Introduction to

Quantum Information Processing

—
Draft

W. Lücke

SS 2005

E
Institute for Physics and Physical Technologies
Clausthal University Of Technology
Leibnizstraße 4
D-38678 Clausthal–Zellerfeld
 3

Preface
Quantum information processing is one of the most fascinating and active fields
of contemporary physics. Its central topic is the coherent control of quantum states
in order to perform tasks — like quantum teleportation, absolutely secure data trans-
mission and efficient factorization of large integers — that do not seem possible by
means of classical systems alone. The vast possibilities of physical implementations
are currently being extensively studied and evaluated. Various proof-of-principle
experiments have already been performed. However, in the present note only some
possiblities can be indicated. Main emphasis will be on quantum optical methods,
indispensable for transmission of quantum information.
For more complete information on achivements and latest proposals concerning
quantum information processing the Los Alamos preprint server

http://xxx.lanl.gov/archive/quant-ph

is highly recommended.

Recommended Literature: (Alber et al., 2001; Bowmeester et al., 2000; Ekert et al., 2000;
Nielsen and Chuang, 2001; Preskill, 01; Shannon, 1949; Bertlmann and Zeilinger, 2002;
Audretsch, 2002; Bruß, 2003)
4
Contents

I Idealized Quantum Gates and Algorithms 9

1 Basics of Quantum Computation 11
1.1 Classical Logic Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Quantum Computational Networks . . . . . . . . . . . . . . . . . . . 17
1.2.1 Quantum Gates . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.2 Quantum Teleportation . . . . . . . . . . . . . . . . . . . . . 24
1.2.3 Universality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2 Quantum Algorithms 33
2.1 Quantum Data Base Search . . . . . . . . . . . . . . . . . . . . . . . 33
2.1.1 Grover’s Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1.2 Network for Grover’s Algorithm . . . . . . . . . . . . . . . . . 34
2.1.3 Details and Generalization . . . . . . . . . . . . . . . . . . . . 35
2.2 Factoring Large Integers . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.2 The Quantum Fourier Transform . . . . . . . . . . . . . . . . 40
2.2.3 Quantum Order Finding . . . . . . . . . . . . . . . . . . . . . 44

3 Physical Realizations of Quantum Gates 51

3.1 Quantum Optical Implementation . . . . . . . . . . . . . . . . . . . . 51
3.1.1 Photons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.1.2 Photonic n-Qubit Systems . . . . . . . . . . . . . . . . . . . . 54
3.1.3 Nonlinear Optics Quantum Gates . . . . . . . . . . . . . . . . 58
3.1.4 Linear Optics Quantum Gates . . . . . . . . . . . . . . . . . . 60
3.2 Measurement-Based Quantum Computation . . . . . . . . . . . . . . 68
3.3 Cold Trapped Ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3.1 General Considerations . . . . . . . . . . . . . . . . . . . . . . 72
3.3.2 Linear Paul Trap . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.3 Implementing Quantum Gates by Laser Pulses . . . . . . . . . 79
3.3.4 Laser Cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

5
6 CONTENTS

II Fault Tolerant Quantum Information Processing 89

4 General Aspects of Quantum Information 91
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 Quantum Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.1 Open Quantum Systems and Quantum Operations . . . . . . 93
4.2.2 Quantum Noise and Error Correction . . . . . . . . . . . . . . 101
4.3 Error Correcting Codes . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.3.1 General Apects . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.3.2 Classical Codes . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3.3 Quantum Codes . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.3.4 Reliable Quantum Computation . . . . . . . . . . . . . . . . . 116
4.4 Entanglement Assisted Channels . . . . . . . . . . . . . . . . . . . . 120
4.4.1 Quantum Dense Coding . . . . . . . . . . . . . . . . . . . . . 120
4.4.2 Quantum Teleportation . . . . . . . . . . . . . . . . . . . . . 121
4.4.3 Entanglement Swapping . . . . . . . . . . . . . . . . . . . . . 123
4.4.4 Quantum Cryptography . . . . . . . . . . . . . . . . . . . . . 124

5 Quantifying Quantum Information 127

5.1 Shannon Theory for Pedestrians . . . . . . . . . . . . . . . . . . . . . 127
5.2 Adaption to Quantum Communication . . . . . . . . . . . . . . . . . 131
5.2.1 Von Neumann Entropy . . . . . . . . . . . . . . . . . . . . . . 131
5.2.2 Accessible Information . . . . . . . . . . . . . . . . . . . . . . 135
5.2.3 Distance Measures for Quantum States . . . . . . . . . . . . . 137
5.2.4 Schumacher Encoding . . . . . . . . . . . . . . . . . . . . . . 144
5.2.5 A la Nielsen/Chuang . . . . . . . . . . . . . . . . . . . . . . . 145
5.2.6 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

6 Handling Entanglement 147

6.1 Detecting Entanglement . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.1.1 Entanglement Witnesses . . . . . . . . . . . . . . . . . . . . . 147
6.1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.1.3 Other Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.2 Local Operations and Classical Communication . . . . . . . . . . . . 156
6.2.1 General Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.2.2 Entanglement Dilution . . . . . . . . . . . . . . . . . . . . . . 161
6.2.3 Entanglement Distillation . . . . . . . . . . . . . . . . . . . . 161
6.3 Quantification of Entanglement . . . . . . . . . . . . . . . . . . . . . 161

A 165
A.1 Turing’s Halting Problem . . . . . . . . . . . . . . . . . . . . . . . . 165
A.2 Some Remarks on Quantum Teleportation . . . . . . . . . . . . . . . 166
A.3 Quantum Phase Estimation and Order Finding . . . . . . . . . . . . 167
A.4 Finite-Dimensional Quantum Kinematics . . . . . . . . . . . . . . . . 171
A.4.1 General Description . . . . . . . . . . . . . . . . . . . . . . . . 171
CONTENTS 7

A.4.2 Qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

A.4.3 Bipartite Systems . . . . . . . . . . . . . . . . . . . . . . . . . 176

Bibliography 183

Index 199
8 CONTENTS
 Part I

Idealized Quantum Gates and

Algorithms

9
Chapter 1

Basics of Quantum Computation

1.1 Classical Logic Circuits

The smallest entity of classical information theory (Shannon, 1949) is the bit (binary
digit), i.e. the decision on a classical binary alternative. Usually bits are identified
with the numbers 0 (for wrong) or 1 (for true) and typically correspond to the
position of some simple switch. Every definite statement may be encoded into a
sufficiently long but finite sequence (b1 , . . . , bn ) of bits.1 In this sense the essence of
a calculations may be described as the transformations of a finite sequence of input
bits (encoding the task) into a finite sequence of output bits (encoding the result).
This suggests the following model for actual calculators:
1. An input register (array of switches) will be put into a state corresponding to
the n1 -tuple (b1 , . . . , bn1 ) ∈ {0, 1}n1 encoding the task.
2. A computational circuit, the elementary components of which are called gates,2
transforms (b1 , . . . , bn1 ) into an n2 -tuple (b′1 , . . . , b′n2 ) of bits encoding the result
to be stored into an output register.
From the mathematical point of view it is only important which element of Fn1 ,n2 ,
denoting the set of all mappings from {0, 1}n1 into {0, 1}n2 , is implemented by
the circuit. Therefore, computational circuits implementing the same mapping are
called equivalent.

Every element of Fn1 ,n2 can be implemented by some assembly of gates listed in
Table 1.1:

Lemma 1.1.1 For arbitrary positive integer n1 , n2 all elements of Fn1 ,n2 can be
represented as compositions of tensor products of functions from Tabular 1.1.
Proof: See below.
DRAFT, October 17, 2007
1
An important consequence of this fact is the halting problem (see Appendix A.1).
2
For simple hardware implementations see (Pütz, 1971, pp. 244–252).

11
12 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

Name Symbol Class Action

ID F1,1 b 7→ b
FANOUT r F1,2 b 7→ (b, b)
H
Hb
NOT 
 F1,1 b 7→ 1−b
&
AND F2,1 (b1 , b2 ) 7→ b1 b2
≥1
OR F2,1 (b1 , b2 ) 7→ b1 + b2 − b1 b2

Table 1.1: Elementary gates

Thus every classical logic circuit corresponds to a graph consisting of symbols from
Tabular 1.1. For instance, the graph

.......
..........
&
.. .
..........
.......
u ≥1
&

corresponds to the mapping

def
SWITCH = OR ◦ (AND ⊗ AND) ◦ (IDNOT ⊗ ⊗ID ⊗ ID) ◦ (ID ⊗ FANOUT ⊗ ID) ,
acting as 
b0 if s = 0 ,
(b0 , s, b1 ) 7−→
b1 if s = 1 .
Another example is the graph

u &
.......
& ..........
.. .
..........
.......
u ≥1
u &

corresponding to
def
 ◦ (AND ⊗ AND) ◦ (ID ⊗
XOR = OR  FANOUT ⊗ ID)
◦ ID ⊗ (NOT ◦ AND) ⊗ ID ◦ (FANOUT ⊗ FANOUT)
and acting as

def NOT(b2 ) if b1 = 1
(b1 , b2 ) 7−→ b1 ⊕ b2 =
b2 if b1 = 0
= b1 + b2 − 2b1 b2
= b1 + b2 mod 2 .
1.1. CLASSICAL LOGIC CIRCUITS 13

Name Symbol Class Action

s
CNOT F2,2 (b1 , b2 ) 7→ (b1 , b1 ⊕ b2 )
h

h
TCNOT F2,2 (b1 , b2 ) 7→ (b1 ⊕ b2 , b2 )
s

SWAP \ F2,2 (b1 , b2 ) 7→ (b2 , b1 )

\

s
(0, b1 , b2 ) 7→ (0, b1 , b2 )
CSWAP3 F3,3
\
\ (1, b1 , b2 ) 7→ (1, b2 , b1 )

CCNOT4 s F3,3 (b1 , b2 , b3 ) 7→ (b1 , b2 , b1 b2 ⊕ b3 )

h

Table 1.2: Some reversible gates

14 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

Of course, also for the gates listed in Table 1.2 there are equivalent networks, e.g.:5

CNOT = (ID ⊗ XOR) ◦ (FANOUT ⊗ ID) , (1.1)

h s
≡ \ \ , (1.2)
s \ h \
s h s
\ ≡ . (1.3)
\ h s h
s s

≡ h s h (1.4)
\
\ s h s

Now we are prepared for the

Proof of Lemma 1.1.1: Thanks to FANOUT and SWAP it is sufficient to proof
the lemma for decision functions, i.e. for n2 = 1 . Obviously, then, the statement
of the lemma holds for n1 = 1 , since the four elements of F1,1 are ID,
def
TRUE = OR ◦ (ID ⊗ NOT) ◦ FANOUT ,

and their compositions with NOT (applied last). Now, assume that the statement of
the lemma has already been proved for n1 = n and consider an arbitrary f ∈ Fn+1,1 .
Then both f0 and f1 , where
def
fs (b1 , . . . , bn ) = f (b1 , . . . , bn , s) ,

can be represented as compositions of tensor products of functions from Tabular 1.1.

There is a composition of FANOUTs and SWAPs acting as

(b1 , . . . , bn , s) 7−→ (b1 , . . . , bn , s, b1 , . . . , bn ).

Composing this with

SWITCH ◦ (f0 ⊗ ID ⊗ f1 )
(to be applied last) gives f . This proves the statement of the lemma for n1 = n + 1 .

According to Lemma 1.1.1 we may perform arbitrarily complex computations by

composing simple hardware components of very small variety. Of course, given
f ∈ Fn1 ,n2 , there are infinitely many representations of f as composition of tensor
products of elementary components. Therefore, the interesting problem arises how
to simplify a given gate (logic circuit) without changing its action.6
DRAFT, October 17, 2007
3
The CSWAP gate is also called Fredkin gate.
4
The CCNOT gate is also called Toffoli gate.
5
See (Tucci, 2004) for more equivalences of classical and/or quantum networks.
6
See (Lindner et al., 1999, Sect. 8.2.3) for n1 ≤ 6 , n2 = 1 and (Lee et al., 1999;
Shende et al., 2003) for quantum gates.
1.1. CLASSICAL LOGIC CIRCUITS 15

From the technological point of view it is also of interest that

def
NAND = NOT ◦ AND

is universal in the sense that it can replace NOT, AND, and OR as elementary
gates:7
NOT = NAND ◦ FANOUT ,
AND = NOT ◦ NAND ,
OR = NAND ◦ (NOT ⊗ NOT) .
In the same sense
def
NOR = NOT ◦ OR
is universal:
NOT = NOR ◦ FANOUT ,
AND = NOR ◦ (NOT ⊗ NOT) ,
OR = NOT ◦ NOR .
Alternatively, in order to minimize dissipation of energy (Landauer, 1961; Landauer, 1998;
Plenio and Vitelli, 2001; Bub, 2001; Parker and Walker, 2003), one may execute all
calculations using only reversible networks8 (Toffoli, 1980a):

Since9
CCNOT3 (b1 , b2 , 1) = NAND(b1 , b2 ) ∀ b1 , b2 ∈ {0, 1}
and  
CCNOT1 (b, 1, 0)
= FANOUT(b) ∀ b ∈ {0, 1} ,
CCNOT3 (b, 1, 0)
the CCNOT gate is universal for reversible classical computation in the following
sense:

For every mapping φ ∈ Fn1 ,n2 there is a reversible n-bit network com-
posed of only CCNOT gates,10 SWAP gates, and ID gates (wires) im-
plementing a mapping f ∈ Fn,n (n ≥ n1 , n2 ) fulfilling
 
f1 (b1 , . . . , bn1 , cn1 +1 , . . . , cn )
 .. 
 .  = φ(b1 , . . . , bn1 ) ∀ b1 , . . . , bn1 ∈ {0, 1}
fn2 (b1 , . . . , bn1 , cn1 +1 , . . . , cn )

for suitably chosen constant bits cn1 +1 , . . . , cn .

DRAFT, October 17, 2007
7
I.e., every classical logic circuit corresponds to a composition of tensor products of IDs,
FANOUTs, and NANDs.
8
Reversible classical networks (logic circuits) are those corresponding to bijections f ∈ Fn,n
for some n ∈ IN .
9
Toffoli called his gate the AND/NAND gate to indicate that also CCNOT3 (j, k, 0) =
AND(j, k) holds. Correspondingly, he called CNOT the XOR/FANOUT gate.
10
The CNOT gate cannot fulfill this purpose for classical computation.
16 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

Of course, it is a nontrivial task to optimize such networks.11

Theorem1.1.2 (Toffoli) For all n1 , n2 ∈ IN and for every φ ∈ Fn1 ,n2 there is
some n ∈ max {n1 , n2 } , . . . , n1 + n2 and some bijection f ∈ Fn,n with
 
f1 (b1 , . . . , bn1 , 0, . . . , 0)
 .. 
∀ b1 , . . . , bn1 ∈ {0, 1} .
 .  = φ(b1 , . . . , bn1 )
fn2 (b1 , . . . , bn1 , 0, . . . , 0)

Proof: See (Toffoli, 1980b, Theorem 4.1).

Exercise 1 Show that CSWAP acts as indicated and, therefore, is universal for
classical reversible computation:

b1 s b s b s b
b2 1 1⊕b 0 b
\ \ \
0 \ b1 b2 0 \ 1 \

Exercise 2 Show that the following networks act as indicated:12

a) Multiplication by 2:
0 a1
\
a1 \ a2
\
a2 \ ··· a3
.. ..
. .
an−1 ··· an
\
an \ 0

b) Adder13 for a, b ∈ {0, 1} :

0 h h cout = a cin ⊕ b (a ⊕ cin )

a s s a
+ ≡
b s s b
cin s h s h a ⊕ b ⊕ cin

DRAFT, October 17, 2007

11
See (Shende et al., 2003) and (Tsai and Kuo, 2001; Younes and Miller, 2003;
Shende et al., 2004a), in this connection.
12
See also (Vedral et al., 1996; Draper, 2000; Tsai and Kuo, 2001; Cheng and Tseng, 2002).
13
Note that a cin ⊕ b (a ⊕ cin ) = 0 iff a + b + cin ≤ 1 .
1.2. QUANTUM COMPUTATIONAL NETWORKS 17

c) Adder for a, b ∈ {0, 1}n :

0 ··· c0
a1 ··· a1
+
b1 ··· b1

0 ··· c1 a1 ⊕ b1 ⊕ c1
.. .. ..
. . .
.. .. ..
. . .
0 cn−2 ··· an−2 ⊕ bn−2 ⊕ cn−2
an−1 ··· an−1
+
bn−1 ··· bn−1

0 cn−1 ··· an−1 ⊕ bn−1 ⊕ cn−1

an ··· an
+
bn ··· bn
an ⊕ bn ⊕ cn
0 ··· |{z}
=0

n
X n
X n
X
aν 2n−ν + bν 2n−ν = c0 2n + (aν ⊕ bν ⊕ cν ) 2n−ν .
ν=1 ν=1 ν=1
| {z } | {z } | {z }
x y x+y

Exercise 3 Show that for every reversible classical 2-bit network there is an equiv-
alent one composed only of CNOTs, TCNOTs and NOTs.14

1.2 Quantum Computational Networks

1.2.1 Quantum Gates
If the computational registers are made smaller and smaller you will finally have to
take into account the quantum behavior of the devices. Then a n-bit register has
to be considered as an array of quantum mechanical systems to be ‘switched’ —
for classical computation — into one of two selected (pure) states corresponding
to two orthonormal state vectors, usually denoted |0i and |1i . This way the n-bit
information (b1 , . . . , bn ) will be encoded into the state vector

|b1 , . . . , bn i ≡ |b1 i ⊗ · · · |bn i (1.5)

corresponding to the situation:

DRAFT, October 17, 2007
14
Hint: Check the action of (ID ⊗ NOT) ◦ TCNOT on the ordered set of 2-bits.
18 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

‘Switch’ number ν being in the state corresponding to |bν i for ν = 1, . . . , n .

Then reversible classical n-bit gates correspond to permutations of the 2n states
corresponding to the orthonormal computational basis
{|bi : b ∈ {0, 1}n }
of the registers state space. Since such permutations are special unitary transfor-
mations there is a chance to implement them by mathematically simple quantum
mechanical evolution (governed by some Schrödinger equation). Of course the
interpolating states do no longer correspond to some element of the computational
basis even if this is the case for both the input state and the output state. Moreover,
quantum mechanics allows coherent superpositions as input states,15 correspond-
ing to complex linear combinations of the elements of the computational basis. Then
the gate causes the transition
X X
λb |bi 7−→ λb |f (b)i ,
n
|{z} n
b∈{0,1} ∈C b∈{0,1}

where f ∈ Fn,n is the mapping corresponding to the classical action of the gate.
This means that, in a way, the gate is able to perform all the 2n transitions
|bi 7−→ |f (b)i , b ∈ {0, 1}n ,
simultaneously — thanks to quantum mechanical evolution. Of course, one would
like to exploit this massive quantum parallelism for more efficient computation.
Unfortunately quantum mechanics imposes severe restrictions:
1. Unknown coherent superpositions cannot be copied with arbitrary precision
(Wootters and Zurek, 1982; Peres, 2002). Otherwise a device for superluminal
communication could be constructed (Werner, 2001, Chapter 3).
2. Every measurement of an unknown state destroys most of the information
carried by that state (quantum state collapse).
3. It is extremely difficult to correct errors caused by unwanted interaction with
the environment.
Nevertheless quantum computational networks can be devised, at least in principle,
which are much more efficient, for certain tasks, than classical computational net-
works. Their general structure is as follows:
• The information is usually processed on one and the same quantum register16
realized as an array of qubits,17 i.e. quantum mechanical systems with a
preselected simple quantum alternative corresponding to orthonormal state
vectors, usually denoted |0i and |1i .
DRAFT, October 17, 2007
15
See (Long and Sun, 2001) for an efficient preparation of these superpositions.
16
Thanks to the SWAP gate this is not a necessity but this point of view simplifies the treatment.
17
Usually qubits are treated as distinguishable, due to their localization in (essentially) disjoint
regions; see (Eckert et al., 2002) for a refined description.
1.2. QUANTUM COMPUTATIONAL NETWORKS 19

• The possible (pure) states of such an n-qubit register correspond to the (nor-
malized) complex linear combinations of the elements (1.5) of the computa-
tional basis.

• ‘Simple’ quantum computational steps are depicted in the network model by

quantum gates with an equal number (≤ n) of quantum wires (horizontal
lines) attached on both sides. These quantum wires represent the qubits on
which the gate acts.

• Several quantum gates may be assembled as in the classical reversible case

(without loopbacks, of course).

• The whole network itself is a (more complicated) n-qubit quantum gate acting
corresponding to some unitary operator Ûnet .

• The action of this operator on the initial state vector |b1 , . . . , bn i representing
the task (encoded in the bit sequence (b1 , . . . , bn )) has to be checked — i.e. the
output state Ûnet |b1 , . . . , bn i has to be measured — to yield a result.

Even though only the probability of a certain outcome of a quantum computation

is predictable (according to the rules of quantum mechanics) quantum computation
may be very useful for problems of the type

“solution easy to check but difficult to find”.

This will be demonstrated by several quantum algorithms in Chapter 2.

Remarks:

1. We use the naive tensor product formalism of quantum mechanics to describe

coupled systems. The latter is very problematic if the interaction of matter
with radiation is to be described; see Section 6.2.1 of (Lücke, nlqo).
2. Of course, the network model described above is just the simplest model for
quantum computing. Some possible generalizations, not to be discussed in this
chapter, are:
• Computation with non-unitarily evolving mixed states (Tarasov, 2002).
• Quantum computation via application of sequences of one-qubit projective
measurements to suitably prepared initial states (Raussendorf et al., 2002).
• Use of non-deterministic gates, i.e. those succeeding only with proba-
bility (considerably) less than 1 (Ralph et al., 2002; Bartlett et al., 2002).
3. We are not going to discuss oddities like18 “quantum computation even before
its quantum input is defined” (Brukner et al., 2003) or “counterfactual compu-
tation” (Mitchison and Jozsa, 2001). For quantum programming we refer to
(Bettelli et al., 2001).
DRAFT, October 17, 2007
18
See 3.1.4, however.
20 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

Name Symbol Operator Matrix Action

 
u11 u12 |0i 7→ u11 |0i + u21 |1i
Û gate Û Û
u21 u22 |1i 7→ u12 |0i + u22 |1i

 
Table 1.3: General one-qubit gate u1j u1k + u2j u2k = δjk

Name Symbol Operator Matrix Action

 
1 0
ID gate 1̂ |bi 7→ |bi

0 1
H
Hb
0 1
NOT gate 

¬
ˆ |0i ⇀
↽ |1i
1 0
 
1 0 |0i 7→ |0i
phase shift gate δ Ŝδ
0 eiδ |1i 7→ eiδ |1i
  √1
1 1 |0i 7→ (|0i + |1i)
Hadamard gate H ÛH √1 2
2 1 −1 |1i 7→ √1 (|0i − |1i)
2

Table 1.4: Special one-qubit gates

The elements |b1 , . . . , bn i of the computational basis of an n-qubit system are

naturally ordered by the corresponding integers
n
X
def
I(b) = bν 2n−ν ∀ b ∈ {0, 1}n . (1.6)
ν=1

It is relative to this ordering that the actions of quantum gates are usually repre-
sented by unitary matrices as in Tables 1.3–1.5.

Remark: Note that every (complex) unitary 2 × 2 matrix corresponds to a spin

rotation; see, e.g., Exercise 19 of (Lücke, tdst) and Sect. 4.2.1 of (Lücke, qft).

Note that a quantum gate may be used for classical computation iff the entries of
its matrix take only values from {0, 1} . Whenever this is the case we use the same
symbol and name for the quantum gate as for its classical analog. In this sense we
have, e.g.,

V s s
¬)
1 (ˆ = CNOT : = ,
¬
ˆ h
1.2. QUANTUM COMPUTATIONAL NETWORKS 21

Name Symbol Matrix Action

s
..   |b1 , . . . , bn , ci
V   . 1l2n 0 (
n Û gate
s 0 Û |b1 , . . . bn i ⊗ Û |ci if b1 = . . . = bn = 1
7→
|b1 , . . . , bn , ci else
Û

Table 1.5: Special (n + 1)-qubit gates

s s
V s s
¬)
2 (ˆ = CCNOT : = , (1.7)
¬
ˆ h

i.e.
h = ¬
ˆ . (1.8)

Obviously, the phase shift gate (for δ 6= 0 mod π) and the Hadamard gate have no
classical analog.

Sometimes it is more convenient to sketch the action of a gate as done in Figure

1.1 for the f -CNOT gate.

|b1 i |b1 i
.. f ..
. .
|bn i |bn i
|ci h |c ⊕ f (b)i

Figure 1.1: f -CNOT gate for f ∈ Fn,1

Quantum computational networks are called equivalent if they implement the

same mapping up to a phase factor.

Exercise 4
a) For arbitrary a ∈ {0, 1}n , n ∈ IN , show that
22 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

N̂a1 s N̂a1
.. fa .. .. .. ..
. . ≡ . . .
N̂an s N̂an
h h ,

where
def
fa (b) = δa,b ∀ b ∈ {0, 1}n
and 
def ¬
ˆ if b = 0 ,
N̂b =
1̂ else .
b) Show for arbitrary f, g ∈ Fn,1 that

.. h .. .. f .. g ..
h=f ⊕g =⇒ . . ≡ . . .

h h h .

A first example showing the superiority of quantum networks is the Deutsch-

Jozsa problem:

Assume you are given an (n + 1) qubit gate which is only known to be

the f -CNOT gate of some f ∈ Fn,1 that is either constant or balanced,
i.e. fulfills X
(−1)f (b) = 0 .
b∈{0,1}n
Find out by ‘asking’ this Deutsch-Jozsa oracle whether f is balanced
or constant.

