Quantum Communication

Dieter Jaksch
TT 2009

Clarendon Laboratory
University of Oxford
Contents
1 Introduction 1

2 Basics of Information Theory 1

2.1 Quantifying Classical Information . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1.1 The setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1.2 The Sender and Shannon’s Noiseless Coding Theorem . . . . . . . . . . . 2
2.1.3 The Receiver and Mutual Information . . . . . . . . . . . . . . . . . . . . 4
2.1.4 The communication channel . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.5 Connection to Statistical Physics . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Quantifying Quantum Information . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Information content of a density operator . . . . . . . . . . . . . . . . . . 9
2.2.3 Joint entropy and mutual information . . . . . . . . . . . . . . . . . . . . 10
2.2.4 Quantum channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Quantum dense coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Photon techniques 14
3.1 Polarization and spatial mode encoding . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.1 Spatial encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.2 Polarization encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.3 Parametric down conversion . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1.4 Single photon sources and single photon detectors . . . . . . . . . . . . . 17
3.1.5 Quantum dense coding: Experimental setup . . . . . . . . . . . . . . . . . 17

4 Testing EPR 18
4.1 Bell inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1.1 The CHSH inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1.2 The Aspect experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.3 Loopholes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 GHZ states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2.1 Violations of local realism . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2 Experimental realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5 Quantum communication 22
5.1 Quantum teleportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2 Entanglement swapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

6 Quantum cryptography 24
6.1 One time pads and the Vernam cipher . . . . . . . . . . . . . . . . . . . . . . . . 25
6.2 The BB84 protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.2.1 Intercept - resend strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.3 Quantum key distribution using EPR pairs . . . . . . . . . . . . . . . . . . . . . 27
6.4 Experimental setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.4.1 Phase encoded fibre systems . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.5 More about communication schemes . . . . . . . . . . . . . . . . . . . . . . . . . 29
6.5.1 The quantum telephone exchange . . . . . . . . . . . . . . . . . . . . . . . 29
6.5.2 Speeding up distribution of entanglement . . . . . . . . . . . . . . . . . . 30

I
 6.5.3 Privacy amplification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

7 Further reading 31

II
1 Introduction
Scientific progress in physics and mathematics has led to the development of efficient technology
for communicating and distributing information in the 20th century. This technology forms one
of the cornerstones of the information society and our global economy which are highly reliant
on secure methods to transfer and distribute information quickly. To satisfy ever increasing
demands for speed and security encryption and communication methods are constantly improved
and there is an ongoing effort to develop corresponding technology further.
Classical information theory is not usually part of a physics undergraduate degree. It as-
sumes simple classical properties of physical systems and based on those a largely mathematical
theory of information is established. The obtained results are mostly independent of the chosen
implementation and its physical details. However, one still has to acknowledge that information
is physical, i.e. information carriers, senders, and receivers obey the laws of physics. The notion
of quantum information comes about since it turns out that the rules of quantum mechanics
violate some of the basic physical assumptions of classical information theory. The consequences
of this are many, ranging from improved channel capacities, the possibility of physically secure
communication protocols to invalidating some assumptions about the security of classical com-
munication protocols. The field of quantum information processing is still at its infancy and
is currently closely linked with physics. However, one can expect that quantum information
theory will develop into a field of its own if simple quantum physical properties determining the
behaviour of quantum information can be identified.
This lecture course starts by introducing the basics of classical information theory and some
of the most important quantum counterparts. This is followed by a discussion of photon tech-
nologies for realizing quantum communication. The violation of basic classical assumptions by
quantum systems is then exemplified by showing how entangled states violate Bell’s inequalities
and local realism. Finally, schemes for efficient quantum communication based on entangled
states and physically secure cryptographic communication methods are introduced.

2 Basics of Information Theory

As already mentioned in the introduction information is physical. It must be embodied in the
state of a physical system and processing of information must be accomplished by dynamical
evolution of a physical system. Information is thereby defined by the ability to perform a certain
task and quantified by how many resources are required to perform a specific task successfully.

Examples:

E1. How many data CD’s (the resources storing the information) are needed to store a map
of the UK (the task) which specifies the position of each address (success)?

E2. Which physical resources are required to transmit a state |Ψi from sender Alice to receiver
Bob with entanglement fidelity F = 0.999?

As can already be seen from these simple examples numerous different types of resources
exist and success can be defined in a number of different ways. It is thus desirable to quantify
information in terms which are to a large extent independent of the physical realization.

1
2.1 Quantifying Classical Information
2.1.1 The setup
We consider the situation where a sender (Alice) communicates with a receiver (Bob) over
a communication channel. We do not wish to make assumptions on how the messages are
embodied. We assume the following general setup.

• The sender Alice:

– She can send one out of N messages x1 · · · xN per use of the channel.
– The probability that Alice chooses message xj is known and given by pj . We do not
know the physical laws which allow to calculate the message chosen by Alice.

• The communication channel:

– The channel is capable of transmitting one of Alice’s N messages to the receiver in

each use.
– It can introduce noise and be susceptible to eavesdropping by Eve.
– We consider classical and (later) quantum channels.

• The receiver: Bob

– The channel provides Bob with one of the messages y1 · · · yM .

– The probability to receive message yn is denoted by qn .

To quantify the amount of information transmitted between sender and receiver in this
scenario we describe Alice’s messages by a random variable X which can take the values x1 · · · xN .
X takes on the value xj with probability pj and the probabilities sum to one
X
pj = 1.
j

The amount of information contained in a message (or in X) is defined as the number of bits
which are at least required to store an outcome of a measurement of X. It tells us how much we
learn from a perfect measurement of (reading) X. When measuring X the uncertainty about
its content is reduced. The information gained about the message is therefore defined as the
reduction in information content induced by the measurement. After a perfect measurement of
X we know for sure which message was sent and subsequent measurements on this message will
not tell us anything new. This process thus reduces the information content to zero and the
gained information equals the original information content. For the general case of imperfect
measurements the original information content and the gained information do not agree and
some residual uncertainty about the message is left. In the following sections we discuss the
properties of the sender, receiver and communication channel, respectively.

2.1.2 The Sender and Shannon’s Noiseless Coding Theorem

The information content of a message sent out by Alice is given by the Shannon entropy H(X)
defined as
XN
H(X) = − pj log2 (pj ).
j=1

2
 1
2.
0.8

0.6

H(p) 0.4

0.2
1. 3.
0
0 0.2 0.4 0.6 0.8 1
p

Figure 1: Shannon entropy H(p) as a function of probability p. Message 0 is sent with probability
p and message 1 with probability 1 − p. A maximum is reached for p = 1/2 at point 2. The
information content goes to zero if only one message is ever sent at points 1 and 3.
Monday, 29 January 2007 10

The Shannon entropy H(X) does not depend on the values xj of X but just on the probabilities
pj . It is thus applicable to any kind of message if the probability distribution of messages is
known. A message which occurs with probability zero does not add to H(X) since 0 log2 (0) = 0.
We cannot gain any information from messages which are never sent. If only one message
appears with certainty it does also not contain any information since 1 log2 1 = 0. Nothing can
be learnt from measuring one message only since the outcome can be predicted with certainty.
The Shannon entropy is bounded by 0 ≤ H(X) ≤ log2 (N ) with H(X) = log2 (N ) if and only if
(iff) pj = 1/N ∀ j.

Example:

E3. Alice can send two messages 0 and 1. She chooses 0 with probability p and 1 with
probability 1 − p. How much information does one of her messages contain?
Using the above expression for H we find

H(p) = −p log2 (p) − (1 − p) log2 (1 − p).

The maximum information content is reached for

dH(p)
= − log2 (p) + log2 (1 − p) = 0 ,
dp

which yields pmax = 1/2 and minima are obtained for pmin,1 = 0 and for pmin,2 = 1 since
the possible range of values of p is bounded by 0 ≤ p ≤ 1. At pmax each message contains
one bit of information while at pmin the information content is zero as expected. The
shape of H(p) is shown in Fig. 1.

If both messages are sent with equal probability in the above example one bit is necessary
to store which message was sent. If Alice always sends the same message nothing needs to be

3
stored to know which message was sent. For all other values of p the information content has
to be interpreted as the average number of bits required to store a message if a large number of
messages (strictly speaking infinitely many) are sent. That this is indeed the case is the content
of Shannon’s noiseless coding theorem which we state here without proof: A message xj can on
average be compressed to H(X) bits using an optimal code for message compression.

Example:

E4. Alice can send messages X with values a,b,c. The probability for a is pa = 1/2 while b, c
have a probability of pb = pc = 1/4. How much information is contained in one message?
1 3
H(X) = −pa log2 pa − pb log2 pb − pc log2 pc = (log2 2 + log2 4) = .
2 2
By encoding a using the bit string 0, b as 10 and c as 11 the average length L of a bit
string representing a value of X will be
1 1
L=1× × 1 + 2 × × 2 = 3/2 = H(X) .
2 4
This code is optimal, the messages cannot be be compressed further without losing in-
formation about the original messages. Note that the bit string of optimally encoded
messages contains a 0 or a 1 at each position with the same probability 1/2.

Finding the optimal encoding for given message probabilities is a non-trivial task (c.f. the
varying performance of data compression software). The following example illustrates the per-
formance of a very simple data compression procedure.

Example:

E5. Alice sends message a with probability p and message b with probability 1 − p. Assuming
p > 1/2 we choose to encode aa as 0, ab as 10 and b as 11. The message string aa occurs
with probability p2 , ab with p(1 − p) and b with (1 − p). The average number of bits
required to store a message L is given by
1
L= × p2 + 1 × p(1 − p) + 2 × (1 − p) .
2
The comparison of the average length L with the Shannon entropy H(p) shown in Fig. 2
reveals that this encoding is never optimal and gets best for p ≈ √
0.774. This encoding is
only better than simply identifying a with 0 and b with 1 if p > 3 − 1.

The noiseless coding theorem quantifies the amount of information contained in the messages
sent out by Alice. It is thus also often called Shannon’s source coding theorem. When the
messages Y are received by Bob they will most often have been subject to noise and thus not
be identical to X. Therefore the question arises what Bob learns about the messages X when
reading the received messages Y .

2.1.3 The Receiver and Mutual Information

The messages Y transmitted to Bob via the communication channel take on values yn with
probabilities qn . Bob reads the messages Y and gains information H(Y ). To quantify the

4
 Hp, Lp
2
L(p)

H(p)

p
0.5 1

Figure 2: Shannon entropy H(p) and average bit string length L(p) as a function of probability
p. The vertical dashed line indicates the value of p where the bit string length is closest to the
entropy H(p).

amount of information gained about the original messages X by reading Y we first introduce
the joint entropy of the variables X and Y as
X
H(X, Y ) = − p(xj , yn ) log2 (p(xj , yn )) ,
j,n

where p(xj , yn ) is the probability that X takes the value xj and Y takes the value yn . This
joint entropy is the total information content of variables X and Y . Furthermore we define the
entropy of X conditional on knowing Y by

H(X|Y ) = H(X, Y ) − H(Y ) .

