Chapter 4

Quantum entanglement

In this chapter we study the nature of a special type of correlation displayed

by the entangled states. These correlations have no classical counterpart, in
other words, they cannot be described by classical probability distributions.
They are genuine quantum mechanical correlations built up in the states of
composite quantum systems.
We first take a close look at the so-called Bell states which violate the
famous Bell inequalities1 . These states display the essence of entanglement
and the CHSH inequality provides an experimentally testable signature of
it. We then describe three applications: a quantum key distribution protocol
(Ekert 1991), quantum teleportation and dense coding. We stress here that
all three of them have been experimentally realized, and form important
primitive protocols for quantum communication.
In quantum information processing one tries to use entanglement as a
quantifiable resource, much like energy or information, and it would be very
convenient to be able to measure the degree or quantity of entanglement.
Finding such a measure is however non-trivial. We will come back to this
point in later chapters.

4.1 Bell states

Production of Bell states. We have seen in chapter 2 that in order to
produce entangled states the Qbits must “interact”, at some point in time.
The prototypical example of entangled states are the Bell states which form
1
There is a class of such inequalities named after John Bell who derived the first
ones. In this chapter we derive the more transparent Clauser-Horne-Shimony-Holt (CHSH)
inequality.

1
2 CHAPTER 4. QUANTUM ENTANGLEMENT

|x> H
| |B >
xy
|y>

Figure 4.1: Quantum circuit producing Bell states.

a basis of C 2 ⊗ C 2 . These can be produced from the unitary gate

U = (CN OT )(H ⊗ I) (4.1)

This is a 4 × 4 matrix equal to the usual matrix product of the two 4 × 4

matrices CN OT and H ⊗ I. The Control Not gate provides the interaction
between the two bits. It is defined as the NOT gate acting on the second bit
provided the first one2 equals 1

CN OT |x, yi = |x, y ⊕ xi

The matrices H and I are the usual 2 × 2 Hadamard and identity matri-
ces. The circuit representation of the unitary gate U = (CN OT )(H ⊗ I) is
depicted in figure 1.
Let us calculate the action of this circuit on a tensor product state |xi ⊗
|yi = |x, yi.

1
(CN OT )(H ⊗ I)|x, yi = (CN OT ) √ |0i + (−1)x |1i ⊗ |yi


2
1 (−1)x
= √ CN OT |0, yi + √ CN OT |1, yi
2 2
x
1 (−1)
= √ |0, yi + √ |1, y ⊕ 1i)
2 2
= |Bxy i

More explicitly we have

1

|B00 i = √ |00i + |11i = U |00i
2
1

|B01 i = √ |01i + |10i = U |01i
2
2
called the control bit
4.1. BELL STATES 3

1

|B10 i = √ |00i − |11i = U |10i
2
1

|B11 i = √ |01i − |10i = U |11i
2
These four states are a unitary “rotation” of the four canonical basis states
of C 2 ⊗ C 2 and thus also form a basis, called the Bell basis.
Here the interaction is effected by the CN OT gate: building such a gate
in a laboratory requires bringing two particles supporting the Qbits |xi and
|yi close enough in space and time (interactions are local). Photons do not
interact directly with one another (Maxwell equations are linear) but they can
interact indirectly through their direct interaction with matter (one speaks
of non-linear optics). Localized sources producing pairs of entangled photons
are excited atoms or nuclei, emitting photons when they fall in their ground
state. Electron spin can also be entangled because the combination of the
Coulomb interaction with the Pauli principle can produce special magnetic
correlations. In fact this kind of entanglement is very common place: in a
hydrogen molecule the spin part of the chemical valence bond*3 between two
hydrogen atoms is the state

1
|B11 i = √ (| ↑↓i − | ↓↑i)
2

The reader should check that

1
1

|B00 i = √ |00i + |11i = √ |γγi + |γ⊥ γ⊥ i (4.2)
2 2

where |γi is any state of C2 . This has remarkable consequences as the follow-
ing discussion will show. For the sake of the argument we suppose that Alice
has captured one photon in her lab and Bob has captured the other photon
in his lab (figure 2). Irrespective how remote the two labs are, it is always
true that the two photons have come from a common localized source. Now
we look at the outcome of several simple measurements that Alice and Bob
might do each in their own lab. We are assuming that they cannot communi-
cate the outcomes of these measurements. We will consider the three specific
situations where: Alice measures first/Bob measures after; Bob measures
first/Alice measures after; Alice and Bob measure simultaneously4 .
3
the anti-symmetry of the spin part allows the orbital part to be in the symmetric
energetically favorable state (Heitler-London theory)
4
For definiteness we have in mind a Galilean picture of space-time. However the dis-
cussion is essentially the same for a relativistic picture for space-time.
4 CHAPTER 4. QUANTUM ENTANGLEMENT

BOB

ALICE

|B>

Figure 4.2: Alice and Bob share an entangled pair.