Note that in classical computation

|b1 i |b1 i
.. f ..
. .
|bn i |bn i
|0i h |f (b)i

you may have to ask the Deutsch-Jozsa oracle 2n−1 + 1 times (in the worst case)
to find the answer. Already for n = 60 that would take more than

259
years > 18 years
60 · 60 · 24 · 365 · 109
1.2. QUANTUM COMPUTATIONAL NETWORKS 23

to get the answer if the oracle is asked at a frequency of 1 GHz. In quantum com-
putation, however, we may take advantage of coherent superpositions:

|b1 i |b1 i |0i H 

 X
.. f .. .. f .. ∼ (−1)f (b) |bi
. . . . 
|bn i |bn i |0i b∈{0,1} n
H
f (b)
√1 (|0i − |1i) h (−1) |1i H h H |1i .
2 √
2
(|0i − |1i) ,

Since  
ÛH |bi = √1
2
|0i + (−1)b |1i
1 X ′
= √ (−1)b b |b′ i ∀ b ∈ {0, 1}
2 b′ ∈{0,1}
and hence
X ′
ÛH⊗n |bi = 2−n/2 (−1)b ·b |b′ i ∀ b ∈ {0, 1}n (1.9)
′ n
b ∈{0,1}
we have
X ′
(−1)f (b)+b ·b |b′ , 1i .
⊗(n+1) ⊗(n+1)
ÛH ◦ f -CNOT ◦ ÛH |0, . . . , 0, 1i = 2−n
b,b′ ∈{0,1}n
(1.10)
For the Deutsch-Jozsa oracle this means

⊗(n+1) ⊗(n+1) ∼ |0, . . . , 0, 1i if f is constant,
ÛH ◦ f -CNOT ◦ ÛH |0, . . . , 0, 1i =
⊥ |0, . . . , 0, 1i else.
Therefore, the following quantum gate has to be used only19 once in order to solve
the Deutsch-Jozsa problem (Cleve et al., 1998, Sect. 3):

|0i 
H H  
 ∼ |0, . . . , 0i if f is constant,
.. .. f .. Ψ=
. . .  ⊥ |0, . . . , 0i else.

|0i H H
|1i H h H |1i

Remark: (1.10) holds for every f ∈ F2,1 and can therefore be applied
also to the Bernstein-Vazirani oracle, i.e. the f -CNOT gate with
f (b) = a · b for some a ∈ {0, 1}n . Then Ψ becomes
X ′
2−n (−1)(a+b )·b |b′ i = |ai .
b,b′ ∈{0,1}n
DRAFT, October 17, 2007
19
Actually, since there is always some tiny probability for getting the wrong answer, the
quantum test should be repeated a few times. For a physical realization of the algorithm see
(Gulde et al., 2003).
24 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

1.2.2 Quantum Teleportation

Obviously, the Bell network

H s
h

acts according to20

|0, 0i + |1, 1i  
def def
|0, 0i 7−→ Ψ0,0 = Φ+ = √ = Û0,0 ⊗ 1̂ Ψ0,0 ,
2
 
def def |0, 1i + |1, 0i
|0, 1i 7−→ Ψ0,1 = Ψ+ = √ = Û0,1 ⊗ 1̂ Ψ0,0 ,
2
 
def def |0, 0i − |1, 1i
|1, 0i 7−→ Ψ1,0 = Φ− = √ = Û1,0 ⊗ 1̂ Ψ0,0 ,
2
 
def def |0, 1i − |1, 0i
|1, 1i 7−→ Ψ1,1 = Ψ− = √ = Û1,1 ⊗ 1̂ Ψ0,0 ,
2
where
def
Û0,0 = + |0ih0| + |1ih1| = 1̂ ,
def
Û0,1 = + |1ih0| + |0ih1| = ¬
ˆ,
def
Û1,0 = + |0ih0| − |1ih1| = Ŝπ ,
def
Û1,1 = − |1ih0| + |0ih1| = Ŝπ ¬ˆ.

This indicates the possibility of quantum dense coding:21

Bob prepares the entangled 22 state Ψ0,0 by applying the Bell network
to the easily available state |0, 0i and sends his first qubit to Alice (ar-
bitrarily far away). Now Alice may transfer a 2-bit message to Bob by
applying one of the operators Û0,0 , Û0,1 , Û1,0 , Û1,1 to this single qubit
and sending it back to Bob. Bob can ‘read’ this message by performing
DRAFT, October 17, 2007
20
The states on the r.h.s are usually called Bell
 states.
 They exhibit maximal correlation
α def −β
between the two qubits. We use the notation = . Thus
β ⊥ +α
 
1 Ψ Ψ⊥ Ψ⊥ Ψ
Ψ− = √ ⊗ − ⊗ ∀ Ψ ∈ C2 \ {0} .
2 kΨk kΨ⊥ k kΨ⊥ k kΨk

21
See (Mermin, 2002) for an interesting discussion of dense coding.
22
Entanglement of vector states Ψ means
i.g.
hΨ| Â ⊗ B̂ |Ψi 6= hΨ| Â ⊗ 1̂ |Ψi hΨ| 1̂ ⊗ B̂ |Ψi ,

i.e. (non-classical) correlations between the subsystems. See (Brukner et al., 2001) in this context.
1.2. QUANTUM COMPUTATIONAL NETWORKS 25

a Bell measurement, i.e. checking whether the state of the 2-qubit

system is Ψ0,0 , Ψ0,1 , Ψ1,0 , or Ψ1,1 .
The whole procedure is described by the following network action:

|b1 i s |b1 i

|b2 i s |b2 i

|0i H s h π s H |b1 i
|0i h h |b2 i .

Remarks:

1. As pointed out in (Mermin, 2002), the network

s s
s s s
≡
H h s h π s H h
h h h

— although unsuitable for dense coding — has the same effect on the considered
special input.
2. We use abbreviations like

s
\ \
for \ s \

 Â

without explicit definition.

Moreover, checking the special cases Ψ ∈ {|0i , |1i}, we see that the teleportation
network
s H
H s h
h

acts on ψ ⊗ |0, 0i in the following way:

1 X
ψ ⊗ |0, 0i 7−→ |bi ⊗ Ûb−1 ψ .
2 b∈{0,1}2
26 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

(for every one-qubit state vector ψ). This indicates the possibility of quantum
teleportation:23
Qubits 2 and 3 are prepared in the state Φ0,0 (indicated by the Bell
subnetwork on the left acting on |0, 0i). Then qubit 2 is sent to Alice
and qubit 3 to Bob (far apart). Since, now, Alice and Bob share an
entangled pair of qubits, Alice may teleport the unknown state ψ of
qubit 1 to Bob in the following way:
Alice performs a Bell measurement on the system formed by
qubits 1 and 2 and sends Bob the classical 2-bit information
b if the result is Ψb (corresponding to the output |bi of the
teleportation network for qubits 1 and 2). After receiving this
information Bob transforms the (collapsed) state of qubit 3
into ψ by applying Ûb .
Note that the actions taken by Alice and Bob, sharing the entangled pair, have the
same effect on qubits 1–3 as the following post selection scheme:24

s H s ⌢
/
h s ⌢
/
¬
ˆ π

In this sense the following scheme describes teleportation of entanglement, also called
entanglement swapping:25

|0i H s

|0i h s H s ⌢
/
|0i H s h s ⌢
/
|0i h ¬
ˆ π

This possibility is very important for creating entanglement for teleportation over
very large distances.

Exercise 5 Show that the entanglement swapping scheme prepares the subsystem
formed by qubits 1 and 4 (arbitrarily far apart) in the state Ψ0,0 .

DRAFT, October 17, 2007

23
The possibility of teleportation, further discussed in Appendix A.2, was first pointed out
in (Bennet et al., 1993). Generalization to n-qubit states is straightforward (Dı́az-Caro, 2005).
Concerning the experimental realization of quantum teleportation see (Giacomini et al., 2002).
24
The symbol ⌢ / represents an ideal test (projective measurement) whether the corresponding
qubit is in state |0i or |1i .
25
See (Bowmeester et al., 2000, Sect. 3.10) and (Gisin and Gisin, 2002).
1.2. QUANTUM COMPUTATIONAL NETWORKS 27

1.2.3 Universality
A 2-qubit gate corresponding to the unitary operator Û (2) is called universal if for
every quantum network there is an equivalent one composed only of one-qubit gates
and 2-qubit gates corresponding to Û (2) .
Ordering the computational basis vectors
def
|I(b)in = |bi ∀ b ∈ {0, 1}n (1.11)
as |0in , |1in , . . . |2n − 1in then λn−1 (ˆ
¬) just interchanges the last two of these vec-
tors, the latter being |1, . . . , 1, 0i and |1, . . . , 1, 1i . Also cyclic permutations of the
computational basis vectors can be achieved by suitable composition of Λν (ˆ ¬) gates,
as the following exercise shows.

Exercise 6 Show that the n-qubit network

h ···
s h ···
.. .. ..
. . .
s s ···
s s ··· h
s s ··· s h
s s ··· s s ¬
ˆ

acts according to
|xin 7−→ |(x + 1) mod 2n in ∀ x ∈ {0, . . . , 2n − 1} ,
where +
n
X def
bν 2 ν−1
= |bi ∀ b ∈ {0, 1}n . (1.12)
ν=1 n

Therefore:
For every quantum gate that has a classical analogue there is an equiv-
alent quantum network composed only of Λν (ˆ ¬) (and ID) gates.

This together with the following theorem shows that the CNOT gate is universal if
the following holds for every ν ∈ IN :
For every Û ∈ U (2) there is a network composed only of single qubit
gates and CNOT gates (and ID gates) that is equivalent to the Λν (Û ) (1.13)
gate.
28 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

Theorem 1.2.1 Let 2 < N ∈ IN . Then every unitary N × N -matrix may be

represented as a product of permutation matrices and N × N -matrices of the form
 
1̂ 0
, Û ∈ U (2) .
0 Û

Outline of proof:26 Given 2 < N ∈ IN , choose some orthonormal basis {e1 , . . . , eN }

of CN . Then, for arbitrary
 
u11 u12
Û = ∈ U (2) ,
u21 u22
N ′ ∈ {2, . . . , N } , and ν ∈ {1, . . . , N } define
(
u11 eN ′ −1 + u21 eN ′ for ν = N ′ − 1 ,
def
ÛN ′ eν = u12 eN ′ −1 + u22 eN ′ for ν = N ′ ,
eν else .
Then, for arbitrary normed
N
X
z= z ν eν ∈ CN ,
ν=1
N ′ ∈ {2, . . . , N } , and ζ ∈ C with
N
X
2 2
|ζ| + |z ν | = 1
ν=N ′ +1

we have    
0 0
  ..  ..
   .  .
   
 0   0 
   ′ 
(N ′ )  0   ζ 
ÛN ′   =  N′ 
 ζ′   z ′ 
 N +1   N +1 
z  z 
 .   . 
 ..   .. 
zN zN
′
for suitable Û (N ) ∈ U (2) and ζ ′ ∈ C . Thus, by iteration, we see that there are
Û (N ) , . . . , Û (2) ∈ U (2) with
(2) (N )
z = Û2 · · · ÛN eN
and hence  −1  −1
(N ) (2)
ÛN · · · Û2 z = eN .
ˆ
Identifying z with the last column of an arbitrarily given unitary N × N -matrix Û
we get  −1  −1 
(N ) (2) ˆ
ÛN · · · Û2 Û = δν,N .
ν,N
Thanks to unitarity, the latter also implies
 −1  −1 
(N ) (2) ˆ
ÛN · · · Û2 Û = δN,ν .
N,ν

Iteration of this argument, if necessary, proves the theorem.

DRAFT, October 17, 2007
26
Compare (Reck et al., 1994; Diţă, 2001).
1.2. QUANTUM COMPUTATIONAL NETWORKS 29

That (1.13) holds for ν = 1 is a simple consequence of the following lemma

(Barenco et al., 1995):

Lemma 1.2.2 For every 27 Û ∈ SU(2) there are Â, B̂, Ĉ ∈ SU(2) with

ÂB̂ Ĉ = 1̂ , ¬B̂ ¬
ˆ ˆ Ĉ = Û .

Proof: Let Û ∈SU(2) . Then one may easily show (see Exercise 28 of (Lücke, eine))
that there are angles ψ, θ, φ with

Û = R̂3 (ψ) R̂2 (θ) R̂3 (φ) ,

where    
+ cos θ2 + sin θ2
ψ
def e+i 2 0 def
R̂3 (ψ) = ψ , R̂2 (θ) = .
0 e−i 2 − sin θ2 + cos θ2
With the definitions
       
def θ def θ ψ+φ def φ−ψ
 = R̂3 (ψ) R̂2 , B̂ = R̂2 − R̂3 − , Ĉ = R3
2 2 2 2

this gives


   
 B̂ Ĉ = R̂3 (ψ) R̂2 θ
2 R̂2 − θ2 R̂3 − ψ+φ
2 R3 φ−ψ
2
= R̂3 (ψ) R̂3 (−ψ)
= 1̂

and


   
¬B̂ ¬
ˆ ˆ Ĉ = R̂3 (ψ) R̂2 θ
2 ¬ˆ R̂2 − θ2 R̂3 − ψ+φ
2 ˆ R3 φ−ψ
¬ 2


     
= R̂3 (ψ) R̂2 θ
2 ˆ R̂2 − θ2 ¬
¬ ˆ ¬ ˆ R̂3 − ψ+φ2 ˆ R3 φ−ψ
¬ 2
θ

θ

 ψ+φ  
φ−ψ

= R̂3 (ψ) R̂2 2 ˆR̂ 2 2 R̂ 3 2 R 3 2

= R̂3 (ψ) R̂2 (θ) R̂3 (φ)

= Û .

Remark: Note that ψ, θ, φ in the above proof correspond to the well-known

Euler angles; see Sect. 2.1.1.3 of (Lücke, mech) and — for generalization — also
(D’Alessandro, 2001).

Because of
|0i s s |0i
|bi Ĉ h B̂ h  ÂB̂ Ĉ |bi

and
DRAFT, October 17, 2007
27
As usual, we denote by U(2) the set of all (complex) unitary 2 × 2-matrices and by SU(2) the
set of all Û ∈U(2) with det Û = 1 .
30 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

|1i s s |1i
|bi Ĉ h B̂ h  ¬B̂ ¬
ˆ ˆ Ĉ |bi

this has the following consequence:

Corollary 1.2.3 For Â, B̂, Ĉ, Û according to Lemma 1.2.2 we have:
s s s
≡ .
Û Ĉ h B̂ h Â

Corollary 1.2.3 together with

s δ s
≡ , Â′ = eiδ Â ,
′
 Â

shows that that (1.13) holds for ν = 1 , indeed.

Exercise 7 Show that

h H s H
≡
s H h H

and28
s H s H s
\ ≡
\ h H h H h .

Moreover, because of
|0i s s s |0i
|0i s h s h |0i
|bi Ĉ B̂ Â |bi

|0i s s s |0i
|1i s h s h |1i
|bi Ĉ B̂ Â B̂ Ĉ |bi
DRAFT, October 17, 2007
28
Recall (1.3).
1.2. QUANTUM COMPUTATIONAL NETWORKS 31

|1i s s s |1i
|0i s h s h |0i
|bi Ĉ B̂ Â ÂB̂ |bi

|1i s s s |1i
|1i s h s h |1i
|bi Ĉ B̂ Â ÂĈ |bi

we have (Sleator and Weinfurter, 1994):

s s s s
s ≡ s h s h

V̂ 2 V̂ V̂ ∗ V̂

V 2
This Sleator-Weinfurter construction may be generalized for n (V̂ ) with
arbitrary n ∈ IN :29

s s s s
s s h s h
s s s s
..
≡ .. .. ..
. . . .
s s s s

V̂ 2 V̂ V̂ ∗ V̂

Since n o
U (2) = V̂ 2 : V̂ ∈ U(2)
this that (1.13) holds for every ν ∈ IN , hence:

The CNOT gate is universal.30

DRAFT, October 17, 2007
29
Of course, one should look for more efficient implementations; see, e.g. Exercise 8 and
(Aho and Svore, 2003; Vatan and Williams, 2004; Shende et al., 2004b). For a nice introduction
into the general theory of computational complexity see (Mertens, 2002).
30
Therefore, if CNOT is implementable together with all one-qubit gates and projective mea-
surements w.r.t. the computational basis, all observables can actually be measured.
 32 CHAPTER 1. BASICS OF QUANTUM COMPUTATION

Exercise 8 Show that the following network acts as indicated:

|b1 i s ··· ··· s |b1 i

|b2 i s ··· ··· s |b2 i
|b3 i s ··· ··· s |b3 i
|b4 i s ··· ··· s |b3 i
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
|bn−1 i ··· s s ··· |bn−1 i
|bn i ··· s s ··· |bn i
|0i h s ··· ··· s h |0i

|0i h s ··· ··· s h |0i

|0i h ··· ··· h |0i

.. .. ··· .. .. .. .. ··· .. ..
. . . . . . . .
|0i ··· s s ··· |0i

|0i ··· h s s h ··· |0i

|0i ··· h s h ··· |0i

|bn+1 i ··· Û ··· ψ

(
Û |bn+1 i if b1 = . . . = bn = 1 ,
ψ=
|bn+1 i else.

Exercise 9 Show that31

s s
≡
h H π H

and that
s π
≡
π s

acts according to
|j, ki 7−→ (−1)δj,1 δk,1 |j, ki .

DRAFT, October 17, 2007

31
This equivalence is exploited in most suggestions for physical realization of CNOT.
Chapter 2

Quantum Algorithms1

So far, we have only discovered a few techniques which can produce speed up versus
classical algorithms. It is not clear yet whether the reason for this is that we do
not have enough intuition to discover more techniques, or that there are only a few
problems for which quantum computers can significantly speed up the solution.
(Shor, 2000)

2.1 Quantum Data Base Search

2.1.1 Grover’s Algorithm
Let us assume that the b ∈ {0, 1}n are the indices of the entries of some unstructured
data base. Moreover let us assume we are given a search machine that provides an
implementation of the fa -CNOT gate, where

1 if b = a
fa (b) = ∀ b ∈ {0, 1}n ,
0 else
when fed with a unique characterization of some entry indexed by a . Now consider
the following problem:
Find a with probability ≥ 50% by testing the behavior of the fa -CNOT
gate.
In classical computing (recall 1.2.2) one has to test fa -CNOT at least 2n−1 -times2 in
order to find a with probability of 50% . A substantial speedup,3 exploiting quantum
DRAFT, October 17, 2007
1
Algorithms are general, step-by-step procedures for solving general problems.
2
Moreover, expectation value for the necessery number of tests for finding a is
n
X2
ν 2n + 1
N= = .
ν=1
N 2

3
See (Aaronson and Gottesman, 2004), however.

33
34 CHAPTER 2. QUANTUM ALGORITHMS

|ai
.
.....
..........
..
....
..
...
...
..
.
... (n)
...
pppppppppp
pppppp pppp Ĝa Φ0
.
... .
.... p p p .
.. pp p p p . .

pppp pppp
... .
.... . .
pppppp pp
... ... ..
. .
.
....
.
p p p p p p ... .
pppppp 2θ . ..
pp pppppp
.. ... .
...
pppp pppp
. .
... pppppppppppp (n) . .
................
..................... pp Φ .. ....
.... .
.
.
0
....... .............
.......
.......
.......
.............
.............
.............
θ ... .. ..
............... .... ..
.
....... .................
.......
.......
.......
.......
.......
...........
. ..
..
..
⊥ |ai
....... .
.......... ....
.....................

(n)
Figure 2.1: Action of Ĝ on Φ0 .

parallelism, was suggested in (Grover, 1996). The basic ingredients of Grover’s

algorithm are the initial state
(n) def X
Φ0 = ÛH⊗n |0, . . . , 0i = 2−n/2 |bi = 2−n/2 Φ1,1 (2.1)
b∈{0,1}n
and the unitary operator
def
Ĝa = −R̂Φ(n) R̂|ai , (2.2)
0

where
def
R̂Ψ = 1̂ − 2 P̂Ψ ∀ Ψ ∈ Hn .
(n)
Since both reflections R̂|ai and R̂Φ(n) leave the |ai-Φ0 -plane invariant, Ĝa acts as
0
a rotation in this plane. To determine this rotation it suffices to check its effect on
(n)
Φ0 .
As explained in Figure 2.1 this action is a rotation by the angle 2 θ towards |ai ,
(n)
where π/2 − θ is the angle between Φ0 and |ai . Therefore:
(n)
Applying Ĝa an appropriate number of times to Φ0 and testing the
result with respect to the computational basis solves the posed problem.

2.1.2 Network for Grover’s Algorithm

Exercise 10 Show that the following (n + 1)-qubit networks act as indicated:4

 
 
Ψ .. fa .. δ0b Ψ + δ1b R̂|ai Ψ
. . 

|bi H h H |bi ,

DRAFT, October 17, 2007

4
Recall Exercise 4a).
2.1. QUANTUM DATA BASE SEARCH 35

 H H 
 
Ψ .. f0 .. δ0b Ψ − δ1b R̂Φ(n) Ψ
. .  0
H H
|bi H h H π |bi .

V
Using the obvious notation and generalization of 1 (Û ) for n-qubit unitary operators
Û , we get from Exercise 10

H H
.. Ĝ .. .. fa .. f0 ..
. a . . . .
≡
H H
s H h h H π

and therefore:

|0i H ··· 

.. .. .. .. Ĝ µ (n)
. . Ĝa . . a Ĝ Φ
 a 0
|0i H ···
|1i s ··· s |1i .
| {z }
µ-times

2.1.3 Details and Generalization

For actual application of Grover’s one has to know the number of times Ga should
be applied. This can be read off from5
    1 X
(n)
Gaµ Φ0 = sin (2µ + 1) θ |ai + cos (2µ + 1) θ √ |bi , (2.3)
2n − 1 b6=a

where !
def 1
θ = arcsin √ n . (2.4)
2
Since
θ ≈ 2−n/2 for 2−n/2 ≪ 1
DRAFT, October 17, 2007
5
Formula (2.3) was presented first in (Boyer et al., 1998).
36 CHAPTER 2. QUANTUM ALGORITHMS

we have  √ 
µ (n) π
Ga Φ0 ≈ |ai for µ = 2n and 2n ≫ 1 .
4
In this sense Grover’s algorithm provides a quadratic speedup compared to clas-
sical computation.

Let us now consider the case that there are exactly t data base entries,6 indexed
by a1 , . . . , at ∈ {0, 1}n , meeting the search criteria and that the search engine,
therefore, provides an implementation of the (fa1 + . . . fat )-CNOT gate. In order to
find at least one of these aν we just have to replace Ĝa by
def
Ĝa1 ,...,at = −R̂Φ(n) R̂|a1 i · · · R̂|at i ,
0

which may be implemented as described in 2.1.2 with the fa -CNOT gate replaced
by the (fa1 + . . . fat )-CNOT gate. Correspondingly, (2.3)/(2.4) have to be replaced
by
  |a i + . . . |a i   1 X
(n) 1 t
Ĝµa1 ,...,at Φ0 = sin (2µ + 1) θt √ +cos (2µ + 1) θt √ |bi
t 2n − t b6∈{a1 ,...,at }
(2.5)
and !
def t
θt = arcsin √ n . (2.6)
2
 
Choosing µ such that sin2 (2µ + 1) θt is close to 1 we get a state that is essential
a superposition of only those states of the computational basis which correspond
to data base entries meeting the search criteria. Performing a test we select one
solution at random.
Unfortunately, we only know how to choose qµ nif we
 know t . If t is unknown we
2 −1
find a solution after an expected number of O t
applications of the described
′
procedure with suitably chosen µ s (Boyer et al., 1998, Sect. 4).

For interesting modifications of Grover’s algorithm see (Ambainis, 2005; Korepin and Grover, 2
Tulsi et al., 2005) and references given there. For a few-qubit experimental imple-
mentation of the algorithm see (Walther et al., 2005) and references given there.
For application to robots see (Dong et al., 2005).

Final Remark: Presumably Grover’s algorithm will not be useful

for searching a standard database, because transferring the database to
quantum memory would require too much effort (Deutsch and Ekert, 1998).

DRAFT, October 17, 2007

6
For instance, we may be interested in only a subset of bit-values
(Grover and Radhakrishnan, 2004)
2.2. FACTORING LARGE INTEGERS 37

2.2 Factoring Large Integers

2.2.1 Basics
The greatest common divisor gcd(n0 , n1 ) for given n0 , n1 ∈ ZZ may be efficiently
determined via Euclid’s algorithm:7
Defining 
nν+2 = nν − nν+1 ⌊ nnν+1
def ν
⌋ if nν+1 6= 0 (2.7)
0 else
successively for ν = 0, 1, 2, . . . we get

gcd(n0 , n1 ) = ns for s = sup {ν ∈ IN : nν 6= 0} < ∞ . (2.8)

Thus, factoring a given product N = p1 p2 of unknown large integers p1 , p2 is a task

of the type
“A solution is easy to check but extremely difficult to find.”
Classical encryption schemes rely on this fact.

The most popular public-key encryption algorithm8 is RSA, named after its three
inventors Ron Rivest, Adi Shamir, and Leonhard Adleman. It works as follows
(Rivest et al., 1978):

• Messages and keys are represented by natural numbers corresponding to binary

strings.
• Messages M are encoded as

C = M e mod N ,

where N > M and e are two public keys created in the following way:
1. Two large prime numbers p and q of comparable size are ran-
domly chosen9 and kept secret. Only their product

N =p·q

is publicly announced.
2. e is chosen as a large random number having 1 as largest com-
mon divisor with (p − 1) · (q − 1) — to be checked by Euclid’s
algorithm.
DRAFT, October 17, 2007
7
As usual, we use the notation
def
⌊x⌋ = sup {n ∈ ZZ : n ≤ x} ∀ x ∈ IR .

Thus, for nν+1 6= 0 , nν+2 is the remainder of the integer division of nν by nν+1 . For details
concerning Euclid’s algorithm see Section 2.2.3.
8
See (Singh, 2002; Kahn, 1967) for the history of classical cryptography and (Schneier, 1996)
for applications.
9
See (Rivest et al., 1978, Section VII.B) how to find large prime numbers without testing pri-
mality by factorization.
38 CHAPTER 2. QUANTUM ALGORITHMS

• The message may be decrypted in the form

M = C d mod N .

where d ∈ {1, . . . , (p − 1) · (q − 1) − 1} is the private key to be determined

from10
e · d = 1 mod (p − 1) · (q − 1) .

Of course, d has to be kept secret as well as the prime numbers p, q which then may
be forgotten.

The task of factorizing N in the form

N =p·q with p, q ∈ {2, . . . , N − 1}

is essentially solved if a factorization

n+ · n− = 0 (modN ) (2.9)

of an integer multiple of N is found that fulfills the conditions

n± 6= 0 (modN ) (2.10)

Then11
gcd (n± , N ) ∈ {2, . . . , N − 1} (2.11)
and these factors may be efficiently determined using Euclid’s algorithm.

Outline of proof for (2.11): Obviously, every prime factor of N must be a factor
of either n+ or n− (or both) and neither n+ nor n− can be the product of all these
(not necessarily pairwise different) factors.

Finding a factorization of type (2.9),(2.10) is facilitated by the following theorem.

Theorem 2.2.1 (Euler’s Theorem) Let x, N ∈ IN . If x and N are coprime,

i.e. if gcd(x, N ) = 1 , then
xϕ(N ) = 1 mod N
holds, where Euler’s ϕ function is defined as 12
def
ϕ(N ) = {y ∈ IN : y < N , gcd(y, N ) = 1} .
DRAFT, October 17, 2007
10
The essential point is that
(p − 1) · (q − 1) = ϕ(p · q) ,
where ϕ denotes the Euler function introduced in Theorem 2.2.1, below. d may be determined
modulo (p − 1) · (q − 1) running Euclid’s algorithm (2.7) for n0 = (p − 1) · (q − 1) and n1 = e
and resubstituting iteratively the expressions for ns = 1 , s given by (2.8), using (2.7) to yield the
representation 1 = e · x + y · N with certain integers x, y .
11
As usual, we denote by gcd(n1 , n2 ) the greatest common divisor of two integers n1 , n2 .
2.2. FACTORING LARGE INTEGERS 39

Proof: See, e.g., (Schroeder, 1997, Sect. 8.3) or (Nielsen and Chuang, 2001, Theo-
rem A4.9).