H(X|Y ) tells us how uncertain we are about the value of X after measuring Y .
If X and Y are uncorrelated, i.e. the probability qn of Y taking on the value yn is independent
of the value of X, then Bob does not learn anything about X from measuring Y . In this case

p(xj , yn ) = pj qn

and we have
H(X, Y ) = H(X) + H(Y ) ,
(see class problems for a proof). Thus the information content of X is not decreased by measuring
Y . In other words, Bob’s uncertainty on the value of X does not decrease when measuring Y ,
i.e. H(X|Y ) = H(X). However, if the value of X is fixed by measuring Y , i.e.1

qn for j = n
p(xj , yn ) = ,
0 otherwise

we find that H(X, Y ) = H(Y ). When Bob measures yn he knows that X certainly has the
value xn . In this case he learns all he can about X, i.e. H(X|Y ) = 0. X and Y are perfectly
correlated. The measurement reduced the information content H(X, Y ) = H(Y ) by H(Y ) to
zero and thus no uncertainty about either X or Y is left after measuring Y .
1
Here we assume for simplicity that M = N and that the ordering of messages is preserved in the transmission

5
 Based on these two observations we introduce the mutual information content of X and Y
as
H(X : Y ) = H(X) + H(Y ) − H(X, Y ) .
This can also be written as H(X : Y ) = H(X) − H(X|Y ). We find that no mutual information
is contained in X and Y if they are uncorrelated since then H(X, Y ) = H(X) + H(Y ). In this
case no information is transmitted over the channel. If X and Y are perfectly correlated we
have H(X, Y ) = H(Y ) and the mutual information between X and Y takes its maximal value
H(X : Y ) = H(X). The information sent by Alice can be completely restored at the receiver
side in this case.

2.1.4 The communication channel

We can now quantify the communication channel in terms of its channel capacity. This defines
the amount of information which is transmitted by the channel in a single use. Note that it is
not important that the messages arrive without being altered. The only criterion is whether
Bob can reconstruct Alice’s message from the output. For instance, for a channel which maps
1 → 1 and 0 → 0 we find H(X : Y ) = H(X) and a channel with 1 → 0 and 0 → 1 also has
H(X : Y ) = H(X). However, a channel with 1 → 1 and 0 → 1 does not transmit information
as can be seen by working out H(X : Y ) = 0.
A noiseless channel N produces messages Y which are perfectly correlated with the initial
messages X and thus H(X : Y ) = H(X). In this case Alice can transmit H(X) bits of informa-
tion with every use of N . By exploiting Shannon’s noiseless coding theorem H(X) = log2 (N )
can be achieved. The channel capacity C(N ) is therefore given by
C(N ) = log2 (N ).
If a noisy channel is used X and Y will be correlated but not perfectly. The question then
is whether by redundant encoding one can ensure arbitrarily good reliability of the channel.
Shannon’s noisy channel coding theorem states that this is possible with a channel capacity of
C(N ) = max{H(X : Y )} ,
{pj }

where the maximum is taken over all possible input probability distributions pj of X. We do
not give a proof of this theorem.
Remark: Sometimes channel capacity is given in bits/sec which is C(N ) times the number
of possible channel uses per second (e.g. for internet connections).

2.1.5 Connection to Statistical Physics

The Shannon entropy is a generalization of the equation S = kB ln W from statistical physics
where kB is the Boltzmann constant and W the number of microstates accessible to the system.
This equation follows from our definition of the Shannon entropy by identifying each microstate
with one of the messages and assuming that each of them is occupied with the same probability
1/W . The Shannon entropy for this type of ‘sender’ takes on its maximum value and equals S
up to an unimportant constant factor. The thermodynamic entropy S is thus a measure of our
ignorance of the thermal state assuming that all microstates are equally likely populated.

2.2 Quantifying Quantum Information

It is useful to revise some quantum mechanics before defining quantum information in a quantum
setup similar to the classical setup discussed in the previous section.

6
2.2.1 Quantum Mechanics
The trace of an operator In linear algebra the trace of a matrix M is defined as the sum
over all diagonal elements X
Tr {M } = Mnn .
n
This definition extends to operators in a linear Hilbert space as follows. Given an operator
M̂ and an orthonormal basis |φn i the operator can be rewritten as a matrix M with matrix
elements
Mnm = hφn | M̂ |φm i .
P
This matrix represents the operator M̂ in basis |φn i and we can write M̂ = nm |φn i Mnm hφm |.
The trace of an operator is defined as the sum over all diagonal elements of M
n o X X
Tr M̂ = Mnn = hφn | M̂ |φn i .
n n

For this definition to be physically relevant we need to show that it is independent

P of the chosen
basis. Two bases are related by a unitary matrix U according to |ψm i = n Umn |φn i. Thus in
the new basis the trace of M̂ is given by the trace of the matrix
n o n o
Tr M̂ = Tr U † M U .

Since the trace does not change under cyclic permutation and U is unitary we have
n o n o
Tr M̂ = Tr M U U † = Tr {M } ,

and thus the definition of the trace is basis independent.

Analytical function of an operator For working out P a function of an operator we first

write it in terms of its eigenvalues and eigenvectors M̂ = m |φm i Mm hφm |. Any power of this
operator can then be written as M̂ n = m |φm i Mm n hφ |. Any analytical function F of this
P
m
P
operator is then well defined as F{M̂ } = m |φm i F{Mm } hφm |.

Example:

E6. We calculate the logarithm of the operator σx +3I = 4 |+i h+|+2 |−i h−| with eigenvalues
2 and 4. We find log2 (σx + 3I) = log2 (4) |+i h+| + log2 (2) |−i h−| = 2 |+i h+| + |−i h−|

The partial trace of an operator The partial trace of an operator is defined as the trace
over a subspace of the total Hilbert space. In working it out an operator acting on the total
Hilbert space is mapped into one acting on a subspace of the Hilbert space only. In our case we
will e.g. be interested in tracing over the part of the Hilbert space pertaining to the sender and
will be left with an operator that only acts on the degrees of freedom accessible to the receiver.
The total Hilbert space is broken up into two parts H = HA ⊗ HB . The basis of H can be
written as |nmi = |niA ⊗ |miB where |niA describes degrees of freedom of the sender HA and
|miB those of of the receiver HB . The partial trace of an operator M̂ over subspace HA is given
by n o X X
TrA M̂ = A hn| M̂ |niA = |miB (Mnm,no ) B ho| .
n n

7
Here we have used the notation Mnm,lo = hnm| M̂ |loi and used the rule2 A hn| loi = δnl |oiB .
The resulting operator contains only basis elements corresponding to degrees of freedom of the
receiver. We have thus traced out the sender. Note that subsequently tracing over both Hilbert
spaces yields the trace of the operator.

The density operator All quantum mechanical expectation values for a system in the pure
state |ψi can be rewritten in terms of a trace
n o
hM̂ i = hψ| M̂ |ψi = Tr |ψi hψ| M̂ .

We can thus define the operator ρ = |ψi hψ| and use it replace the state vector |ψi. This operator
is called the density operator or state of the system. All observable quantities can be worked
out from the density operator3 .
Sometimes the state vector of a system is not known but it is known that the system will be
in state |ψn i with probability pn . This can e.g. happen if the system preparation is imperfect,
after a measurement if the outcome is not revealed, in decoherence processes, when a system is
in a thermal state, or when it produces quantum messages |ψn i with probability pn . Then the
expectation value of an operator has to be worked out for each possible state and to be weighted
with the corresponding probability4 . This yields
X n o
hM̂ i = pn Tr |ψn (t)i hψn (t)| M̂ .
n

The trace is linear and therefore we can define the (mixed) density operator
X
ρ= pn |ψn (t)i hψn (t)| ,
n
n o
and in general write hM̂ i = Tr M̂ ρ . Using a density operator to describe a system thus
allows a compact and efficient way to include classical uncertainty about its wave function into
quantum mechanical calculations.

Global measurement In second year quantum mechanics measurement of a hermitian op-

erator M̂ with eigenvectors |φn i and non-degenerate eigenvalues an is introduced. The average
measurement outcome for a system in pure state |Ψi is given by hM̂ i = hΨ| M̂ |Ψi. In a single
measurement the eigenvalue an is obtained and the system collapsed into the state |φn i with
probability pn = | hφn | Ψi |2 = hφn | Ψi hΨ| φn i. For a mixed state ρ this extends using identical
arguments as above. Outcome an is obtained with probability pn = hφn | ρ |φn i. If outcome an
is obtained the mixed state is collapsed into |φn i hφn | ρ |φn i hφn | /pn = |φn i hφn |.
For an operator with degenerate eigenvalues an and eigenvectors |φn,l i, where l enumerates
the setPof degenerate eigenvectors for eigenvalue an , the probability
P of outcome an is given by
pn = l hφn,l | ρ |φn,l i. The mixed state is collapsed into l |φn,l i hφn,l | ρ |φn,l i hφn,l | /pn . Note
that pn is included here to obtain a normalized state. The observer cannot distinguish states
with degenerate eigenvalues. The corresponding degrees of freedom are not accessible in this
measurement.
2
This is not a proper scalar product and slightly sloppy, though widely used, notation.
3
The unmeasurable global phase is gone.
4
This is a classical uncertainty of the state and has nothing to do with the quantum uncertainty.

8
 In most cases we will not be concerned with the eigenvalues an obtained in a measurement
of an operator M̂ . We instead specify a measurement via a set of orthnormal states |φn i and ask
about the probability pn of projecting the system into state |φn i. The corresponding eigenvalues
are assumed to be non-degenerate so that these state are distinguishable for the observer. The
operator M̂ and its eigenvalues an are not required to calculatePthe probabilities pn . However,
a corresponding observable M̂ could be constructed as M̂ = n an |φn i hφn | with arbitrarily
chosen real an 6= al for n 6= l.

