alice and Bob can locally produce a given state
a | 0 0ñ 1 b | 1 1ñ with an asym ptotic e ciency of m ln (1) For any separable state r the m easure of entangle-
2 = nS vN ( q A). T his w ill be the basis of one of the measures m ent should be zero, i.e.
of entan glem ent introduced by Bennett et al. [7]. Of course,
E( r ) 5 0. ( 33)
Alice and Bob cannot do better than this limit, since both
440
T he next condition concerns the behaviour of the
entanglement under simple local transformations, i.e.
local unitary transformations. A local unitary
transformation simply represents a change of the
basis in which w e consider the given entan gled state.
B ut a change of basis should not change the am ount
of entan glem ent that is accessible to us, because at
any tim e we could just reverse the basis change.
T herefore in both bases the entanglement should be
the same.
(2) For any state r and any local unitary transforma-
tion, i.e. a unitary transformation of the form
U A R U B , the entan glem ent remains unchanged.
T herefore
² ²
E( r ) 5 E(U A Ä UBr UA Ä U B ).
T he third condition is the one that really restricts the
class of possible entan glement m easures. Unfortu-
nately it is usually also the property that is the m ost
di cult to prove for potential measures of entangle-
m ent. W e have seen in section 1 that A lice and Bob
cannot create entan glem ent from nothing, i.e. using
only local operations and classical comm unication.
In section 2 we have seen that given some initial
entanglement w e are able to select a subensemble of
states that have higher entanglement. This can be
done using only local operations and classical
com munication. H owever, what we cannot do is to
increase the total am ount of entan glem ent. W e can
calculate the total am ount of entan glem ent by
sum ming up the entan glem ent of all system s afte r
w e have applied our local operations, classical
com munications and subselection. That m eans that
in ® gure 5 we tak e the probability p i that a system
w ill be in particular subensemble e i and multiply it by
the average entan glement of that subensemble. This
result w e then sum up over all possible subensembles.
T he num ber w e obtain should be sm aller than the
entanglement of the original ensemble.
(3) Local operations, classical comm unication and sub-
selection cannot increase the expected entanglement,
i.e. if w e start with an ensemble in state r and end up
w ith probability p i in subensembles in state r i then
w e w ill have
E(r ) ³ p i E( r i ).
i
T his last condition has an im portan t implication as it
tells us something about the e ciency of the m ost
general entanglement puri® cation method. T o see
this w e need to ® nd out what the m ost e cient
puri® cation procedure will look like. Certainly it w ill
select one subensemble, w hich is described by a
m axim ally entan gled state. As w e w ant to make sure
M. B. Plenio and V. Vedral
that we have as m any pairs as possible in this
subensemble, we assum e that the entanglement in all
the other subensembles van ishes. Then the prob-
ability that we obtain a m axim ally entangled state
from our optim al quantum state puri® cation proce-
dure is bounded by
E( r )
p singlet £ . ( 36)
E singletstate
T he considerations leading to equation (36) show
that every entanglement m easure that satis® es the
three conditions presented in this section can be used
to bound the e ciency of entan glement puri® cation
procedures from above. Before the reader accepts
this statem ent (s)he should, how ever, carefully
( 34)
reconsider the above argum ent. In fact, w e have
m ade a hidden assumption in this argum ent which is
not quite trivial. W e have assum ed that the
entanglement measures have the property that the
entanglement of two pairs of particles is just the sum
of the entanglements of the individual pairs. This
sounds like a reasonable assum ption but we should
note that the entanglement m easures that we
construct are initially purely math ematical objects
and that we need to prove that they behave reason-
ably. Therefore we demand this additivity property
as a fourth condition
(4) Given two pairs of entan gled particles in the total
state r = r 1 R r 2 then we have
E( r ) 5 E( r 1) 1 E( r 2). ( 37)
N ow we have speci® ed reasonable conditions that
any `decent’ m easure of entan glem ent should satisfy
and in the next section we will brie¯ y explain some
possible measures of entan glem ent.
4.2. T hree m easures of entangl em ent
In this subsection we w ill present three m easures of
entan glement. O ne of them, the entropy of entan glem ent,
will be de® ned only for pure states. Nevertheless it is of
great importance because there are good reasons to accept
it as the unique measure of entanglement for pure states.