By Euler’s theorem, for N ∈ IN and

x ∈ {2, . . . , N − 1} , gcd (x, N ) = 1 (2.12)

the function
def
f (ν) = xν ∀ν ∈ ZZ (2.13)
has a minimal period
def
r = inf {a ∈ IN : xa = 1 (modN )} , (2.14)

called the order 13 of x modulo N . If r is even and

xr/2 6= −1 (modN ) (2.15)

then the conditions (2.9) and (2.10) are fulfilled for

n± = xr/2 ± 1 .

Outline of proof: (2.9) is a consequence of (2.14) and

  
xr/2 + 1 xr/2 − 1 = xr − 1 .

(2.10) follows from (2.15) and the corresponding property

xr/2 6= +1 (modN )

implied by (2.14).

Summarizing, we have the following factoring algorithm:

1. Randomly choose some x ∈ {2, . . . , N − 1}.

2. If gcd(x, N ) 6= 1 then x is already a nontrivial factor of N .

3. If (2.12) holds, determine the order r of x modulo N .

4. If r is odd or xr/2 = −1 (modN ), restart the algorithm.

 
5. If r is even and xr/2 6= −1 (modN ), determine the factors gcd xr/2 ± 1, N
using Euler’s algorithm.
DRAFT, October 17, 2007
12
By |M | we denote the number of elements of a finite set M .
13
The numbers x ∈ {1, . . . , N − 1} form a group w.r.t. multiplication modulo N . Every element
x of this group generates a cyclic subgroup of order r .
40 CHAPTER 2. QUANTUM ALGORITHMS

Thanks to the following theorem (and Euler’s algorithm) the efficiency of this
factoring algorithm depends solely on the available techniques for determining (2.14).

Theorem 2.2.2 Let m be the number of different prime factors of the positive
integer N and let x ∈ {1, . . . , N − 1} be randomly chosen. If gcd(x, N ) = 1 , then
the (conditional) probability for (2.14) being even and xr/2 6= −1 (modN ) is not
smaller than 1 − 2−m .

Proof: See, e.g., (Nielsen and Chuang, 2001, Theorem A.4.13).

In order to gain exponential speed up for the determination of (2.14), Peter

W. Shor suggested the following (Shor, 1994):

Instead of calculating xa (modN ) for a = 1, 2, . . . until the result is

1 (modN ), transform the state
2L
1 2 X−1
|ai2L ⊗ |0iL ,
2L a=0

where n o
def
L = min l ∈ IN : N ≤ 2l ,
into the state14
2L
1 2 X−1
|ai2L ⊗ |xa (modN )iL
2L a=0
(exploiting quantum parallelism) and evaluate the latter by means of the
quantum Fourier transform applied to the first 2L qubits..

2.2.2 The Quantum Fourier Transform

In order to plot, over the interval [−Ω, +Ω] , the Fourier transform

1 Z T0
fe(ω) = √ f (t) ei ω t dt
2π 0

of a (sufficiently well-behaved) signal restricted to the time interval [0, T0 ] it is suf-

ficient to have the discrete values
 
2π 2π
fe k , k ∈ ZZ , k ≤ Ω,
T0 T0
DRAFT, October 17, 2007
14
An efficient implementation is described in (Vedral et al., 1996).
2.2. FACTORING LARGE INTEGERS 41

if T0 is large enough (depending on the required precision). In order to determine

these values approximately it is sufficient to know the sampling values
 
T0
f j , j ∈ {0, 1, . . . , N − 1} ,
N
for sufficiently large N ∈ IN (depending on T0 and Ω):
  −1  
2π 1 NX T0 2π T0 2π
fe k ≈√ f j ei k N j for k ≤ Ω.
T0 2π j=0 N N T0

Remark: In order to estimate the quality of this approximation note that

N
X −1   Z T0
T0 2π
f j ei k N j = ∆T0 /N (t) f (t) ei ω t dt ,
j=0
N 0

where
def
X  T0

∆T0 /N (t) = δ t−ν
N
ν∈ZZ
15
and hence
√ N X  2N 
e T /N (ω) =
∆ 2π δ ω − µπ .
0
T0 T0
µ∈ZZ |{z}
Nyquist-frequency

Hence, using the so-called discrete Fourier transform

 
 1 NX
−1
2π

def
{xj }j∈{0,...,N −1} 7 → xek = √
− xj ei k N j (2.16)
 N j=0 
k∈{0,...,N −1}

we get  
e 2π T0 2π
f k ≈√ xek( mod N ) if k ≤Ω
T0 2π N T0
for the sampling values
 
T0
xj = f j , j ∈ {0, . . . , N − 1} .
N
Since
N −1
X 2π 1 − ei 2π m
ei k N
m
= 2π = 0 ∀ m ∈ {1, . . . , N − 1} , (2.17)
k=0 1 − ei N m

the inverse of the transformation (2.16) is

( )
def1 NX−1
2π
{xek }k∈{0,...,N −1} 7−→ xj = √ xek e−i k N j . (2.18)
N k=0 j∈{0,...,N −1}

DRAFT, October 17, 2007

15
See, e.g., (Lücke, musi, Anhang A.1).
42 CHAPTER 2. QUANTUM ALGORITHMS

Especially for N = 2n we define the quantum Fourier transform F̂n by

X def X n
F̂n x(b) |bi = xe(b) |bi ∀ (x0 , . . . , x2n −1 ) ∈ C2 , (2.19)
n n
b∈{0,1} b∈{0,1}

where16 
def
x(b) = xI(b) 
def ∀ b ∈ {0, 1}n .
xe(b) = xeI(b) 
n
Obviously, F̂n is a linear operator on C2 and, therefore,
!
  1 2π
x(a) = δa,b ∀ a ∈ {0, 1} n
=⇒ xe(a) = √ n ei I(a) 2n I(b) ∀ a ∈ {0, 1} n
(2.16) 2
implies17
√ X 2π
2n F̂n |bi = ei I(a) 2n I(b) |ai (2.20)
n
a∈{0,1}
X O n  
2π
n−ν a I(b)
= ei 2 ν 2n
|aν i ∀ b ∈ {0, 1}n .
a∈{0,1}n ν=1

The latter implies18

n  
1 O i 2π 2−ν I(b)
F̂n |bi = √ |0i + e |1i ∀ b ∈ {0, 1}n . (2.21)
2n ν=1
Thanks to
n−µ−ν
ei 2π bµ 2 = 1 for µ ≤ n − ν
we may rewrite this as
n  Pn 
1 O i 2π bµ 2n−µ−ν
F̂n |bi = √ n |0i + e µ=n−ν+1 |1i ∀ b ∈ {0, 1}n (2.22)
2 ν=1
or as
n Yν  !
1 O −(α−1) b
F̂n |bi = √ n |0i + ei 2π 2 (n−ν)+α
|1i ∀ b ∈ {0, 1}n (2.23)
2 ν=1 α=1

showing, by the way, that F̂n is isometric.

Exercise 11 Using (2.23), show that the n-qubit network

DRAFT, October 17, 2007
16
Recall the definition of I(b) in (1.6).
17
Note the ordering of
def
⊗nν=1 χν = χ1 ⊗ . . . ⊗ χn .

18
An import special case is F̂n |0i = ÛH⊗n |0i .
2.2. FACTORING LARGE INTEGERS 43

H ŝ2 ··· ŝn−1 ŝn

s ··· H ··· ŝn−2 ŝn−1 ···
.. .. m
. .
s s ··· H ŝ2
s s ··· s H

with

  |b1 i |bn i
def 1 0 .. ..
ŝν = i 2π/2ν , . m
. ,
0 e
|bn i |b1 i

implements the quantum Fourier transform.19 Moreover, using Corollary 1.2.3,

show that

ŝν Ĉν h Ĉ −1 h
ν
≡
s s s δν

holds for δν = −π/2ν and Ĉν = R̂3 (δν ) .

Remarks:
1. If the crossings are ignored then the above implementation of the quan-
tum Fourier transform uses n/2 SWAP gates, n Hadamard gates and n2 /2
Λ1 (ŝν ) gates.
2. Since
X D E
x
eI(a) = xI(b) a F̂n b
(2.19)
b∈{0,1}n D E
X n
= xI(b) F̂n−1 a b ∀ a ∈ {0, 1} ,
b∈{0,1}n

the above network implementation for F̂n yields a nice factorization of the
matrix corresponding to the discrete Fourier transformation. This factoriza-
tion is the core of the radix-2 version of the fast Fourier transform (FFT)
algorithm.20

DRAFT, October 17, 2007

19
Note that
1

√ |0i + ei π b |1i = ÛH |bi ∀ b ∈ {0, 1} .
2
20
See, e.g., (Brigham, 1974; Nussbaumer, 1982).
44 CHAPTER 2. QUANTUM ALGORITHMS

Exercise 12 Prove the identities21

√
2 Fn′ (0, b2 , . . . , bn ) = +Fn′ −1 (b2 , . . . , bn′ , 0, bn′ +1 , . . . , bn )
Pn′
2−ν
+ei 2π b
ν=2 ν Fn′ −1 (b2 , . . . , bn′ , 1, bn′ +1 , . . . , bn ) ,
√
2 Fn′ (1, b2 , . . . , bn ) = +Fn′ −1 (b2 , . . . , bn′ , 0, bn′ +1 , . . . , bn )
Pn′
2−ν
−ei 2π b
ν=2 ν Fn′ −1 (b2 , . . . , bn′ , 1, bn′ +1 , . . . , bn )
for the partial discrete Fourier transform Fn′ defined by
 ′

X n
X
def n′ −ν−ν ′
(Fn′ x) (b) = x(b′1 , . . . , b′n′ , bn′ +1 , . . . , bn ) exp+i 2π bν bν ′ 2 
′ n′ ν,ν ′ =1
b ∈{0,1}
for n′ < n ∈ IN and
def def
(Fn x) (b) = xe(b) , (F0 x) (b) = x(b) .

2.2.3 Quantum Order Finding

As already mentioned at the end of 2.2.1, to find the order (2.14) of a given integer
x ∈ {2, . . . , N − 1} , Shor suggested to exploit the effectively implementable22 state
2L
def 1 2 X−1  
ΨShor = L
F̂2L |ai 2L ⊗ |xa (modN )iL
2 a=0
2L
1 2 X−1 i a 2π
= 2L
e 22L c |ci2L ⊗ |xa (modN )iL .
(2.19),(2.16) 2 a,c=0
where n o
def
L = min l ∈ IN : N ≤ 2l .
The essential point is the following:

′
xa (modN ) = xa (modN ) ⇐⇒ a′ = a mod r .
Therefore,
D E 2
def
p(a, c) = |ci2L ⊗ |xa (modN )iL ΨShor
2
1 X 2π n o
i a′ c
= e 22L ∀ a, c ∈ 0, . . . , 22L − 1 ,
24L a′ ∈Ma

DRAFT, October 17, 2007

21
Let us point out that
Pn′ I(b2 ,...,b ′ )
n
bν 2−ν iπ
ei 2π ν=2 =e ′
2n −1 ∀ b2 , . . . , bn′ ∈ {0, 1} .

22
See also (Coppersmith, 1994), in this connection.
2.2. FACTORING LARGE INTEGERS 45

where  n o 
def ′ 2L ′
Ma = a ∈ 0, . . . , 2 − 1 : a = a mod r ,

is negligible unless all the phases

brc
2π (mod2π) ∈ [0, 2π)
22L
with n j  ko
b ∈ 0, . . . , 22L − 1 − min Ma /r ,
are predominantly almost the same, i.e. (assuming L sufficiently large)
unless  
rc r
[0, 2π) ∋ 2π 2L (mod2π) = O 2L .
2 2
23
The latter means that
 
c d 1
2L
− = O 2L
2 r 2
holds for some integer d .
More precisely, one can show:24
A projective measurement of ΨShor w.r.t. the computational basis is likely
to find the first 2L-qubit register in a state |ci2L with

c d 1
− ≤ , gcd(d, r) = 1 (2.24)
22L r 2 r2
being fulfilled for integer d .

If we have found a fraction c/22L fulfilling (2.24) for some (unknown) integer d
with gcd(c, d) = 1 , then r may be efficiently determined using the continued fraction
algorithm:25

The motivation for the definition (2.7) — given n0 , n1 ∈ IN — is the observation

that $ % $ %!
nν nν 1 nν
= + nν − nν+1 ∀ nν , nν+1 ∈ IN
nν+1 nν+1 nν+1 nν+1
| {z }
remainder
DRAFT, October 17, 2007
2L
23
If, by chance, r divides 22L then b runs from 0 to 2 r − 1 and, therefore, an adaption of (2.17)
shows that p(a, c) vanishes exactly unless 2r2Lc = d holds for some integer d .
24
See (Shor, 1997, Section 5) for details and Appendix A.3 for an improved search algorithm.
Note that r ≤ ϕ(N ) < N .
′
25
Otherwise the prescription will yield an integer r′ that fails the test xr = 1(modN ) almost
certainly. Then the whole procedure has to be repeated.
46 CHAPTER 2. QUANTUM ALGORITHMS

since, then, we have

 
n0 n0 1
= + n 
n1 n1 1
n2
 
n0 1
= + jn k
n1 1
+1
n2 n2
n3

 
n0 1
= + jn k
n1 1
+ j k1
n2 n2 1
+
n3
( nn43 )
etc.,
hence the continued fraction expansion26
 ν−2 
n0 n0 X 1 | 1 |
= + j
nµ
k +
nν−1
∀ ν ∈ {1, . . . , s} (2.25)
n1 n1 µ=1 nµ+1 nν

with s given by (2.8).

Remark: Formally, i.e. without specification of the aµ and bµ , finite continued

fractions27
a1 | a2 | aν |
b0 + + + ... + ,
| b1 | b2 | bν
are recursively defined by
a1 | def a1
b0 + = b0 +
| b1 b1
and
a1 | a2 | aν+1 | def a1 | a2 | aν |
b0 + + + ... = b0 + + + ... .
| b1 | b2 | bν+1 | b1 | b2 aν+1
bν + bν+1

Due to
nν+1 6= 0 =⇒ nν+2 < nν+2 < nν+1
(2.7)
j k
ns−1 ns−1
s must be finite and fulfill the equation ns
= ns
. Since
gcd (nν−1 , nν ) 6= 0 ⇐⇒ gcd (nν+1 , nν )
we see that gcd(n0 , n1 ) divides ns and that ns , dividing ns−1 , also divides n0 and
n1 . Hence (2.8) holds, indeed.

d
Given c, L ∈ IN fulfilling (2.24), the set of possible fractions c
is strongly re-
stricted by the following lemma.
DRAFT, October 17, 2007
26
Note that the continued fraction expansion does not change if n0 and n1 are replaced by
n′0 = p n0 and n′1 = p n1 , where p ∈ IN .
27
See, e.g., (Perron, 1954; Perron, 1957) or (Brezinski, 1991) for the general theory of continued
fractions. See also (Baladi and Vallee, 2003).
2.2. FACTORING LARGE INTEGERS 47

Lemma 2.2.3 Given n0 , n1 , d, r ∈ IN fulfilling

n0 d 1
− < ,
n1 r 2 r2

using definition (2.7) and (2.8), we have28

 t
d n0 X 1 |
= + j
nµ
k
r n1 µ=1 nµ+1

for some t ∈ {1, . . . , s − 1} .

Proof: See (Nielsen and Chuang, 2001, Theorem A4.16).

Therefore, using (2.7) for

n0 = c , n1 = 22L
and determining — for t = 1, 2, . . . , s — the numbers At , Bt ∈ IN characterized by
  t
n0 X 1 | At
+ j
nµ
k = gcd(At , Bt ) = 1 ,
n1 µ=1 Bt
nµ+1

we have
r = Bt for some t ≤ s .
Also these At , Bt can be efficiently determined via Euclid’s algorithm:

If
a1 | a2 | an |
b0 + + + ... + ,
| b1 | b2 | bn
is well-defined then
ν+1 ν−1
X aµ | X aµ | aν bν+1 |
b0 + = b0 + + ∀ ν = 1, . . . , n − 1
µ=1 | bµ µ=1 | bµ | bν bν+1 + aν+1

and, therefore,
a1 | a2 | aν | Aν (a1 , . . . , aν ; b0 , . . . , bν )
b0 + + + ... + = ∀ ν = 1, . . . , n ,
| b1 | b2 | bν Bν (a1 , . . . , aν ; b0 , . . . , bν )
DRAFT, October 17, 2007
28
Note, however, that

Xt t−1
1 | X 1 | 1 |
at = 1 =⇒ = + ∀ a0 , . . . , at ∈ IN .
µ=1
| aµ µ=1
| aµ | at−1 + 1
48 CHAPTER 2. QUANTUM ALGORITHMS

if the Aν and Bν are recursively defined by29


def 
Aν = bν Aν−1 + aν Aν−2
def for ν = 1, 2, . . . , n ,
Bν = bν Bν−1 + aν Bν−2 

where
def def def def
A−1 = 1 , A0 = b 0 , B−1 = 0 , B0 = 1 .
Fortunately, these definitions also imply
)
aµ = 1 ∀ µ ∈ {1, . . . , n}
=⇒ gcd (Aν , Bν ) = 1 ∀ ν ∈ {1, . . . , n} .
bµ ∈ IN ∀ µ ∈ {0, . . . , n}

Outline of proof:

Aν+1 Bν = bν+1 Aν Bν + Aν−1 Bν

= bν+1 Aν Bν + Aν−1 Bν−2 + bν Aν−1 Bν−1

and the corresponding equation with A, B interchanged imply

Aν+1 Bν − Bν+1 Aν = − (Aν Bν−1 − Bν Aν−1 ) .

By induction, starting from

A0 B−1 − B0 A−1 = −1 ,

this gives
ν+1
Aν Bν−1 − Bν Aν−1 = (−1) ∀ ν ∈ {0, . . . , n} . (2.26)
In case
Aν = dν cν , Bν = eν cν
the positive integer cν would divide the l.h.s. of (2.26), hence also the r.h.s. The
latter, however, is only possible for cν = 1 .

Example30 N = 899 , L = 10 : If, for instance, the projective measurement gives

|267137i2L for the first 2L qubits, then we get31
c
= [0, 3, 1, 12, 2, 1, 1, 1, 26, 1, 1, 22, 2]
22L
def 1| 1| 1 | 1| 1| 1| 1| 1 | 1| 1| 1 | 1|
= 0+ + + + + + + + + + + +
| 3 | 1 | 12 | 2 | 1 | 1 | 1 | 26 | 1 | 1 | 22 | 2
DRAFT, October 17, 2007
29
Thus
(bν bν+1 + aν+1 ) Aν−1 + (aν bν+1 ) Aν−2 = bν+1 Aν + aν+1 Aν−1
and similarly for the Bµ .
30
Compare (Rosé et al., 2004, Section III).
31
In Maple, after the command ”with(numtheory);” this result will be produced by the com-
mand ”convert(267137/220 , cfrag);” and the subconvergent [0, 3, 1, 12, 2, 1, 1, 1] can be evalu-
ated by the command ”nthconver([0,3,1,12,2,1,1,1],7);”.
2.2. FACTORING LARGE INTEGERS 49

and succeed with t = 7 :

107 d
[0, 3, 1, 12, 2, 1, 1, 1] = = ,
420 r
i.e. testing r = 420 , we get32 the factors
899
gcd(11210 + 1, 899) = 29 , gcd(11210 − 1, 899) = 31 = .
29
If, for instance, we unfortunately measure c = 801411 then we get
801411
= [0, 1, 3, 4, 7, 1, 80, 1, 1, 7, 6]
220
with the subconvergent
 
107 321
[0, 3, 4, 7, 1] = = .
140 420
In this case, although the condition

c d 1
− ≤
22L r 2 r2

(but not gcd(d, r) = 1) is fulfilled (with d = 321) r = 420 will presumably not be
detected and the whole procedure will be repeated, maybe with a different random
value for x .

DRAFT, October 17, 2007

32
E.g., by the Maple commands ”gcd(112 10,899);” and ”gcd(112 10,899);”.
50 CHAPTER 2. QUANTUM ALGORITHMS
Chapter 3

Physical Realizations of Quantum

Gates1

A set of necessary conditions to be fulfilled for the physical implementation of quan-

tum computation is given by DiVincenzo’s checklist (DiVincenzo, 2000):
1. Qubits have to be well characterized and scalable.2
2. The standard states |0, . . . , 0i must be preparable.3
3. The duration of a gate operation must be much smaller than the decoherence
time.
4. A universal set of gates must be implementable.
5. The qubits must be measurable in order to be able to ‘read out the result’ of
a quantum computation.
Remark: For quantum communication the following two requirements have to be
added:

• It must be possible to convert static qubits into flying qubits (typically photons).
• It must be possible to protect flying qubits against decoherence.

3.1 Quantum Optical Implementations

Optical systems currently constitute the only realistic proposal for long-distance quan-
tum communication and underly implementations of quantum cryptography.
(Knill et al., 2001)
DRAFT, October 17, 2007
1
See the special volume Fortschr. Phys. 48 (2000) No. 9–11.
2
Thus, 1-photon realizations of n-qubit systems should be considered as qudits with d = 2n
rather than n-qubit systems proper.
3
This is still difficult for n-photon systems.

51
52 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

3.1.1 Photons
In the Coulomb gauge the free electromagnetic field E(x, t) , B(x, t) is given in
the form
∂
B(x, t) = curl A(x, t) , E(x, t) = − A(x, t) ,
∂t
 ∗
A(x, t) = A(+) (x, t) + A(+) (x, t)
by some complex vector potential
 
Z 2
X dk
A(+) (x, t) =  ǫj (k) fˇj (k) e−i(c|k|t−k·x) q ,
j=1 2 |k|

where, for every k 6= 0 , the vectors ǫj (k) form a

( )
k
right handed orthonormal basis ǫ1 (k), ǫ2 (k),
|k|

and thus guarantee div A(+) = 0 .

Remark: We use SI conventions; see Apendix A.3.3 of (Lücke, edyn).

In the Heisenberg picture of the quantized theory the classical fields E(x, t) and
B(x, t) have to be replaced by corresponding observables on the state space Hfield ,
i.e. by operator-valued (generalized) functions Ê(x, t) and B̂(x, t) to be interpreted
in the following way:

If Φ ∈ Hfield is a sufficiently well-behaved (and kΦk = 1) then

D E D E
Φ Ê(x, t) Φ resp. Φ B̂(x, t) Φ

is the expectation value for E(x, t) resp. B(x, t) in the Heisenberg

state (corresponding to) Φ .

Up to unitary equivalence these observables are given by

∂
B̂(x, t) = curl Â(x, t) , Ê(x, t) = − Â(x, t) ,
∂t
 †
(+) (+)
Â(x, t) = Â (x, t) + Â (x, t) ,
3.1. QUANTUM OPTICAL IMPLEMENTATION 53

where4
 
q Z 2
X
(+) dk
ǫj (k) âj (k) e−i(c|k|t−k·x) q
def
 (x, t) = (2π)−3/2 µ0 h̄c  (3.1)
j=1 2 |k|
and the âj (k) are annihilation operators5 fulfilling the commutation relations6
 †
[âj (k), âj ′ (k′ )]− = 0 , [âj (k), âj ′ (k′ ) ]− = δjj ′ δ(k − k′ ) (3.2)
on a suitable dense subspace D0 of the Hilbert space H containing a cyclic nor-
malized vacuum state vector Ω characterized (up to a constant phase factor) by
âj (k) Ω = 0 (3.3)
(in the distributional sense). This also fixes the inner product on H .

The operators â of the form

X Z  ∗
â = âj (k) fˇj (k) dk , fˇ1 , fˇ2 ∈ L2 (IR3 ) (3.4)
j=1,2

with7 h i
â , ↠= 1̂ (3.5)
−
characterize modes of the quantized electromagnetic field corresponding to the clas-
sical complex vector potentials
* † +
(+) (+) †
A (x, t) = Â (x, t) Ω â Ω
*   +
(+)
= Ω Â (x, t) , ↠Ω
−
 
q Z 2
X dk
= (2π)−3/2 µ0 h̄c  ǫj (k) fˇj (k) e−i(c|k|t−k·x) q .
(3.4),(3.1),(3.2) j=1 2 |k|
(3.6)
DRAFT, October 17, 2007
4
Thanks to the special choice of the factor in front of the integral we get the desired expression
Z  
1 1
Ĥ = ǫ0 : Ê(x, t) · Ê(x, t) : + : B̂(x, t) · B̂(x, t) : dx
2 µ0
for the Hamilton operator, characterized (up to an additive constant) by
∂ (+) i h (+)
i
 (x, t) = Ĥ ,  (x, t) .
∂t h̄ −

5
See (Mizrahi and Dodonov, 2002), however.
6
Of course, the notation † includes the requirement
 † 
hΦ | âj (k) Φ′ i = âj (k) Φ Φ′ .

7
Condition (3.5) is equivalent to â† Ω = 1 .
54 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

With these modes the subspaces H(n) ⊂ H of n-photon state vectors may be defined
recursively by
def
H(0) = {λ Ω : λ ∈ C}
and n o
def
H(n+1) = ↠Φ(n) : Φ(n) ∈ H(n) , â mode for n = 0, 1, 2, . . . .

Physical characterization of n-photon states: n (ideal) detectors but no more

can be made fire by an incoming n-photon state.

Modes âν , âµ are called orthogonal, iff the states â†ν Ω , ↵ Ω are orthogonal, i.e. iff
[âν , âµ ]− = 0 .

Fortunately, already classical electrodynamics tells us how photons are affected

by passive linear optical components:

The change of the mode â of a photon, caused by a passive linear opti-

cal component, is such that the corresponding complex vector potential
changes as predicted by classical electrodynamics.

3.1.2 Photonic n-Qubit Systems

Single-Photon Realization
For every n ∈ IN , using sufficiently many beam splitters a single-photon state ↠Ω
can be changed into a coherent superposition8
X
λb â†b Ω
b∈{0,1}n

of (essentially) orthogonal 1-photon states

def
|bi = a†b Ω

which may be chosen as elements of the computational basis of a (simulated) n-qubit

system.

Remarks:

1. Since the single qubits of the 1-photon realization cannot exist independently
from each other we should better call the system a qu2n it system.
DRAFT, October 17, 2007
8
For the special case n = 3 an example is illustrated in Figure 3.1.
3.1. QUANTUM OPTICAL IMPLEMENTATION 55

qqqqqqqqq
.....
.. ..........
. √1 |0, 0, 0i
..... 8
.....
. . .......
..... ........... .
.... ........
..... .....
..... qqqqqqqqqq √1
.....
.....
................. ..... ..........
.
8
|0, 0, 1i
... .
.
. ..
.................... .
..... .........
..... .....
.....
qqqqqqqqq √1
.....