Local observables and measurement A local observable of the receiver can be written as
M̂ = IA ⊗ M̂B since the receiver cannot directly measure degrees of freedom located at the
sender. We work out the expectation value of M̂ by first tracing over the sender and find
D E n n oo n o n o
M̂ = TrB TrA M̂ ρ = TrB M̂B TrA {IA ρ} = TrB M̂B ρB

where ρB = TrA {ρ} is the reduced density operator of the receiver. Thus all information
about observables which can be measured locally by the receiver is contained in the reduced
density operator ρB . The same argument holds for the sender with reduced density operator
ρA = TrB {ρ}.
The measurement of a local observable M̂ = IA ⊗ M̂B can be described as above by specifying
the eigenstates of MB written as |φm iB and using the reduced density operator ρB . The mea-
surement collapses the state of the receiver into |φm iB and the total system into ρA ⊗|φm iB hφm |
with probability pm = B hφm | ρB |φm iB . The operator M̂ has degenerate eigenvalues5 since the
receiver cannot
P measure degrees of freedom of the sender. Alternatively, we can write the oper-
ator as M̂ = n,m |niA hn| ⊗ am |φm iB hφm | with am the (assumed) non-degenerate eigenvalues
of MB and n labeling the degeneracy6 . Using the above definitions for measuring a global de-
generate observable M̂ we again find that the system is projected into state ρA ⊗ |φm iB hφm |
with probability pm . Both approaches are equivalent.

Example:

E7. Alice and Bob share a Bell state |ψ − i. What is the reduced density operator for each of
them? What are the probabilities for a local measurement at Alice’s site to project the
system into the orthogonal states |φ1 iA and |φ2 iA ? In this case the Hilbert space is split
into H = C2 ⊗ C2 and basis states |nmi = |niA ⊗ |miB with n, m = 0, 1. The reduced
density operator of the receiver Bob is
1 IB
ρB = A h0| ψ − ψ − 0iA + A h1| ψ − ψ − 1iA = (|0iB h0| + |1iB h1|) = .
2 2
This is the maximally mixed state of a qubit. By symmetry we obtain ρA = |0iA h0| +
|1iA h1| = IA /2. The probability to project into either of two states in a local measurement
is 1/2.

2.2.2 Information content of a density operator

In quantum mechanics the classical messages X are replaced by the density operator. Alice
prepares quantum states to be sent to Bob. These encode the messages and are described by a
5
Except forP
the trivial case where the sender only consists of a one dimensional Hilbert space
6
Note that n |niA hn| is the identity operator on the sender subspace.

9
 ry system is and can thus be viewed as being represented by S(ρ) qubits.
• Like in classical compression this only works on average, i.e. if the source
produces a large number m of quantum messages.
en questions • Reliably in this case means that the entanglement fidelity of the original
state ρ⊗m after compression Cm and decompression Dm tends to 1 for large
on tasks? m. The entanglement fidelity tells us how well the state ρ⊗m preserves its
entanglement with an environment during compression and decompression.
We do not define the entanglement fidelity here (see NC page 420).
sing quantum
Cm Dm
ρ ρ‘ ρ”

m log2(d) qubits mS(ρ) qubits m log2(d) qubits

17 Friday, 25 November 2005 18

Figure 3: Compression Cm and decompression Dm of m copies of a quantum state ρ. The size

of the original Hilbert space corresponds to m log2 (d) qubits which are compressed to mS(ρ)
qubits.
Photons as spatial mode and polarization encoded qubits

density operator ρ.• The

Spatial mode encoding
density operator ρ is hermitian and can thus be written as
– Two spatial modes a and b (direction, momentum) are chosen to
represent the qubit states
X |0i and |1i
– Single qubit gates ρ pj |xj by
are=implemented i hxj | ,
• a phase shifter in one jspatial mode Æ phase gate
• beam splitter Æ Hadamard gate
where |xj i are orthogonal
– Twonormalized
qubit gates caneigenvectors (i.e.nonlinearities
be realized by Kerr the quantum messages) and pj are the
probabilities of Alice state |xj i to Bob7 . The von Neumann entropy of the state ρ is
sending encoding
• Polarization
defined by – The qubit is encoded in the photon polarization e.g. |0i = |Hi, |1i = |Vi
X
– Single qubit gates are implemented by
S(ρ) = Tr {ρ log2 (ρ)} = − pj log2 (pj ) .
• polarization rotators and polarization phase shifters
j
• polarizing beam splitter Æ spatially separate |Hi and |Vi components
The von Neumann–entropy
Two qubitisgates
a measure of our ignorance
e.g. with polarizing about
beam splitters thenonlinearities
and Kerr quantum state. It plays a
• Linear
similar role for quantum opticsas
states quantum computing entropy
the Shannon by entanglement
does forcreation via measurement
classical random variables. Using
• Photon number encoding: |0i Æ no photon |1i Æ 1 photon
the von Neumann entropy quantum states can almost be treated as if they were information.
• Spatial + polarization encoding allows to store two qubits in one photon
The von Neumann entropy also plays an important role in quantum statistical mechanics. Up
– This encoding is not scalable
19 to a constant factorFriday,
it 25reduces to the familiar entropy S for thermal states where
November 2005 20 each accessible
microstate (message) is occupied (sent) with the same probability.

Schumacher’s quantum noiseless channel coding theorem In analogy to Shannon’s

noiseless coding theorem Schumacher showed that states ρ in a d dimensional Hilbert space H
produced by a quantum information source can be compressed. In particular it is possible to
reliably compress and decompress ρ to a quantum state in a Hilbert space HA with dimension

dim(HA ) = 2S(ρ) ,

and can thus be viewed as being represented by S(ρ) qubits. Like in classical compression this
only works on average, i.e. if the source produces a large number m of quantum messages. The
procedure is schematically shown in Fig. 3. Reliably in this case means that the entanglement
fidelity of the original state ρ⊗m after compression Cm and decompression Dm tends to 1 for
large m. The entanglement fidelity tells us how well the state ρ⊗m preserves its entanglement
with an environment during compression and decompression8 .

2.2.3 Joint entropy and mutual information

We define the joint entropy for the quantum state of Alice and Bob ρAB in analogy to classical
information as
S(ρAB ) = −Tr {ρAB log2 (ρAB )} .
7
We assume here for simplicity that different messages are orthogonal.
8
We do not define the entanglement fidelity here (see Nielsen and Chuang, page 420 for details).

10
The reduced density operators ρA for Alice and ρB for Bob yield the corresponding information
contents S(ρA ) and S(ρB ). From those the conditional entropy and mutual information follow
as in the classical case
S(ρA |ρB ) = S(ρAB ) − S(ρB ) ,
S(ρA : ρB ) = S(ρA ) + S(ρB ) − S(ρAB ) .
These quantities replace those introduced in classical information theory. Note that they do,
however, not have the classically expected properties. For instance, the conditional entropy can
become negative as we will explicitly work out later for the case of two entangled qubits.
In contrast to the classical case Bob cannot read his messages without affecting the quantum
state. The process of measuring the received messages can thus influence the entropy of the
state. We study different possibilities by considering measurement on one qubit.

Example:
√
E8. A qubit is prepared in the pure state |+i = (|0i+|1i)/ 2. It is measured in the computa-
tional basis and we consider two different scenarios: (i) the outcome of the measurement
is |0i and (ii) the outcome of the measurement is not revealed. Case (ii) can e.g. happen
in a decoherence process where the measurement is performed by the environment and
the outcome is not accessible. How does the information content of the qubit change?
(i) The initial state is ρi = |+i h+|. This is a pure state (state |+i is prepared with
certainty) and thus has entropy S(ρi ) = 0. After the measurement the state is ρf = |0i h0|
which is again pure and thus S(ρf ) = 0. While the quantum state changes in this process
the qubit remains in a pure quantum state and does not contain information. Still, each
of the two possible measurement outcomes occurs with probability 1/2.
(ii) The possible states after the measurement are |0i and |1i each with probability of
p0 = p1 = 1/2. Not knowing the actual outcome we thus have to write
1 1
ρf = |0i h0| + |1i h1| .
2 2
This state has entropy S(ρf ) = 1. Lacking the knowledge of the measurement outcome
turns the initial pure state into a mixed state. Its information content (uncertainty about
the qubit state) increases.
Finally we consider what happens if a measurement in the computational basis is per-
formed on the state ρf and the outcome is |0i. In this process the entropy is reduced
from 1 to 0. The mixed state is turned into a pure state by the measurement and one bit
of information about the original state is gained. This type of measurement is consistent
with measuring in a classical system.

2.2.4 Quantum channels

A quantum channel transforms input systems described by a Hilbert space H1 into output
systems described by Hilbert space H2 .9 It is represented mathematically by a completely
positive, unital map T which acts on the density operator ρ as
n
Ej ρEj† .
X
T (ρ) =
j=1
9
H2 is often assumed to be identical to H1 .

11
 P †
Here Fj are the so-called Kraus operators which fulfill j Ej Ej < I. The channel capacity
quantifies the number of qubits which can be faithfully transmitted. For an ideal channel we
have T = I, the identity operation. The channel capacities of quantum channels are fully
understood only for special cases. For instance the Holevo-Schumacher-Westmoreland (HSW)
theorem gives the channel capacity if only product input states are used10 . A detailed discussion
of channel capacity is beyond the scope of this lecture course. Instead we consider two simple
examples of how noise and decoherence affect the information content of quantum systems.

Examples:

E9. A classical bit is stored in two atomic states |0i and |1i. The state |0i is stable while
the state |1i is metastable and spontaneously emits photons at a rate γ. How does the
mutual information between the state of the atom and the initial bit change with time?
The original bit is in a maximally mixed state ρA = (|1i h1| + |0i h0|)/2 and contains
S(A) = 1 bit of information. The joint bit-atom system is initially prepared in state
ρAB = (|11i h11| + |00i h00|)/2. If the atom is initially in state |1i the probability p1 of
finding it in state |1i at a later time is determined by the equation ṗ1 = −γp1 and thus
we find
e−γt 1 − e−γt 1
ρAB = |11i h11| + |10i h10| + |00i h00| .
2 2 2
This state has entropy

e−γt 1 − e−γt 1 − e−γt

 
1
S(AB) = (γt log2 (e) + 1) − log2 + .
2 2 2 2

We find the entropy of the atomic state by tracing out the initial bit obtaining the reduced
density operator
e−γt 2 − e−γt
ρB = |1i h1| + |0i h0| .
2 2
This state has entropy

e−γt 2 − e−γt 2 − e−γt

 
S(B) = (γt log2 (e) + 1) − log2 .
2 2 2

Both entropies are shown in Fig. 4. Initially the entropy solely arises from the unknown
state of the bit. The uncertainty of the atomic state due to spontaneous emission then
increases the overall entropy for short times t ≤ 1/γ. For times t  1/γ the atom
will be in the pure state |0i and the entropy of the system is again solely due to the
state of the initial bit. However, in this process the correlation between the state of
the atom and the bit is lost as becomes evident by looking at the mutual information
S(A : B) = S(A) + S(B) − S(AB) also shown in Fig. 4. Note: This example can equally
well be described using classical information theory.

E10. A photonic channel is used to transmit two messages |0i ≡ no photon present and
|1i ≡ one photon present. 20% of the photons are lost in the channel. We investigate
the following classical and quantum scenarios for using this channel to establish mutual
information between sender Alice an receiver Bob.
10
See Nielsen and Chuang page 555.