• Alice measures first and Bob after. The “measurement apparatus” of

Alice is formed by the projectors {|αihα| ⊗ I, |α⊥ ihα⊥ | ⊗ I} so that,
according to the measurement postulate, the Bell state collapses on one
of the projections (remember we have to normalize after projecting)
1 1
|αihα| ⊗ I|B00 i = √ |ααi → |αi ⊗ |αi, with prob
2 2
1 1
|α⊥ ihα⊥ | ⊗ I|B00 i = √ |α⊥ α⊥ i → |α⊥ i ⊗ α⊥ i, with prob
2 2
Therefore Alice observes her photon in the collapsed state |αi or |α⊥ i.
Bob, on his side, does not know anything, and doesn’t even know that
Alice has performed measurements ! In order for him to learn some-
thing he can try to perform a measurement on his photon. But he has
to choose a basis {|βi, |β⊥ i}. Given that his photon is in the state |αi,
his photon collapses to |βi with prob cos2 (α − β) or to |β⊥ i with prob
sin2 (α − β). Similarly, given that his photon is in the state α⊥ we get
the same result with cos2 and sin2 interchanged. The fact that Bob
does not know the initial state of his photon or that he does not even
known what Alice has done should not bother you: the point is that he
does a specific experiment (measurement in the β, β⊥ basis) and finds
a net outcome. The net outcome in Bob’s lab is that the photon is in
the state |βi with prob 21 or |β⊥ i with prob 12 .

• Bob measures first and Alice after. The same discussion shows that, if
Bob performs measurements first (in the β, β⊥ basis) while Alice sleeps
and Alice measures after (in the (α, α⊥ basis) the net outcome of each
party is the same.

• Bob and Alice measure simultaneously. You might think (?) that if
both parties perform simultaneous local measurements the whole sce-
4.1. BELL STATES 5

nario is different. Let us try. Suppose Alice and Bob perform simulta-
neous measurements in the basis
{|α, βi, |α, β⊥ i, |α⊥ , βi, |α⊥ , β⊥ i}
The Bell state
1 1
|B00 i = √ (|00i + |11i) = √ (|γγi + |γ⊥ γ⊥ i)
2 2
will collapse to one of the four basis states. So Alice will be in possession
of a photon in state |αi or |α⊥ i and Bob in possession of a photon in
the state |βi or |β⊥ i. The situation is exactly the same than in the
previous situations ! It is very instructive to compute the probabilities
of the respective collapsed states (which are nothing else than the basis
states). One finds that these are5
1 1 1 1
cos2 (α − β), sin2 (α − β), sin2 (α − β), cos2 (α − β)
2 2 2 2
Alice finds that the probability of her outcomes |αi (resp |α⊥ )
1 1 1
cos2 (α − β) sin2 (α − β) =
2 2 2
(for both cases) as in the previous scenarios; and the same holds true
for Bob. Therefore the conclusions that Alice and Bob infer from
their simultaneous local measurements are the same than in the non-
simultaneous cases above.
To summarize the situation, we see that when Alice and/or Bob perform
successive or simultaneous local measurements on their photons, whatever is
their choice of basis they find the photon in one of the two chosen basis states
with probability 12 . In other words the entropy of the probability distribution
of their local outcomes is maximal (it equals ln 2 bits) and they may infer that
their photon is in a “maximally disordered state“. In fact if they don’t know
that the source produced an entangled pair or if nobody tells them that the
two photons are entangled they have no way of even noticing that the pair
is entangled. It seems that we have no way of knowing if we are entangled
to some distant Alien in the universe, just by performing local experiments
in our part of the universe (scary no ?!). We will see in the next section that
Alice and Bob can assert that their photons are entangled if they are allowed
to communicate. Here by communicate we mean the perfect or approximate
transmission of a message.
5
fortunately independent of γ
6 CHAPTER 4. QUANTUM ENTANGLEMENT

time GALILEE time MINKOWSKI

B
B

A A

A B
A B

space space

Figure 4.3: Galilean space-time: B is in the future of A (left); B and A are

simultaneous (right). Minkowski space-time: the slopes of the light-cones equal
1/c (in natural relativistic units the speed of light is c = 1 as on the picture). B is
in the future light cone of A (left); B and A are space-like separated and cannot
be causally related. Note that for c → +∞ the slope of the light cone vanishes
and one recovers the Galilean picture.

Let us also point out that here we have discussed the situation having
in mind a Galilean picture of space-time. In other words the meaning of
the words ”before”, ”simultaneous” and ”after” is the ”usual” one. However
this is only an approximation and one might question if a proper account of
Minkowskian space-time (see figure 4.3) would change our conclusions. Ac-
cording to the theory of special relativity these words are relative to each ob-
server’s frame of reference. What has an absolute meaning is the space-time
interval which may be space-like, time-like (or zero). If the local measurement
events (events are points in space-time) of Alice and Bob are separated by
a space-like vector there cannot possibly be a causal connection between the
events, and in particular it is guaranteed that Alice and Bob cannot establish
a classical communication link during the experiment. On the other hand if
the measurement events are separated by a time-like vector it is conceivable
that there is a causal connection between the events, however unless Alice
and Bob set up such a communication link, there is no reason to believe that
there is a causal connection between the outcomes since they are exactly the
same as in the case of space-like separation.