Then we will present the entanglement of formation which
was the ® rst measure of entanglement for m ixed states and
( 35)
whose de® nition is based on the entropy of entan glem ent.
Finally we introduce the relative entropy of entan glem ent
which was developed from a completely diŒerent view -
point. F inally we will compare the relative entropy of
entan glement w ith the entanglement of formation.
The ® rst measure we are going to discuss here is the
entropy of entanglement. It is de® ned in the following w ay.
Assume that Alice and B ob share an entangled pair of
particles in a state r . Then if Bob considers his particle
 Teleportation, entanglement and thermodynamics in the quantum world
alone he holds a particle whose state is described by the
reduced density operator r B = tr A {r }. The entropy of
entan glement is then de® ned as the von Neumann entropy
of the reduced density operator r B , i.e.
2 tr {r
E vN 5 S vN ( r B) 5 B ln r
One could think that the de® nition of the entropy of
entan glement depends on whether A lice or Bob calculate
the entropy of their reduced density operator. H ow ever, it
can be shown that for a pure state r this is not the case, i.e.
both w ill ® nd the same result. It can be shown that this
measure of entan glement, when applied to pure states,
satis® es all the conditions that we have form ulated in the
previous section. T his certainly m akes it a good m easure of
entan glement. In fac t many people believe that it is the only
measure of entan glem ent for pure states. W hy is that so? In
the previous section we have learnt that an entan glem ent
measure provides an upper bound to the e ciency of any
puri® cation procedure. For pure states it has been shown
that there is a puri® cation procedure that achieves the limit
given by the entropy of entanglement [7]. W e reviewed this
procedure in the previous section. In addition the inverse
property has also been shown. Assume that we want to
create N copies of a quantum state r of two particles purely
by local operations and classical com munication. As local
operations cannot create entan glem ent, it will usually be
necessary for Alice and B ob to share some singlets before
they can create the state r . How many singlet states do they
have to share beforehand? The answer, again , is given by
the entropy of entan glem ent, i.e. to create N copies of a
state r of two particles one needs to share N E( r ) singlet
states beforehand. T herefore we have a very interesting
result. The entanglement of pure states can be concentrated
and subsequently be diluted again in a reversible fashion.
One should note, however, that this result holds only when
we have many (actu ally in® nitely m any) copies of entan gled
pairs at once at our disposal. For ® nite N it is not possible
to achieve the theoretical limit exactly [40]. This observa-
tion suggests a close relationship betw een entan glem ent
transformations of pure states and thermodynam ics. W e
will see in the following to what extent this relationship
extends to mixed entangled states.
W e will now generalize the entropy of entan glem ent to
mixed states. It will turn out that for m ixed states there is
not one unique measure of entan glem ent but that there are
several diŒerent measures of entanglement.
H ow can we de® ne a m easure of entan glement for m ixed
states? As we now have agreed that the entropy of
entan glement is a good m easure of entanglement for pure
states, it is natu ral to reduce the de® nition of mixed state
entan glement to that of pure state entan glem ent. One way
of doing that is to consider the am ount of entan glem ent
that we have to invest to create a given quantum state r of a
pair of particles. By creating the state w e m ean that w e
441
represent the state r by a statistical mixture of pure states.
It is important in this representation that we do not restrict
ourselves to pure states that are orthonorm al. If w e want to
attr ibute an am ount of entan glem ent to the state r in this
}. w ay then this should be the smallest am ount of entangle-
B ( 38)
m ent that is required to produce the state r by mixing pure
states together. If we m easure the entan glem ent of pure
states by the entropy of entan glem ent, then we can de® ne
the entanglement of formation by
EF(r ) 5 min p i E vN ( | w i ñ á w i | ). ( 39)
r 5 p i |w i ñ á w i|
i i
T he m inimization in equation (39) is taken over all possible
decom positions of the density operator r into pure states
| w ñ . In general, this minimization is extrem ely di cult to
perform. Luckily for pairs of two-level system s one can
solve the minim ization analytically and write down a closed
expression for the entanglement of formation which can be
w ritten entirely in term s of the density operator r and does
not need any reference to the states of the optimal
decom position. In addition the optimal decomposition of
r can be constructed for pairs of tw o-level systems. To
ensure that equation (39) really de® nes a m easure of
entanglement, one has to show that it satis® es the four
conditions w e have stated in the previous section. The ® rst
three conditions can actually be proven analytically (we do
not present the proof here) while the fourth condition (the
additivity of the entanglement) has so far only been
con® rm ed numerically. Nevertheless the entan glem ent of
formation is a very im portant m easure of entan glem ent
especially because there exists a closed analytical form for it
[41].