12
 SHABL SHBL SHA:BL
1.5 1 1
1.4 0.8 0.8
1.3 0.6 0.6
1.2 0.4 0.4
1.1 0.2 0.2
Γ t Γ t Γ t
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

Figure 4: Entropies of an atom B storing a classical bit A. The atom undergoes spontaneous
emission.

(i) Classical messages X with values |0i and |1i are sent with probabilities p0 = p1 = 1/2
and received as messages Y as |0i and |1i. How much mutual information is created in
one use of the channel?
We find H(X) = 1. The joint entropy can be worked out from the probabilities p(0, 0) =
1/2, p(0, 1) = 0, p(1, 1) = 4/10 and p(1, 0) = 1/10. It is given by H(X, Y ) = 1.361. The
probabilities on the receiver side are q0 = 6/10 and q1 = 4/10 and thus H(Y ) = 0.971. We
therefore find H(X|Y ) = H(X, Y ) − H(Y ) = 0.39 and H(X : Y ) = H(X) − H(X|Y ) =
0.61. The noise in the channel (photon loss) significantly reduces the maximally achiev-
able mutual information from 1 bit to 0.61 bits per use of the channel. Note: phase noise
leading to |1i → − |1i does not influence this scheme (it only leads to timing errors and
may slightly reduce the maximum possible number of uses of the channel per second).
(ii) Alice creates an entangled state |Ψ− i and sends one qubit to Bob. The resulting
quantum state is
9 1
ρ= Ψ− Ψ− + |00i h00| .
10 10
Since the states |Ψ− i and |00i are orthogonal this density operator has eigenvalues 9/10,
1/10, 0 and 0 giving an entropy of S(ρ) = 0.469. The reduced density operator on Alice’s
side is ρA = (11 |0i h0|+9 |1i h1|)/20 with entropy S(ρA ) = 0.9928. By symmetry S(ρB ) =
0.9928. We thus find for the conditional entropy S(ρA |ρB ) = −0.524 and a mutual
information of S(ρA : ρB ) = 1.517. If no noise were present the mutual information
would increase to 2 bits. Note: This scheme is not resistant against phase noise. It is
crucial to keep coherence in the transmission process in order to achieve the increased
mutual information compared to the classical case.
As we see from the last example the mutual information established via distributing an
entangled state can be larger than in the classical case and negative conditional entropies, which
are not possible in classical schemes, may arise. However, there is no (known) scheme on how to
solely use this kind of mutual information to transmit messages from Alice to Bob. Alice initially
prepares a pure state which does not contain information. Entanglement alone is thus not
sufficient for communication (and faster than light communication is not possible). However, by
combining entanglement and classical communication improvements over conventional classical
communication schemes can be achieved e.g. in quantum dense coding.

2.3 Quantum dense coding

The idea behind quantum dense coding is that Alice can distribute entanglement via the state
|Ψ− i before she has decided which message to send to Bob. The communication scheme thus

13
starts with the state |Ψ− i shared between Alice and Bob. Alice then sends one of four possible
messages 00, 01, 10, 11 by applying a local quantum operation on her qubit of the entangled
pair. In detail she applies 00: I, 01: σz , 10: σx and 11: σx σz . This operation turns the initial
Bell state into the orthogonal Bell states 00: |Ψ− i, 01: |Ψ+ i, 10: |Φ− i and 11: |Φ+ i. She then
sends her qubit to Bob via the quantum channel and Bob measures both bits in the Bell basis.
A Bell state analyzer for photon states will be discussed in detail in Sec. 3.1.5. Since the Bell
states are orthogonal they can be distinguished having only one copy of the state.
Note: At some point before the communication takes place Alice and Bob need to share the
entangled pair. This requires Alice sending a qubit to Bob or sending qubits to Alice and Bob
from a source of entangled qubits. The entangled state is independent of the message to be sent
in both cases and the distribution can thus be done before Alice decides on which message to
send. The initial entangled pair thus acts as a resource for communication between Alice and
Bob.

3 Photon techniques
Here we discuss methods for realizing quantum communication and computation with photons.
The experimental setups discussed in this section are shown in the appendix.

3.1 Polarization and spatial mode encoding

The wave function of the photon (see optics part of this lecture course) has spatial and po-
larization degrees of freedom. Both can be used to encode qubits. In spatial mode encoding
orthogonal direction/momentum modes a and b are chosen to represent the qubit states |0i and
|1i. In polarization encoding the qubit is encoded in the photon polarization e.g. |0i = |Hi and
|1i = |V i. In both cases single qubit gates can be implemented by linear optical elements and
two qubit gates via nonlinear media. We will next discuss these conventional ‘network’ quantum
computing techniques. In addition proposals for quantum computing using purely linear optics
exist. There entanglement is created by postselection of photons after a measurement is carried
out. More recently the creation of highly entangled graph states of photons has attracted a lot of
interest. These entanglement of these states serves as resource for carrying quantum information
processing. These methods will not be discussed here in more detail.

3.1.1 Spatial encoding

If two spatial paths impinge on a beam splitter (BS) a single qubit gate changing the amplitudes
in the two paths is realized. For instance a simple 50/50 BS maps an input state√|Ψiin =
α|0iin + β|1iin into the output state |Ψiout = H|Ψiin = [(α + β)|0iout + (α − β)|1iout ]/ 2 which
realizes a Hadamard gate. For general beam splitters the transformation is

eiφ sin(ξ)
 
cos(ξ)
BS(ξ, φ) =
e−iφ sin(ξ) − cos(ξ)

where cos2 (ξ) and sin2 (ξ) are the reflectivity and transmitivity of the beam splitter, respectively.
The simple 50/50 BS corresponds to BS(π/4, 0) = H. The value of φ is the phase shift experienced
in a transmission through the BS similar to the phase gate discussed next.
A single qubit phase gate Φ is implemented by putting a slab of transparent medium with
refractive index n and length L into the path of one spatial mode. This causes a phase shift
φ = (n − n0 )Lω/c0 with respect to the second arm, which we assume to be in air. Here n0 and

14
c0 are refractive index and speed of light in air, respectively. This maps the qubit wave function
according to the truth table |0iout → |0iin and |1iout → eiφ |1iin .
These two linear optical elements allow the realization of all necessary single qubit operations.
Kerr nonlinearities χ are necessary to create a two qubit phase gate where a phase shift is induced
if two photons are traveling a distance L in the Kerr medium (see nonlinear optics part of this
course). Since this phase is conditional on both photons being present it is an entanglement
phase given by ϕ = χL and can be used to realize a controlled two qubit phase gate

|00iin → |00iout
|01iin → eiφ |01iout
|10iin → eiφ |10iout
|11iin → ei(ϕ+2φ) |11iout ,

where the phase φ is caused by the linear refractive index of the Kerr medium. Together these
gates form a universal set of gates for quantum computing.

Example:

E11. We can now interpret the familiar setup of a Mach-Zehnder interferometer in terms of
single qubit gates. The setup consists of a 50/50 BS followed by a phase shifter Φ in
one arm and then a second 50/50 BS. This maps the input state |Ψiin according to
|Ψiout = HΦH|Ψiin which in matrix form is given by
    
1 1 1 1 0 1 1 α
|Ψiout =
2 1 −1 0 eiφ 1 −1 β

and gives  
cos(φ/2)α − i sin(φ/2)β
|Ψiout = eiφ/2 .
−i sin(φ/2)α + cos(φ/2)β
Note that the phase shifter can be written as Φ = eiφ/2 (cos(φ/2)I − i sin(φ/2)σz ) and thus
HΦH= eiφ/2 (cos(φ/2)I − i sin(φ/2)σx ), using Hσz H= σx .

3.1.2 Polarization encoding

Single qubit gates are implemented by polarization rotators and polarization phase shifters
acting on the polarization degrees of freedom instead of the spatial degrees of freedom discussed
in the previous section. For realizing two qubit gates polarizing beam splitter (PBS) can be used
to first spatially separate |Hi and |V i components. The spatially separated paths are then used
to realize a phase gate as before. After a PBS the qubit state can be measured by two spatially
separated photo detectors.

Example:

E12. A quarter-wave plate with its fast axis at an angle of π/4 to the vertical and horizontal
directions of polarization is placed into the path of a polarization encoded photon. Which
gate does this wave plate implement? What happens for a half-wave plate placed at an
angle of φ to the vertical axis?
√ √
The fast axes is (|V i + |Hi)/ 2 and the slow axis is (|V i − |Hi)/ 2. The fast axis
will pick up a phase of e−iπ/2 relative to the slow axis. An initially vertically polarized

15
 √
photon will end up in state |V i → √ (|V i − i |Hi)/ 2 while a horizontally polarized photon
is turned into |Hi → (|V i+i |Hi)/ 2 (up to global phases). For a half-wave plate a phase
of eiπ = −1 is picked up. Up to global phases this leads to a rotation of the polarization
angle by 2φ, i.e. |V i → cos(2φ) |V i + sin(2φ) |Hi and |Hi → − cos(2φ) |Hi + sin(2φ) |V i.
This is a σx or NOT gate for φ = π/4 and a Hadamard gate for φ = π/8.

3.1.3 Parametric down conversion

In nonlinear optical media a single photon can be converted into a pair of photons as discussed
in the nonlinear optics part of this lecture course. Energy, momentum and angular momentum
conservation rules obeyed in this process determine the correlations between the two created
photons. Experimental setups can be designed to create the following types of entangled states.

Time entanglement This only relies on the fact that the two photons in a pair are created
simultaneously and satisfy energy conservation laws. The time entangled states can be measured
in a Franson type interferometer with two short |Si1,2 and two long arms |Li1,2 . The created
state is given by
1h i
|ψi = |Si1 |Si2 + ei(φ1 +φ2 ) |Li1 |Li2 + eiφ2 |Si1 |Li2 + eiφ1 |Li1 |Si2 .
2
Here φ1,2 is a phase introduced in the long arms of the interferometer. The path difference
between the L and the S arms is much longer than the coherence length of the photons and
thus no interference fringes are observed inside the two interferometers. While the overall state
is a product state appropriate time gating can be used to detect only the first two terms of
the state and discard the SL and LS terms. The LL and SS terms are truly coincident and
indistinguishable because the time of creation of the photon pair is not known. These two parts
of the wavefunction will thus interfere and show fringes as a function of the phase φ1 + φ2 . This
interference indicates time entanglement and can be observed by measuring coincidences in the
outputs of the two interferometers.

Momentum entanglement This can be created in non-collinear down-conversion and ful-

filling the phase matching conditions. The state created in this process is given by
1 h i
|Ψi = √ eiφb |ai1 |bi2 + eiφa |ai2 |bi1 ,
2
where |ai1,2 and |bi1,2 are different photon paths. Combining these paths with BSs interference
fringes are detected as a function of φa,b indicating entanglement.