4.2 Bell inequalities and Aspect experiment

We saw in the last section that if there is no communication between Alice
and Bob they can only infer that the photons are in a maximally disordered
state. In this section we will see that by doing repeated measurements and
by communicating the results afterwards, Alice and Bob can assert if the
4.2. BELL INEQUALITIES AND ASPECT EXPERIMENT 7

ALICE BOB

|B>

Figure 4.4: Experimental set up

state produced by the source is entangled or not.

The procedure that we are going to describe was initially invented by
John Bell, and motivated by a famous paper of Einstein-Podolsky-Rosen.
The later claimed that the entangled states do not provide a ”complete“
description of the correlations present in the system, and where seeking a
”classical“ theory of these correlations. Bell’s approach to the problem is to
try to decide if the correlations in a real pair of entangled photons (produced
by an excited atomic source say) can be described or cannot be described by a
classical theory. The general idea is that if a pair of photons is described by a
classical theory then appropriate correlation functions of the measurements of
Alice and Bob satisfy very special constraints. These constraints are violated
if the pair is described by a quantum mechanical Bell state. We will see that
Bell’s approach is able to discriminate between a huge set of classical theories
and QM. Famous experiments of Aspect-Grangier-Roger have shown that
standard QM wins!
The experimental protocol. A source S produces, at each instant of time
n, a pair of photons. We do not have any prejudice as to what is the state
or the description of the pair. One photon flies to Alice’s lab and the other
flies to Bob’s lab. In each lab our two protagonists operate independently:
they do not communicate and do not care what the other one does.

• At each time instant n, Alice randomly uses analyzers

{|αi, |α⊥ i} or {|α0 i, |α⊥

0
i}

to measure the polarization of her photon. When she records a click in

the detector she sets an = +1 or a0n = +1 and when the detector does
not click she sets an = −1 or a0n = −1. She keeps track of her choices
for the analyzer at each n.
8 CHAPTER 4. QUANTUM ENTANGLEMENT

• At each time instant n, Bob randomly uses analyzers

{|βi, |β⊥ i} or {|β 0 i, |β⊥0 i}
to measure the polarization of his photon. When he records a click in
the detector he sets bn = +1 or b0n = 1 and when the detector does
not click he sets bn = −1 or b0n = −1. He keeps track of his choices of
analyzers for each n.
• Now there is a classical communication phase. Alice and Bob meet
and discuss all their measurements. They classify them according to
the four experimental setups. At each time instant n the possible ar-
rangements of analyzers were
1 = (α, β), 2 = (α, β 0 ), 3 = (α0 , β), 4 = (α0 , β 0 )
For each arrangement they compute the following empirical averages
1 X 1 X 1 X 0 1 X 0 0
an1 bn1 , an2 b0n2 , an3 bn3 , a b
N1 n N2 n N3 n N4 n n4 n4
1 2 1 4

Then they compute the following correlation function

1 X 1 X 1 X 0 1 X 0 0
Xexp = an1 bn1 + an2 b0n2 − an3 bn3 + a b
N1 n N2 n N3 n N4 n n4 n4
1 2 1 4

Prediction of classical theories. We assume that the quantities that Alice

and Bob measure correspond to well defined observables A, A0 , B, B 0 that
have simultaneous definite values a, a0 , b, b0 even when there is no observer
(i.e. even before the measurement). This is the hypothesis of ”realism“
and is discussed in more detail later. Basically, this is like saying that a
particle has a definite position and velocity (here analogous to a and a0 ) even
when these quantities are not observed or measured. Furthermore we assume
that the outcomes of Alice and Bob can be modeled by a joint probability
distribution6
Pclass (a, a0 , b, b0 )
Here by a, b, a0 and b0 we mean the random variables modeling the measure-
ment outcomes. The expectation with respect to Pclass is denoted by Eclass .
The corresponding theoretical prediction for each empirical average above is
Eclass [ab], Eclass [ab0 ], Eclass [a0 b], Eclass (a0 b0 )
6
this second assumption follows from the assumption of ”realism” combined with ”lo-
cality“. This is explained in the next paragraph
4.2. BELL INEQUALITIES AND ASPECT EXPERIMENT 9

and using only the linearity of expectation

Xclass = Eclass [ab + ab0 − a0 b + a0 b0 ]