As the entan glement of form ation is a m easure of
entanglement it represents an upper bound on the e ciency
of puri® cation procedures. H ow ever, in addition it also
gives the am ount of entan glem ent that has to be used to
create a given quantum state. This de® nition of the
entanglement of formation alone guarantees already that
it will be an upper bound on the e ciency of entan glem ent
puri® cation. This can be seen easily, because if there w ould
be a puri® cation procedure that produces, from N pairs in
state r , m ore entanglement than N E F ( r ) then we would be
able to use this entan glem ent to create m ore than N pairs in
the state r . Then we could repeat the puri® cation procedure
and we w ould get even more entanglement out. This w ould
imply that we w ould be able to generate arbitrarily large
am ounts of entan glement by purely local operations and
classical communication. T his is im possible and therefore
the entanglement of form ation is an upper bound on the
e ciency of entan glem ent puri® cation. W hat is much m ore
di cult to see is whether this upper bound can actually be
achieved by any entan glem ent puri® cation procedure. On
the one hand w e have seen that for pure states it is possible
to achieve the e ciency bound given by the entropy of
442 M. B. Plenio and V. Vedral
entan glement. On the other hand for m ixed states the
situation is m uch m ore complicated because we have the
additional statistical uncertainty in the mixed state. W e
would expect that we have to make local measurements in
order to remove this statistical uncertainty and these
measurements w ould then destroy som e of the entangle-
ment. On the other hand we have seen that in the pure state
case we could recover all the entanglement despite the
application of measurements. This question w as unresolved
for some time and it w as possible to solve it when yet
another m easure of entan glem ent, the relative entropy of
entan glement, w as discovered.
The relative entropy of entanglement has been intro-
duced in a diŒerent w ay than the two entan glem ent
measures presented above [13,15]. T he basic ideas in the
relative entropy of entanglem ent are based on distinguish-
ability and geom etrical distance. The idea is to compare a
given quantum state r of a pair of particles with
disentan gled states. A canonical disentan gled state that
one can form from r is the state r A R r B where r A ( r B ) is
the reduced density operator that A lice (Bob) are obser-
ving. Now one could try to de® ne the entanglement of r by
any distance between r and r A R r B . T he larger the
distance the larger is the entan glem ent of r . U nfortunately
it is not quite so easy to make an entan glem ent m easure.
The problem is that we have picked a particular (although
Figure 6. A geometric way to quantify entanglement. The set
of all density matrices T is represented by the outer circle. Its
subset of disentangled (separable) states D, is represented by the
* calculate both the entan glement of formation and the
inner circle. A state r belongs to the entangled states, and q is
the disentangled state that minimizes the distance D( r | | q ). This
minimal distance can be de® ned as the amount of entanglement
in r .