........ .
.
.
....
... .....
.....
.....
.....
.....
...
..........
.
8
|0, 1, 0i
..
..... .....
..... ........
......
. .
.. . . ..
..... ..... .......... .
..... .....
..... .....
..... .....
qqqqqqqqqq
.....
.....
....... . .....
. . .
.....
. .....
.. ..........
. √1 |0, 1, 1i
.........
..... .
......
. 8
..... ........
..... ............ .
.....
.....
.....
qqqqqqqqqq √1
.....
.....
.....
..... .....
.....
..........
.
8
|1, 0, 0i
..... .....
.....
..... .. .......
..... .................. .
..... .. ...
..... ..... .........
..... ..... .....
.....
.....
.............. qqqqqqqqqq
.....
. ..........
. √1 |1, 0, 1i
..... ..
..... ....... 8
..... ..
..... ........
..... ............. .
.....
.....
.....
.....
.....
..... qqqqqqqqqq
....
........... √1 |1, 1, 0i
..... ..... 8
..... .....
..... .........
.
..... ............. .
.....
.....
.....
qqqqqqqqqq
.....
... ..........
. √1 |1, 1, 1i
8

(3)
Figure 3.1: Preparation of Φ0 in a single-photon realization.

2. Cascades of beam splitters may also be used to implement approximate mea-

surement of the number of photons:9
E.g., replace the single-photon input in Figure 3.1 by a 2-photon state and
direct the output rays into separate ideal detectors. Then the probability that
exactly two of these detectors ‘fire’ is 15/16 (≈ 94 %).

For such a choice of computational bases all unitary transformations can be (essen-
tially) effected by linear optical components. Thanks to Theorem 1.2.1 it is sufficient
to show this for n = 1 :
Assume, for instance, that that |0i and |1i describe horizontally polarized (almost
monochromatic) photons. Then the 1-qubit gates may be implemented by linear
optical elements corresponding to Jones matrices Û in the following way:
ppppp p
λ1 |0i ppppp pp pp λ′1 |0i
ppppp p p ppp p
..... ........................... ..... .....
D̂ ....
π ..... ....................... ..... ..... ....... ppppp p ppp
p ............
......... D̂− 2 ....
π ..... ..... ..................... .....
2 ppp ppppp
..
.
...
.
...
.
.. ..
....... ......
......... ... ...
...
...
.
...
.
ppppp
...
′ p ppp
λ |1i 2...... ppppp
.
... χ χ pp pp λ′2 |1i
ppppp ..p..p. ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ....................... .....
. ...... .......
..... ..................... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... .....
ppppp .............. Û ..
........... p p
ppp
ppp pp ppp

All unitary transformations of the polarization state of a photon (with almost sharp
momentum) can be (almost accurately) performed by proper use of only λ/2-blades
DRAFT, October 17, 2007
9
See (Bartlett et al., 2002) for details. See also (Haderka et al., 2003; Waks et al., 2003).
56 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

and polarization-dependent phase shifters.

General remark: Instead of performing the n-qubit transformations

Û(n) one may choose a fixed n-qubit transformation V̂(n) perform the
n-qubit transformations Û(n) V̂(n) and interprete the result w.r.t. the
n o
def
new computational bases |bi′ = V̂(n) |bi : b ∈ {0, 1}n . Usually, in
practice, this freedom is tacitly made use of.

Exercise 13 Consider the single-photon realization of a 2-qubit system with:

(
upper path and
âI(0,0) =
b
vertical polarization ,
(
upper path and
âI(0,1) =
b
horizontal polarization ,
(
lower path and
âI(1,0) =
b
vertical polarization ,
(
lower path and
âI(1,1) =
b
horizontal polarization .

a) Show that the CNOT gate may be implemented by placing a 90◦ polarization
rotator into the lower path.

b) Show that the TCNOT gate may be implemented by applying a polarization

beam splitter — reflecting the vertically polarized components and transmit-
ting the horizontally polarized components — in the following way:
qqqqqqqqqqq ppp
α00 |0, 0i + α01 |0, 1i .......... ..
....... ..... qqqq ppp ppp p ppppppppp qqqqqqqqqqqqq .....
...
..........
....... α00 |0, 0i + α11 |0, 1i
.....
..... pppp pp
.....
ppppp
ppppp ......
.....
pp .....
.....
.
..
pppp pp p ppppp
.....
..... .....
.....
ppppppp
.....
..... ........
........
ppppp
ppppp .....
........
..... ......... pp ppp
p p
p pp
.....
ppppp .....
..... .....
pp pp
.....
.
.
....ppppp
...
p .....
.....
.
..
p p p
qqq pp p qqqqqqqqqqqqqqp p .....
.....
qqqqqqqqqqq ppppppp ppp pp
.......... ........ .....
α10 |1, 0i + α11 |1, 1i ....... q .. ..........
....... α10 |1, 0i + α01 |1, 1i .

The single-photon ‘realization’ has a serious disadvantage spoiling the eventual

speed-up of quantum computation:

The number of optical devices required for the single-photon simula-

tion of n-qubit systems grows exponentially with n (Cerf et al., 1998;
Kwiat et al., 2000).

Therefore, we will consider only many-photon realizations in the following.

3.1. QUANTUM OPTICAL IMPLEMENTATION 57

Multi-Photon Realization
For given n ∈ IN we may choose a fixed10 set
n o
âν,j : ν ∈ {1, . . . , n} , j ∈ {0, 1}

of 2n pairwise orthogonal modes and consider the n-photon states

def
|bi = â†1,b1 · · · â†n,bn Ω ∀ b ∈ {0, 1}n

as elements of the computational basis of a true n-qubit system.11

Here, while the 1-qubit gates may still be easily implemented by linear optical
components, the physical realization of universal 2-qubit gates is quite a technolog-
ical challenge.12

Fock Realization
Another possibility is to choose some fixed set {â1 , . . . , ân } of pairwise orthogonal
modes and consider the states

|bi = |biF ∀ b ∈ {0, 1}n , (3.7)

where
1  ν 1  ν n
def
|ν1 , . . . , νn iF = √ â†1 · · · â†n Ω ∀ ν1 , . . . , νn ∈ ZZ+ , (3.8)
ν1 ! · · · ν n !
as elements of the computational basis of the n-qubit system.
The states â†b Ω of the single-photon realization for n-qubit systems form the
subset n o
|biF : b1 + . . . + bn = 1
of the computational basis of the Fock realization for 2n -qubit systems using the
2n modes
âI(b)+1 = âb , b ∈ {0, 1}n .
Obviously, for n=1 the Fock realization does not coincide with the single-photon
realization. Actually, 1-qubit gates like the Hadamard gate cannot be implemented
by linear optical components, since |0iF represents the vacuum state. Nevertheless,
the Fock realization has certain advantages as, e.g., those to be discussed in 3.1.4.
DRAFT, October 17, 2007
10
Recall the above general remark, however.
11
Of course, for n = 1 the n-photon realization coincides with the single-photon realization.
12
Concerning recent progress in detector technologies see (Rosenberg et al., 2005).
58 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

3.1.3 Nonlinear Optics Quantum Gates

For both the n-photon and the Fock realization of n-qubit systems the (universal)
CPHASE gate Λ1 (Ŝπ )

s s
≡
π H h H

is nonlinear in the sense that — contrary to the action of linear optics components
— the modes are are not transformed independently of each other.

If a non-linear sign gate NS is available,13 i.e. an optical one-way gate acting

according to14
   
α (â∗ )0 + β (â∗ )1 + γ (â∗ )2 Ω .......................... NS ......................... α (â∗ )0 + β (â∗ )1 − γ (â∗ )2 Ω

then the CPHASE gate can be easily implemented by proper use of linear optical
components like the Hadamard beam splitters (with deflecting mirrors) charac-
terized in Figure 3.2.

|0i...... √1 |0i + √12 |0i

..... ............. .........
2
..... ..................... ..... ..... ..................... .....
. ..... .....
..... .... ....
............ ....... .......
........ ........
..... . .
.... ....
.....
..... .....
..... .... ..
..
. .
..... .... ..
.
.......................... .......................... ....
..........................
.....
..........................
.......................... ..........................
..........................
.......................... ........
. .....
..... ....
.....
..... .....
.....
..... .....
........
......
....
√1 |1i |1i .....
............. ......
.........
... − √1 |1i
......... ..................2..... ..
..... .................. ........
..... 2
..... ..................... .....
...

Figure 3.2: Hadamard beam splitter.

One such implementation,15 suggested in (Ralph et al., 2002), is sketched in Figure

3.3. The essential idea is to exploiting two-photon interference (see Section 5.2.3
of (Lücke, nlqo)) at two Hadamard beam splitters forming a balanced Mach-
Zehnder interferometer as sketched in Figure 3.4.

One would like to realize the necessary NS gates by means of optical nonlinear-
ities. Unfortunately, sufficiently strong nonlinearities of crystals are accompanied
DRAFT, October 17, 2007
13
See (Sanaka et al., 2003) for an experimental realization.
14
The gate is non-linear in the sense that its action cannot be reduced to a linear transformation
of the modes â .
15
The modes âν,µ are assumed to differ only by vertical translation and to describe photons with
(almost) sharp momenta.
3.1. QUANTUM OPTICAL IMPLEMENTATION 59

â.....1,0
.................... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ...................
......... .........

â1,1
.. ....
..... .......... ...... ..... ..... ..........
..... ....
.........
......... ..... ..... ............ ...... ..... NS ..... ..... ........................ ..... ..........
..... ....
.........
.......... ..... ..... ................ .....
..... ..... . ..... ..... .
..
.......................................
.......................................
....................................... ..
.......................................
.......................................
.......................................
..... ..... . ..... ..... .
â 2,1
.......................................
.......................................
.......................................
..... ....
.......................................
.......................................
.......................................
..... ....
. . ..
..
..... .......... ..... ..... ..... ..... ..... .......... ..... ..... ................... .....
........ NS ........
.. .
..... ..... ............. ..... .......
.. .......... ..... ..... ................. .....
........

â2,0
..... .................... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... .........................
..........

Figure 3.3: Optical implementation of a CPHASE gate.

α |0i
......
α |0i
..... ............... ...... ..... ........ ...... ..
. .......... ..... .....
..... .....
..... ....
.......
............ .....
.....
........
. ..... .....
..... .....
..... . .....
.... .....
..... .....
..... .....
..... ..... ..... .....
..........................
.......................... .. ..........................
......................... ..
..........................
.......................... .... .... .... ....
..........................
.........................
. ..... . .....
.... ....
..... .....
..... .....
.
..
. ..... ..
.. .....
. .
........... .....
.....
.........
β |1i
...... ...
.....
.....
.......
.....
......
.... β |1i
......... ................ ..
..... ............... .. ...

Figure 3.4: Action of a balanced Mach-Zehnder interferometer.

by too strong absorption and therefore are not suitable. A way out may eventu-
ally be provided by electromagnetically induced transparency16 (EIT), discussed in
Section 8.3.2 of (Lücke, nlqo).

Using beam splitters characterized by

|0i
..... ..... ................. ......
cos ϑ |0i
................ ..... ..... .....
−e+i ϕ sin ϑ |0i
................ ..... ..... .....
...... .... ..... ....... ..... .......
.....
..... .....
...... ... ...
............ ........... ...........
.. ..... .....
..... . .
.... ....
.....
..... .....
..... ..... .....
..... . .
.... ....
..... .... .
....
ϑ .
.....
..... ϕ ϑ .....
.........
.....
ϕ
..... .. .....
...
.....
..... .....
.....
..... .....
..... .. .....
..
.....
................. . . ..
.
......... .................
....
.
..... e−i ϕ sin ϑ |1i |1i
.. ....................... ..... ..... .....
..... ..... ................. .......
.....
.....
....
..
..... cos ϑ |1i
. ....................... ..... ..... .....
......

indeterministic17 NSx -gates, i.e. gates acting according to

       
† † 2 † † 2
α 1̂ + β â + γ â Ω 7−→ α 1̂ + β â + x γ â Ω
DRAFT, October 17, 2007
16
See (Ottaviani et al., 2005) and concluding discussion of (Munro et al., 2005a). See also
(Munro et al., 2005b) concerning the use of ‘weak’ cross Kerr nonlinearities.
17
A gate is called indeterministic if it acts correctly with nonzero probability < 1 not depending
on the input state and if, after its action, it is known whether the action was correct or not. An
indeterministic gate is called near-deterministic if its probability of success is near to 1.
60 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

NSx
..... ......................... ..... ..... ..... ...
ϕ4 ..... ..... ..... .............................. ..... ..... . ..... . .....
..
....... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ...................... ..
.... .....
.....
..... .....
.....
....... ....
.........
...... ............
..... ..... .
.
..... ....
.
.... .
′ † .....
..... .....
.....
(â ...)....... Ω . .........
......... .
.....
..................
.......... .....
.....
.....
ϑ .
2 . ..... ..... ..
...
ϕ 2 ..... ..
. ...
....
D
.
1
..... .... .
...... ..... ......
................. ..... . ...............
..... .
.... ..
.. .
.....
..... ....
.
................ .....
..... .......... ..... .....
.....
..... ....... ..
..... .....
.
.....
..... .... ..... .....
.
..... .... ..... .....
..... .
..... . .... ..... .....
ϑ 1
........
...
.....
ϕ 1 ϑ .
3 ..... .... ..... ... ϕ
3
..... . ..
.... .....
..... ..... .....
..... ...
. ...... .....
............... ............... .....
...... .
.. ..... ........
..... .
.... ........
...... .
..... ..... ..... ....
..... .....
.
..................
..... 2
.....
D
.....
..

Figure 3.5: An indeterministic NSx -gate

if successful, can be implemented18 as sketched in Figure 3.5. According to (Knill et al., 2001,
Fig. 1) these gates act successfully iff a single-photon state is detected by D1 and
the vacuum state is detected by D2 . For
   
   ◦
 ϕ1 0
ϑ1 +22.5 ϕ   0 
     2  
 ϑ2  =  65.5302◦  and  = 
 ϕ3   0 
ϑ3 −22.5◦
ϕ4 180◦
we have x=-1 and the probability of success is 0.25. For
   
   ◦
 ϕ1 88, 24◦
ϑ1 +36, 53 ϕ   −66, 52◦ 
     2  
 ϑ2  =  62.25◦  and  = 
◦  ϕ3   −11, 25◦ 
ϑ3 −36.53
ϕ4 102, 24◦
we have x=i and the probability of success is 0.18082.

3.1.4 Linear Optics Quantum Gates19

Indeterministic Gates
Fortunately, using photonic memory and quantum teleportation, (deterministic)
CPHASE gates may by implemented using indeterministic ones:
DRAFT, October 17, 2007
18
See also (Rudolph and Pan, 2001; Ralph et al., 2002; Hofmann and Takeuchi, 2002) and espe-
cially (Gilchrist and Milburn, 2002) for other possibilities.
19
See (Dowling et al., 2004).
3.1. QUANTUM OPTICAL IMPLEMENTATION 61

The essential observation, due to (Gottesman and Chuang, 1999), is the equivalence

s ⌢
/
 B
s ⌢
/
Φ+
h π s s
 ≡ ,
(3.9)
h π π π
Φ+
s ⌢
/
B s ⌢
/

where
s H
B ≡
h

is the inverse of the Bell network and Φ+ the state defined in 1.2.2. This equivalence
is obvious from the discussion of quantum teleportation in 1.2.2 and, thanks to

s s
h s ≡ s h

π π π

implies the equivalence20

s ⌢
/
 B
s ⌢
/
Φ+
s π h π s
 ≡
π h π π π .
Φ+
s ⌢
/
B s ⌢
/

Therefore:
If the auxiliary 4-qubit state
 
(4) def
ΦB = 1̂ ⊗ CPHASE ⊗ 1̂ (Φ+ ⊗ Φ+ )
is available then the (deterministic) CNOT gate can be replaced by ap-
propriately modified double quantum teleportation.
DRAFT, October 17, 2007
20
See also (Brukner et al., 2003) in this connection.
62 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

(4)
If ΦB can be stored for later use21 then it is sufficient to have a nondeter-
(4)
ministic gate producing ΦB with nonzero probability from standard input. One
such possibility, obviously, is to replace the deterministic CNOT gates in the above
(4)
characterization of ΦB by nondeterministic ones, if the latter are available.

In principle (Kim et al., 2001) , (deterministic) teleportation is possible by ex-

ploiting sum frequency generation of both type I and type II for a complete Bell
measurement (see, e.g., Section 3.2.2 of (Lücke, nlqo) for the explanation of sum
frequency generation). However, also indeterministic teleportation by only partial
Bell measurement — as, e.g. sketched in Figure 3.6 and explained by Exercise 14 —
may be useful for improving the probability of success, at least , of an indeterministic
CPHASE gate.

......
..... ........................ ..... ..... ..... ..... ................. ..... . .....
.
..... ..... ......
.
..... ....
..... .....
...
..... .. ..
.
........
.................................
..................................
.................................
... ...
..................................
.................................
..... ..
.....
..... .....
.
...
. ......
. .....
..... ........................ ..... .... ..... ..... ................. ..... . ....
......

Figure 3.6: Partial Bell measurement.

Exercise 14 Consider a Hadamard beam splitter for the orthogonal modes â1 â2 ,
i.a a beam splitter acting as
1 1
â1 7→ √ (â1 + â2 ) , â2 7→ √ (â1 − â2 ) .
2 2
Show that the beam splitter, applied to the Bell states
1   1  
Φ± = √ Ω ± â†1 â†2 Ω , Ψ± = √ â†1 ± â†2 Ω
2 2
in the corresponding Fock representation as sketched in Figure 3.6, acts as follows:
1  1 
Φ± 7−→ √ |0, 0iF ± |2, 0iF − |0, 2iF ,
2 2
F
Ψ01 = Ψ+ 7−→ |1, 0i ,
Ψ11 = Ψ− 7−→ |0, 1iF .

According to (3.9) and Exercise 14 a so-called CZ 1 gate, i.e. an indeterministic

4
CPHASE gate working with probability of success 1/4 , can be implemented in the
Fock realization as sketched in Figure 3.7 This gate succeeds if the (photon number
3.1. QUANTUM OPTICAL IMPLEMENTATION 63

..... ..... ........................ ..... ..... ..... ..... ..... ..... ..... .....
.......D1
..
...... .... .... .... .... .... .... .... ..................... . ....
..... .... ......
..... ...
.
....
..... ....
..... ...
..
...........
.......................................
.......................................
...
...
.. ...
.......................................
.......................................
.......................................
.......................................
..... ..
...
..... ....
....
.
..
..

( ..... ......................... ..... ..... ..... .....

.
.... ....
.... D
......
...... .... .... .... .... .... .... .... .................... . ....
u 2
. ..... ..
. .......
.
.
Ψ01 .
.
.
u
..... ........................ ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... .....
π ..... ..... ..... ..... ..... ..... ....................... .....

( ..... .......................... π ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... .....
π ..... ..... ..... ..... ..... ..... ........................ .....
.
.
Ψ01 .
.

..... ......................... ..... ..... ..... ..... ...

. .
... .
...
.
. .....
................ ......... .... ..................... . .....
u D2′
. .. ....... ..... ..
..... .
.. .......
..... ....
...
.....
. ....
..... .
..
....... ...
.......................................
.......................................
.......................................
.......................................
....................................... .
....... ...
..... ..
...
..... ....
....
.....
........ .
..... ..... ................ ..... ..... ..... ..... ..... ..... ..
..
.....
....
.... ......
...... .... .... .... .... .... .... .... ...................... . ....
D1′
...... ..
.......

Figure 3.7: Implementation of a CZ 1 gate

4

resolving) detectors D1 , . . . , D2′ indicate that the common state of the first two qubits
as well as that of the last two qubits is a single-photon state.22 The ancillary states

|0, 1i + |1, 0i â†1 + â†2

Ψ01 = √ = √ Ω
2 2
may be easily prepared using a single-photon source and a Hadamard beam split-
ter:23
â†1 Ω.......... . .... .... .................... .... ... 
........
.....
..... ... ........
.....
..... ...
.... 

.....
..... ....
.... 

.....
.....
..... ....
... 
.......................................
.......................................
.......................................
.......................................
........
. ...
. .....
.......................................
.......................................
. 
Ψ01
.
. ...
.... 

.
. .... 

Ω
. . . . . . . . . ..
.
.
. ....
....
...... .... ..................... .... ...
......

Near-Deterministic Quantum Gates

The construction sketched in Figure 3.7 can be generalized in the following way
using 4n ancillary qubits24 instead of only 4 (Knill et al., 2000; Knill et al., 2001):
DRAFT, October 17, 2007
21
See (Pittman and Franson, 2002) in this connection.
22
The dotted vertical lines indicate that the effect of the corresponding CPHASE gates on the
output qubits can be achieved via LOCC (Local Operations and Classical Communication).
23
Note that the single-photon state Ψ01 is entangled when interpreted as Fock realization of
a 2-qubit state.
24
The case n = 2 is of special interest (Nielsen, 2004).
64 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

Let â0 , . . . , â2n be pairwise orthonormal modes describing photons with (almost)
sharp momenta and prepare the first 2n ancillary qubits in the state
  
Xn j
Y Y2n
(n) def 1
Φtele = √  â†ν  ↵  Ω ,
n + 1 j=0 ν=1 µ=n+j+1

(instead of Ψ01 ). Let F̂(n) be the linear operator on Hâ0 ,...,â2n characterized by
F̂(n) Ω = Ω and
(
−1
â for â ⊥ {â0 , . . . , ân } ,
F̂(n) â F̂(n) = √1
Pn −i k 2π
j (3.10)
n+1 j=0 e n+1 âj for â = âk , k ∈ {0, . . . , n} ,

where — as in Section 1.2.3 of (Lücke, nlqo) — we denote by Hâ0 ,...,â2n the smallest
closed subspace of Hfield that contains Ω and is invariant under â†0 , . . . , â†2n . Then,
as explained in connection with the single-photon simulation of n-qubit systems, the
transformation
−1
âν 7→ F̂(n) âν F̂(n) ∀ ν ∈ {0, . . . , 2n} (3.11)
can be implemented using by linear optics.

Remark: For n + 1 = 2m and25

|kim = â†k Ω ∀ k ∈ {0, . . . , 2m − 1}
the corresponding state transformation
X n
1 2π
|kim 7−→ F̂(n) |kim = √ e+i k n+1 j |jim ∀ k ∈ {0, . . . , 2m − 1}
n + 1 j=0

simulates the m-qubit quantum Fourier transform.

Exercise 15 Using the Campbell-Hausdorff formula26

e B̂e− = exp(ad ) B̂ , (3.12)

where
def
ad Ĉ = [Â, Ĉ]− ,
show the following:
a)
† †
e−i âj âj ϕ âk e+i âj âj ϕ = e+i ϕ δj,k âk ∀ j, j ∈ {0, . . . , 2n} , ϕ ∈ IR .
b)

−i(â†1 â2 +â†2 â1 )θ +i(â†1 â2 +â†2 â1 )θ
e â1 e = cos θ â1 + i sin θ â2 
−i(â†1 â2 +â†2 â1 )θ † † ∀ θ ∈ IR .
e â2 e+i(â1 â2 +â2 â1 )θ = i sin θ â + cos θ â 
1 2

DRAFT, October 17, 2007

25
Recall (1.12).
26
Compare Footnote 50.
3.1. QUANTUM OPTICAL IMPLEMENTATION 65

c) F̂(n) is a unitary operator on Hâ0 ,...,â2n .

Since
 
j 
Y  j
Y n
X 2π
−1
F̂(n) âν F̂(n) =  e+i ν l
n+1 ν â†lν 
ν=1 (3.10) ν=1 lν =0
Xn Yj  
2π
= e+i ν n+1 lν â†lν ∀ j ∈ {1, . . . , n}
l1 ,...,lj =0 ν=1

and
j 
Y  n
X j
Y 2π
−1
F̂(n) âν F̂(n) = e+i ν n+1 lν â†lν
ν=0 l0 ,...,lj =0 ν=0
X n j+1
Y 2π 2π
= e−i n+1 lν e+i ν n+1 lν â†lν ∀ j ∈ {0, . . . , n − 1}
l1 ,...,lj+1 =0 ν=1
X n j+1
Y −i 2π ′ 2π ′
= e l
n+1 ν e+i ν n+1 lν â†lν′ ∀ j ∈ {0, . . . , n − 1} ,
l0′ ,...,lj′ =0 ν=1

we get
 
(n)
F̂(n) α 1̂ + β â†0 Φtele
 
 Y
2n  Y
n 
1
= √ α â†ν Ω + β F̂(n) â†ν Ω
n+1 ν=n+1  ν=0
 
n
X n
Y  N j 2n
Y
+α λN0 ,...,Nn  â†j  â†ν  Ω (3.13)
N0 ,...,Nn =0 j=0 ν=n+1+N0 +...+Nn
0<N0 +...+Nn ≤n
  
n
X n 
Y N j 2n
Y
2π
j †
+β λN0 ,...,Nn  e−i n+1 âj  â†ν  Ω
N0 ,...,Nn =0 j=0 ν=n+N0 +...+Nn
0<N0 +...+Nn ≤n

with suitable λN0 ,...,Nn ∈ IR , not depending on α, β ∈  C . Now, if the number

† (n)
Nν of photons of mode âν in the state F̂(n) α 1̂ + β â0 F(n) Φtele is checked for all
ν ∈ {0, . . . , n} by projective measurement then:

• If
0 < N0 + . . . + N n ≤ n (3.14)
then the state is projected onto
 
Y
n 
2π
ÂN0 ,...,Nn α 1̂ + β e−i n+1
j Nj
â†n+N0 +...+Nn  B̂N0 ,...,Nn Ω,
j=1
66 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

where n 
Y N ν
def
ÂN0 ,...,Nn = λN0 ,...,Nn â†ν ,
ν=0
2n
Y
def
B̂N0 ,...,Nn = â†ν .
ν=n+1+N0 +...+Nn

n
• The probability for (3.14) — thanks to unitarity of F̂(n) — is .
n+1
Qn 2π
Since the phase factor j=1 e−i n+1 j Nj is fixed by the measurement result, this shows:
 
With probability arbitrarily close to 1 the state α 1̂ + β â†0 Ω with un-
 
known α , β ∈ C can be teleported27 into α 1̂ + β â†n+N0 +...+Nn Ω with
random N0 , . . . , Nn fulfilling (3.14) resulting from — typically destruc-
tive — measurement of the corresponding photon numbers in the state
(n)
F̂(n) α 1̂ + β â†0 F(n) Φtele .