Polarization entanglement Non-collinear type-II down-conversion phase matching can be

used to achieve photons entangled in polarization. Photons at certain angles with the optical
axis such that they are emitted along cones with no common axis are used. One cone is ordi-
narily, the other extraordinarily polarized. They intersect along two directions where the light
is unpolarized. At this intersection the state of two photons is
1 h i
|Ψi = √ |V i1 |Hi2 + eiφ |Hi1 |V i2 .
2
The phase φ can be controlled by a compensator crystal.

16
 PBS D3

D4
|Ψi23 BS

PBS

D2 D1

Figure 5: A partial Bell state analyzer identifying the polarization encoded Bell states |Ψ+ i and
|Ψ− i.

3.1.4 Single photon sources and single photon detectors

The simplest way to achieve a single photon source is to use laser light which is a superposition
|αi (a coherent state) of different photon number states |ni given by
∞
2 /2
X αn
|αi = e−|α| √ |ni .
n=0
n!

The first few terms in this sum are

α2 α3
|αi ∝ |0i + α|1i + √ |2i + √ |3i .
2 6

By attenuating this beam the laser intensity and thus the parameter |α|2 goes down. Then the
dominant contributions to the state of the laser are vacuum |0i and the single photon
√ state |1i
while all higher order terms decrease with higher powers of α. For instance for α = 0.1 if light
ever makes it through the attenuator it is a single photon with 95% probability. The drawback
of this simple method is that the source does not indicate whether a photon is present or not.
Instead, in parametric down conversion measuring one photon indicates the presence of the other.
More sophisticated schemes are currently developed which allow the on demand generation of
a photon with well defined polarization, wave length and direction. Such single photon sources
are important for improved implementations of quantum communication schemes.
Single photon detectors required in quantum communication should posses a high quantum
efficiency, detect photons over a broad frequency range (100nm to 2000nm), have low dark count
rates (NO false counts, NO afterpulsing). They should recover quickly after detecting a photon
and have fast rise/pulse pair resolution. Single photon detectors are further separated into
photon number resolving and photon counting devices.

3.1.5 Quantum dense coding: Experimental setup

Parametric down conversion is used to create a pair of polarization entangled photons realizing
two qubits in a Bell state. One qubit is sent to the receiver and the other is manipulated by

17
the sender before being submitted to the receiver: A λ/4 and λ/2 plate are used to realize σx
and σz operations on this qubit. When both qubits have arrived at the receiver side a Bell state
measurement needs to be carried out.
A setup for partially achieving a Bell state measurement is shown in Fig. 5. Two polar-
ization entangled photons are incident onto the BS. The overall wave function of the two pho-
tons (polarization + spatial wave function) has to be symmetric since the photons are bosons.
Therefore, if the polarization part of the state is (anti)symmetric the spatial part also has to
be (anti)symmetric. For Bell state |Ψ− i with antisymmetric polarization part one photon has
to follow the upper arm and the other photon the lower arm after the BS. Thus a coincidence
between D3 and D2 or D4 and D1 is registered. For |Ψ+ i both photons follow the same arm
and thus a coincidence between D1 and D2 or between D3 and D4 is registered. In the other
two cases |Φ+ i , |Φ− i two photons are detected in the same detector and they cannot be dis-
tinguished. This analyzer therefore identifies two of the four Bell states and distinguishes them
from the other two Bell states. It does not allow to identify all four of them. Once can show
that using only linear optics it is not possible to distinguish all four Bell states.

Example:

E13. Two identical photons impinge on a 50/50 BS one arriving at the upper arm |ui and the
other at the lower arm |li. The photons have a symmetric polarization wave function.
Which path will the photons take after the BS?
√
The initial symmetric wave function is (|uli + |lui)/ 2. Each photon undergoes a
Hadamard gate. This turns the wave function into

(|ui + |li)(|ui − |li) + (|ui − |li)(|ui + |li) |uui − |lli

√ = √ .
8 2
Therefore both photons follow the upper or lower arm after the BS. Because of interference
the |uli and |lui terms cancel and thus the two photons will never follow different paths.

4 Testing EPR
We discuss violations of local realistic assumptions by quantum mechanics as first predicted
by Einstein Podolsky and Rosen (EPR). We consider the two most prominent examples of
violating Bell inequalities and measurements on Greenberger-Horne-Zeilinger (GHZ) states. The
experimental setups discussed in this section are shown in the appendix.

4.1 Bell inequalities

Violations of Bell inequalities can be used to demonstrate non-classical properties of entangled
states and test quantum mechanics. Systems which follow classical common sense are described
by a local and realistic theory and can be shown to obey inequalities for the maximum strengths
of correlations between their constituents. We will show that these inequalities are violated by
quantum mechanics and discuss experiments demonstrating such violations.

4.1.1 The CHSH inequality

We derive a Bell-type inequality (the so called CHSH inequality) by analyzing the Gendanken-
experiment shown in Fig. 6 using common sense (not quantum theory). We assume

18
 Q=±1 S=±1
A C B
R=±1 T=±1
Alice Charlie Bob

Figure 6: Schematic experimental setup to detect violations of the CHSH inequality

• Charlie prepares two systems (possibly correlated) and sends one to Alice and the other
one to Bob.

• After receiving their respective particles Alice and Bob both randomly choose to measure
one of two properties of their particle. Then they simultaneously perform their measure-
ment.

• They repeat this experiment many times and record their outcomes

• Alice and Bob meet and investigate the correlations between their experimental results.

What can they expect to obtain? For simplicity we assume that each measurement can only yield
a value of ±1. We describe the possible measurements of Alice by random variables Q and R and
those of Bob by random variables S and T . By common sense we assume that the measurement
values of Q, R, S, and T exist independent of observation. This is the assumption of realism.
Furthermore, Alice’s measurement does not influence the outcome of Bob’s measurement. They
are performed in a causally disconnected manner, so it is reasonable to assume this. This is the
assumption of locality. We investigate the expression

QS + RS + RT − QT = (Q + R)S + (R − Q)T = ±2 ,

since either Q + R or Q − R is zero. We assume that the probability for measurement values
Q = q, R = r, S = s, T = t before the measurement is p(q, r, s, t) and using this probability
distribution we find the expectation value
X X
E(QS + RS + RT − QT ) = p(q, r, s, t)(QS + RS + RT − QT ) ≤ 2 p(q, r, s, t) = 2 .
q,r,s,t q,r,s,t

This yields the CHSH inequality obeyed if the assumptions of local realism hold for the mea-
surements carried out by Alice and Bob

E(QS) + E(RS) + E(RT ) − E(QT ) ≤ 2 .

We now analyze the same experiment using quantum mechanics for the case where Charlie
generates an entangled Bell state |Ψ− i. He sends one qubit to Alice and the other to Bob. Alice
chooses between measuring the operators
√ Q = σz and R = √ σx on her qubit. Bob measures one
of the operators S = −(σz + σx√ )/ 2 and T = (σ √z − σ x )/ 2. We√find the quantum mechanical
√
expectation values hQSi = 1/ 2, hRSi = 1/ 2, hRT i = 1/ 2, and hQT i = −1/ 2 and
therefore √
E(QS) + E(RS) + E(RT ) − E(QT ) = 2 2 > 2 .
The experiment violates the CHSH inequality and the assumptions of local realism. Entangle-
ment between Alice’s and Bob’s states yields correlations stronger than allowed by local realism.

19
 α β
D1 D3

PBS EPR source PBS

D2 D4

Figure 7: Schematic experimental setup of the Aspect experiments

4.1.2 The Aspect experiments

The first experiments demonstrating the violation of Bell inequalities were carried out by A. As-
pect and co-workers11 . A schematic setup of the experiment is shown in Fig. 7. The mea-
surements were carried out on polarization entangled photons in state |Φ+ i. One photon was
detected after a polarizing beam splitter (PBS) at angle α or β, the other after a PBS either at
angle β or γ. Assuming local realism one can show that the number of coincidences N to obtain
one photon at output port 0 or 1 (i.e. at detector D1 or D2 on Alice’s side and at detector D3
or D4 on Bob’s side) obeys the inequality

N (1α , 1β ) ≤ N (1α , 1γ ) + N (1β , 0γ ) .

Here N (1α , 1β ) is the probability for Alice to obtain a click in D1 when setting her PBS at angle
α and Bob to obtain a click in D3 when setting his PBS at angle β. This inequality is violated if
the two photons are prepared in the entangled state |Φ+ i when setting α − β = β − γ = 30◦ since
N (1α , 1β ) ∝ cos2 (α − β). By correlating different measurement results the Aspect experiments
managed to violate the above Bell inequality.

4.1.3 Loopholes
The results of the Aspect experiments (and several experiments which followed them) can be
viewed as evidence for the violation of local realism but this is not the only explanation. Var-
ious experiments had several loopholes: a) fair sampling assumption which presumes that the
measured values are a fair reflection of all possible outcomes; b) rather small efficiency of photo
detectors; c) accidental coincidences were removed in the experiment; d) polarizers were set up
(not randomly) before the photons were created; e) strict Einstein locality of the measurements
was not obeyed; f) the quantum system was not truly a bipartite system since it e.g. consisted
of an atom and two photons. Addressing these loopholes requires a)b) 100% detection efficiency
(e.g. achieved in ion trap experiments but only at 3µm distance); c)keeping the accidental co-
incidences in the data; d)e) adjusting the polarizers randomly after the photons are created.
A random quantum process can be used to set up the measurement. Measurements can be
performed in strict Einstein locality and in different moving frames; f) measuring additional
particles in the system and showing that they are not correlated with the two photons.

4.2 GHZ states

Bell inequalities are formulated in terms of expectation values. In principle the measurement of
these expectation values requires infinitely many runs of the experiment. Local realism is only
11
A. Aspect et al.,Phys. Rev. Lett. 47, 460 (1981); A. Aspect et al., Phys. Rev. Lett. 49, 91 (1982).

20
violated on average and not in any single run of an experiment. In contrast quantum mechanics
predicts the violation of local realism with certainty for some entangled states of three particles,
e.g. GHZ states. In these experiments measurement outcomes which are not allowed according
to local realism will be found with certainty using quantum theory. Since we can make definite
predictions rather than statistical ones no inequalities are needed in this setup.