natural) disentan gled state. Under a puri® cation procedure
this product state r A R r B can be turned into a sum of
product states, i.e. a classically correlated state. But what
we know for sure is that under any puri® cation procedure a
separable state of the form equation (11) will be turned into
a separable state. T herefore it w ould be much more natu ral
to com pare a given state r to all separable states and then
® nd that separable state that is closest to r . This idea is
presented in ® gure 6 and can be w ritten in a formal w ay as
E RE ( r ) 5 min D( r || q ). ( 40)
q [ D
Here the D denotes the set of all separable states and D can
be any function that describes a m easure of separation
between two density operato rs. Of course, not all distan ce
measures will generate a `decent’ measure of entan glem ent
that satis® es all the conditions that w e demand from an
entan glement m easure. Fortunately, it is possible to ® nd
some distances D that generate `decent’ m easures of
entan glement and a particularly nice one is the relative
entropy w hich is de® ned as
S( r || q ) 5 {
tr r ln r 2 r ln q }. ( 41)
The relative entropy is a slightly peculiar function and is in
fac t not really a distance in the m athematical sense because
it is not even symm etric. N evertheless it can be proven that
equation (40) together w ith the relative entropy of equation
(41) generates a measure of entanglement that satis® es all
the conditions we were asking for in the previous section. It
should be said here that the additivity of the relative
entropy of entan glem ent has only been con® rmed numeri-
cally as for the entan glem ent of formation. All other
properties can be proven analytically and it should also be
noted that for pure states the relative entropy of entangle-
ment reduces to the entropy of entanglement w hich is of
course a very satisfying property.
But why does the relative entropy of entan glem ent
answer the question w hether the upper bound on the
e ciency of entan glement puri® cation procedures that we
found from the entan glem ent of formation can actually be
achieved or not? T he answer comes from a direct
comparison of the two m easures of entan glem ent for a
particular kind of state. These, called W erner states, are
de® ned as
q F 5 F | w 2 ñ á w 2 |1
12 F 1 1 1 1
( 42)
2 2
( |w ñáw | 1 |u ñáu | 1 |u ñáu | ),
3
where w e have used the Bell basis de® ned in equations (4) ±
(7). The param eter F is the ® delity of the W erner state and
lies in the interval [ 14,1]. F or W erner states it is possible to
relative entropy of entan glem ent analytically. In ® gure 7
the entan glem ent of the W erner states w ith ® delity F is
plotted for both entan glement m easures. One can clearly
 Teleportation, entanglement and thermodynamics in the quantum world
see that the relative entropy of entan glem ent is smaller than
the entanglement of form ation. B ut we know that the
relative entropy of entanglement, because it is an entangle-
ment measure, is an upper bound on the e ciency of any
entan glement puri® cation procedure too. Therefore w e
reach the following very interesting conclusion. Assum e w e
are given a certain am ount of entanglement that we invest
in the most optimal way to create by local means som e
mixed quantum states r of pairs of tw o-level systems. H ow
many pairs in the state r we can produce is determined by
the entanglement of formation. Now we try to recover this
entan glement by an entan glem ent puri® cation method
whose e ciency is certainly bounded from above by the
relative entropy of entanglement. The conclusion is that the
am ount of entanglement that we can recover is always
smaller than the am ount of entanglement that we originally
invested. Therefore w e arrive at an irreversible process, in
stark contrast to the pure state case where we were able to
recover all the invested entan glement by a puri® cation
procedure. This result again sheds some light on the
connection between entan glement m anipulations and ther-
modynam ics and in the next section we w ill elaborate on
this connection further.
5. Thermodynam ics of entangle ment
Here we would like to elucidate further the fundam ental
law of quantum information processing by comparing it to
the Second Law of Thermodynam ics. The reader should
not be surprised that there are connections between the
two. First of all, both laws can be expressed math ematically
Figure 7. Comparison of the entanglement of formation with
the relative entropy of entanglement for Werner states with
® delity F . The relative entropy of entanglement is always smaller
than the entanglement of formation. This proves that in general
entanglement is destroyed by local operations.
443
by using an entropic quantity. The second law says that
thermodynam ical entropy cannot decrease in an isolated
system. The fundam ental law of quantum inform ation
processing, on the other hand, states that entan glem ent
cannot be increased by local operations. Thus both of the
laws serve to prohibit certain typ es of processes w hich are
impossible in nature (this analogy was ® rst em phasized by
Popescu and Rohrlich in [42], but also see [15,43]). The rest
of the section shows the two principles in action by solving
two simple, but im portant problems.
5.1. R eversible and irreversible processes
W e begin by stating more form ally a form of the Second
L aw of thermodynam ics. T his form is due to C lausius, but
it is com pletely analogous to the no increase of entropy
statem ent w e gave above. In particular it will be m ore
useful for what we are about to investigate.
The Second Law of Therm odynam ics (C lausius).