Notice that
ab + ab0 − a0 b + a0 b0 = a(b + b0 ) + a0 (b0 − b)
and that
−2 ≤ a(b + b0 ) + a0 (b0 − b) ≤ 2
Indeed if b = b0 then only the first term survives which leads to the inequal-
ity; while if b 6= b0 only the second term survives which again leads to the
inequality. Thus we have for the expectation,

−2 ≤ Xclass ≤ 2

This is one of the simplest Bell type inequalities which was derived by
Clauser-Horne-Shimony-Holt and is called the CHSH inequality.
In order to derive this result we haven’t assumed anything about the
state of preparation of the source. We have only assumed that the experi-
mental results can be cast into a joint probability distribution. In fact this is
not a priory so obvious. There are four experimental arrangements so that
when Alice and Bob meet they have four histograms that can be fitted to 4
probability distributions:

P1 (a, b), P2 (a0 , b), P3 (a, b0 ), P4 (a0 , b0 )

Are these the marginals of a common Pclass (a, a0 , b, b0 ) ? It is not a priory

clear that, in this experiment, nature gives us histograms that are marginals
of a common joint distribution. In fact this is not always the case. Indeed
any of us can construct four probability distributions that are not marginals
of a common one, and this is an outcome of our brains (viewed as a physical
systems). So why is the assumption leading to the CHSH inequality very
reasonable ? We answer this question below, but do not attempt to provide
the most general argument.
Let us admit that the laws of physics are ”local”. By this we mean that
when Alice (resp. Bob) perform measurements that are space-like separated
Alice’s experimental outcomes (resp. Bob’s) depend only on her own local
choice of analyzers. As far as we know, this is an assumption that underlies
all the known (i.e. experimentally verified) fundamental laws of physics.
Furthermore let us suppose, following our classical intuition, or following
Einstein, that the outcomes of experiments should be well defined preexisting
10 CHAPTER 4. QUANTUM ENTANGLEMENT

functions7 of the system state and the experimental set-up. This is sometimes
called ”realism“.
In mathematical terms ”local realism“ means that there should be a func-
tion, such that

a = fA (α; λ), a0 = fA (α0 ; λ), b = fB (β; λ), b0 = fB (β 0 ; λ)

Here λ is a set of variables accounting for the state of the system and whatever
is needed to compute the experimental outcome. It has become customary
to call them ”hidden variables”.
The hidden variables may be random or deterministic8 and their set of
values is described by a probability distribution h(λ). According to ”local
realism“ the histograms of Alice and Bob are modeled by
Z
P1 (a, b) = dλh(λ)δ(a − fA (α, λ))δ(b − fB (β, λ))
Z
0
P2 (a, b ) = dλh(λ)δ(a − fA (α, λ))δ(b0 − fB (β 0 , λ))
Z
0
P3 (a , b) = dλh(λ)δ(a0 − fA (α0 , λ)δ(b − fB (β, λ))
Z
0 0
P4 (a , b ) = dλh(λ)δ(a0 − fA (α0 , λ)δ(b0 − fB (β 0 , λ))

Evidently these are the marginals of a joint probability distribution

Z
Pclass (a, a , b, b ) = dλh(λ)δ(a−f (α, λ))δ(a0 −f (α0 , λ))δ(b−f (β, λ))δ(b0 −f (β, λ))
0 0

Prediction of QM for a Bell state. First of all we notice that according

to the quantum formalism the measurements of Alice and Bob are measure-
ments of the 4 observables (hermitian matrices)

A = (+1)|αihα| + (−1)|α⊥ ihα⊥ |, A0 = (+1)|α0 ihα0 | + (−1)|α⊥

0 0
ihα⊥ |

and

B = (+1)|βihβ| + (−1)|β⊥ ihβ⊥ |, B 0 = (+1)|β 0 ihβ 0 | + (−1)|β⊥0 ihβ⊥0 |

7
we could also frame the discussion in a slightly more general context where the outcome
is described by a probability distribution pA (a|α, λ). Here we have pA (a|α, λ) = δ(a −
fA (α, λ). The conclusions are however the same, but this remark is interesting because it
shows that it is not determinism that is at stake here.
8
in this case the distribution is simply a Dirac h(λ) = δ(λ − λ0 )
4.2. BELL INEQUALITIES AND ASPECT EXPERIMENT 11