In order to implement a CZn2 /(n+1)2 gate, i.e. an indeterministic CPHASE gate

working with probability of success n2 /(n + 1)2 , 4n + 2 pairwise orthogonal modes
â0 , . . . , â2n , b̂0 , . . . , b̂2n are needed. â0 resp. b̂0 is used for the Fock realization of the
first resp. second input qubit of the gate. The ancillary Fock qubits corresponding
to the modes â1 , . . . , â2n , b̂1 , . . . , b̂2n have to be prepared in the 4n-qubit state
2n
Y
def (n) (n)
Ψ̌(n)
anc = Ŝπ(ν,µ) Φtele ⊗ Ψtele ,
ν,µ=n+1

(n) (n)
where Ψtele is defined similarly to Φtele with modes âν replaced by modes b̂ν and
Ŝπ(ν,µ) means action of the CPHASE gate on the pair of Fock qubits corresponding
′
to the modes âν , b̂µ . Defining F̂(n) resp. Â′N0 ,...,Nn resp. B̂N
′
0 ,...,Nn
similarly to F̂(n)
resp. ÂN0 ,...,Nn resp. B̂N0 ,...,Nn with â-modes replaced by b̂-modes and
 
Y
n 
def 2π
t̂α,β (N) = α 1̂ + β e−i n+1
j Nj
â†n+N0 +...+Nn  ,
j=1
  (3.15)
Yn 
def 2π
t̂′α,β (N) = α 1̂ + β e−i n+1
j Nj
b̂†n+N0 +...+Nn 
j=1

we get
   
′
F̂(n) F̂(n) α 1̂ + β â†0 α′ 1̂ + β ′ b̂†0 Ψ̌(n)
anc
2n
Y        
(n) (n)
= Ŝπ(ν,µ) F̂(n) α 1̂ + β â†0 Φtele ⊗ ′
F̂(n) ′
α 1̂ + β ′ b̂†0 Ψtele
ν,µ=n+1

DRAFT, October 17, 2007

27
See (Fattal et al., 2003) for an experimental demonstration of the basic version.
3.1. QUANTUM OPTICAL IMPLEMENTATION 67

and hence, by (3.13):

   
′
F̂(n) F̂(n) α 1̂ + β â†0 α′ 1̂ + β ′ b̂†0 Ψ̌(n)
anc

2n
Y  Y
2n  Yn 
 α β
= Ŝπ(ν,µ) 
 √ â†ν Ω + √ F̂(n) †
âν Ω
ν,µ=n+1 n+1 ν=n+1 n+1 ν=0
n
!
X
+ λN0 ,...,Nn ÂN0 ,...,Nn t̂α,β (N) B̂N0 ,...,Nn Ω
N0 ,...,Nn =0
0<N0 +...+Nn ≤n
 Y
2n  Yn 
α′ β′
⊗ √ b̂†ν Ω + √ ′
F̂(n) b̂†ν Ω
n+1 ν=n+1 n+1 ν=0

n
!
X 
+ λN0′ ,...,Nn′ Â′N0′ ,...,Nn′ t̂′α′ ,β ′ (N′ ) B̂N
′
′ ,...,N ′ Ω
0 n


N0′ ,...,Nn′ =0
0<N ′ +...+Nn ′ ≤n
0

Therefore, if the numbers N0 , . . . , Nn , N0′ , . .. , Nn′ of photons in the modes â
0 , . . . , ân ,
  
′
b̂0 , . . . , b̂n are checked for the state F̂(n) F̂(n) α 1̂ + β â†0 α′ 1̂ + β ′ b̂†0 Ψ̌(n)
anc by pro-
jective measurement then:

• If
0 < N0 + . . . + Nn ≤ n and 0 < N0′ + . . . + Nn′ ≤ n (3.16)
then the state is projected onto a state with the factor
2n
Y
def
Ψout = Ŝπ(ν,µ) tα,β (N) t′α′ ,β ′ (N′ ) Ω
ν,µ=n+1
′ ′
= Ŝπ(n+N0 +...+Nn ,n+N0 +...+Nn ) tα,β (N) t′α′ ,β ′ (N′ ) Ω

which may be easily transformed, using only linear optical components, into
the desired output
  
Ŝπ(0,0) α 1̂ + β ↠α′ 1̂ + β ′ b̂† Ω .

• The probability for (3.16) is n2 / (n + 1)2 .

Especially for n = 2 , according to (Knill et al., 2001, Supplementary Informa-

(n)
tion, Fig. 4), the ancillary state Φtele can be prepared (indeterministically) using
only linear optics as sketched in Figure 3.8.
68 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

â†1 ...Ω
........ ..... ...... ..... ..... ..... ..... ..... ..... ..... ..... ..... .................... ..... ..... ..... ..... ..... ..... ..... ..... ..... ......
1
...... ..... ..... ..... ..... ..... ................ .
... .....
..... ..... ..... .....
..... ..... .. ....
.
...... ... .
............... .. ............. ...
..... ......... .. .........
..... ...... .... ......
..... ..... ..... .....
.....
.....
..... ..... ..... .....
.. .... . ....
◦............... . ◦ ◦.... ........... . ◦
+45 90
.....
..... .
..... NS
...... ..................... ..
i
..... ..................... .......
..... −45 .
..... ..... 90 2
.... ..... .... .....
.....
....
. .
............. ....
. ..... ..... ..... ..... ..... .................... .
..... .....
............... .............
..
.... ............. ..... ....
.
..... ... . . . .
.. .. ..... . . .
.......
.....
..... ..... ..... ..... .....
..... .... ....
. ..... .....
◦ .............. . ◦ ◦.... ........... . ◦
+45 0
..... ..... −45 ..... ..... 0 .
.
........
.......
.. .....
.....
..... ..... ..... ..... ..... .....
..... ..... ..... ..... . ..... .....

...
........... . . .
.. .
...............
..... ...
.
............
.. .. . .
.........
..........
...
.. .....
3
.....
..... .................... ..
..... ..... ..... ..... .....
..... ..... ..... ....
. .....
ϑ . ◦ .
.
.
......
..... .. ...
.....
0..
. .
NS
...... .................... .. . .
..... ................. ....
i ...
..
.....
..... .....
.....
..... ..... .....
........ .....
.
.
..
.. .
... .....
.
.
.. ...............
......
. ..... ...
. .....
.
. .
................. ....... .............. .....

â†4 ...Ω
.. ........ .... .....
.....
........ ....
.
.........
. ....
.....
...... ..... ..... ..... ..... ..... ..... ..... ................... ..... ..... ..... ..... ..... ..... ..... ...... .
.. .
.. ..... 4
....... ..... ..... ..... ..... ................... .
... .

(2)
Figure 3.8: Preparation of Φtele

3.2 Measurement-Based Schemes for Quantum Com-

putation28
The implementation of near deterministic CPHASE gates just described shows some
interesting deviations from the conventional network model:
• Many ancillary qubits are used in addition to the input-qubits.

• The total state of all the qubits is suitably prepared using easily implementable
deterministic gates and indeterministic gates, the latter acting only on the
ancillary qubits.

• Then the desired output state can be transported onto the output-qubits using
only (photon number) measurements and applying some easily performable
final correction depending on the measurement results.
Meanwhile it turned out that for every n-qubit gate (n ∈ IN) and standard input
state the corresponding output state can be efficiently produced,29 e.g., as follows
(Childs et al., 2005; Walther et al., 2005; Nielsen, 2005):
1. Prepare a n×m-qubit cluster state30 with sufficiently large m in the following
way:
DRAFT, October 17, 2007
28
See also (Browne and Rudolph, 2004; Rudolph and Virmani, 2005; Varnava et al., 2005;
Raussendorf, 2005) and (Lim et al., 2004).
29
Concerning the preparation of explitly known states see (Kaye and Mosca, 2004).
30
We use the notion cluster state in a more general sense than originally introduced in
(Raussendorf and Briegel, 2001; Raussendorf, 2003).
3.2. MEASUREMENT-BASED QUANTUM COMPUTATION 69

(a) Prepare n × m qubits in the states

def |0iνµ + |1iνµ

|+iνµ = √ , (ν, µ) ∈ {1, . . . n} × {1, . . . m} ,
2
and imagine them arranged in a n × m-matrix scheme:
|+i11 |+i12 ··· |+i1m

|+i21 |+i22 ··· |+i2m

.. .. ··· ..
. . .
|+in1 |+in2 ··· |+inm

(b) Join all horizontally neighbouring qubit states and certain vertically neigh-
bouring qubit states, depending on the effective network action desired,
by lines respecting the rule that every state attached to a vertical line
should have only one neighbour attached to a vertical line. Writing
l
νµ instead of |+iνµ

we may get, e.g.

.................. .................. .................. .................. ..................
... ... ... ... ... ... ... ... ... ...
.... .... .... .... ....
... 11 .... ... 12 .... ... 13 .... ... 14 ..... ... 15 ....
. . . .
............... ............... ............... ............... ...............

.................. .................. .................. .................. ..................

... ... ... ... ... ... ... ... ... ...
..... ..... ..... ..... .....
... 21 .... ... 22 .... ... 23 .... ... 24 .... ... 25 ....
. . . . .
............... ............... ............... ............... ...............

................. ................. ................. ................ .................

.... ... .... ... .... ... .... ... .... ...
..... ..... ..... ..... . .....
... 31 ..... ... 32 ..... ... 33 ..... ... 34 ..... ... 35 .....
. . . .
............... ............... ............... ............... ...............

as an example for n = 3 , m = 5 .
(c) For every (horizontal or vertical) line (bond) apply a CPHASE gate to
the pair of qubits connected by the line.
2. Once the cluster state is prepared certain projective single qubit measurements
are performed on the qubits corresponding to the first row (qubits 11, . . . , n1).
3. Once the qubits of the ν-th column have been tested certain projective sin-
gle qubit measurements, depending on the outcome of the previous measure-
ments31 (and the final output desired), are performed on the qubits corre-
sponding to the (ν + 1)-th column.
4. When the measurements on the m − 1-th column are performed the n-qubit
system corresponding to the m-th column is left in the desired output state up
to known single-qubit transformations depending on the results of the previous
measurements.
5. The deviation from the desired output state may be either corrected or may
be taken account of by appropriate change of the computational basis.
DRAFT, October 17, 2007
31
Measurements the outcome of which determine the choice of subsequent measurements are
called feed-forwardable.
70 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

A fascinating aspect of such measurement-based schemes for quantum computation

is that the universal CPHASE gate is needed only for preparing the cluster state
and for this purpose an indeterministic implementation of the phase gate is sufficient
(Nielsen, 2004; Nielsen and Dawson, 2004; Chen et al., 2005):32

Every cluster state can be prepared by successively applying single-bond and/or

double-bond operations:

I. Single-bond operations: E.g. for the case sketched in Figure 3.9 the
two teleportations (measurements of N0 , N1 , N2 resp. N0′ , N1′ , N2′ ) are at-
tempted one after the other.
.................. .................. .................. .................. ......
...... .........
... ... ... .. ... ... ... ... ..
..... ..... ..... ..... .... ..
... .. ... ... ... .. ... ... ..
.................. .................. .................. .................. ....
.................

.. .. .. .........
........ .......... ........ .......... ........ .......... ..... .......
... ... ... ...
....
...
..
..
..................
....
...
..
..
..................
....
A
...
..
....... ..........
..................
C ..
..
..................
.

.................. .................. ..................

... ... ... ... ... ...
....
...
..................
.. ....
...
..................
.. ....
...B
..................
..

Figure 3.9: Trying to add qubit C with a single bond.

The teleportation that does not affect A is tried first. If it does not suc-
ceed one has to restart with a newly prepared qubit C . If it does succeed
then the second teleportation is tried. If the latter also succeeds C is
added with a bond to A . If it does not succeed then a projective single-
qubit measurement w.r.t. the computational basis has been performed
on A and this qubit has to be removed from the cluster and single-qubit
correction have to be performed on the qubits connected to A (for the
sketched special case only one) depending on the measurement result
indicated by the corresponding detectors. As soon as the size of the
cluster has been changed by this procedure we say that a single-bond
operation has been performed. Thus:
A single-bond operation adds resp. removes a qubit with prob-
ability 2/3 resp. 1/3.

II. Double-bond operations: E.g. for the case sketched in Figure 3.10 one
first try to connect D to C by a single-bond operation. If this adds D
to the cluster we apply CZ4 /9 to the pair B, D . If the corresponding
teleportation not affecting B fails then D has to be removed from the
cluster and we have to restart with a newly prepared qubit D . If the
first teleportation succeeds also the second teleportation is tried. If the
DRAFT, October 17, 2007
32
Moreover, such schemes circumvent the problem of programmable deterministic quantum gate
arrays (Nielsen and Chuang, 1997).
3.3. COLD TRAPPED IONS 71
.................. .................. .................. .................. ..................
... ... ... ... ... ... ... ... ... ...
.... .... .... ..... .....
... ... ... ... ... ... ... .. ...
.
.................. .................. .................. .................. ....
................

.................. .................. .................. ..................

... ... ... ... ... ... ... ...
.....
... ...
..................
.....
... ...
..................
.....
A
... ...
..................
.....
... C ..
..................
.
...
..
..
....... ..........
..
....... ..........
..
....... .......... .....
....... .........
... ... ... ... ... ... ... ...
....
... ..
..................
....
... ..
..................
....
B
... .
..................
...... .........
... D ..
..................

Figure 3.10: Trying to add qubit D with two bonds.

latter succeeds D is connected to both C and B if not then B and D

have to be removed from the cluster. As soon as either C or B has
been removed or D has been connected to both C and D we say that a
double-bond operation has been performed. Thus:

A double-bond operation adds resp. removes a qubit with

probability 23 · 32 resp. 31 + 23 · 31 .

In order to create the cluster, after every double-bond operation that removed a
qubit a single-bond operation can be applied. Then:

On average, 2N successive operations add at least

      
2 1 2 2 2 1 2N
− N+ · − · N =
3 3 3 3 3 3 9
qubits to the cluster.

Easier implementable indeterministic CPHASE gates33 with lower probabilities

of success are sufficient if the cluster is built from microclusters in which a single
qubits are connected to several dangling qubits in order to allow for multiple gluing
attemps.34

3.3 Cold Trapped Ions35

Of all the proposed technologies for quantum information processing devices, arguably
one of the most promising and certainly one of the most popular is trapped ions.
(James, 2000)

DRAFT, October 17, 2007

33
See, e.g., (Gasparoni et al., 2004; Zhao et al., 2005).
34
See end of (Nielsen, 2004).
35
See (Bužek and Šašura, 2002; Ghosh, 1995; Wunderlich and Balzer, 2003).
72 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

3.3.1 General Considerations

One of the earliest proposals to fulfill DiVincenzo’s requirements for quantum
computation is the following:

1. • Qubits are identified with ions of some specified kind being in a superposi-
tion of two specified (at least) metastable energy eigenstates |gi resp. |ei
representing the computational basis states |0i resp. |1i . Typically, |gi
is the ground state.
• The ions are bound to specified places inside an ion trap and their col-
lective oscillation is cooled to the quantum mechanical ground state.

2. The states |g1 , . . . , gn i ≡ |0, . . . , 0i may be prepared by applying Laser ra-

diation tuned to the transition of |ei into a rapidly decaying higher energy
eigenstate.

3. The decoherence time is of the order 10−1 seconds while the duration of gate
operations is of the order 10−14 seconds.36

4. • 1-qubit rotations are implemented by laser pulses — of appropriate du-

ration and phase — tuned to the transition between |gi and |ei.
 
• The controlled sign gate Λ1 Ŝπ is implemented by a suitable sequence
of laser pulses exploiting one of the collective translational modes as data
bus in the following way:
(i) A first pulse on the control qubit (Ion j1 ) acts according to

|gj1 i|0iosc −
7 → |gj1 i|0iosc ,
(3.17)
|ej1 i|0iosc −7 → −i |gj1 i|1iosc .

(ii) A second laser pulse on the target qubit (Ion j2 ) — tuned to the
transition of |gi into an exited energy eigenstate |ěi different from
|ei — acts for b ∈ {0, 1} according to

|gj2 i|biosc −7 → (−1)b |gj2 i|biosc ,

(3.18)
|ej2 i|biosc 7−→ |ej2 i|biosc .

(iii) A third pulse of the same type as used in (i) acts according to

|gj1 i|0iosc −
7 → |gj1 i|0iosc ,
(3.19)
|gj1 i|1iosc −7 → −i |ej1 i|0iosc .

These laser pulses have no effect if the control qubit (Ion j1 ) is originally
in the state |gj1 i. On the other hand, if the control qubit (Ion j1 ) is
originally in the state |ej1 i then the action is as follows:
DRAFT, October 17, 2007
36
See, e.g. (Nielsen and Chuang, 2001, Fig. 7.1) for a comparison with other implementations.
3.3. COLD TRAPPED IONS 73
...
. ...... .....
....... ..................... ...............
z
....
.. ... ............
.......................... .. ....................
... .... . ...... ........... .........
.... .... .... .... ..... . .
.....
.....
. . ... ..
..... ........................................................................................................................................................................................................................................................................................... ........ .......... ....... ....
. . .. ...
...
. .... ... ..
................................ ... ...
.... .
.............................................................................................................................................................................................................................................................................................. ... ...
...
...
.. ..
. ... ... .. . ...
. . . . .. .. . ..
... .... ... ..
. .
. ..
..
.
.......
.
. ..
. ..
. ... ..
.. ... ..... .. ..
.
....
..
..
.
....
..
.
..
. ..
..
...
..
...
...
.. ....
......
.. .
.....
.
y ..
.
.
.
.
..
.
.
.
.
...
...
. .. ...
...
..
..
..
..
.. . . ..
... ... .. .
..
.
..
... .. ..
..... .
.. . . .
.. .. ....
....... ..... ...
...
...
.. . . ..... ... ...
. .
. . ... . .. . .. . . . . . .......................
... .. .... . .. .. . .
.... ....
... ... ... ..
..
. .
. .
.
.. .... ..... .. . .. ..
..... ....
.. ..
..
...
..
..
...... ..
.
. ............................................................................................................
.
... ..
. .. .. .. ..... .. ... ..
... ..
..
..... ...
.. .
. ..
......
. ..
.
. ..... . ..
.. .. .. .. .. ..
. .. .. ... ..... ..
..
... ..
.... ... .. . ...
...... .
. .. ..
..
.. .. .. . ..
..... ..
. ...
..
..
. ..
.
... .. ..
.....
. . . ...
.......... .
.. ..
.
.... ...
..
...
...
...
... .
..
.
..
.
.
..
.
..
..
..
.
....
.
..
.
....
.
..
.
.
..
.
..
.
.
...
...
..
x .
.
.
.
.
.
.
.
.
.
.
..
..
..
... ...
..
.
... .... ... .. ...
... ... ..... . .. ..
... ......................................................................................................................................................................................................................................................................................... .... ..............
.. .. .. .. ..
..
.. ... .. .. .. .. ...
..
... .. ..
..
.. .
. .. .. ..
. . . ...
. .
.
... ... ... ...............
.. ... ..... .. ... ...
..
.. ... .
.. .. ........ ..
.. .. ... ...
... ... .. ... . ... ..
.. .. ...
.. .
. ..
. ..
. ... .
. ... ..
.
.. ... . . .. ... . .
.. ..
.
... .
.. ... ..
...
.. ... ...
..
... ... .. ..
... .. ... ..
... .. .. .. .. ..
...
... ... .. .. ...
.
.. .................
. .. . ... ..
. ...
.
... ... ... .
. ... .. .
..
...
.... ... .... ... .... ... ... ...
.... ...
.... .... ....
......................... ... ....
.... .... ... ...
..... ....
..
. . ..
.... . .. .... .....
..
....
. . ... .... . .......
...... . . .
. ...... .. . .
....... .............................. ............. ....... ................................................
.. ....... ..

Figure 3.11: A linear Paul trap

– The first pulse transfers this information into the data bus by exciting
it to |1iosc and applies −i σ̂1 to the control qubit.
– Then the second pulse multiplies the |gj2 i-component of the target
qubit’s state by −1.
– Finally, the third pulse returns the data bus into its ground state and
acts on the |gj2 i-component of the control qubit’s state by −i σ̂1 , once
more.
 
Obviously, the resulting action is that of a Λ1 Ŝπ gate.

5. Measurement of the final state of the computation may be done by irradiating

by continuous laser radiation tuned to the transition of |gi into a rapidly
decaying higher energy eigenstate. Then only those ions ‘found’ in the ground
state show strong fluorescence.

All this will be explained in more detail — for the special case37 of 40
Ca+ ions in a
linear Paul trap — in the subsequent sections.

3.3.2 Linear Paul Trap

The linear Paul trap used by the Innsbruck group is sketched in Figure 3.11.
DRAFT, October 17, 2007
37
See http://heart-c704.uibk.ac.at/quantumcomputation.html
74 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

There are four rod electrodes, namely along

n o
def
C1 = (x, y, z) ∈ IR : y = −r0 , z = 0 , x ∈ {− x20 , + x20 } ,
n o
def
C2 = (x, y, z) ∈ IR : z = +r0 , y = 0 , x ∈ {− x20 , + x20 } ,
n o
def
C3 = (x, y, z) ∈ IR : y = +r0 , z = 0 , x ∈ {− x20 , + x20 } ,
n o
def
C4 = (x, y, z) ∈ IR : z = −r0 , y = 0 , x ∈ {− x20 , + x20 } ,

and two ring electrodes (end caps). The electric potential of the rod electrodes
along C1 and C3 is38 Φ(t) while the rod electrodes C2 and C4 are grounded. Both ring
electrodes have the constant electric potential Φring > 0 .

Trapping in Radial direction

If the rod electrodes were infinitely long and infinitely thin then they would con-
tribute the electric potential
!
Φ0 y2 − z2
Φideal (x, y, z, t) = 1+ if Φ(t) = Φ0 = const
2 r02

since this fulfills the Laplace equation as well as the boundary conditions along
the rods. If

Φ(t) = Φ0 cos(Ω t) , Ω ≈ 16 − 18 MHz (radio frequency) ,

then the quasi stationary approximation

!
Φ0 cos(Ω t) y2 − z2
Φ⊥ (x, y, z, t) ≈ 1+
2 r02
y2 − z2
= Φ0 cos(Ω t) + 21 Φ0 cos(Ω t) (3.20)
2 r02
can be used near the center of the trap — also for finite rods, if sufficiently long.

The evolution equations of a particle of mass m and electric charge q in the

electric potential (3.20) are39

q Φ0
ÿ(t) + cos(Ω t) y(t) = 0 , (3.21)
m r02
q Φ0
z̈(t) − cos(Ω t) z(t) = 0 , (3.22)
m r02
ẍ(t) = 0 .
DRAFT, October 17, 2007
38
Note that static electric potentials cannot have minima in regions free of electric charge.
39 Ω
In the Innsbruck experiment: Φ0 ≈ 300 − 800 V , , r0 ≈ 1.2 mm .
2π
3.3. COLD TRAPPED IONS 75

With
def 2 q Φ0 def Ωt
b= , ζ = (3.23)
m r02 Ω2 2
equations (3.21) and (3.22) become equivalent to the special cases
!2
d
y(ζ) + 2 b cos(2 ζ) y(ζ) = 0 , (3.24)
dζ
!2
d
z(ζ) − 2 b cos(2 ζ) z(ζ) = 0 (3.25)
dζ

of the Mathieu differential equation

!2
d  
y(ζ) + a + 2 b cos(2 ζ) y(ζ) = 0 . (3.26)
dζ

The general solution of (3.26) is40

X  
y(ζ) = C2n λ+ e+µ ζ e+2i n ζ + λ− e−µ ζ e−2i n ζ ,
n∈ZZ

with general integration constants λ± (to be adapted to the initial conditions) and
certain constants C2n and µ depending an a, b . We are interested in stable solutions,
only, and therefore have to require

µ = iβ , β ∈ IR .

Then X     
y(ζ) = C2n λ1 cos (2n + β) ζ + λ2 sin (2n + β) ζ , (3.27)
n∈ZZ

where
def
λ1 = λ+ + λ− , λ2 = i (λ+ − λ− ) .
Inserting this into (3.26) gives the recursion formula

a − (2n + β)
C2n+2 − C2n + C2n−2 = 0 . (3.28)
b
Defining
def C2n def def
G2n = , A = λ1 C0 , B = λ2 C0
C0
and exploiting the well-known theorems for sin and cos we can rewrite (3.27) in the
equivalent form
y(ζ) = Y + (ζ) + δy (ζ) ,
DRAFT, October 17, 2007
40
See (Ghosh, 1995, Section 2.3).
76 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

where:
def
Y ± (ζ) = A cos(β ζ) ± B sin(β ζ) ,
∞ 
X 
def
δy (ζ) = Y + (ζ) (G2n + G−2n ) cos(2nβζ) − Y − (ζ) (G2n − G−2n ) sin(2nβζ) .
n=1

For sufficiently small41 |b| one may show:

1. Solution of (3.28) gives a stable solution (3.27).
2. The micromotion δy (ζ) is negligible.
3. The corresponding statements hold also for z(ζ) .

In the following the micromotion will be neglected.

Trapping in Axial Direction

The positive potential of the ring electrodes in Figure 3.11 servers to confine positive
ions also in axial direction. The additional (time-independent) electric potential
contributed by these electrodes is
 2
x
Φk (x, y, z) ≈ ξ Φring for x/x0 ≪ 1 ,
x0
where ξ is some geometric factor of order 1 characterizing the contribution of Φring
on the center of the ring electrodes.

In the Innsbruck experiment we have

Φring ≈ 2000 V , x0 ≈ 5 mm
and
m 2 2
q Φk (x, y, z) ≈ ω x
2 x
with
ωx
≈ 500 − 700 kHz for 40 Ca+ .
2π
Since ωx is definitely smaller than β ζ ≈ 1, 4 − 2 MHz , in this experiment, it is
sufficient — for what follows — to consider only the motion along the x-axis in the
potential contributed by the ring electrodes and the repulsive Coulomb potential of
the ions. The total mechanical potential for N identical ions then is (approximately)
N N
m 2X q2 X 1
V̌ (x1 , . . . , xN ) = ωx x2j + . (3.29)
2 j=1 4π ǫ0 j,k=1 |xj − xk |
j<k

DRAFT, October 17, 2007

41
Note that b ∼ q/m .
3.3. COLD TRAPPED IONS 77

Mean Values of the Ion Positions

For j ∈ {1, . . . , N } , let x̌j be the mean value of the j-th ion’s x-coordinate. Obvi-
ously, for
def def
x = (x1 , . . . , xN ) = x̌ = (x̌1 , . . . , x̌N )
the potential of the N -ion system has to be minimal:
!
∂
V̌ (x1 , . . . , xN ) = 0 ∀ j ∈ {1, . . . , N } . (3.30)
∂xj |x=x̌

Without loss of generality we may assume

x̌1 < x̌2 < . . . < x̌N . (3.31)

under this condition (3.30) is equivalent to

N
X XN
1 1
Xj − 2 + 2 = 0 ∀ j ∈ {1, . . . , N } , (3.32)
k,j=1 (Xk − Xj ) k,j=1 (Xk − Xj )
k<j k>j

where42 !1
def def q2 3
Xj = x̌j /γ , γ = . (3.33)
4πǫ0 m ωx2
For N ≤ 3 the solutions are easily determined:

N = 1 : X1 = 0 ,
q q
1
N = 2 : X1 = − 3 4
, X2 = + 3 14 ,
q q
5
N = 3 : X1 = − 3 4
, X2 = 0 , X3 = + 3 54 .