There exists no thermodynam ic process the sole eŒect
of which is to extract a quantity of heat from the colder
of two reservoirs and deliver it to the hotter of the two
reservoirs.
Suppose now that we have a thermodynam ical system.
W e w ant to invest som e heat into it so that at the end our
system does as m uch w ork as possible w ith this heat input.
T he e ciency is therefore de® ned as
W out
g 5 . ( 43)
Q in
N ow it is a well known fact that the above e ciency is
m axim ized if we have a reversible process (sim ply because
an irreversible process wastes useful work on friction or
som e other lossy mechanism). In fact, w e know the
e ciency of one such process, called the C arnot cycle.
W ith the Second Law on our m ind, we can now prove that
no other process can perform better than the C arnot cycle.
T his boils dow n to the fact that we only need to prove that
no other reversible process performs better than the Carnot
cycle. The argument for this can be found in any under-
graduate book on Thermodynam ics and brie¯ y runs as
follows (again reductio ad absur dum ). T he Carnot engine
tak es some heat input from a hotter reservoir, does some
w ork and delivers an am ount of heat to the colder
reservoir. Suppose that there is a better engine, E, that is
operating betw een the same tw o reservoirs (we have to be
fair when comparing the e ciency). Suppose also that w e
run this better m achine backwards (as a refrigerato r): w e
w ould do som e w ork on it, and it would tak e a quantity of
heat from the cold reservoir and bring som e heat to the hot
444 M. B. Plenio and V. Vedral
reservoir. For simplicity w e assume that the work done by a
Carnot engine is the same as the w ork that E needs to run
in reverse (this can always be arranged and w e lose nothing
in generality). Then w e look at the two m achines together,
which is just another thermodynam ical process: they
extract a quantity of heat from the colder reservoir and
deliver it to the hot reservoir with all other things being
equal. B ut this contradicts the Second Law, and therefore
no m achine is m ore e cient than the Carnot engine.
In the previous section we have learnt about the
puri® cation schem e of Bennett et al. [7] for pure states.
E ciency of any scheme w as de® ned as the num ber of
maxim ally entangled states we can obtain from a given N
pairs in some initial state, divided by N . This scheme is in
addition reversible, and we would suppose, guided by the
above thermodynam ic argument, that no other reversible
puri® cation schem e could do better than that of Bennett et
al. Suppose that there is a more e cient (reversible)
process. Now Alice and Bob start from a certain num ber
N of m axim ally entan gled pairs. They apply a reverse of the
scheme of B ennett et al. [7] to get a certain num ber of less
entan gled states. But then they can run the m ore e cient
puri® cation to get M m axim ally entan gled states out.
However, since the second puri® cation is m ore e cient
than the ® rst one, then w e have that M > N. So, locally
Alice and Bob can increase entanglement, which contra-
dicts the fundam ental law of quantum information proces-
sing. W e have to stress that as far as the mixed states are
concerned there are no results regarding the best puri® ca-
tion schem e, and it is not completely understood w hether
the sam e strategy as above could be applied (for m ore
discussion see [15]).
In any case, the above reasoning shows that the
conceptual ideas behind the Second Law and the funda-
mental law are similar in natu re. N ext we show another
attr active application of the fundam ental law. W e return to
the question at the beginning of the article that started the
whole discussion: can Alice teleport to B ob as many qubits
as she likes using only one entan gled pair shared betw een
them?
5.2. W hat can we learn from the non-inc rease of
entanglement unde r local operations?
If the scheme that we are proposing could be utilized then it
would be of great technological advantag e, because to
create and maintain entan gled qubits is at present very
hard. If a single maxim ally entan gled pair could transfer a
large am ount of inform ation (i.e. teleport a num ber of
qubits), then this w ould be very useful. However, there is
no free lunch. In the same w ay that we cannot have an
unlimited am ount of useful work and no heat dissipation,
we cannot have arbitrarily m any teleportations w ith a
single maxim ally entangled pair. In fact, we can prove a
much stronger statem ent: in order to teleport N qubits
Alice and Bob need to share N maxim ally entan gled pairs!