At each time instant n the state of the photon pair is described by some ket
|Ψi ∈ C 2 ⊗ C 2 . The quantum mechanical prediction for the four empirical
averages of Alice and Bob is

hΨ|A ⊗ B|Ψi, hΨ|A ⊗ B 0 |Ψi, hΨ|A0 ⊗ B|Ψi, hΨ|A0 ⊗ B 0 |Ψi

and for the correlation function

XQM = hΨ|A ⊗ B|Ψi + hΨ|A ⊗ B 0 |Ψi − hΨ|A0 ⊗ B|Ψi + hΨ|A0 ⊗ B 0 |Ψi

Now let us compute this quantity for the Bell state

|Ψi = |B00 i

The first average is best computed by expressing the Bell state as √1 (|ααi +
2
|α⊥ α⊥ i).
1 1
hB00 |A ⊗ B|B00 i = hαα|A ⊗ B|ααi + hα⊥ α⊥ |A ⊗ B|α⊥ α⊥ i
2 2
1 1
+ hαα|A ⊗ B|α⊥ α⊥ i + hα⊥ α⊥ |A ⊗ B|ααi
2 2
1 1
= hα|A|αihα|B|αi + hα⊥ |A|α⊥ ihα⊥ |B|α⊥ i
2 2
1
1
· 1 · |hα|βi| − |hα|β⊥ i|2 + · (−1) · |hα⊥ |βi|2 − |hα⊥ |β⊥ i|2
2


=
2 2
1
1
cos (α − β) − sin (α − β) − sin2 (α − β) − cos2 (α − β)
2 2


=
2 2
= cos2 (α − β) − sin2 (α − β) = cos 2(α − β)

Performing similar calculations for the other averages we find

XQM = cos 2(α − β) + cos 2(α − β 0 ) − cos 2(α0 − β) + cos 2(α0 − β 0 )

This quantity is maximized for the following choice of angles (and all global
rotations of this choice of course, figure 4),
π π π
α = 0, α0 = − , β = , β 0 = −
4 8 8
and equals
π π 3π π √
XQM = cos + cos − cos + cos = 2 2
4 4 4 4
We see that the CHSH inequality is violated ! For the three other Bell states
on finds the same result. In the exercises you will show that this is the
12 CHAPTER 4. QUANTUM ENTANGLEMENT

b = pi/8

a =0

b’ = −pi/8

a’ = − pi/4

Figure 4.5: Optimal choice of analyzer orientation

maximum possible violation over all quantum states of C 2 ⊗ C 2 . In this

sense the Bell states are maximally entangled.
QM predicts that the four histograms of Bob and Alice are

1
P1 (a, b) = (1 + ab cos 2(α − β))
4
1
P2 (a, b0 ) = (1 + ab0 cos 2(α − β 0 ))
4
1
P3 (a0 , b) = (1 + a0 b cos 2(α0 − β))
4
1
P4 (a0 , b0 ) = (1 + a0 b0 cos 2(α0 − β 0 ))
4

For example: P2 (+1, −1) = |hα, β⊥0 |B00 i|2 = 41 (1 − cos 2(α − β)). There are
special choices of the angles α, β, α0 , β 0 for which these are not the marginals
of a common distribution Pclass (a, b, a0 , b0 ) otherwise we would have |X| ≤ 2:
this is just a mathematical fact9 . Now, nature produces these four histograms
in an experiment satisfying locality in the sense that all analyzer choices of
Alice and Bob are independent. But she plays a very subtle magic trick with
us: the correlations that are built up in Bell’s states are non-local in the sense
that correlations are present in the measurement outcomes even though the
measurements on the photons are purely local. Alice and Bob cannot notice
these non local correlations by purely local means in their own lab. They
have to meet or to communicate by exchanging matter.
Experiments. In a famous set of experiments performed in the 80’s Aspect-
Grangier-Roger showed that experiment agrees with QM and not with clas-
sical theories. The difficulty of these experiments is that, one wants to rotate
the analyzers of Alice and Bob fast enough so that the measurement events
9
In some sense they are the marginals of a quantum state
4.3. EKERT PROTOCOL FOR QKD 13

are separated by a space-like interval. Otherwise, one may always argue that
some form of classical communication or interaction conspires to make up
the results (on speaks of locality loophole). This is the challenge that the
Aspect experiments were the first to address, as compared with other slightly
earlier experiments. This locality loophole has been since then conclusively
settled by more recent experiments10 . There are other issues, that one has
to address in principle, such as the efficiency of coincident detections (called
detection loophole). So far there are no experiments that completly address
all loopholes at the same time.
The Aspect experiments tell us that we have to abandon the ”local re-
alism“. QM does not give up locality in the sense that fundamental models
of interactions are local. Rather it gives up realism in the sense that e.g one
cannot assert position and velocity (or a and a0 ) are predetermined before
a measurement. Indeed, when Alice performs a measurement her outcome
does not depend on what Bob does (this is locality) but at the same time it
is not a well defined function f (·, λ) independent of her choice of analyzers
(in this sense realism doesn’t hold). There cannot exist such a function de-
pending on hidden variables λ with distribution h(λ), which accounts for the
experimental results.
It is sometimes said that QM is non-local: this has to be understood in the
sense that quantum mechanical states of the Hilbert space can be non-local, in
other words arbitrarily extended; however the physical laws of interactions
are, as far as we know, local.
As you can begin to suspect it does not make much sense to stick to
”classical intuitions“, and words like ”local realism“ or anything of this sort.
We have to face new concepts and develop new intuitions.