For N > 3 the Xj have to be determined numerically.

Of course, the distance between the ions is minimal at the center of the trap.
Numerical calculations show that
2.018
∆x̌min ≈ γ
N 0.559
(James, 1998). Therefore, in the Innsbruck experiment, the overlap of the ions’ wave
functions is negligible43 and the ions are individually addressable by laser beams.
DRAFT, October 17, 2007
42
For 40 Ca+ and ω2πx ≈ 700 kHz: γ ≈ 4.85 µm .
43
Recall Footnote 17 of Chapter 1.
78 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

Collective Oscillations
Near its minimum the potential (3.29) may be approximated as
N
def m 2 X
V̌ (x1 , . . . , xN ) ≈ V (q1 , . . . , qN ) = ωx Vjk qj qk ,
2 j,k=1

where
def
qj = xj − x̌j ∀ j ∈ {1, . . . , N } , (3.34)
!
def 1 ∂ ∂
Vjk = V (x1 , . . . , xN ) ∀ j, k ∈ {1, . . . , N } .
m ωx2 ∂xj ∂xk |x=x̌

Explicitly, by (3.29) and (3.33), we have

 N

 X 2

 1+ for j = k ,

 3
l=1 |Xl − Xj |
Vjk = Vkj = l6=j (3.35)



 2
−
 for j 6= k .
|Xk − Xj |3

The corresponding system of evolution equations is

n
X
q̈j (t) + ωx2 Vjk qk (t) = 0 . (3.36)
k=1

Since the matrix (Vjk ) is positive and symmetric, there is an orthonormal system of
eigenvectors  
Cl1
. 
Cl =  ..  , l ∈ {1, . . . , N } ,
ClN
of this matrix in IRN with positive eigenvalues:
    
V11 ... V1N Cl1   Cl1
 .. ..   ..  ω̌l 2  .. 
 . .  .  =  .  l ∈ {1, . . . , N } . (3.37)
ωx
vN 1 . . . VN N ClN ClN

Hence, every solution of (3.36) is a superposition

N 
X 
(+l) (−l)
qj (t) = λ+
l qj (t) + λ−
l qj (t) (3.38)
l=1

of the special collective motions (eigenmodes)

(±l) def
qj (t) = Clj e∓i ω̌l t . (3.39)
3.3. COLD TRAPPED IONS 79

The first two eigenvectors may always be chosen as

 
1
1 .
C1 = √  ..  with ω̌1 = ωx ,
N 1
  (3.40)
X1
1  ..  √
C2 = q  .  with ω̌2 = 3 ωx .
X12 + . . . + XN2 X
N

PN
Proof: (3.35) directly implies k=1 Vjk = 1 and, therefore, (3.37) for l = 1 with
ω̌1 = ωx . On the other hand, we have
N
X N
X N
X
2 Xl 2 Xj
Vjk Xk = Xj + 3 − 3
(3.35) |Xl − Xj | |Xl − Xj |
k=1 j6=l=1 j6=l=1
XN
Xj − Xl
= Xj + 2 3
j6=l=1
|Xl − Xj |
N
X N
X
2 2
= Xj + 2 − 2
(3.31) |Xl − Xj | |Xl − Xj |
l=1 l=1
l<j l>j

= 3 Xj .
(3.32)
√
The latter implies (3.37) for l = 2 with ω̌2 = 3 ωx .

   
(1) (1) (2) (2)
Then q1 , . . . , qN is called the center of mass mode and q1 , . . . , qN the
breathing mode.

3.3.3 Implementing Quantum Gates by Laser Pulses

Quantization of the Collective Oscillations
Using the generalized coordinates
 
Ql (t) = ℜ λ+
l e
−i ω̌l t
+ λ−
l e
+i ω̌l t
(3.41)
we get
N
X
xj (t) = x̌j + Clj Ql (t) ∀ j ∈ {1, . . . , N } (3.42)
l=1
for the x-coordinates of the ions and
N  
mX
L= Q̇2l − ω̌l2 Q2l (3.43)
2 l=1
for the Lagrangian corresponding to (3.36). With the canonically conjugate mo-
menta
def ∂
Pj = L = m Q̇j
∂ Q̇j
80 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

we get the Hamiltonian

N N
1 X mX
H(Q, P, t) = Pl2 + ω̌ 2 Q2
2m l=1 2 l=1 j l
and standard quantization of the system amounts to selecting a (normalized) ground
state (vector) |0iosc and replacing the Ql , Pl by operators
s  
h̄ ˇ† ,
ˇl + â
Q̂l = â l
2m ω̌l
s
 
(3.44)
h̄ m ω̌l ˇ ˇ† ,
P̂l = −i âl − â l
2
†
ˇl and creation operators â
with annihilation operators â ˇ obeying the canon-
l
44
ical commutation relations
h i  
ˇl , â
â ˇk =0, â ˇ†
ˇl , â = δlk (3.45)
− k
−

and the Fock condition

ˇl |0i = 0 ∀ l ∈ {1, . . . , N } .
â (3.46)
osc

The quantized Hamiltonian, then, is

N N
1 X mX
Ĥosc = P̂l2 + ω̌j2 Q̂2l
2m l=1 2 l=1
N
X  
ˇ† â
ˇ 1
= h̄ ω̌l âl l + (3.47)
l=1 2
and a maximal orthonormal system of the state space (for the quantized oscillatory
motion) is given by the Fock states
YN  nl
def 1 ˇ†
|n1 , . . . , nN iosc = √ â |0iosc , n1 , . . . , nN ∈ ZZ+ , (3.48)
n1 ! . . . nN ! l=1 l
fulfilling
ˇ† â
ˇ
âl l |n1 , . . . , nN iosc = nl |n1 , . . . , nN iosc . (3.49)
The corresponding observables for the x-coordinates of the ions are
N
X
x̂j = x̌j 1̂osc + Clj Q̂l (t)
(3.42) l=1
s  
N
X h̄ ˇ† .
ˇl + â
= x̌j 1̂osc + Clj â l (3.50)
(3.44) l=1 2m ω̌l
DRAFT, October 17, 2007
44
These commutation relations are easily seen to be equivalent to
[Q̂l , Q̂k ]− = 0 , [P̂l , P̂k ]− = 0 , [Q̂l , P̂k ]− = i h̄ δlk .
3.3. COLD TRAPPED IONS 81

Of course, a more detailed description of the ions’ motion would require inclusion
of the modes describing radial oscillations.

Laser-Ion Interaction
In the following we assume that all the ions are 40 Ca+ ions. Moreover we select a
special mode âˇl (typically l = 1 or 2) as data bus. All other modes are assumed
0
cooled into their ground state and will not be included in the description. The
Hamiltonian describing the collective motion of all ions the the internal state of the
j-th ion not interacting with an external electromagnetic field then is
 
ˇ† â
h̄ ω̌l0 â ˇ 1
Eej |ej ihej | + Egj |gj ihgj |
l0 l0 + 2
+ ,
| {z }
Eej −Egj Eej +Egj
= 2
(|ej ihej |−|gj ihgj |)+ 2
(|ej ihej | + |gj ihgj |)
| {z }
=1̂int

where Eej resp. Egj is the energy level of the qubit state |ej i resp. |gj i . This
Hamiltonian describes the same time evolution as
def h̄ ω0j ˇ† âˇ
Ĥ0j = − σ̂3j + h̄ ω̌l0 â l0 l0 , (3.51)
2
where
Eej − Egj
def def
ω0j = , σ̂3j = |gj ihgj | − |ej ihej | . (3.52)
2 h̄
(3.52) will be used in the following. We assume the laser radiation to be strong
enough for the exterior field formalism to be adequate (see, e.g., Sections 7.1.1
and 7.2.1 of (Lücke, nlqo) in this connection). Interaction with the classical laser
radiation will be described in the dipole approximation 45 by adding
 
def
V̂j = −qel r̂j · E (x̂j , 0, 0) t ,

where
def
qel = ( electric charge of the valence electron ,
def observable of the valence electron’s
r̂j =
position relative to the center of the ion ,
def
E(x, t) = external electric field at position x and time t .
We assume that the exterior field is of the form
 
E(x, t) = E0 e−i(ωL t−kL ·x+φ̌) + e+i(ωL t−kL ·x+φ̌) . (3.53)
DRAFT, October 17, 2007
45
In the quadrupole approximation we would have to add
 
′ def ′
V̂j = −qel r̂j · (r̂j · ∇r′ ) E(x , t) .
|r′ =(x̂
j ,0,0)
82 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

Replacing r̂j by

1̂int r̂j 1̂int = (|gj ihgj | + |ej ihej |) r̂j (|gj ihgj | + |ej ihej |)
def − def
σ̂j− = reg,j =
def
σ̂j =
z }| { z }| { z }| {
= hgj | r̂j ej i |gj ihej | + hgj | r̂j ej i |gj ihej |

we get
    
  ˇ†
ˇl +â
−i ωL t−ηl0 j â +φj
†
reg,j σ̂j+ σ̂j− 0 l
V̂j = −qel + (reg,j ) · E0 e 0

   
+i ωL t−ηl0 j â ˇ†
ˇl +â +φj
0 l
+e 0
,

where s
def h̄ def
η l0 j = kL1 , φj = φ̌ − kL1 x̌j .
2m ω̌l0
In the interaction picture (see, e.g., Section 7.1.1 of (Lücke, nlqo))the time evolution
is determined by
i i
V̂jI = e+ h̄ Ĥ0j t V̂j e− h̄ Ĥ0j t
instead of Ĥ0j + V̂j . Using the Campbell-Hausdorff formula,46 here in the form
i i
 
e+ h̄ Ĥ0j t V̂j e− h̄ Ĥ0j t = exp ad i Ĥ0j t V̂j ,
h̄

and
ˇ† ]− = +h̄ ω̌l â
[Ĥ0j , â ˇ† ,
l0 0 l0
ˇl ]− = −h̄ ω̌l â
[Ĥ0j , â ˇl ,
0 0 0

h̄ ω 0j
[Ĥ0j , σ̂j± ]− = [σ̂3j , σ̂j± ]−
2
= ±h̄ ω0j σ̂j± ,
we get47
    !
  ˇl e−i ω̌l0 t +H.c. +φj
−i ΩL t−ηl0 j â
V̂jI = −qel reg,j · E0 σ̂j+ e+i ω0j t + H.c. e 0
+ H.c. .

Hence in the rotating wave approximation,48 i.e. if we neglect the contributions

of the higher frequencies ± (ωL + ω0j ) :
 
h̄ +i ηl j ˇl e−i ω̌l0 t +H.c.
â
V̂jI ≈ λj σ̂j+ e 0 0
e−i(ωL −ω0j )t + H.c. ,
2
DRAFT, October 17, 2007
46
Recall Exercise 15.
47
By ‘H.c.’ we denote the hermitian conjugate of the preceding term.
48
See (Aniello et al., 2003) for some criticism of this approximation.
3.3. COLD TRAPPED IONS 83

where49
def 2
λj = − qel reg,j · E0 e−i φj . (3.54)
h̄
Using the Baker-Hausdorff formula 50
h i h i 1
Â, [Â, B̂]− = B̂, [Â, B̂]− = 0 =⇒ eÂ+B̂ = e− 2 [Â,B̂]− e eB̂ (3.55)
− −

and (3.45) we get

 
ˇl e−i ω̌l0 t +H.c.
+i ηl0 j â ˇ† +i ω̌l t −i ω̌l t
ˇ − 12 ηl2
= e+i ηl0 j âl0 e e+i ηl0 j âl0 e
0 0 0
e e 0j .

Therefore, if
ωL − ω0j = k ω̌l0 , k ∈ ZZ , (3.56)
then the rotating wave approximation becomes
 µ
 ν
ˇ†
â ˇl
h̄ − 1 η2 X∞ l0 â 0
V̂jI ≈ λj σ̂j+ e 2 l0 j (i ηl0 j )µ+ν e+i (µ−ν−k) ω0j t + H.c.
2 µ,ν=0 µ! ν!

For sufficiently small λj non-resonant transitions and hence terms with µ−ν −k 6= 0
may be neglected. Then we may use
  

 h̄ + ˇ† |k| ˇ ˇ† ˇ
 λj σ̂j âl0 F̂ k (âl0 âl0 ) + H.c. for k ≥ 0 ,
V̂jI =  2 (3.57)
 h̄ ˇ ˇ† ˇ ˇ |k|
 λj σ̂ + F̂ k (â âl ) âl
j l0 + H.c.
0 0 for k ≥ 0 ,
2
where  ν  ν
∞  ν ˇ†
â ˇl
â
ˇ ˇ† ˇ def − 21 ηl20 j X l0 0
F̂ k (â l0 âl0 ) = e (i ηl0 j )|k| −ηl20 j . (3.58)
ν=0 ν! (ν + |k|)!
DRAFT, October 17, 2007
49
In the quadrupole approximation (recall Footnote 45) we have to add
2 D E
def
λ′j = − qel ej (r̂j · E0 )(r̂j · kL ) gj .
h̄

50
For operators in finite dimensional vector spaces (3.55) may be proved as follows: Since


λÂ
e B̂ e−λ and exp adλ B̂ fulfill the same first order differential equation and initial condition
(for λ = 0), the Campbell-Hausdorff formula (3.12) holds for arbitrary  , B̂ . Therefore, also
def
f1 (λ) = eλ Â eλ B̂

and 2

f2 (λ) = e (
def λ Â+B̂ )+ λ2 [Â,B̂ ]
−

fulfill the same first order differential equation and initial condition (for λ = 0), if the l.h.s. of
(3.55) holds, and hence f1 = f2 .
84 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

Multiplying V̂jI from the right with

∞
X
1̂osc = |niosc oschn|
n=0

and inserting
σ̂j+ = |ej ihgj |
finally yields
∞    
h̄ X n,k n,k †
V̂jI = Ωn,k n,k
j Âj + Ωj Âj , (3.59)
2 n=0
with (
def |ej ihgj | ⊗ |n + |k|iosc oschn| for k ≥ 0 ,
Ân,k
j = (3.60)
|ej ihgj | ⊗ |niosc oschn + |k|| for k ≤ 0 ,
and
s
n!  
def − 21 ηl2 |k|
Ωn,k
j = λj e 0j (i ηl0 j ) L|k| η 2
, (3.61)
(n + |k|)! n l0 j
!
def
n
X (−x)ν n+α
Lαn (x) = (generalized Laguerre polynomials) .
ν=0 ν! n−ν

Outline of proof for (3.61): We have to show

 ν+|k|
ν s
∞
X ˇ†
â ˇl
â
ν l0 0 n!
(−x) |niosc oschn| = L|k| (x) |n + |k|iosc oschn|
ν=0
ν! (ν + |k|)! (n + |k|)! n

and
 ν
ν+|k| s
∞
X ˇ†
â ˇl
â
ν l0 0 n!
(−c) |n + |k|iosc oschn + |k|| = L|k| (x) |niosc oschn + |k|| .
ν=0
ν! (ν + |k|)! (n + |k|)! n

Since  ν
ν  †  †   † 
ˇ† ˇl ˇ â ˇ ˇ ⡠ˇ ˇ
âl0 â 0
= â l0 l0 âl0 l0 − 1 . . . âl0 âl0 − (n − 1)

(see Equation 1.63 of (Lücke, nlqo)), the first of these equations follows from
 ν+|k|
ν  |k|

ˇ†
â ˇl
â |niosc = ˇ†
â n (n − 1) . . . n − (ν − 1) |niosc
l0 0 l0

q (n+|k|)!
= n (n − 1) . . . n − (ν − 1) n! |n + |k|iosc
(3.48)
s

 n! (n + |k|)!
|n + |k|iosc for n ≥ ν
= (n + |k|)! (n − ν)!


0 else ,
3.3. COLD TRAPPED IONS 85

the second from

 † ν
 † ν

ˇ
â ˇl ν+|k| |n + |k|i
â ˇ
= |niosc oschn| â ˇl ν+|k| |n + |k|i
â
l0 0 osc l0 0 osc

   † ν+|k|  !†

ν
= n + |k| â ˇ ˇl
â n |ni
osc l0 0 osc
s osc

 n! (n + |k|)!
 |niosc for n ≥ ν
= (n + |k|)! (n − ν)!


0 else .

The propagator in the interaction picture is the exponential51

∞  
i I X t  n,k 
e− h̄ V̂j t = cos Ωn,k
j B̂j + Ĉjn,k
n=0 2
  ! (3.62)
n,k t ek,j
n,k −i Φ
− i sin Ωj Âj e + H.c. + D̂jn,k ,
2

where  
e def n,k π
Φ k,j = arg Ωj = φj − |k| (3.63)
(3.61),(3.54) 2
and (
def |ej ihej | ⊗ |n + |k|iosc oschn + |k|| for k ≥ 0 ,
B̂jn,k =
|ej ihej | ⊗ |niosc oschn| for k ≤ 0 ,
(
def |gj ihgj | ⊗ |niosc oschn| for k ≥ 0 ,
Ĉjn,k =
|gj ihgj | ⊗ |n + |k|iosc oschn + |k|| for k ≤ 0 ,

 |k|−1
 X

 |ej ihej | ⊗ |niosc oschn| for k ≥ 0 ,


def
D̂jn,k = n=0
 |k|−1

 X

 |gj ihgj | ⊗ |niosc oschn| for k ≤ 0 .

n=0

Outline of proof for (3.62): By (3.60) we have

 †  |gj ihej | ⊗ |ni
osc oschn + |k|| for k ≥ 0
Ân,k
j =
|gj ihej | ⊗ |n + |k|iosc oschn| for k ≤ 0

DRAFT, October 17, 2007

51
Usually Ωn,k
j is called the Rabi frequency for the transition

|gj i ⊗ |n + |k|iosc ⇀
↽ |ej i ⊗ |niosc if k ≤ 0 ,

resp.
|gj i ⊗ |niosc ⇀
↽ |ej i ⊗ |n + |k|iosc if k ≥ 0 .
86 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

and therefore
Ânj 1 ,k Ânj 2 ,k = 0,
 †
Ânj 1 ,k Ânj 2 ,k = δn1 n2 B̂jn1 ,k , (3.64)
 †
Ânj 1 ,k Ânj 2 ,k = δn1 n2 Ĉjn1 ,k

for k ≥ 0 as well as for k ≤ 0 . This implies, first of all,

 
t  n,k n,k
∞
Y
i I
e− h̄ V̂j t = exp −i Ωj Âj + H.c. . (3.65)
n=0
2

Moreover, (3.64) implies

 2 2  
Ωn,k
j Ân,k
j + H.c. = Ωn,k
j B̂jn,k + Ĉjn,k ,
 2ν+1 2ν+1  
Ωn,k Ân,k + H.c. = Ωn,k Ân,k ek,j + H.c.
e−i Φ
j j j j

and hence
   
t  n,k n,k 
n,k t

exp −i Ωj Âj + H.c. = 1̂ + 1 − cos Ωj B̂jn,k + Ĉjn,k
2   2 
n,k t
−i sin Ωj Ân,k
j e−i φej + H.c. .
2
Inserting this into (3.65) yields (3.62).

Especially, we have
i I
e− h̄ V̂j t
∞  
X t  
= cos Ωn,0
j |ej ihej | ⊗ |niosc oschn| + |gj ihgj | ⊗ |niosc oschn| (3.66)
n=0 2
  !
t  
− i sin Ωn,0
j |ej ihgj | ⊗ |niosc oschn| e−i φj + H.c. for k = 0
2

and
i I
e− h̄ V̂j t
∞  
X t  
= cos Ωn,1
j|ej ihej | ⊗ |niosc oschn| + |gj ihgj | ⊗ |n + 1iosc oschn + 1|
n=0 2
  !

n,1 t −i (φj −π/2)
− i sin Ωj |ej ihgj | ⊗ |niosc oschn + 1| e + H.c.
2
+ |gj ihgj | ⊗ |0iosc osch0| for k = −1 .
(3.67)
3.3. COLD TRAPPED IONS 87

Laser Pulses for Quantum Computation

An exterior electromagnetic field of the form52 (plane-wave), (3.56) acting on the
j-th ion during the time interval
∆t = 2 ϕ/ Ωn,k
j
 
is called a ϕ-pulse for (n, k) . Examples for the action exp − h̄i V̂jI ∆t of these
pulses are:
1. ϕ-pulse for (0,0) with φj = ψ :
 
α |gj i ⊗ |0iosc + β |ej i ⊗ |0iosc 7−→ α cos ϕ − i e+i ψ β sin ϕ |gj i ⊗ |0iosc
 
+ β cos ϕ − i e−i ψ α sin ϕ |ej i ⊗ |0iosc .
π π
2. 2
-pulse for (0,-1) with φj = 2
:

|gj i ⊗ |0iosc 7−→ + |gj i ⊗ |0iosc ,

|ej i ⊗ |0iosc 7−→ −i |gj i ⊗ |1iosc .

3. π-pulse for (0,-1) with φj = 0 :

|gj i ⊗ |0iosc 7−→ + |gj i ⊗ |0iosc ,
|gj i ⊗ |1iosc 7−→ − |gj i ⊗ |1iosc ,
|ej i ⊗ |0iosc 7−→ − |gj i ⊗ |0iosc ,
|ej i ⊗ |1iosc 7−→ + |gj i ⊗ |1iosc .

Obviously, the action of a ϕ-pulse is that of a 1-qubit rotation, i.e. it is de-

scribed by the matrix
  n o
cos ϕ −i e+iψ sin ϕ
w.r.t. |gj i =
b |0i , |ej i =
b |1i ,
−i e−iψ sin ϕ cos ϕ
as long as nl0 = 0 . Moreover, suitable laser pulses for implementing the CNOT gate
as described in 3.3.1 are the following:
π
(3.17) : a -pulse for (0, −1) with φj = π2 on ion j1
2
(3.18) : a π-pulse for (0, −1) with φj2 = 0 on ion Ej2 ,
tuned to a transition |gj2 i ⊗ |1iosc ⇀
↽ e′j2 |0iosc
E
with e′j2 sufficiently different from |ej2 i ,
π
(3.19) : a -pulse for (0, −1) with φj = π2 on ion j1 .
2
Exercise 16 DRAFT, October 17, 2007
52
Note that for electromagnetic radiation the electric field uniquely fixes the magnetic field and
vice versa.
88 CHAPTER 3. PHYSICAL REALIZATIONS OF QUANTUM GATES

This shows that a sufficiently large class of quantum gates may be implemented,
provided that the oscillatory modes can be cooled to their ground state.

Show the following:

a) The matrices ei σ̂1 ϕ and ei σ̂2 ϕ correspond to special 1-qubit rotations, for all
ϕ ∈ IR .
π π
b) e−i σ̂1 4 σ̂2 e+i σ̂1 4 = σ̂3 .

c) Every Û ∈ SU(2) may be represented in the form53

ψ θ φ
Û = ei σ̂2 2 ei σ̂1 2 ei σ̂2 2 , ψ , θ , φ ∈ IR .

3.3.4 Laser Cooling

See (Metcalf and van der Straten, 1999).

DRAFT, October 17, 2007

53
Recall the beginning of the proof of Lemma 1.2.2.
 Part II

Fault Tolerant
Quantum Information Processing

89
Chapter 4

General Aspects of Quantum

Information

An open system is nothing more than one which has interactions with
some other environment system, whose dynamics we wish to neglect, or
average over.

(Nielsen and Chuang, 2001, p. 353)

4.1 Introduction
Quantum information theory deals with transmission and quantification of quantum
information. Roughly speaking, quantum information is the information carried
by n-qubit systems (n ∈ IN). Of the utmost importance for quantum information
— as opposed to classical information,1 — are:

1. The quantum mechanical superposition principle:

The pure states of a quantum system are in 1-1-correspondence with

the 1-dimensional subspaces of a complex Hilbert space (unless
there are superselection rules2 ).

2. The no-cloning theorem (Dieks, 1982; Wootters and Zurek, 1982; Peres, 2002):

There cannot exist any recipe for preparing any two or more systems
such that each of them carries the same quantum information as a
given quantum system — unless the state of the latter is completely
known, of course.

DRAFT, October 17, 2007

1
See (Shannon, 1949) for the theory of classical information.
2
See (Verstraete and Cirac, 2003; Bartlett and Wiseman, 2003) for the case with superselection
rules.

91
92 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

Remark: There are several justifications for the no-cloning theorem,3 e.g.:

• Given Ψ0 ∈ H , there is no linear extension of the mapping

def
C(Ψ ⊗ Ψ0 ) = Ψ ⊗ Ψ for all normalized Ψ ∈ H

to all of H ⊗ H .
• Cloning would allow measurement of incommensurable quantum observables
— impossible according to quantum mechanics.
• Cloning would, by proper use of one of the available Bell sources, allow for
superluminal communication4 — impossible according to special relativity.

While the superposition principle opens up the fascinating possibilities of quan-

tum computation, the impossibility of cloning unknown quantum states strongly
limits the amount of information that can be read out from quantum states.5 Thus,
e.g., it is impossible to distinguish nonorthogonal states6 by a single measurement.
Nevertheless, quantum information seems to offer a wealth of useful applications
without any classical equivalent.

We say that quantum information is transmitted — to whatever degree intact —

through a quantum (rather than classical ) channel if the information is sent
using systems whose quantum character cannot be neglected in this respect. In
other words:
We identify quantum channels with open7 quantum systems, used
for quantum communication.
For practical communication it is important that information can be transferred
in a reversible way from one physical system to another. The no-cloning theorem
implies that unknown quantum information cannot be transferred from quantum
to purely classical systems in a reversible way. This is why quantum channels and
entanglement-assisted classical channels are the main topic of these lectures.