In order to prove this we need to understand another
simple concept from quantum mechanics. N am ely, if we
can teleport a pure unknown quantum state then w e can
teleport an unknow n mixed quantum state (this is obvious
since a mixed state is just a com bination of pure states). But
now comes a crucial result: every mixed state of a single
qubit can be thought of as a part of a pure state of two
entangled qubits (this result is more general, and applies to
any quantum state of any quantum system , but we do not
need the generalization here). Suppose that we have a single
qubit in a state
q 5 a 2| 0ñ á 0| 1 b 2| 1ñ á 1| . ( 44)
This single qubit can then be viewed as a part of a pair of
qubits in state
| w ñ 5 a | 00ñ 1 b | 11ñ . ( 45)
One obtains equation (44) from equation (45) simply by
tak ing the partial trace over the second particle. Bearing
this in m ind we now envisage the following teleportation
protocol. Alice and B ob share a maxim ally entangled pair,
and in addition Bob has a qubit prepared in some state, say
| 0ñ . A lice than receives a qubit to teleport in a general (to
her unknown) state q . After the teleportation we want
Bob’ s extra qubit to be in the state q and the maxim ally
entan gled pair to stay intact (or at least not to be
completely destroyed). This is shown in ® gure 2.
Now we wish to prove this protocol impossibleÐ entan-
glem ent sim ply has to be com pletely destroyed at the end.
Suppose it is not, i.e. suppose that the above teleportation
is possible. Then Alice can teleport any unknown (m ixed)
state to B ob using this protocol. B ut this m ixed state can
arise from an entan gled state where the second qubit (the
one to be traced out) is on Alice’s side. So initially A lice
and B ob share one entangled pair, but after the teleporta-
tion they have increased their entanglement as in ® gure 8.
Since the initial state can be a maxim ally mixed state
(a = b = 2 / ) the ® nal entanglement can grow to be twice
Ð 1 2
the maxim ally entangled state. B ut, as this would violate
the fundam ental law of quantum information processing it
is impossible and the initial maxim ally entangled pair has
to be destroyed. In fac t, this argum ent shows that it has to
be destroyed completely. Thus we see that a simple
application of the fundam ental law can be used to rule
out a whole class of impossible teleportation protocols.
Otherwise every teleportation protocol w ould have to be
checked separately and this w ould be a very hard problem.
6. C onclusions
Let us brie¯ y recapitulate what w e have learnt. Quantum
teleportation is a procedure w hereby an unknown state of a
 Teleportation, entanglement and thermodynamics in the quantum world
Figure 8. A diagramatical proof that the teleportation protocol
in ® gure 2 is impossible. Alice is on the left of the dashed line,
Bob on the right. Initially Alice is holding a mixed state q and
Bob a particle in state | 0ñ . In addition Alice and Bob share a pair
of maximally entangled particles in state |w 1 ñ . The particle in
the mixed state q that Alice is holding can be part of a pair of
entangled particles. The aim is that ® nally, after the teleporta-
tion Bob holds the state q and Alice and Bob still have their two
particles in a maximally entangled state |w 1 ñ . However, not only
the state q will be transferred to Bob but also its entanglement
with other particles. Therefore after the envisaged teleportation
Alice and Bob would be sharing more entanglement than
initially. This contradicts the fundamental law of quantum
information processing that entanglement cannot be increased.
quantum system is transferred from a particle at a place A
to a particle at a place B. The w hole protocol uses only
local operations and classical com munication between A
and B. In addition, A and B have to share a m axim ally
entan gled state. E ntanglement is central for the whole
teleportation: if that state is not maxim ally entan gled then
teleportation is less e cient and if the state is disentangled
(an d only classically correlated) then teleportation is
im possible. W e have then derived a fundam ental law of
quantum inform ation processing which stipulated that
entan glement cannot be increased by local operations and
classical com munication only. This law w as then investi-
gated in the light of puri® cation procedures: local protocols
for increasing entan glem ent of a subensemble of particles.
W e discussed bounds on the e ciency of such protocols
and emphasized the links betw een this kind of physics and
the theory of thermodynam ics. This led us to formulate
various measures of entanglement for general mixed states
of two quantum bits. A t the end w e returned to the
problem of teleportation, asking how many entan gled pairs
we need in order to teleport N qubits. Using the
fundam ental law of quantum inform ation processing w e
oŒered an elegant argum ent for needing N m axim ally
entan gled pairs for teleporting N qubits, a pair per qubit.