4.3 Ekert protocol for QKD

A nice application of the CHSH inequality is a protocol for the generation
of a secret key by two parties. We assume that a localized source of EPR
particles delivers entangled Qbits to Alice and Bob at each time instant n in
the state
1 1
|B00 i = (|00i + |11i) = (|θθi + |θ⊥ θ⊥ i)
2 2
Moreover they have also established a noiseless communication channel.
The protocol:
10
see the review by Anton Zeilinger ”Experiment and the foundations of quantum
physics“, in Reviews of Modern Physics 71, S288-S297 (1999)
14 CHAPTER 4. QUANTUM ENTANGLEMENT

b_3=pi/8
b_2=0
a_3 =0
b_1=−pi/8
a_2 = −pi/8

a_1 = −pi/4

Figure 4.6: Alice and Bob’s random choices of analyzers

• Alice has analyzers oriented in directions a1 , a2 , a3 and records the

results of measurements, at each time instant, for the observables

A(a) = (+1)|aiha| + (−1)|a⊥ iha⊥ |

where she chooses a randomly among a1 , a2 , a3 (figure 5).

• Bob has three analyzers oriented along b1 , b2 , b3 and records the

results of measurements, at each time instant, for the observables

B(b) = (+1)|bihb| + (−1)|b⊥ ihb⊥ |

where he chooses b randomly among b1 , b2 , b3 (figure 5).

• Alice and Bob start a public discussion over the communication chan-
nel: they inform each other on what vectors they used at each time
instant.

• They do a security check to ensure that no eavesdropper is present.

Alice and Bob select the time instants when the basis choices were

(a3 , b3 ), (a3 , b1 ), (a1 , b1 ), (a1 , b3 )

Note that these are the same four analyzer arrangements used for the
Bell inequalities (figure 6). For such configurations and only for such
ones they exchange their measurement results. Each party computes
an empirical correlation coefficient

Xexp = Av[an (a3 )bn (b3 )] + Av[an (a3 )bn (b1 )]

− Av[an (a1 )bn (b3 )] + Av[an (a1 )bn (b1 )]

where Av√is the empirical average. In a perfect world they should find
Xexp = 2 2. We will see later that when an eavesdropper is present
4.3. EKERT PROTOCOL FOR QKD 15

b_3=pi/8

a_3=0
b_1=−pi/8

a_1=−pi/4

Figure 4.7: CHSH configuration

they will certainly find Xexp ≤ 2 because the effect of the eavesdropper
is to destroy the entanglement of the EPR pair and the system then
behaves ”classically“. The security check thus consists in checking that

Xexp > 2

If the test passes they conclude there is no eavesdropper and generate

the key, if not they stop communication.

• The key generation process is as follows. For every time n such that
they used the same basis - that is (a3 , b2 ) or (a2 , b1 ) - they know for
sure that
an = bn = 1, or an = bn = −1
(one can also check that in this case hB00 |A⊗B|B00 i = cos 2(a,ˆb) = 1).
Thus they have a common subsequence of ±1’s that they keep secret
and forms their shared secret key.

Attacks from Eve. Let us consider the simplest measurement attack in

which Eve captures each photon of the EPR pair and makes a measurement
(figure 7). Then she sends each photon (in the resulting state) to Alice and
Bob. She measures Alice’s photon in the basis {ea , e⊥ a } and Bob’s photon
in the basis {eb , e⊥
b }. Her strategy for the successive choices of basis at each
time instant is described by a probability distribution
Z Z
ρ(ea , eb ) ≥ 0, d2 ea d2 eb ρ(ea , eb ) = 1

After Eve’s measurement the pair of photons is left in one of the four
tensor product states

|ea , eb i, |ea , e⊥ ⊥ ⊥ ⊥
b i, |ea , eb i, |ea , eb i
16 CHAPTER 4. QUANTUM ENTANGLEMENT

ALICE
|B> BOB
a_1, a_2, a_3 b_1,b_2,b_3

e_a e_b

EVE

Figure 4.8: Eve collapses the pair in a tensor product state

with corresponding probabilities

1 1 2 d
|hea , eb |B00 i|2 = cos2 (ed
a , eb ), |hea , e⊥ 2
b |B00 i| = sin (ea , eb )
2 2
1 2 d 1
|he⊥ 2
a , eb |B00 i| =sin (ea , eb ), |he⊥ ⊥ 2
a , eb |B00 i| = cos2 (ed
a , eb )
2 2
Let us compute the correlation coefficient that Alice and Bob would find
during the security test. Given Eve’s choice (ea , eb ) we have

1 1 2 d⊥
X(ea , eb ) = cos2 (ed
a , eb )S(ea , eb ) + sin (ea , eb )S(ea , e⊥
b )
2 2
1 2 d ⊥ 1 ⊥ ⊥
+ sin (e⊥
a , eb )S(ea , eb ) + cos2 (e⊥
d⊥
a , eb )S(ea , eb )
2 2
where S(v, w) is the correlation coefficient for a pair of photons in the state
|v, wi resulting from Eve’s measurement,

S(v, w) = hv, w|A(a3 )⊗B(b3 )+A(a3 )⊗B(b1 )−A(a1 )⊗B(b3 )+A(a1 )⊗B(b1 )|v, wi

The average correlation coefficient found by Alice and Bob when Eve operates
is Z Z
X= d2 ea d2 eb ρ(ea , eb )X(ea , eb )

We leave it as an exercise to check that |S(v, w)| ≤ 2. This is not be too

surprising since |v, wi is a tensor product state. This immediately leads to,

|X| ≤ 2.