In the following, unless stated otherwise, we will always work in the interaction
picture and make extensive use of Dirac’s bra-ket notation.8 For simplicity, we
consider only finite-dimensional Hilbert spaces. This is sufficient for clarify-
ing the main points.
DRAFT, October 17, 2007
3
See (Werner, 2001, Section 2.3) for a detailed discussion.
4
See (Herbert, 1982). Similarly, joint measurability (without cloning) of non-commuting ob-
servables would enable superluminal communication.
5
On the other hand, the non-cloning theorem constitutes the basis for secure quantum cryp-
tography (see http://www.idquantique.com) and quantum passwords (Gu and Weedbrook, 2005).
6
We tacitly identify states with their density matrices or wave functions (if pure).
7
We have to consider open quantum systems since quantum information is prone to quantum
noise caused by interaction with the environment.
8
See, e.g., (Lücke, eine).
4.2. QUANTUM CHANNELS 93

4.2 Quantum Channels

4.2.1 Open Quantum Systems and Quantum Operations
Let us consider two quantum systems S1 resp. S2 with (finite dimensional) state
spaces H1 resp. H2 and consider S2 as the environment of S1 ; i.e. let us assume the
bipartite System S composed of S1 and S2 to be closed. Moreover, let us assume
that S1 is prepared in a pure state9 ρ̂(1) ∈ S(H1 ) at time 0 . Then, at time 0 there
are no correlations between S1 and S2 and the state of S is of the form
ρ̂0 = ρ̂(1) ⊗ ρ̂(2) (4.1)
with ρ̂(2) ∈ S(H2 ) . At time t , then (see, e.g., Section 6.1.2 of (Lücke, nlqo)), the
state of S is
ρ̂t = Û ρ̂(1) ⊗ ρ̂(2) Û † , (4.2)
where the unitary operator Û on the state space H1 ⊗ H2 of system S is given by
def i i
Û = e+ h̄ Ĥ0 t e− h̄ Ĥ t ,
(1) (2)
Ĥ resp. Ĥ0 = Ĥ0 ⊗ Ĥ0 being the actual resp. free Hamiltonian of S . The partial
state of S1 at time t , therefore, is
(1)
ρ̂t = C(ρ̂(1) ) , (4.3)
where10    
def
C(ρ̂′ ) = trace 2 Û ρ̂′ ⊗ ρ̂(2) Û † ∀ ρ̂′ ∈ S(H1 ) . (4.4)
Thanks to the spectral theorem, the initial state of the environment can be written
in the form n n
X2 ED X2
(2) (2)
ρ̂(2) = λβ φβ φβ , λβ = 1 (4.5)
β=1
|{z} β=1
≥0
n o
(2)
with some orthonormal basis φ1 , . . . , φ(2)
n2 of H2 . (4.4) and (4.5) imply11
n2
X D  ED  E
(2) (2)
C(ρ̂′ ) = λβ ψα(2) Û ρ̂′ ⊗ φβ φβ Û † ψα(2)
α,β=1

n2
X †
= K̂α,β ρ̂′ K̂α,β ∀ ρ̂′ ∈ S(H1 ) (4.6)
α,β=1

DRAFT, October 17, 2007

9
By S(H) we denote the set of all states, i.e. of all positive operators ρ̂ on the Hilbert space
H with trace (ρ̂) = 1 . L(H) , as usual, denotes the set of all bounded linear operators on H .
10
Note that C depends not only on Ĥ and Ĥ0 but also on the initial state of the environment!
11
We use the notation
!
D XN E XN D E
(2) def
φ Âk ⊗ B̂k ψ (2) = Âk φ(2) B̂k ψ (2)
k=1 k=1

for φ(2) , ψ (2) ∈ H2 , Â1 , . . . , ÂN ∈ H1 , and B̂1 , . . . , B̂N ∈ H2 .

94 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION
n o
(2)
for every orthonormal basis ψ1 , . . . , ψn(2)
2
of H2 , where
q D E
def (2)
K̂α,β = λβ ψα(2) Û φβ ∈ L(H1 ) ∀ α, β ∈ {1, . . . , n2 }
and, therefore,
n2
X n2
X D ED E
† (2) (2)
K̂α,β K̂α,β = λβ φβ Û † φ(2)
α φ(2)
α Û φβ
α,β=1 α,β=1
Xn2 D E
(2) (2)
= λβ φβ | Û † Û φβ
β=1 | {z }
=1
= 1.
(4.5)

If, from an ensemble in the state (4.2), those individuals are selected (by projective
ED
measurement of the environment) for which S2 is in the partial state ψα(2) ψα(2)
then (4.6) has to be replaced by12
n2
X †
C(ρ̂′ ) = K̂α,β ρ̂′ K̂α,β ∀ ρ̂′ ∈ S(H1 )
β=1

and  
trace C(ρ̂) ≤ 1
is the relative number of individuals selected. In either case E is just a special
quantum operation:13

Definition 4.2.1 Let H and H′ be Hilbert-spaces. Then by Q(H, H′ ) we denote

the set of all mappings C from S(H) into L(H′ ) of the form
N
X
C(ρ̂) = K̂k ρ̂ K̂k† ∀ ρ̂ ∈ S(H) , (4.7)
k=1

with suitable N ∈ IN and K̂k ∈ L(H, H′ ) fulfilling14

N
X
K̂k† K̂k ≤ 1̂ . (4.8)
k=1

The elements of Q(H, H′ ) are called quantum operations and the K̂k in (4.7) are
called Kraus operators for C .

DRAFT, October 17, 2007

12
For a corresponding representation of general linear maps see (Shabani and Lidar, 2006, The-
orem1).
13
See (Buscemi et al., 2003) for efficient realizations and (Werner, 2001, Section 2.6.2) for the
dual action of quantum operations on the observable algebras.
14
Note that the adjoint K̂ † of a linear mapping K̂ ∈ L(H, H) is characterized by
D E D E
ψ ′ K̂ ψ = K̂ † ψ ′ ψ ∀ ψ ∈ H , ψ ′ ∈ H′ .
H H′
†
Thus, e.g., (|ψ ′ ihψ|) = |ψihψ ′ | .
4.2. QUANTUM CHANNELS 95

Remarks:
1. Note that15
 
0 ≤ trace C(ρ̂) ≤ 1 ∀ ρ̂ ∈ S(H1 )
| {z }
≥0

and that
N
X  
K̂k† K̂k = 1̂ ⇐⇒ trace C(ρ̂) = 1 ∀ ρ̂ ∈ S(H1 ) .
k=1

2. Obviously, we also have

C1 ∈ Q(H1 , H2 ) , C2 ∈ Q(Ĥ2 , H3 ) =⇒ C2 ◦ C1 ∈ Q(H1 , H3 ) .

3. But, given C1 ∈ Q(H1 , H2 ) and C3 ∈ Q(H1 , H3 ) , this does not

guarantee existence of an C2 ∈ Q(H2 , H3 ) with C3 = C2 ◦ C1 .
4. Of course, nonexistence of such C2 is only possible if C1 is not in-
vertible. The latter is obviously the case if, e.g., H1 = H2 and
3
1X
C1 (ρ̂) = τ̂ ν ρ̂ τ̂ ν
4 ν=0
1
= 1̂ ∀ ρ̂ ∈ S(H) .
2
5. The linear extension
N
X
def
C̄(Â) = K̂k  K̂k† ∀  ∈ L(H1 ) ,
k=1

of C to all of L(H1 ) is completely positive,16 i.e.:

 
Â′ ≥ 0 =⇒ 1 ⊗ C̄ (Â′ ) ≥ 0 ∀ Â′ ∈ Cn ⊗H1 , n ∈ IN .

Lemma 4.2.2 Let H1 and H2 be finite-dimensional Hilbert-spaces17 and let C

be a mapping from S(H1 ) into L(H2 ) . Then C is a quantum operation, i.e. C ∈
Q(H1 , H2 ) , iff the following three conditions are fulfilled:
DRAFT, October 17, 2007
 
15
The possibility trace C(ρ̂) < 1 is included to allow for absorption.
| {z }
≥0
16
Every trace preserving affine mapping of S(H) into S(H) can be represented as (the restriction
of) a difference of two completely positive mappings (Kuah and Sudarshan, 2005).
17
Since we agreed to consider only finite-dimensional Hilbert spaces, all operators on H1
resp. H2 are of trace class. For infinite-dimensional Hilbert spaces H1 = H2 see (Davies, 1976,
Theorem 2.3 and notes on page 147).
96 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

1.  
trace C(ρ̂) ≤ 1 ∀ ρ̂ ∈ S(H1 ) .

2.
 
C λ ρ̂1 + (1 − λ) ρ̂2 = λ C(ρ̂1 ) + (1 − λ) C(ρ̂2 ) ∀ λ ∈ [0, 1] , ρ̂1 , ρ̂2 ∈ S(H1 ) .

3.  
(1 ⊗ C̄) |ΨihΨ| ≥ 0 ∀ Ψ ∈ H1 ⊗ H1 ,
where C̄ denotes the unique linear extension18 of C to all of L(H1 ) .

Outline of
n proof: Assume o that C fulfills the requirements 1–3. Choose an orthonor-
(1) (1)
mal basis φ1 , . . . , φn1 of H1 and defining

n1 D
X E
def
ψ∗ = ψ φ(1)
ν φ(1)
ν ∀ ψ ∈ H1 (4.9)
ν=1

we get D E D E
ψ ∗ φ(1)
ν = φ(1)
ν ψ ∀ ψ ∈ H1 , ν ∈ {1, . . . , n1 }

and hence19
  n1 D
X ED E  ED 
C |ψihψ| = φ(1) ψ ψ φ(1) ¯ φ(1) φ(1)
C
ν µ ν µ
2. req.
ν,µ=1
Xn1  ED  ED 
= hψ |∗
φ(1) φ(1) ¯
⊗ C φν(1) (1)
φµ |ψ ∗ i
ν µ
ν,µ=1

= hψ ∗ | Â |ψ ∗ i ∀ ψ ∈ H1 , kψk = 1 , (4.10)

where
n1
X ED  ED 
Â
def
= φ(1) φ(1) ¯ φ(1) φ(1)
⊗C
ν µ ν µ
ν,µ=1 !
  n1
X ED ED
= ¯
1⊗C φ(1) φ(1) ⊗ φ (1)
φ(1)
ν µ ν µ
ν,µ=1
| {z }
Pn1 (1) (1)
Pn 1 (1) (1)
= φν ⊗φν φν ⊗φν ≥0
ν=1 ν=1

DRAFT, October 17, 2007

18
Existence of this extension, is guaranteed by the second condition. For self-adjoint  it is
given by
   
¯ (Â) def Â+ Â−
C = trace (Â+ ) C + trace (Â− ) C ,
(trace Â+ ) (trace Â− )
where Â+ resp. Â− denotes the positive resp. negative part of  :

 = Â+ + Â− ± ± ≥ 0 , Â+ Â− = 0̂ .

Linear mappings from L(H1 ) into L(H2 ) are also called superoperators.
19
We use the notation explained in Footnote 11, with the roles of the tensor factors H1 , H2
interchanged.
4.2. QUANTUM CHANNELS 97

and hence
0 ≤ Â ∈ L(H1 ⊗ H2 ) .
3. req.

This, together with the spectral theorem implies

N
X
 = |Ψk ihΨk | for suitable Ψ1 , . . . , ΨN ∈ H1 ⊗ H2 .
k=1

The latter, together with (4.10) gives

  N
X
¯ |ψihψ| =
C hψ ∗ | |Ψk ihΨk | |ψ ∗ i ∀ ψ ∈ H1 .
k=1

Thus, defining the linear mappings20

def
K̂k ψ = hψ ∗ | |Ψk i ∀ ψ ∈ H1 , k ∈ {1, . . . , N }

from H1 into H2 we get

  XN
C |ψihψ| = K̂k |ψihψ| K̂k† ∀ ψ ∈ H1 , kψk = 1 .
k=1

By condition 2 this gives (4.7). The latter together with condition 1, finally, gives21
(4.8).
Conversely, conditions 1–3 are easily seen to be fulfilled for C ∈ Q(H1 , H2 ) .

Remark: At first glance one might consider conditions 1–3 of Lemma

4.2.2 as natural requirements on the action of quantum channels. Note,
however, that the output state of a quantum channel need not even be
determined by the input state in case of initial correlations with the
environment.22

Definition 4.2.1 and Lemma 4.2.1 describe the most general quantum operations.
Simple versions are, among others:
DRAFT, October 17, 2007
20
Here, we use the notation
′ + N D ′
N
X X E
(1) (2) def (1) (2)
hψ| ψj ⊗ ψj = ψ ψj ψj
j=1 j=1

(1) (1) (2) (2)

for ψ, ψ1 , . . . , ψN ′ ∈ H1 and ψ1 , . . . , ψN ′ ∈ H2 .
PN
21
Violation of (4.8) would imply existence  of  a normalized eigenvector ψ+ of k=1 K̂k† K̂k with
eigenvalue greater than 1 and thus trace C(ρ̂) > 1 for ρ̂ = |ψ+ ihψ+ | .
22
Recall Footnote 10. See also (Kuah and Sudarshan, 2005), in this connection.
98 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

1. Unitary transformations

ρ̂′ 7−→ C(ρ̂′ ) = V̂ ρ̂ V̂ †

given, e.g., by (4.4) for

Û = V̂ × V̂ ′ , V̂ unitary .

2. Complete projective measurement operations

n1
X n o
(1)
ρ̂′ 7−→ C(ρ̂′ ) = P̂φ(1) ρ̂ P̂φ(1) , φ1 , . . . , φ(1)
n1 orthonormal basis of H1 ,
j j
j=1

(2)
given, e.g., by (4.4) if there are ρ̂1 , . . . , ρ̂(2)
n1 ∈ S(H2 ) with pairwise orthogonal
23
supports and such that
(2)
Û P̂φ(1) ⊗ ρ̂(2) Û † = P̂φ(1) ⊗ ρ̂j ∀ j ∈ {1, . . . , n1 } .
j j

Example:24
Û = action of CNOT ,
S1 = control qubit ,
S2 = target qubit , 
ρ̂(2) = |0ih0| or |1ih1| .

3. Cascaded complete projective measurement operations

n1
X
ρ̂′ 7−→ C(ρ̂′ ) = K̂(j1 ,...,jr ) ρ̂′ K̂(j† 1 ,...,jr ) ,
j1 ,...,jr =1

where
K̂(j1 ,...,jr ) = P̂r,jr · · · P̂1,j1 ∀ (j1 , . . . , jr ) ∈ {1, . . . , n1 }r
with P̂k,1 , . . . , P̂k,n1 being the projectors of the k-th projective measurement.
Remark: Note that
n1
X †
K̂(j1 ,...,jr )
K̂(j1 ,...,jr ) = 1
j1 ,...,jr =1

but that, in general, the K̂(j1 ,...,jr ) are no longer projection operators.

DRAFT, October 17, 2007

23
In this case nondemolition measurements on S1 can be performed by corresponding (usually
destructive) measurements on S1 . Note, however, that for the resulting partial state of S1 it does
not matter whether the state of the environment, including the measurement apparatus, is checked
or not.
24
See, e.g., (Lücke, 2002, Chapter 1) for a detailed discussion of the CNOT gate. Another
example would be the Stern-Gerlach measurement if it really worked as usually described.
4.2. QUANTUM CHANNELS 99

Theorem 4.2.3 Let H1 and H2 be Hilbert-spaces and let K̂1 , . . . , K̂N , K̂1′ , . . . ,
K̂N ∈ L(H1 , H2 ) . Then
N
X N
X
K̂k ρ̂ (K̂k )† = K̂k′ ρ̂ (K̂k′ )† ∀ ρ̂ ∈ S(H1 ) (4.11)
k=1 k=1
 
iff there is a unitary N × N -matrix U jk with

N
X
K̂j′ = U jk K̂k ∀ j ∈ {1, . . . , N } . (4.12)
k=1

Outline of proof: Assume that (4.11) holds. Then, if we choose an orthonormal

basis {φ1 , . . . , φn } of H1 and define
n
X  
def
Ψk = φν ⊗ K̂k φν ,
ν=1
Xn  
def
Ψ′k = φν ⊗ K̂k′ φν
ν=1

for k ∈ {1, . . . , N } , we get

N
X n 
N X
X   
|Ψk ihΨk | = |φν ihφµ | ⊗ K̂k |φν ihφµ | (K̂k )†
k=1 k=1 ν,µ=1
XN X n    
= |φν ihφµ | ⊗ K̂k′ |φν ihφµ | (K̂k′ )†
(4.11)
k=1 ν,µ=1
XN
= |Ψ′k ihΨ′k | .
k=1
 
Therefore, by Corollary A.4.3, there is a unitary N × N -matrix Uj k with

N
X
Ψk = Uk j Ψ′j ∀ k ∈ {1, . . . , N } ,
j=1

i.e. with
 
n
X   n
X N
X
φν ⊗ K̂k φν = φν ⊗  Uk j K̂j′ φν  ∀ k ∈ {1, . . . , N } .
ν=1 ν=1 j=1

This implies
N
X
K̂k φν = Uk j K̂j′ φν ∀ ν ∈ {1, . . . , n} , k ∈ {1, . . . , N }
j=1

and hence (4.12).

Conversely, it is obvious that (4.12) implies (4.11).
100 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

The standard example, in case n1 > 1 , for a mapping C of S(H1 ) into S(H1 )
fulfilling conditions 1 and 2, but not 3, of Lemma 4.2.2 is the transposition 25
n1
X ED n1
X ED
(1) (1) def (1) (1)
ρ̂′ = ρj k φj φk 7−→ C(ρ̂′ ) = T(ρ̂′ ) = ρj k φk φj . (4.13)
j,k=1 j,k=1

n o
(1)
depending on the basis φ1 , . . . , φ(1)
n1 of H1 .

Proof of positivity: Thanks to the spectral theorem it is sufficient to show that

Spure (H1 ) is left invariant under transposition. This, however follows from26
 
T |φihψ| = |ψ ∗ ihφ∗ | ∀ φ, ψ ∈ H1 . (4.14)

Disproof of complete positivity: Obviously, it is sufficient to check the case

n1 = 2 . Then, with
def (1) (1)
|jki = φj ⊗ φk ∀ j, k ∈ {1, 2} ,
we have
¯ )(Ŵ ) = 1 − λ 
(1 ⊗ T λ |0, 0ih0, 0| + |1, 1ih1, 1|
4
1 + λ 
+ |0, 1ih0, 1| + |1, 0ih1, 0| (4.15)
4
λ 
− |1, 1ih0, 0| + |0, 0ih1, 1|
2
for the Werner states27
def 1−λ
Ŵλ = 1̂ ⊗ 1̂ + λ Ψ− Ψ− , λ ∈ [0, 1] .
4
But for λ > 1/3 (4.15) cannot be positive, since, e.g.
    
¯ )(Ŵ ) |0, 0i + |1, 1i = 1 − 3λ |0, 0i + |1, 1i .
(1 ⊗ T λ
4

DRAFT, October 17, 2007

25
Note that, for ρ̂′ ∈ S(H1 ) , transposition is equivalent to complex conjugation
n1
X ED n1 
X ∗ ED
(1) (1) (1) (1)
ρ̂′ = ρj k φj φk 7−→ ρj k φj φk .
j,k=1 j,k=1

26
Recall (4.9).
27
These states are distinguished by their invariance property
   †
Û ⊗ Û Ŵλ Û ⊗ Û = Ŵλ for all unitary Û ∈ L(H)

(Werner, 1989, Section II). Note that the flip operator

|0, 0ih0, 0| + |1, 0ih0, 1| + |0, 1ih1, 0| + |1, 1ih1, 1|

coincides with 1̂ ⊗ 1̂ − 2 |Ψ− ihΨ− | .

4.2. QUANTUM CHANNELS 101

4.2.2 Quantum Noise and Error Correction

The action of noisy quantum channels corresponds to non-invertible quantum oper-
ations C . Nevertheless such an operation may be become invertible by restriction
to states with support on a suitable subspace C of H — and thus allow for error
correction on the code space C .

Theorem 4.2.4 Let H be Hilbert space, C ∈ Q(H, H) and let C be a linear

subspace of H . Then the following three statements are equivalent:
1. There are Kraus operators K̂1 , . . . , K̂N for C and a1 , . . . , aN ≥ 0 with

P̂C K̂j† K̂k P̂C = δjk aj P̂C ∀ j, k ∈ {1, . . . , N } . (4.16)

2. There is a trace preserving quantum operation R with

 
(R ◦ C) |ψihψ| ∝ |ψihψ| ∀ ψ ∈ C kψk = 1 . (4.17)

3. There are Kraus operators K̂1′ , . . . , K̂N′ for C with

†
P̂C K̂j′ K̂k′ P̂C = ajk P̂C ∀ j, k ∈ {1, . . . , N } (4.18)

for some self-adjoint matrix (ajk ) .

Outline of proof: Assume the first statement to be true. Then, using the polar
decomposition
q
+ †
K̂j P̂C = Ûj P̂C K̂j K̂j P̂C
√
= + aj Ûj P̂C (4.19)

(see, e.g., Lemma 7.3.20 of (Lücke, eine)) we have

δjk P̂C = P̂C Ûj† Ûk P̂C (4.20)

and hence
P̂Ûj C P̂Ûk C = δjk P̂Ûj C . (4.21)
For
N 
X   †
R(ρ̂) def
= Ûj† P̂Ûj C ρ̂ Ûj† P̂Ûj C + P̂0 ρ̂ P̂0 ∀ ρ̂ ∈ S(H) , (4.22)
j=1

where
N
X
def
P̂0 = 1̂ − P̂Ûj C (4.23)
j=1

and thus
  N
X
P̂0 C P̂C ρ̂ P̂C P̂0 = aj P̂0 Ûj P̂C ρ̂ P̂C Ûj† P̂0
(4.19)
j=1
= 0 ∀ ρ̂ ∈ S(H) ,
(4.21)
102 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

this gives28
  N
X †
(R ◦ C) |ψihψ| = P̂C Ûj† P̂Ûj C K̂k |ψi hψ| K̂k P̂Ûj C Ûj P̂C (4.24)
| {z }
j,k=1 √
= + ak Ûk |ψi
4.19)
(
| {z }
+
√
= δjk ak Ûk |ψi
( 4.21)
XN
= aj |ψihψ| ∀ψ ∈ C .
(4.20)
j=1

Since
N 
X †  
Ûj† P̂Ûj C Ûj† P̂Ûj C + P̂0† P̂0 = 1̂ ,
(4.21)
j=1

this implies the second statement.29 Now assume

  the second
 statement to be
true. Then (4.17) holds and, of course,30 trace C |ψihψ| must be constant for
normalized ψ ∈ C . Therefore,

 
R ◦ C P̂C ρ̂ P̂C = γ P̂C ρ̂ P̂C ∀ ρ̂ ∈ S(H) (4.25)

holds for some γ ≥ 0 . If K̂1′ , . . . , K̂N

′
are Kraus operators for C and R̂1 , . . . , R̂N are
Kraus operators for R then (4.25) is equivalent to
N
X †
R̂j K̂k′ P̂C ρ̂ P̂C K̂k′ R̂j† = γ P̂C ρ̂ P̂C ∀ ρ̂ ∈ S(H) .
j,k=1

Then,31 by Theorem 4.2.3, there are complex numbers λjk with

R̂j K̂k′ P̂C = λjk P̂C ∀ j, k ∈ {1, . . . , N }
and hence
†
P̂C K̂l′ R̂j† R̂j K̂k′ P̂C = λ∗jl λjk P̂C ∀ j, k, l ∈ {1, . . . , N } .
Since R is trace preserving, this implies the third statement with
N
X
alk = λ∗jl λjk ∀ k, l ∈ {1, . . . , N } .
j=1

Finally, assume the third statement to be true. Then, by the spectral theorem,
there are a unitary matrix (ujk ) and real numbers a1 , . . . , aN with
N
X
u∗j ′ j aj ′ k′ uk′ k = aj δj,k ∀ j, k ∈ {1, . . . , N } .
j ′ ,k′ =1

DRAFT, October 17, 2007

28
Note that P̂C Ûj† P̂Ûj C = Ûj† P̂Ûj C .
29
Recall Remark 1 to Definition 4.2.1.
30
Note that for ρ̂1 , ρ̂2 ∈ S(H) and λ1 , λ2 ∈ C we have

λ1 ρ̂1 + λ2 ρ̂2 ∝ ρ̂1 + ρ̂2
=⇒ λ1 = λ2 .
ρ̂1 6= ρ̂2

31 √
We may add N − 1 zeros as Kraus operators to γ P̂C .
4.3. ERROR CORRECTING CODES 103

This, together with (4.18), implies the first statement with

N
X
K̂j = ukj K̂k′ ∀ j ∈ {1, . . . , N } .
k=1

Remark: (4.19), (4.21), and (4.22) show how errors produced by C can
be corrected for the code C :
Perform a projective measurement w.r.t. the orthogonal sub-
spaces Ûj C and apply Ûj† according to the result of this ‘mea-
surement’.

Corollary 4.2.5 Let C be a linear subspace of the Hilbert space H and let K̂1 , . . . ,
K̂N resp. K̂1′ , . . . , K̂N′ be Kraus operators for C ∈ Q(H, H) resp. C′ ∈ Q(H, H) . If
(4.16) holds and if the K̂j′ are complex linear combinations of the K̂j then, with R
as constructed in the proof of Theorem 4.2.4, (4.17) holds also for C replaced by C′ .

Outline of proof: Assume that

N
X
K̂j′ = λjk K̂k
k=1

and let R be defined as in the proof of Theorem 4.2.4. Then

 
(R ◦ C′ ) |ψihψ|
XN   √ √  †
= P̂C Ûj† P̂Ûj C λkl + ak Ûl P̂C |ψihψ| λ∗kr + ar P̂C Ûr† P̂C Ûj† P̂Ûj C
(4.24)
j,k,l,r=1
N
X 2
= aj |λkl | |ψihψ| ∀ψ ∈ C .
(4.21)
j,k=1

4.3 Error Correcting Codes

4.3.1 General Apects
According to standard formulations (Lücke, 1996) the time evolution of closed32
quantum mechanical systems is unitary.
Let us consider the closed system of one qubit together with its environment. A
unitary transformation of the corresponding state space maps separated pure states
to entangled pure states:
 
|χi = α |0i + β |1i ⊗ |Ei
    (4.26)
7−→ |χ′ i = α |0i ⊗ |E0,0 i + |1i ⊗ |E0,1 i + β |0i ⊗ |E1,0 i + |1i ⊗ |E1,1 i
DRAFT, October 17, 2007
32
For open quantum systems see (Alicki, 2003) and references given there.
104 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

(∀ α, β ∈ C). This way the originally pure partial state33

 
hχ| Â ⊗ 1̂ |χi = |α|2 h0| Â |0i + |β|2 h1| Â |1i + 2 ℜ α β h0| Â |1i ∀ Â ∈ Lsa (C2 )

may become a mixture:34

hχ′ | Â ⊗ 1̂ |χ′ i = |α|2 h0| Â |0i + |β|2 h1| Â |1i if 35 hEj,k | Elm i = δjl δk0 δm0 .

In other words:

The environment may cause decoherence.

Therefore, we have to be able to undo unwanted changes caused by the environment.