The analogy betw een thermodynam ics and quantum
inform ation theory m ight be deeper, but this at present
remains unknown. Q uantum inform ation theory is still at a
very early stage of developm ent and, although there are
445
already some extraordinary results, a number of areas is
still untouched. In particular the status of w hat we called
the fundam ental law is unclear. F irst and foremost, it is not
known how it relates to other results in the ® eld, such as,
for exam ple, the no-cloning theorem [36] which states that
an unknown quantum state cannot be duplicated by a
physical process. W e hope that research in this area w ill
prove fruitful in establishing a deeper sym biotic relation-
ship between information theory, quantum physics and
thermodynam ics. Q uantum theory has had a huge input
into information theory and thermodynam ics over the past
few decades. Perhaps by turning this around w e can learn
m uch more about quantum theory by using inform ation-
theoretic and thermodynam ic concepts. Ultimately, this
approach m ight solve som e long standing and di cult
problems in modern physics, such as the m easurement
problem and the arrow of tim e problem. T his is exactly
w hat w as envisaged more that 60 years ago in a statem ent
attr ibuted to Einstein: `T he solution of the problems of
quantum mechanics will be thermodynam ical in natu re’
[44].
A cknow ledgements
T he authors would like to thank Susana F. Huelga and
Peter L . Knight for critical reading of the manuscript. This
w ork was supported in part by Elsag-Bailey, the UK
E ngineering and Physical Sciences Research Council
(EPSR C) and the European TM R R esearch Network
E RBF M RXC T960066 and the European T M R Research
N etw ork E RBF M RXC T960087.
R eferences
[1] Bell, J. S., 1965, Physics, 1 195.
[2] Bell, J. S., 1966, R ev. Mod. Phys., 38, 447.
[3] Bell, J. S., 1987, Speakable and Unspeakable in Quantum M echanics
(Cam bridge: Cambridge University Press).
[4] Clau ser, J. F., and Shimony, A., 1978, Rep. Prog. Phys., 41, 1881.
[5] Aspect, A., D alib ard , J., Grangier, P., and Ro ger, G ., 1982, Phys. Rev.
Lett ., 49, 1804.
[6] Santos, E., 1991, Phys. R ev. Lett., 66, 1388.
[7] Bennett, C. H ., Bernstein, H . J., Popescu, S., and Schumacher, B.,
1996, Phys. Rev. A, 53, 2046.
[8] Bennett, C . H ., Brassard , G ., Popescu, S., Schumacher, B., Smolin, J.
A., and W ootters, W . K., 1996, Phys. Rev. Lett., A 76, 722.
[9] D eutsch, D., Ekert, A., Jozsa, R., M acchiavello, C., Popescu, S., and
Sanpera, A., 1996, Phys. R ev. Lett., 77, 2818.
[10] G isin, N., 1996, Phys. Lett. A, 210, 151.
[11] H orodecki, M ., Horodecki, P., and H orodecki, R., 1997, Phys. Rev.
Lett., 78 574.
[12] Bennett, C . H ., D iV incenzo, D. P., Smolin, J. A., and W ootters, W.
K ., 1996, Phys. Rev. A, 54, 3824.
[13] V edral, V., Plenio, M . B., and Rippin, M . A., and P. L. K night, 1997,
Phys. Rev. Lett., 78, 2275.
[14] V edral, V ., Plenio, M . B., Jacobs, K., and K night, P. L., 1997, Phys.
Rev. A, 56, 4452.
[15] V edral, V ., and Plenio, M . B., 1998, Phys. R ev. A, 57, 1619.
446 M. B. Plenio and V. Vedral
[16] Bennett, C. H ., Brassard , G ., Crepeau, C., Jozsa, R., Peres, A., and
W ootters, W . K ., 1993, Phys. Rev. Lett., 70, 1895.
[17] Bouwmeester, D., Pan, J. W ., M attle, K ., Eibl, M ., Wein furter, H .,
and Zeilinger, A., 1997, Nature, 390, 575.