Thus Alice and Bob notice the presence of Eve. Note that Eve could ma-
nipulate (unitarily) the pair after her measurements in order to send other
photon states to Alice and Bob. However if she re-entangles the photons she
4.4. QUANTUM TELEPORTATION 17

behaves as a new source for Alice and Bob, and she gets no information from
their measurements !
Finally let us note that if Eve copies the EPR pair (this can be done with
a machine that copies the four orthogonal Bell states) and waits for the public
discussion before doing the measurements, she gets no information about the
secret key. Indeed her measurements operate on a different pair and thus
she get the same result than Alice and Bob only half of the time. This is
equivalent to flip a coin at each time instant and cannot yield information.
Experiments. see in Review of Modern Physics 74 p 145-190 (2002) the ex-
tensive article ”Quantum cryptography“ by N. Gisin, G. Ribordy, W. Tittel,
H. Zbinden.

4.4 Quantum teleportation

Suppose that Alice and Bob are spatially separated and that Alice possesses
a Qbit state,
|Φi = α|0i + β|1i, |α|2 + |β|2 = 1
The state (i.e α and β) is not necessarily known to Alice and is not known
to Bob. They also share an EPR pair
1
|B00 i = √ (|00i + |11i)
2
and have at their disposal a classical communication channel.
We are going to explain that by sending only two classical bits of infor-
mation over the classical channel, Alice can teleport the state to Bob. Here
teleportation means that |Φi is destroyed in Alice’s lab and is reconstructed
in Bob’s lab. Note that destruction of |Φi in Alice’s lab is to be expected be-
cause of the no-cloning theorem. After the teleportation process, Bob knows
that he possesses the state |Φi but still does not know the state itself (i.e he
does not know α and β). We stress that the teleportation process involves
physical transport of matter in the classical communication phase between
Alice and Bob. Of course this classical communication phase cannot happen
at speeds greater than that of light, so that the whole teleportation process
does not violate the principles of relativity. We also note that the material
support of the state (e.g. photon polarization, electron spin) |Φi = α|0i+β|1i
is not necessarily the same in Alice’s and Bob’s lab.
Teleportation can be summarized by the following ”law“

teleporting 1 Qbit = sending 2 Cbits + sharing 1 EPR pair

18 CHAPTER 4. QUANTUM ENTANGLEMENT

and can be thought of, as some form of communication between Alice and Bob
which share a classical channel and an ”EPR like channel“. The quantum
state |Φi in Alice’s lab is erased on her side and reproduced in Bob’s lab -
the information contained in α and β has not been communicated.
The protocol.
• A source produces an EPR pair of particles in the Bell state |B00 i23 .
One particle, called particle 2 is sent to Alice and one particle , called
particle 3 is sent to Bob. The Hilbert space of the entangled system 23
is H2 ⊗ H3 = C 2 ⊗ C 2 .
• Alice prepare a particle, called 1, in the state |Φi1 = α|0i + |βi. The
Hilbert space of particle 1 is H1 = C 2 .
• The total Hilbert space of the composite system 123 is H1 ⊗ H2 ⊗ H3 =
C 2 ⊗ C 2 ⊗ C 2 and the total state is
|Ψi = |Φi1 ⊗ |B00 i23
At this point a short calculation will facilitate the subsequent discussion
α β α β
|Ψi = √ |000i + √ |100i + √ |011i + √ |111i
2 2 2 2
• Alice makes a local measurement in her lab, i.e on partiles 12. She uses
an apparatus that has measurement basis of H1 ⊗ H2
{|B00 i12 , |B01 i12 , |B10 i12 , |B11 i12 }
The associated projectors for the total system are
P00 = |B00 ihB00 |⊗I3 , P01 = |B01 ihB01 |⊗I3 , P10 = |B10 ihB10 |⊗I3 , P11 = |B11 ihB11 |⊗I3
As usual the outcome of the measurement is one of the four possible
collapsed states11 (check this calculation and also that the probability
of each outcome is 14 )
1
P00 |Ψi = |B00 i12 ⊗ (α|0i3 + β|1i3 )
2
1
P01 |Ψi = |B01 i12 ⊗ (β|0i3 + α|1i3 )
2
1
P10 |Ψi = |B10 i12 ⊗ (α|0i3 − β|1i3 )
2
1
P11 |Ψi = |B11 i12 ⊗ (−β|0i3 − α|1i3 )
2
11
up to normalization
4.5. DENSE CODING 19

• Depending on the random outcome Bob has one of the four states

α|0i3 + β|1i3 = |Φi

β|0i3 + α|1i3 = X|Φi
α|0i3 − β|1i3 = Z|Φi
β|0i3 − α|1i3 = iY |Φi

but he does not know the state he has.