4.2.1 Violations of local realism

A GHZ state is a three qubit entangled state
1 1
|GHZi = √ (|000i + |111i) = √ (|V V V i + |HHHi) .
2 2
We analyze the state using notation commonly employed for polarization encoded qubits |0i ≡
|V i and |1i ≡ |Hi which are eigenstates of σz . The polarizations rotated through 45◦ with
respect to |Hi and |V i are denoted by |H 0 i and |V 0 i and are eigenstates of σx . Left handed |Li
and right handed |Ri circular polarizations are eigenstates of σy . Rewriting the state |GHZi in
the YYX basis we find
1
RLH 0 + LRH 0 + LLV 0 + RRV 0


|GHZi = .
2
Thus if measuring in the YYX basis we know with certainty the outcome of the third measure-
ment after determining the state of the first two qubits. By cyclic permutation we find analogous
expressions for measuring any two photons in circular polarization and the remaining one in 45◦
basis.

Local realistic analysis From a local realistic point of view these perfect correlations can
only be explained by assuming that each photon carries elements of reality which determine
the outcome for all measurements considered. Let us investigate a measurement in the XXX
basis. Which outcomes are possible if these elements of reality exist? The permutations of
|GHZi imply that if H 0 (V 0 ) is obtained for one photon the other two have to have opposite
(identical) circular polarizations. Imagine we find V 0 and V 0 for photons 2 and 3. Since 3 is V 0 ,
1 and 2 have to have identical circular polarization. Also, since 2 is V 0 , 1 and 3 have to have
identical circular polarization. All of these polarizations are elements of reality so all photons
have identical circular polarization. Thus photon 1 needs to carry polarization V 0 . We conclude
that |V 0 V 0 V 0 i is a possible outcome. Using similar arguments one can verify that the only four
possible outcomes are

V 0V 0V 0 H 0H 0V 0 H 0V 0H 0 V 0H 0H 0 .

Quantum theoretical analysis In the XXX basis the state |GHZi reads
1
H 0H 0H 0 + H 0V 0V 0 + V 0H 0V 0 + V 0V 0H 0


|GHZi = .
2
Local realism and quantum theory predict opposite results in all cases!

21
 Alice classical channel |Ψi3

BSM U

|Ψ−i23 Bob

|Ψi1

EPR source

Figure 8: Quantum teleportation setup.

4.2.2 Experimental realization

Polarization entangled pairs of photons are created in the BBO crystal such that
1  
|Ψi = √ |HV i + eiφ |V Hi .
2
In the rare event that two pairs are created with one UV pulse the fourfold coincidence corre-
sponds to the observation of the state
1
|GHZi = √ (|HHV i + |V V Hi) ,
2
at three detectors D1, D2, D3 and the detection of the fourth photon at a detector T acts as
the trigger. Note that in this experiment the coherence of the photons needs to be substantially
longer than the length of the UV pulse so that the two pairs created in the downconversion
processes are not distinguishable. See the classes for a detailed analysis of the experimental setup.
Experimental results obtained from this setup agree with the quantum mechanical predictions
and violate the local realistic predictions.

5 Quantum communication
We have already discussed quantum dense coding and seen that entanglement can serve as a re-
source for communication between Alice and Bob when combined with classical communication.
In quantum dense coding two bits of information are transmitted via a shared pair of qubits and
transmission of one qubit between sender and receiver. We will now consider more sophisticated
quantum communication schemes. The experimental setups discussed in this section are shown
in the appendix.

5.1 Quantum teleportation

The aim of quantum teleportation is to send an unknown quantum state of qubit 1 from Alice
to Bob using classical communication and an entangled pair of qubits as shown in Fig. 8. The
sender Alice receives qubit 1 in an unknown state |Ψi1 = α |0i + β |1i and qubit 2 which is

22
part of an entangled Bell state |Ψ− i23 with qubit 3. The state of the three qubits is |Ψi123 =
[α(|001i − |010i) + β(|101i − |110i)]. Alice performs a Bell state measurement (BSM) on her
two qubits and tells Bob the result. The measurement projects qubits 1 and 2 onto one of the
four Bell states. The quantum state is projected according to the measurement outcome:

Outcome: Quantum state:

|Ψ− i12 |Ψ− i12 (α |0i + β |1i)3
|Ψ+ i12 |Ψ+ i12 (α |0i − β |1i)3 .
|Φ− i12 |Φ− i12 (α |1i + β |0i)3
|Φ+ i12 |Φ+ i12 (α |1i − β |0i)3

Note that qubit 3 is never entangled with qubits 1 and 2 after the measurement.
Bob receives qubit 3 and the measurement result from Alice. He applies a unitary U to
particle 3 conditional on the measurement result as follows.

Outcome: Quantum operation:

|Ψ− i12 U =I
|Ψ+ i12 U = σz .
|Φ− i12 U = σx
|Φ+ i12 U = σx σz

For any measurement outcome the operation U turns the state of qubit 3 into the original state
of qubit 1 yielding |Ψi3 = α |0i + β |1i up to a irrelevant global phase.
We remark that the quantum state of all three qubits before Alice’s measurement can be
written as
1
|Ψi123 = Ψ− 12 (α |0i + β |1i)3 + Ψ+ 12
(α |0i − β |1i)3 + Φ− 12
(α |1i + β |0i)3
2
+ Φ+ 12 (α |1i − β |0i)3 .

Thus each of the four Bell states will be found with probability 1/4 in Alice’s measurement. If
the outcome is not revealed the measurement turns the state into a mixed state given by

ρ123 = Ψ− Ψ− (α |0i + β |1i)(α∗ h0| + β ∗ h1|) + Ψ+ Ψ+ (α |0i − β |1i)(α∗ h0| − β ∗ h1|)+

1
Φ− Φ− (α |1i + β |0i)(α∗ h1| + β ∗ h0|) + Φ+ Φ+ (α |1i − β |0i)(α∗ h1| − β ∗ h0|) .
4
At this stage the reduced density operator of Bob’s particle is given by
1
ρ3 = (|0i h0| + |1i h1|) .
2
This is the maximally mixed state and has no correlations with the initial state of particle
1. Once the measurement outcome has been communicated from Alice to Bob he applies the
conditional unitary operation U . We analyze its effect by studying what happens to the different
parts of ρ123 and find
1
ρ0123 = Ψ− Ψ− + Ψ+ Ψ+ + Φ− Φ− + Φ + Φ+ (α |0i + β |1i)(α∗ h0| + β ∗ h1|)


4
I
= (α |0i + β |1i)(α∗ h0| + β ∗ h1|) .
4
Alice now possesses a maximally mixed state which is not correlated with Bob’s particle. Bob is
in possession of the state to be teleported. The teleportation of 1 qubit starts with an entangled

23
 |Ψ−
14
classical channel

4 1 BSM 2 3 U

|Ψ−
14 |Ψ−
23 Bob

Alice

EPR source EPR source

Figure 9: Schematic entanglement swapping setup.

state of negative conditional entropy S(ρA |ρB ) = −1 which is independent of the qubit state
to be sent. The transmission of 2 bits of classical information from Alice to Bob via a classical
channel is required to complete the protocol. Classical communication and entanglement are
both essential resources for teleportation but none of them is sufficient on its own to teleport
|Ψi1 from Alice to Bob.

5.2 Entanglement swapping

The state of qubit 1 is teleported with all its quantum properties using the setup described in the
previous section. In particular any entanglement of qubit 1 with another system is preserved.
We can see this by assuming α and β being state vectors instead of just c-numbers. This fact
can be used to extend the teleportation setup to teleport entanglement. In this case the state
of qubit 1 is not well defined (e.g. qubit 1 is in a Bell state with another qubit 4). The initial
state of the system can thus be written as |Ψi = (|αi4 |0i1 + |βi4 |1i1 ) |Ψ− i23 . The experimental
setup for entanglement swapping is shown in Fig. 9. The analysis of this setup using the same
method as in the teleportation protocol is left as an exercise. Also, the same entanglement and
information resources as for teleportation are necessary.

6 Quantum cryptography
Cryptographic protocols can be classified by the type of security against eavesdropping. There
exist mathematically secure schemes (like public key RSA encryption) whose security relies
on assumptions12 about the mathematical complexity of decrypting the cipher text without
possessing the correct key. The majority of nowadays secure public internet connections relies
on such schemes. Alternatively a cryptographic setup may provide a physically secure method
for communicating. In such setups the security is provided by the physical laws13 governing
the communication protocol. Here we first discuss a provably secure classical communication
protocol and then quantum methods for distributing the necessary keys. The experimental
setups discussed in this section are shown in the appendix.
12
These assumptions are sometimes unproven.
13
Physical laws are not provably correct.

24
6.1 One time pads and the Vernam cipher
The Vernam cipher is a cryptographic protocol which allows the encryption and decryption
protocol to be publicly known. The security of the protocol relies entirely on the key which is
private and not publicly known. Alice and Bob share identical n-bit secret key strings (the one
time pad). Alice encodes her message by adding message and key using a classical XOR gate
on each pair of bits. Bob decodes by subtracting the key again, i.e. by applying another XOR
operation with his key bit. As long as the key is of the same length as the message and can be
securely distributed to Alice and Bob the Vernam cipher is provably secure and Eve’s mutual
information with the sent message can be made arbitrarily small. This means that one needs a
secure method for distributing a large number of key bits. Key bits must be delivered in advance
of the message. Otherwise one could deliver the message itself by secure means. Furthermore
the key bits must be guarded until they are used and the key must be destroyed after the bits
were used. These difficulties in key distribution make the Vernam cipher impractical for general
use. However, it is used e.g. in military applications.
We note as an aside that the problem of key distribution is circumvented in public key
cryptography. The public key can easily be used to encrypt a message (like a box can be
locked using a padlock without possessing the key). To decrypt the message a private key
(corresponding to the key for the padlock) needs to be used. In public key cryptography Alice
sends out public keys to everyone and whoever wants to securely communicate with Alice may
use her key. The security of this protocol relies on the assumption that decrypting the message
without possessing the private key is difficult.