[18] Boschi, D ., Branca, S., DeM artini, F., H ardy, L., and Popescu, S.,
1998, Phys. Rev. Lett., 80, 1121.
[19] V edral, V ., and Plenio, M . B., 1998, Prog. Quant. Electron., to be
publish ed.
[20] Ekert, A., and Jozsa, R., 1996, Rev. mod. Phys., 68, 733.
[21] Barenco, A., 1996, Contemp. Phys., 37, 375.
[22] Jozsa, R., 1997, eprint quant-ph /9707034.
[23] Plenio, M . B., and Knight, P. L., 1996, Phys. Rev. A, 53, 2986.
[24] Plenio, M . B., and Knight, P. L., 1997, Proc. R. Soc. Lond. A, 453,
2017.
[25] C alderbank, A. R., and Shor, P. W ., 1996, Phys. Rev. A, 54, 1098.
[26] Ekert, A., and Macchiavello, C., 1996, Phys. R ev. Lett., 77, 2585.
[27] Shor, P. W ., 1995, Phys. R ev. A, 52, 2493.
[28] Steane, A. M., 1996, Proc. R. Soc. Lond. A, 452, 2551.
[29] H uelga, S. F., Macchiavello , C., Pellizzari, T., Ek ert, A. K ., Plenio, M.
B., and C irac, J. I., 1997, Phys. Rev. Lett., 79, 3865.
[30] G rover, L. K., 1997, lanl e-print quant-ph /9704012.
[31] C irac, J. I., Ek ert, A., Huelga, S. F., and M acchiavello, C., Lanl e-
print quant-ph /9803017.
[32] Bose, S., Vedral, V ., and K night, P. L., 1998, Phys. Rev. A, 57, 822.
[33] M urao, M ., Plenio, M . B., Popescu, S., V edral, V., and K night, P. L.,
1998, Phys. Rev. A, 57, R4075.
[34] Schumacher, B., 1995, Phys. Rev. A, 51, 2738.
[35] V edral, V ., Rippin , M. A., and Plenio, M . B., 1997, J. mod. Optics , 44,
2185.
[36] W ootters, W . K ., and Zu rek, W . H ., 1992, Nature, 299, 802.
[37] Popescu, S., 1994, Phys. Rev. Lett., 72, 797.
[38] Lin den, N ., M assar, S., and Popescu, S., Lanl e-print quant-ph/
9805001.
[39] C over, T. M ., and Thomas, J. A., 1991, Elements of Information
Theory (New York: John W iley and Sons Inc.).
[40] Lo, H . W ., and Popescu, S., 1997, lanl e-print quant-ph /9707038.
[41] W ootters, W. K ., 1998, Phys. Rev. Lett., 80, 2245.
[42] Popescu, S., and Ro hrlich, D., 1997, Phys. R ev. A, 56, R3319.
[43] H orodecki, M ., and Horodecki, R., 1997, lan l e-print quant-ph /
9705003.
[44] Einstein, A., this quote is attributed to Einstein, however, w e were
unable to trace the original reference.
M artin Plenio studied in G oÈ ttingen (Germany)
where he obtained both his D iplom a (1992) and
his PhD (1994) in Theoretical Physics. H is m ain
research area at that tim e was Quantum Optics
and in particular the propertie s of single quantum
systems such as single trapped ions irradiated by
laser light. After his PhD he joined the Theore-
tical Quantum Optics group at Imperial College
as a postdoc. It was here that he started to
become interested in quantum com puting, quan-
tum communication and quantum information
theory. Since January 1998 he is now a lecturer in
the Optics Section of Imperial College.
Vlatko Vedral obtained both his ® rst degree
(1995) and PhD (1998) in Theoretical Physics
from Imperial College. He is now an Elsag-Bailey
Postdocto ral Research Fellow at the Center for
Quantum Computin g in Oxford. From October
1998 he will take up a Junior Research Fellow-
ship at M erton College in Oxford. H is m ain
research interests are in connectio ns between
inform ation theory and quantum mechanics,
including quantum com puting, error correction
and quantum theory of communication.