• Alice knows that the outcome of the measurement (in her lab) is one of
the four Bell states. She can thus use the Bell basis to re-measure (this
will not perturb Bob’s particle this time) and determine her outcome.
This outcome can be encoded by two classical bits

00, 01, 10, 11

that she sends to Bob over the classical communication channel. As

soon as Bob receives Alice’s message he knows that she has finished
her operations and he has the two bits of information needed to decide
which unitary operation he has to perform on his state in order to
recover |Φi,

I(α|0i3 + β|1i3 ) = |Φi

X(β|0i3 + α|1i3 ) = |Φi
Z(α|0i3 − β|1i3 ) = |Φi
−iY (β|0i3 − α|1i3 ) = |Φi

4.5 Dense coding

Suppose Alice and Bob have established a quantum channel over which they
can send Qbits (for example a optic fiber over which photons travel). We
will study the capacity of such a noisy channel later in the course but for
the moment let us address a simpler question. Assume that Alice and Bob
share an EPR pair. How much information does one Qbit convey over the
quantum channel ?
The answer is that 2 classical bits of information can be transmitted by
Alice to Bob, by sending only 1 Qbit as long as they share an EPR pair. The
protocol that achieves this is called dense coding.
We will come back to the problem of communicating classical/quantum
messages over noisy quantum channels assisted/or not by entanglement in
20 CHAPTER 4. QUANTUM ENTANGLEMENT

later chapters. As we will see even for simple analogs of Shannon’s channel
coding theorem there are various open questions.
Dense coding can be summarized as follows:

communicating 2 Cbits = sending 1 Qbit + sharing 1 EPR pair

This ”law“ may seem complementary to the one of teleportation. Note how-
ever that here only two particles are involved and it is the Qbit that is
physicaly transported form Alice to Bob.
Protocol.

• An EPR pair in the state |B00 i is prepared by a source and each particle
sent to Alice and Bob.

• Alice wants to communicate two bits of information to Bob:

– To send 00 she leaves her particle intact (or applies the unitary
gate I) and physically sends her particle to Bob. Bob receives the
particle and is now in possession of the whole state

|B00 i

– To send 01 she applies the unitary gate X to her particle and then
physically sends her particle to Bob. Bob is now in possession of
the pair in the state

X1 ⊗ I2 |B00 i = |B01 i

– To send 10 she applies the unitary gate Z to her particle and then
physically sends her particle. Bob is now in possession of the pair
in the state
Z1 ⊗ I2 |B00 i = |B10 i
– To send 11 she applies the unitary gate iY to her particle and
then physically sends her particle. Bob is now in possession of the
pair in the state
(iY )1 ⊗ I2 |B00 i = |B11 i

• Bob now has the EPR pair 12 in some state |Bxy i. In order to determine
the two Cbits that Alice sent he must decide which Bell state he has.
Since he knows that he has one of the four Bell states in his lab, he can
do a local measurement in the Bell basis, and access the information
xy.
4.5. DENSE CODING 21

H H

Figure 4.9: Device for Bell basis measurements

Measurement in the Bell basis. One might think that measuring in

the Bell basis is a theoretician’s wishful thinking. In fact this has been
realized experimentally, and although explaining how is beyond the scope
of this course, we give here an argument that shows that, in principle, it
suffices to have H and CN OT gates (the simplest unitary gates) together
with polarization analyzers (the simplest measurement apparatus).
We have seen at the beginning of this chapter that Bell states can be
generated as |Bxy i = (CN OT )(H ⊗ I)|xyi. The projectors on the Bell basis
states are therefore related to the ones over the Z basis,

|Bxy ihBxy | = (CN OT )(H ⊗ I)|xyihxy|(H ⊗ I)(CN OT )

(here we have used that the Hadamard and control not matrices are her-
mitian). The projectors |xyihxy| correspond to the analyzer-photo-detector
apparatus for photons or to spin analyzers (Stern-Gerlach analyzer) for spins
(Z basis). The circuit representation of a measurement device in the Bell
basis in given on figure 8. The input is any state |Ψi, and the output is one
of the four states
hBxy |Ψi
|Bxy i
|hBxy |Ψi|

Experiments. Quantum teleportation and dense coding have been realized

experimentally. A summary of the subject can be found in ”Les dossiers de la
recherche“ no 18, février 2005, ”L’étrange pouvoir de l’intrication quantique“,
by N. Gisin.