6.2 The BB84 protocol

The BB84 protocol (see also the short option Quantum ideas for a qualitative description) is
a physically secure way to distribute a secret key. It also allows to detect the presence of an
eavesdropper Eve. Alice begins with two strings A and B each consisting of (4 + δ)n bits. She
encodes these strings as a block of (4 + δ)n qubits
(4+δ)n
O
|ψi = |ψak ,bk i ,
k=1

where ak is the k th bit of A and bk is the k th bit of B. Each qubit is in one of the four states

|ψ00 i = |0i
|ψ10 i = |1i
|ψ01 i = |+i
|ψ11 i = |−i

The bits in A are encoded in the basis X or Z as determined by B. These four states are not
mutually orthogonal and cannot be distinguished with certainty. Bob receives E(|ψi hψ|), where
E describes the action of the channel and an eventual eavesdropper. He publicly announces
the fact that he has received the state. At this point Alice, Bob and a possibly present Eve
have their own states each with separate density matrices. Note that Alice has not revealed
B thus Eve has no knowledge on which basis she should have used when trying to eavesdrop
the communication by measuring qubits. At best she can guess and if her guess is wrong she
will disturb the states received by Bob. Note that noise in the channel also contributes to E.
Bob now measures each qubit in basis X or Z depending on a random (4 + δ)n bit string B 0

25
which he creates on his own. We call Bob’s measurement results A0 . After this Alice announces
B over a public channel and Bob and Alice discard all bits in {A, A0 } except for those where
the bits in B and B 0 are equal. We assume that δ is sufficiently big so that they can keep 2n
bits. It is important that Alice does not publish B before Bob has received the message to
ensure security of the scheme! To check for noise and eavesdropping Alice now selects n bits
and publicly announces the selection. Alice and Bob publicly compare these n bits and if more
than t bits disagree they abort and retry the protocol. t is selected so that they can apply
information reconciliation and privacy amplification (see Sec. 6.5.3) to obtain m < n acceptably
secret shared key bits. This protocol can be generalized to other states and bases. For instance
the B92 protocol only uses two non-orthogonal states |0i and |+i for the communication.
For the BB84 protocol qubits need to be sent via a quantum channel and also classical
bits are transmitted from Alice to Bob. However, it does not require any entanglement. The
protocol relies on the fact that non-orthogonal quantum states cannot be perfectly distinguished
by an eavesdropper whose actions will necessarily affect some of the states received by Bob. This
reduces the mutual information between Alice and Bob for those cases where they have measured
in the same basis. By detecting this reduction in mutual information they can identify Eve as
we will now investigate for a simple eavesdropping strategy.

6.2.1 Intercept - resend strategy

The security of the BB84 protocol relies on the impossibility for any eavesdropper to distinguish
between Alice’s states without disrupting the correlations between the bits in A and A0 . We
investigate one particular eavesdropping strategy where Eve intercepts the sent qubits, measures
them and the resends them. This is called the intercept/resend strategy and we will see how
Alice and Bob can detect Eve in this case and abort their communication.
Let us assume that Eve intercepts each qubit. She chooses the X or Z basis at random to
measure the qubit. Then she prepares a qubit in the state she had measured and sends it to
Bob. With 50% probability Eve will choose the wrong basis. Each time Eve’s basis is wrong
she will get a result which is completely uncorrelated with the bit that Alice has sent. If the
channel is otherwise perfect this leads to the following outcomes and probabilities provided that
Alice sent message 0 in basis X

Alice Eve probability Bob probability

0X 0X 1/2 0X 1/4
0Z 1/8
1Z 1/8
1Z 1/4 1Z 1/8
0X 1/16
1X 1/16
0Z 1/4 0Z 1/8
0X 1/16
1X 1/16

The cases where Alice sends 1X, 0Z, or 1Z can be worked out in the same way. Eve guesses the
correct value of the bit with 75% probability. If Alice and Bob measure in the same basis then
their results will disagree with a probability of 1/4. For a perfect noiseless channel the mutual
information between Alice’s and Bob’s messages obtained when measuring in the same basis has
thus been reduced from H(X : Y ) = 1 to H(X : Y ) = 0.456 by Eve. The probability for Alice

26
and Bob to find disagreement and thus identifying Eve when comparing n of their key bits is
given by  n
3
Pd = 1 − .
4
Thus the number of bits n that need to be compared for detecting an eavesdropper with a
probability Pd is
log2 (1 − Pd )
n= .
log2 (3/4)
By sacrificing n bits from their key Alice and Bob detect Eve with probability Pd .

6.3 Quantum key distribution using EPR pairs

A quantum channel emits pairs of photons in the singlet state of polarizations
1
|Ψi = √ [|V i1 |Hi2 − |Hi1 |V i2 ] .
2
The two photons fly along the z-axis to Alice and Bob, respectively. They perform measurements
and register the outcome in one of three bases, obtained by rotating around the z-axis by angles
φa1 = 0, φa2 = π/4, φa3 = π/8 for Alice and by φb1 = 0, φb2 = −π/8, φb3 = π/8 for Bob. These angles
are chosen independently and randomly for each pair. The outcomes can be ±1 depending on
which polarization is measured. The correlation coefficient of the measurements is given by
E(φai , φbj ) = P++ (φai , φbj ) + P−− (φai , φbj ) − P+− (φai , φbj ) − P−+ (φai , φbj ) ,
and quantum mechanics predicts E(φai , φbj ) = − cos[2(φai − φbj )], where the P’s are the probabil-
ities that ±1 is obtained in the respective bases. For the two pairs of bases 1 and 3 quantum
mechanics predicts perfect anti-correlations E(φa1 , φb1 ) = E(φa3 , φb3 ) = −1. Alice and Bob now
define S as
S = E(φa1 , φb3 ) + E(φa1 , φb2 ) + E(φa2 , φb3 ) − E(φa2 , φb2 ) ,
√
which should be (see CHSH inequality) S = −2 2. They discard measurements in which either
or both failed to register a qubit. Alice and Bob can now publicly announce the orientations
of the analyzers. Then they announce all results for which their orientations √ were different.
This allows them to establish the result for S which will only be S = −2 2 if the particles
were not disturbed. This ensures that the remaining measurements are perfectly anti-correlated
and use them to establish a secret key. Eve cannot elicit any information from the particles
while in transit from the source to the legitimate user since no information is encoded there.
The information ‘comes into being’ after the legitimate users perform the measurements and
communicate in public afterwards. In each case an eavesdropper will introduce elements of
physical reality to the particles and will lower S below its quantum limit. Thus the Bell theorem
can expose an eavesdropper.
In this scheme classical communication is only necessary to expose and eavesdropper but
not for establishing the secret key. If no eavesdropper could be present all measurements could
be carried out in the same basis. Before the measurement an entangled state with negative
conditional entropy S(ρA |ρB ) = −1 is created. The classical information gained from the mea-
surements is perfectly correlated with H(X : Y ) = 1. However, neither Alice nor Bob have the
ability to engineer their measurement outcome and choose which message to send. Quantum
mechanics does not tell us how to influence a measurement outcome and we do not have a more
advanced theory to do this. Thus, while correlated classical information comes into being dur-
ing the measurement process, it cannot be used to transmit messages between Alice and Bob.
However, the random bits generated in this scheme are very useful as a secret key.

27
 a)

Laser PM D0

PM D1

Alice Bob
b)

Laser PM D0

PM D1

Alice Bob

Figure 10: BB84 using phase encoding in optical fibres setups. a) Extended Mach-Zehnder
setup. b) Collapsed Mach-Zehnder setup. The circles denote delay loops of ∆.

6.4 Experimental setups

We can distinguish between free space cryptography where polarization encoded qubits are used
and fibre systems. Free space experiments require robust optical setups at the sender and
receiver side. Recent developments include the realization of small and highly efficient devices
which achieved distances of up to 23.4km. The predicted maximum distance is ≈ 1000km so
that connections to satellites are in principle possible. Here we consider in more detail two fibre
setups to realize the BB84 protocol.

6.4.1 Phase encoded fibre systems

Optical fibres do not conserve the polarization because of randomly fluctuating birefringence
(1 hour timescale). Polarization tracking is possible but would make a polarization scheme
cumbersome. Instead we consider an extended Mach-Zehnder setup used for phase encoding
and shown in Fig. 10a). Alice uses her phase modulator (PM) to encode 0, 1 in phases 0 and π
or in phases π/2 and 3π/2. Bob also chooses between 0 phase shift and π/2 phase shift for his
measurements. This scheme is equivalent to polarization encoding but replaces the polarization
with a relative phase in the wave function. The drawback is that keeping the phase constant over
large distances is very difficult due to temperature variations and other imperfections induced
by the environment.

Example:

E14. We analyze the setup shown in Fig. 10a). Alice produces the state |0i, which is turned
into |+i by the BS and the PM √ induces a relative phase φA so that the state which leaves
iφ
the sender is (|0i + e |1i)/ 2. In the ideal case the delay ∆ accumulated between Alice
A

and Bob acts equally on both arms and gives |0i → |0, ∆i, and |1i → |1, ∆i. Since both
arms are equally affected we leave ∆ out in the following. Bob’s PM √ introduces another
relative phase φB so that the state turns into (e iφB iφ
|0i + e |1i)/ 2. The BS turns this
A

state into [(eiφB + eiφA ) |0i + (eiφB − eiφA ) |1i]/2. If Alice chooses φA = 0 (φA = π) and

28
 7 D C 6

8 5
A, B, C and D are the O is the central
users of the network O telephone exchange

2 3

1 A B 4

Figure 11: A quantum telephone exchange.

Bob chooses φB = 0 he will get a click in D0 (D1 ) with certainty. If he chooses φB = π/2
a click in each detector is equally likely. If Alice chooses phases φA = π/2 (φA = 3π/2)
and Bob selects φB = π/2 D0 (D1 ) will click with certainty. Otherwise he will get equal
probability for a click in one of the detectors. Thus Alice encodes 0 and 1 in the two
bases by choosing phases 0 and π or π/2 and 3π/2 to use this setup for realizing the
BB84 protocol.

A more practical scheme is realized by collapsing the interferometer as shown in Fig. 10b).
Two pulses are propagating down the single fibre. They are denoted by S (short path, no delay
at the sender side) and L (long path, delay ∆ at the sender side). The delay ∆ is assumed to be
much longer than the duration of the photon wave packet. After traveling through Bob’s part
of the Mach-Zehnder they create three different outputs: SS (which only experiences the delay
between sender and receiver station) and LL (going through the delay lines at sender and receiver
side in addition to the delay of the connecting fibre) are not relevant as they show no interference
effects. SL and LS (going through exactly one of the delay lines at sender and receiver in addition
to the delay of the fibre) are indistinguishable and thus interfere. Experimentally they can be
selected by time gating. The choice of phase shifts by Alice and Bob gives the encoding-decoding
in the SL and LS components exactly as in the previous scheme. This setup is much more stable
since the pulses follow the same path for most of the setup. Any phase fluctuations which
happen on time scales much longer than the delay ∆ will only affect the global phase of the
wave function which is irrelevant. The major drawback of the scheme is that half of the signal
is lost in the SS and LL path. The analysis of this scheme is similar to the analysis presented
for the above setup and left as an exercise. Note that the delay loops ∆ at sender and receiver
are present in one arm only. Their action can thus not be ignored in the analysis.

6.5 More about communication schemes

The setups discussed in this section are included for completeness and will only be discussed in
the lectures if there is sufficient time left.

6.5.1 The quantum telephone exchange

Entanglement swapping can be used to realize a quantum telephone exchange. Imagine there
are N users in a communication network. Each user shares a Bell state with a central exchange
as shown in Fig. 11. Projecting the particles at the exchange O into an entangled multi-particle

29
 a)

A O B

b)

A C O D B
Figure 12: Speeding up entanglement distribution. a) Entanglement distribution without en-
tanglement swapping. b) Entanglement distribution using an entanglement swapping station.

state by doing a measurement will immediately project their partner particles at the users site
into the same type of entangled state.

Example:
E15. Imagine that all the Bell states
√ are of the type |Φ+ i. A, B and C would like to share
a |GHZi = (|000i + |111i)/ 2 for further quantum communication. If the exchange O
projects the particles 2, 3 and 5 into the GHZ state then we find
+ + +
√
235 hGHZ| ( Φ 12
Φ 34
Φ 56
) = (|000i146 + |111i146 )/ 2.

In this process the quantum telephone exchange provides the communicating parties with
an entangled state and becomes disentangled from all of them. The advantages of this scheme
are that pure Bell states between the users and O can be created by state purification protocols
i.e. using a large number of not maximally entangled state to distil fewer better entangled states.
The preparation of the shared Bell pairs between users and O is independent of the states to be
shared later between the users. The telephone exchange becomes disentangled from the users
and cannot eavesdrop in later communication. The drawback is that the operations to be carried
out by the telephone exchange are difficult to realize.

6.5.2 Speeding up distribution of entanglement

Entanglement swapping can save a significant amount of time in providing distant parties A
and B with entangled pairs of particles. For this several Bell state producing and Bell state
measuring substations are put into the route between them as shown in Fig. 12b). In Fig. 12a)
it takes a time ta = L/2v with v < c the speed of the entangled particles and L the distance
between A and B. Case Fig. 12b) uses entanglement swapping at O which is an entanglement
measuring station. At t = 0 C and D send off Bell pairs. The particles arrive at time t = L/4v.
The Bell measurement at O takes a time tm . Thus it takes a time tb = L/4v+tm to distribute the
Bell pair. Note that one has to add the time needed to classically communicate the measurement
result at O to A and B and therefore there is no advantage when the distributed qubits are
photons.

30
6.5.3 Privacy amplification
Privacy amplification is a purely classical technique to reduce the amount of mutual information
an eavesdropper can gain when distributing an encryption key. It also reduces the number of
message which Alice and Bob can send. The amplification process starts once Alice and Bob
share with high probability an identical reconciled key of length n and know the error rate  of
the transmission. They assume that all the errors are due to an eavesdropper Eve. Then they
deduce t, the number of bits by which the key has to be shortened for privacy. Alice picks a
random (n − t) × n binary matrix K and publicly transmits K to Bob. Using K Alice and Bob
obtain the final private key as
kfinal = K · kreconciled
.
While implementing privacy amplification is simple, finding t and proving security is very
difficult. In general one distinguishes between different types of eavesdropping attacks:

• Incoherent attacks: Eve entangles quantum probes with one photon at a time. She then
stores (quantum memory) and measures her probes after Alice and Bob have made their
public announcements.

• Collective attacks: Eve only entangles her probes with one photon but has a quantum
computer to further process her states after the public communication.

• Coherent attacks: Eve can entangle her probe with any dimension of the whole state in
the transmission. She has a quantum computer to process the resulting states at any time
she wishes to do so.

Answers have been provided for more and more powerful attacks but there are still open questions
about the security of quantum communication.

7 Further reading
A comprehensive introduction to quantum computing can be found in 14 . For further details on
quantum communication schemes and Bell inequalities discussed in this manuscript see 15 .

14
Michael A. Nielsen and Isaac L. Chuang, Quantum Computation and Quantum Information, Cambridge
University Press (2000)
15
D. Bouwmeester, A. Ekert, A. Zeilinger, The Physics of Quantum Information, Springer (Berlin) (2000); R.A.
Bertlmann and A. Zeilinger, Quantum [Un]speakables, Springer (Berlin) (2002).

31
 Appendix

Schematics and
Experimental setups

Thursday, 30 April 2009 1

 3. Photon techniques

Thursday, 30 April 2009 2

 3.1.1 Spatial encoding: Beam splitter
• A simple 50/50 BS for spatial mode encoded qubits

• Matrix representation of the dynamics of a general beam splitter

This time evolution is unitary. BS(45◦,0)=H is a simple 50/50 beam splitter.

Thursday, 30 April 2009 3

 3.1.1 Spatial encoding: phase gate
• A slab of transparent medium put into the path of one mode

A medium of length L with refractive

index n yields a phase shift φ

• The resulting quantum gate is a phase gate with the truth table

• Kerr nonlinearities χ allow to create a two qubit phase gate where a phase
shift is induced if two photons are travelling a distance L in the Kerr
medium. The resulting entanglement phase is

Qubit 1:

Qubit 2:

Thursday, 30 April 2009 4

 3.1.1 Mach-Zehnder interferometer

eiφ

• The Mach-Zehnder interferometer evolves the input state |Ψiin according to

H φ H

Thursday, 30 April 2009 5

 3.1.2 Polarization encoding
• Implement a two qubit gate e.g. a CNOT gate

• Measure a qubit

Photon H Polarizing beam splitter

source
V

Thursday, 30 April 2009 6

 3.1.3 Parametric down conversion: Time entanglement
• The emission time is uncertain within the coherence time of the pump laser.
• The photons are simultaneous since they are broadband with coherence
times of order 100fs.
• Two-photon Franson-interferometry

α β
1 2

D1a D2a
EPR source
D1b D2b

• Inside the interferometer

Thursday, 30 April 2009 7

 3.1.3 Parametric down conversion: Momentum
entanglement

• The pairs are emitted in either modes a1, b2 or in modes a2, b1.
• Before the beam splitters we thus have the entangled state

• Behind the BSs the two paths cannot be distinguished Æ interference

• Coincident detections in a and b detectors vary cosinusoidally on changing
the phase difference φ = φa - φb

Thursday, 30 April 2009 8

 3.1.3 Parametric down conversion: Polarization
entanglement

• Different light speed of ordinary and extraordinary beam Æ distinguishable

photons Æ compensation by inserting crystals of half the thickness at 90◦

• Phase φ can be controlled by compensator crystal, additional half wave

plate for creating the other two Bell states

Thursday, 30 April 2009 9

 3.1.4 Single photon source

Thursday, 30 April 2009 10

 3.1.4 Single photon source

coincidence detection
at DA and DB

Thursday, 30 April 2009 11

 3.1.4 Single photon detectors

Thursday, 30 April 2009 12

 3.1.5 Quantum dense coding: Experimental setup

Thursday, 30 April 2009 13

 4. Testing EPR

Thursday, 30 April 2009 14

 4.1.2 The Aspect experiments: Setup

• Experimental setup

• Atomic cascade

Thursday, 30 April 2009 15

 4.2.2 GHZ: Source for three-photon GHZ states
• Polarization entangled pairs of photons
are created in the BBO crystal such that

• In the rare event that two pairs are

created with one UV pulse the four
fold coincidence corresponds to
the observation of the state

at the detectors D1, D2, D3.

• The detection of a photon at detector
T acts as the trigger
• Note: The coherence of the photons needs
to be substantially longer than the length of
the UV pulse so that the two pairs are
not distinguishable

Thursday, 30 April 2009 16

 4.2.2 GHZ: Experimental proof of GHZ entanglement
• As a first step |GHZi entanglement has to be confirmed experimentally.
Four fold coincidences are detected for variable delays in path a

Graph (a) polarization analysis at D3 (two curves ± 45◦), conditioned on T,

and the detection of one photon at D1 polarized at 45◦ and one photon at
detector D2 polarized at -45◦. In (b) no such intensity difference is predicted
if the polarizer in front of detector D1 is set at 0◦

Thursday, 30 April 2009 17

 4.2.2 GHZ: Measurements in different bases
• Performing the measurements in the YYX (a), YXY (b), and XYY (c) basis
confirms the entanglement properties of the |GHZi state
• The experiment yields a visibility of 71%.
• Based on these results one can identify the terms which are supposed to be
absent and those which should be present.
• Thus one can compare the quantum mechanical and local realistic results
for measurements in the XXX basis.

Thursday, 30 April 2009 18

 4.2.2 GHZ: Local realism vs. quantum mechanics
• The measurements in the XXX basis yield the following results: (a) XXX
quantum mechanics; (b) XXX local realism; (c) XXX experimental results:

Thursday, 30 April 2009 19

 5. Quantum communication

Thursday, 30 April 2009 20

 5.1 Quantum teleportation: Setup

Thursday, 30 April 2009 21

 5.1 Quantum teleportation: Analysis
• A UV pulse creates the ancillary pair of photons 2 and 3. After retroflection it
can create a second pair of photons 1 and 4 where 1 is prepared in the
initial state to be teleported. Photon 4 serves as a trigger at detector p.
• Alice looks for coincidences after the BS where the photon to be teleported
and photon 2 are superposed. She identifies only the state |Ψ-i12 by finding
coincidence counts at f1 and f2.
• Bob then knows that his photon 3 is in the initial state of photon 1.
• Bob checks this state using polarization analysis using the PBS (a) d1f1f2
coincidence for -45◦ and (b) d2f1f2 coincidence for 45◦ for a 45◦ photon 1.

0.25 0.25
3fold coinc. prob.

a) b)
0 0
0 0
Thursday, 30 April 2009 delay delay 22
 5.1 Quantum teleportation: Experimental results
• Experimental results for a 45◦ and 90◦ photon state

Thursday, 30 April 2009 23

 6. Quantum cryptography

Thursday, 30 April 2009 24

 6.4 Free space cryptography: Polarization encoding
• Polarization is conserved in free space.
• Strongly attenuated laser pulses are created by sending laser pulses
through two pinholes with a diameter of 100µm separated by 9mm.
• Strong attenuation Æ only very few pulses result in detection events.
• A record of these detected has to be kept and communicated from Bob to
Alice.
• Distance of 23.4km achieved between Karwendelspitze (2244m) and
Zugspitze (2960m).
– Reduced air turbulence effects due to elevated beam path, but high
demands on temperature and weather condition stability of devices.
– Possible range: Up to 1000km (connections to satellites possible!)
• High voltage Pockels cells for rotating the polarization replaced by four
lasers creating different polarizations (indistinguishability of lasers?)
• Non-polarizing beam splitter for choosing the measurement at Bob’s.
• Galilean telescope to produce a near diffraction-limited 40mm beam.
• Bob collects the light in a 25cm aperture Schmidt-Cassegrainian telescope.
• A final net key rate of approx 500Bits/sec was achieved

Thursday, 30 April 2009 25

 6.4 Free space cryptograph: Alice – the sender

Thursday, 30 April 2009 26

 6.4 Free space cryptograph: Bob – the receiver

Thursday, 30 April 2009 27

 6.4 Free space cryptography – real experiment

Alice:

Bob:

Thursday, 30 April 2009 28