Now, for arbitrary vectors |Er i from the state space of the environment we have
3 
X     
σ̂r |0i ⊗ |Er i = |0i ⊗ |E0 i + |E3 i + |1i ⊗ |E1 i + i |E2 i (4.27)
r=0

and
3 
X     
σ̂r |1i ⊗ |Er i = |0i ⊗ |E1 i − i |E2 i + |1i ⊗ |E0 i − |E3 i , (4.28)
r=0

where the σ̂r correspond to the Pauli matrices:36

       
def 1 0 def 0 1 def 0 −i def 1 0
σ0 = , σ1 = , σ2 = , σ3 = .
0 1 1 0 +i 0 0 −1
(4.29)
Equations (4.26)–(4.28) imply
3 
X  
|χ′ i = σr α |0i + β |1i ⊗ |Er i (4.30)
r=0

if
|E0,0 i = |E0 i + |E3 i , |E0,1 i = |E1 i + i |E2 i ,
|E1,0 i = |E1 i − i |E2 i , |E1,1 i = |E0 i − |E3 i ,
i.e. if
|E0,0 i + |E1,1 i |E0,1 i + |E1,0 i
|E0 i = , |E1 i = ,
2 2
|E0,1 i − |E1,0 i |E0,0 i − |E1,1 i
|E2 i = , |E3 i = .
2i 2
DRAFT, October 17, 2007
33 2 2
We assume that |α| + |β| = 1 and hE | Ei = 1 .
34
This is why open quantum quantum mechanical systems (Davies, 1976) do not evolve unitarily.
35
Consider, e.g., the simple example |Ei = |0i , |χ′ i = CNOT |χi.
36
Hence σ̂0 = 1̂ , σ̂1 = ¬
ˆ , σ̂3 = Ŝπ , σ̂2 = i σ̂1 σ̂3 .
4.3. ERROR CORRECTING CODES 105

(4.30) tells us that there are only three types of errors to be corrected, corresponding
to37 σ̂1 , σ̂2 , σ̂3 . In this sense the set of possible errors for single-qubit systems is
discrete.

More generally, a unitary operation of the state space of an n-qubit system and
its environment acts according to38
X X X E
λb |bi ⊗ |Ei 7−→ λb |b′ i ⊗ Eb,b′ , (4.31)
b∈{0,1}n b∈{0,1}n b′ ∈{0,1}n
E
where the Eb,b′ are suitable state vectors of the environment depending on |Ei
(and b, b′ , of course), but not on the λb . Now, from (4.27)/(4.28) we see that for
arbitrary states |Eb,b′ i of the environment there are vectors |Er i with
3 
X  X
σ̂r |bi ⊗ |Er i = |b′ i ⊗ |Eb,b′ i ∀ b ∈ {0, 1} .
r=0 b′ ∈{0,1}
E
Straightforward induction shows that for arbitrary state vectors Eb,b′ there are
corresponding state vectors |Er (b)i ; r ∈ {0, . . . , 3}n ; with
X   X E
σ̂r |bi ⊗ |Er (b)i = |b′ i ⊗ Eb,b′ ∀ b ∈ {0, 1}n ,
n
r∈{0,1,2,3} b′ ∈{0,1}n

where
def
σ̂r = σ̂r1 ⊗ . . . ⊗ σ̂rn ∀ r ∈ {0, . . . , 3}n .
Together with (4.31) this shows that every unitary action on an n-qubit system39
and its environment is of the form
X X X  
λb |bi ⊗ |Ei 7−→ λb σ̂r |bi ⊗ |Er (b)i , ∀ b ∈ {0, 1}n .
b∈{0,1}n b∈{0,1}n r∈{0,1,2,3}n

Usually, in the theory of quantum error correction, only the case

|Er (b)i = |Er i ∀ b ∈ {0, 1}n , r ∈ {0, 1, 2, 3}n

DRAFT, October 17, 2007
37
Of course, the σ̂ν are not the only set of operators serving this purpose:
3
X 3
X 3
X
def def
σ̂r′ = urs σ̂s , |Er′ i = u∗rs |Es i , urq u∗qs = δrs
s=0 s=0 q=0
3
X 3
X



=⇒ σ̂r |bi ⊗ |Er i = σ̂r′ |bi ⊗ |Er′ i ∀ b ∈ {0, 1} .
r=0 r=0

38
This is a simple consequence of linearity and the fact that the |bi form a basis of the n-qubit
state space.
39
We assume that the qubits are not destroyed. For instance, if a qubit is identified with an
atom in a superposition of its ground state and its first excited state, then exciting a higher level
destroys this qubit.
106 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

is considered.40 Then the error action is of the form

X   X
Ψ ⊗ |Ei 7−→ σ̂r Ψ ⊗ |Er i , Ψ= λb |bi , (4.32)
r∈{0,1,2,3}n b∈{0,1}n

and error correction for such quantum noise should be possible along the lines
indicated below.

Remark: Alternatively, (4.32) may be written in the form

X   E
Ψ ⊗ |Ei 7−→ X̂a Ẑb Ψ ⊗ Ea,b , (4.33)
n
a,b∈{0,1}

where:
def
X̂b = (δ0b1 + δ1b1 σ̂1 ) ⊗ . . . ⊗ (δ0bn + δ1bn σ̂1 ) ,
def (4.34)
Ẑb = (δ0b1 + δ1b1 σ̂3 ) ⊗ . . . ⊗ (δ0bn + δ1bn σ̂3 ) .

To explain the essential idea of quantum error correction, let us assume that also
for multi-qubits systems only one-qubit errors corresponding to σ̂3 (phase errors)
occur. In order to conserve an unknown one-qubit state (disentangled from the
environment) we first of all encode

Ψ = α |0i + β |1i

for arbitrary α, β ∈ C into the (pure) three-qubit state

Ψ̂ = α |ŵ0 i + β |ŵ1 i ,

where
def 1  
|ŵ0 i = √ |0, 0, 0i + |0, 1, 1i + |1, 0, 1i + |1, 1, 0i (even parity)
4
(4.35)
def 1  
|ŵ1 i = √ |1, 1, 1i + |1, 0, 0i + |0, 1, 0i + |0, 0, 1i (odd parity) .
4
This may be done in the following way:

α |0i + β |1i h 


|0i H s h α |ŵ0 i + β |ŵ1 i .


|0i H s

DRAFT, October 17, 2007

40
For the general case see (Knill et al., 1999). The b-independence of the |Er (Ψ)i is easily
derived if every qubit interacts only with its own environment. Note, however, that the |Er i need
neither be orthogonal nor normalized nor unique!
4.3. ERROR CORRECTING CODES 107

Decoding is not more difficult:

 h
 α |0i + β |1i

α |ŵ0 i + β |ŵ1 i  h s H |0i

s H |0i .

If the Estate vector of the total system (three-qubit system plus environment) is
Ψ̂⊗ Ê then, according to our assumption, the interaction between both subsystems
can cause only transitions of the form
E 3 
X  E
Ψ̂ ⊗ Ê 7−→ ˆ (ν) Ψ̂ ⊗ Ê3(ν)
σ̂ 3
ν=0
E
(ν)
with suitable state vectors Ê3 of the environment, where

ˆ (ν) |b1 , b2 , b3 i def |b1 , b2 , b3 i if ν = 0
σ̂ 3 = bν ∀ b ∈ {0, 1}3 .
(−1) |b1 , b2 , b3 i else

The essential point is that the subspaces

n o
def ˆ (ν)
Hν = σ̂ 3 α̂ |ŵ0 i + β̂ |ŵ1 i : α̂, β̂ ∈ C

are pairwise orthogonal. Therefore, to restore the original encoded state vector
α |0i + β |1i, it suffices to perform an optimal test (measurement of first kind ) to
(ν)
which of the four subspaces Hν this state vector belongs and apply σ̂3 according
to the outcome.

Exercise 17 Show that the (n + 1)-qubit network

s s ··· s
h ···
h ···
.. ..
. .
h

 
transforms λ0 |0i + λ1 |1i ⊗ |0, . . . , 0i into λ0 |0, . . . , 0i + λ1 |1, . . . , 1i and discuss
its possible use for error correction.
108 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

4.3.2 Classical Codes

The general idea of classical error correction (Hamming, 1950; Pless, 1989) is the
following:
• Consider a channel transmitting n-bit words without changing more than m
bits of any word.
• Then the original words can be uniquely reconstructed from the received ones
if only special code words w = (w1 , . . . , wn ) ∈ {0, 1}n are sent which are
chosen such that the Hamming distance
n
X  
′ def ′ 2
d(w, w ) = |wν − wν′ | = kw − w k
ν=1

between any two code words w, w′ is > 2 m .

Obviously, 2n must be larger than the number of code words (the more the larger
m is) for error correction to work this way.41 Actually:
“Error-correcting coding is the art of adding redundancy efficiently so
that most messages, if distorted, can be correctly decoded.”
(Pless, 1989, p. 2)

Especially convenient are the [n, k] linear classical codes, for which a set C ⊂ {0, 1}n
of 2k code words — the code — is selected by means of an (n − k) × n-matrix Ĥ
as42 n o
def
C = ker(Ĥ) = b ∈ {0, 1}n : Ĥ b = 0 .
Of course, the n − k rows of the so-called parity check 43 matrix Ĥ have to be
independent in order to have
 
dim ker(Ĥ) = k .

Warning: The code C may contain transposed row vectors of the parity
check matrix Ĥ .

Without restriction of generality the parity check matrix can be assumed to be of

the form44  
Ĥ = Â 1ln−k ,
DRAFT, October 17, 2007
41
Of course this reduces the capacity of the communication channel.
42
Here, we identify matrices with the corresponding linear maps. Note that the components of
Ĥ are in {0, 1} and that all arithmetic is to be understood modulo 2. Hence, e.g., Ĥ ≡ −Ĥ .
n
X
43 ?
Every row (h1 , . . . , hn ) gives rise to a parity check hν bν = 0 on the substring of those bits
ν=1
of b in places where the row has 1’s .
44
This is easily seen using Gaußian elimination. Eventually the bits have to be relabeled.
4.3. ERROR CORRECTING CODES 109

where  is some (n − k) × k-matrix. In this form we easily see that

Ĥ Ĝ = 0
holds with the n × k-matrix  
1lk
Ĝ = ,
Â
hence45 n o
C = Ĝ a : a ∈ {0, 1}k .

Remark: A possible coding would be

{0, 1}k ∋ a
|{z} 7−→ Ĝ a
|{z} ∈ {0, 1}n .
word of message corresp. code word

While Ĝ may directly be used as a generator of the code, Ĥ is more convenient

for error detection:
Let E ⊂ {0, 1}n be the set of possible ‘errors’ and let Ĥ/\ E be an
injection. Then the distortion
w 7→ w′ = w + e
of a code word w ∈ C by an error e ∈ E can be identified by checking
the error syndrome
Ĥ w′ = Ĥ e
and corrected by adding (=subtracting) e .

Exercise 18 The rows of the r × (2r − 1) parity check matrix characterizing the
so-called binary Hamming code Ham[r, 2] are the nonzero elements of {0, 1}r ,
ordered46 according to the value of the corresponding binary numbers.47
a) Show for every r ≥ 2 that Ham[r, 2] is suitable for correcting errors on single
bits.
b) Discuss Ham[2, 2] in detail.

Let C be a a [n, k] linear classical code. Then

def
C ⊥ = {b ∈ {0, 1}n : b · b′ = 0 mod 2 ∀ b′ ∈ C}
is called its dual code.

Exercise 19 Let C be a [n, k] linear classical code with parity check matrix Ĥ and
generator Ĝ . Show the following:48
DRAFT, October 17, 2007
45
Note that, thanks to 1lk , the rows of Ĝ are all independent.
46
Actually, different orderings give rise to equivalent codes.
47
See (4.53) for r = 3 .
48
As usual, we denote the number of elements of a finite set C by |C| .
110 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

a) C ⊥ is a [n − k, n] linear classical code with parity check matrix Ĥ ⊥ = ĜT and

generator Ĝ⊥ = Ĥ T .

b)
 ⊥
C⊥ =C.

c) (
X b·b′ |C| if b ∈ C ⊥ ,
(−1) =
0 if b ∈ {0, 1}n \ C ⊥ .
b′ ∈C

4.3.3 Quantum Codes

In classical communication the received message may be inspected and corrected
according to the error syndrome. In quantum communication, however, we should
carefully avoid too detailed ‘measurement’ (associated with uncontrollable ‘collapse’)
of the state before reconstruction. Therefore, in order to be able to correct all errors
corresponding to error operations σ̂ ∈ E we have to look for quantum codes49 of the
following form:

• The n-qubit state space H containing the quantum code words is a direct sum
of specified subspaces Hd .

• Every σ̂ ∈ E is of a definite type d , i.e. σ̂ |ŵi ∈ Hd holds for all quantum

code words50 |ŵi .

• If σ̂, σ̂ ′ ∈ E are of the same type d then σ̂ |wi ∼ σ̂ ′ |wi holds for all quantum
code words |wi (but not necessarily for other state vectors).

Under these conditions — if only errors corresponding to operations σ̂ ∈ E are

superimposed51 — quantum error correction is possible as indicated in 4.3.1:

• The ‘received’ state is forced — via corresponding ‘measurement’ — to ‘col-

lapse’ into a state described by an element of one of the subspaces Hd .

• Since the ‘collapsed’ state is just the sent code word distorted by an error of
type d we only have to apply the inverse of some unitary error operation of
type d to reconstruct the correct code word.

DRAFT, October 17, 2007

49
By n-qubit quantum code we always mean the set of pairwise orthogonal n-qubit state
vectors used as quantum code words. Note, however, that many authors mean by quantum code
the complex linear span of quantum code words.
50
In 4.3.1 we already used linear superpositions |ŵ0 i , |ŵ1 i of the computational base states as
quantum code words, in order to indicate additional possibilities in quantum coding.
51
Of course, E should include the trivial ‘error operation’ σ̂ = 1̂ .
4.3. ERROR CORRECTING CODES 111

Exercise 20
a) Show that for Shor’s 9-qubit code words
E      
ˆ0 def
ŵ = 2−3/2 |0, 0, 0i + |1, 1, 1i ⊗ |0, 0, 0i + |1, 1, 1i ⊗ |0, 0, 0i + |1, 1, 1i ,
E      
ˆ1 def
ŵ = 2−3/2 |0, 0, 0i − |1, 1, 1i ⊗ |0, 0, 0i − |1, 1, 1i ⊗ |0, 0, 0i − |1, 1, 1i

every superposition of single-qubit errors, i.e. every distortion of the form

 E E X E E
α ŵ ˆ 1 ⊗ |Ei 7−→
ˆ 0 + β ŵ ˆ 0 + β σ̂r ŵ
α σ̂r ŵ ˆ 1 ⊗ |Er i ,
r∈R9
3 n
[ o
def
R9 = permutations of (r, 0, 0, 0, 0, 0, 0, 0, 0) ,
r=0
may be corrected as described above.
E E
ˆ 0 + β ŵ
b) Show that encoding Ψ = α |0i + β |1i into α ŵ ˆ 1 may be achieved as
follows:

Ψ s s H s s 




|0i h 





|0i h 





|0i h H s s 


 E E
|0i h ˆ 0 + β ŵ
α ŵ ˆ1


h 

|0i 



h s s 

|0i H 



h 

|0i 



h 
|0i

Recall that, according to (4.33), for every n ∈ IN the possible error n-qubit error
operations are elements of the Pauli group 52
n o
def
Sn = i ν X̂b1 Ẑb3 : ν ∈ {0, . . . , 3} , b1 , b3 ∈ {0, 1}n (4.36)
That the latter is a group w.r.t. operator multiplication follows immediately from
the algebra of Pauli matrices:
σ̂ν σ̂ν = 1̂ ∀ ν ∈ {0, 1, 2, 3} ,
σ̂j σ̂k = − σ̂k σ̂j ∀ j, k ∈ {1, 2, 3} , j 6= k ,
σ̂1 σ̂2 = i σ̂3 , (4.37)
σ̂2 σ̂3 = i σ̂1 ,
σ̂3 σ̂1 = i σ̂2 .
DRAFT, October 17, 2007
52
The X̂b and Ẑb were defined by (4.34) and (4.29).
112 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

The following statements also follow directly from these relations:

σ̂ν = σ̂ν∗ = σ̂ν−1 ∀ ν ∈ {0, 1, 2, 3} , (4.38)

n o
def
Sn ⊃ S0n = σ̂r1 ⊗ · · · ⊗ σ̂rn : r1 , . . . , rn ∈ {0, . . . , 3} is not a group , (4.39)
iν X̂b1 Ẑb3 ∈ S0n ⇐⇒ i ν = i b1 ·b3 ∀ ν ∈ {0, 1, 2, 3} , b1 , b3 ∈ {0, 1}n , (4.40)
σ̂σ̂ ′ ∈ {+σ̂ ′ σ̂ , −σ̂ ′ σ̂} ∀ σ̂, σ̂ ′ ∈ Sn . (4.41)

Theorem 4.3.1 Let W ⊂ H be an n-qubit quantum code and let E ⊂ Sn be a set

of error operations including the trivial operation 1̂ . Assume that the linear span
HW of W is stabilized by the subset SW of Sn , i.e. that
n o
HW = Ψ̂ ∈ H : ĝ Ψ̂ = Ψ̂ ∀ ĝ ∈ SW . (4.42)

Moreover, assume
σ̂ ∗ σ̂ ′ ∈
/ N(SW ) \ SW ∀ σ̂, σ̂ ′ ∈ E , (4.43)
where N(SW ) denotes the normalizer of SW :

N(SW ) = {σ̂ ∈ Sn : σ̂ ĝ σ̂ ∗ = ĝ ∀ ĝ ∈ SW } .

Then there is a unique mapping dσ̂ from SW into {+1, −1} such that:
   
ĝ σ̂ Ψ̂ = dσ̂ (ĝ) σ̂ Ψ̂ ∀ ĝ ∈ SW , σ̂ ∈ E , Ψ̂ ∈ HW , (4.44)
n o
def
σ̂ HW ⊂ Hdσ̂ = Φ̂ ∈ H : ĝ Φ̂ = dσ̂ (ĝ) Φ̂ ∀σ̂ ∈ E , (4.45)
′
dσ̂ 6= dσ̂′ =⇒ Hdσ̂ ⊥ Hdσ̂′ ∀σ̂, σ̂ ∈ E , (4.46)
dσ̂ = dσ̂′ =⇒ σ̂ Ψ̂ = σ̂ ′ Ψ̂ ∀σ̂, σ̂ ′ ∈ E , Ψ̂ ∈ HW . (4.47)

Outline of proof: (4.44) is a direct consequence of (4.41) and (4.42). (4.44) directly
implies (4.45). Since53
ĝ = ĝ ∗ ∀ ĝ ∈ SW (4.48)
— and since eigenvectors corresponding to different eigenvalues of a self-adjoint op-
erator are always orthogonal — (4.44) also implies (4.46). Finally, (4.44) and (4.41)
imply
ĝ σ̂ (∗) = dσ̂ (ĝ) σ̂ (∗) ĝ ∀ ĝ ∈ SW , σ̂ ∈ E
and hence
∗ ∗
(σ̂ ∗ σ̂ ′ ) ĝ (σ̂ ∗ σ̂ ′ ) = dσ̂ (ĝ) dσ̂′ (ĝ) ĝ (σ̂ ∗ σ̂ ′ ) (σ̂ ∗ σ̂ ′ )
= dσ̂ (ĝ) dσ̂′ (ĝ) ĝ ∀ ĝ ∈ SW , σ̂, σ̂ ′ ∈ E .
(4.38)

DRAFT, October 17, 2007

53
This is because for all σ̂ ∈ Sn we have σ̂ 2 = 1̂ ⇐⇒ σ̂ = σ̂ ∗ .
4.3. ERROR CORRECTING CODES 113

Therefore (4.47) follows according to

∗
dσ̂ = dσ̂′ =⇒ (σ̂ ∗ σ̂ ′ ) ĝ (σ̂ ∗ σ̂ ′ ) = ĝ ∀ ĝ ∈ SW
∗ ′
=⇒ σ̂ σ̂ ∈ N(SW )
=⇒ σ̂ ∗ σ̂ ′ ∈ SW
(4.43)
=⇒ σ̂ ∗ σ̂ ′ Ψ̂ = Ψ̂ ∀Ψ̂ ∈ HW
(4.42)
=⇒ σ̂ ′ Ψ̂ = σ̂ Ψ̂ ∀Ψ̂ ∈ HW .
(4.38)

Remarks:

1. In view of (4.43), SW should be chosen as large as possible.

2. The maximal SW fulfilling the requirements of Theorem 4.3.1 for
given HW is an abelian group called the stabilizer of HW .
3. Quantum codes W fulfilling the requirements of Theorem 4.3.1 are
called stabilizer codes.
4. If there are σ̂ , σ̂ ′ ∈ E and |ŵi ∈ W with

σ̂ |ŵi = σ̂ ′ |ŵi but σ̂ 6= σ̂ ′

then the code is called degenerate w.r.t. E .

5. Shor’s 9-qubit code, described in Exercise 20, is degenerate w.r.t.
the set of single qubit error operations. This can be easily seen by
considering phase flip errors on different qubits.
6. A stabilizer code is nondegenerate w.r.t E iff

σ̂ 6= σ̂ ′ =⇒ σ̂ ∗ σ̂ ′ ∈
/ SW ∀ σ̂, σ̂ ′ ∈ E .

7. In classical coding there is no analog for degeneracy.

Lemma 4.3.2 For j ∈ {1, 2} , let the Cj be a [n, kj ] linear classical codes with
C2 ⊂ C1 6= C2 , and define
 
 1 X 
def def
W = CSS (C1 , C2 ) = |ŵb i = q |b + b′ i : b ∈ C1  . (4.49)
|C2 | b′ ∈C2

where H denotes the n-qubit state space. Then (4.42) holds for
n o
SW = X̂a Ẑb : a ∈ C2 , b ∈ C1⊥ . (4.50)
114 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

Outline of proof: Let

X n o
Ψ̂ = λb |bi ∈ Ψ̂ ∈ H : ĝ Ψ̂ = Ψ̂ ∀ ĝ ∈ SW .
b∈{0,1}n

Then X X
1
Ψ̂ = Ẑ ′ λb |bi
(4.50) C1⊥ ′ ⊥ b
b ∈C1 b∈{0,1}n
X 1 X ′
= λb ⊥ (−1)b·b |bi
C1
bX
∈{0,1}n b′ ∈C1⊥
= λb |bi
Ex. 19
b∈C1
1 X X
= X̂b′ λb |bi
(4.50) |C2 | ′
X b ∈C 2 b ∈C 1

= λb ŵb .
b∈C1
Since, obviously,

X̂a Ẑb′ ŵb = ŵb ∀ a ∈ C2 , b′ ∈ C1⊥ , b ∈ C1

this proves the lemma.

Remarks:

1. The quantum codes CSS(C1 , C2 ) described by Lemma 4.3.2 are called

Calderbank-Shor-Steane codes .
2. The number of code words for these codes is
|C1 |
|CSS (C1 , C2 )| = = 2k1 −k2 .
|C2 |

Lemma 4.3.3 Let Cj , W and SW be given as in Lemma 4.3.2. If C1 as well as

C2⊥ is suitable for correcting errors on up to t bits then

σ̂ 6= σ̂ ′ =⇒ σ̂ ∗ σ̂ ′ ∈
/ N(SW ) ∀ σ̂, σ̂ ′ ∈ E (4.51)

holds for ( )
n n
X o
n ν
E = X̂e1 Ẑe3 : e1 , e3 ∈ b ∈ {0, 1} : |b | ≤ t . (4.52)
ν=1

Outline of proof: Consider σ̂, σ̂ ′ ∈ E with σ̂ 6= σ̂ ′ . Then there are

( n
)
n
X
ej , e′j ∈ b ∈ {0, 1} : |bν | ≤ t
ν=1
4.3. ERROR CORRECTING CODES 115

with54
{e1 + e′1 , e3 + e′3 } 6= {0}
and  ∗
σ̂ ∗ σ̂ ′ = X̂e1 Ẑe3 X̂e′1 Ẑe′3
= Ẑe3 X̂e1 +e′1 Ẑe′3

= (−1)e3 ·(e1 +e1 ) X̂

′
e1 +e′1 Ẑe3 +e′3
Hence
∗
(σ̂ ∗ σ̂ ′ ) X̂a Ẑb (σ̂ ∗ σ̂ ′ )
= X̂e1 +e′1 Ẑe3 +e′3 X̂a Ẑb ˆẐe3 +e′3 Xe1 +e′1

= i(e3 +e3 )·a i b·(e1 +e1 ) X̂a Ẑb ∀ a b ∈ {0, 1} .

′ ′ n

If e3 + e′3 =
6 0 then (e3 + e′3 ) · a 6= 0 mod 2 and, consequently,
∗
(σ̂ ∗ σ̂ ′ ) X̂a (σ̂ ∗ σ̂ ′ ) = −X̂a
for some a ∈ C2 since the generator of C2 is the parity check matrix of C2⊥ . On the
other hand, if e1 + e′1 6= 0 then b · (e1 + e′1 ) 6= 0 mod 2 and, consequently,
∗
(σ̂ ∗ σ̂ ′ ) Ẑb (σ̂ ∗ σ̂ ′ ) = −Ẑb
for some b ∈ C1⊥ . Thus σ̂ ∗ σ̂ ′ ∈
/ N (SW ).

Remark: Obviously, CSS(C1 , C2 ) is nondegenerate w.r.t. E specified by

Lemma 4.3.3.

55
In general, the number of operations
!
σ̂r affecting at most t ∈ {0, . . . , n} qubits
t
X n
of an n-qubit system is56 3j . Therefore, in order to correct all corresponding
j=0 j
errors for a nondegenerate n-qubit code according to the scheme described above,
that many subspaces Hd are needed. Moreover, the dimension of each of these
subspaces must not be smaller than the number of code words. Therefore:
Correction of all errors on at most t qubits of a nondegenerate n-
qubit code spanned by 2k orthogonal code words is not possible if the
quantum Hamming bound
t
!
X n j k
3 2 ≤ 2n
j=0 j
is violated.
Note that for k = t = 1 the quantum Hamming bound becomes 2 + 6n ≤ 2n , hence
n ≥ 5.

For further details on quantum codes see (Preskill, 01, Chapter 7), and (Schlingemann and Werner, 20
Schlingemann, 2001; Keyl and Werner, 2002).
DRAFT, October 17, 2007
54
Recall Footnote 42.
55
Recall (4.32).
56
The index j = 0 corresponds to the trivial error operation (unit operator).
116 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

4.3.4 Reliable Quantum Computation

Let us discuss the implementation of error
 correction in more detail.
 For simplicity,
⊥
we consider only the quantum code CSS Ham[3, 2], Ham[3, 2] , called the Steane
code. According to Exercise 18 a parity check matrix for Ham[3, 2] is
 
0 0 0 1 1 1 1
Ĥ3 =  0 1 1 0 0 1 1 
def 
. (4.53)
1 0 1 0 1 0 1

Exercise 21

a) Show that the code words corresponding to the parity check matrix Ĥ3 are
the same as those corresponding to the parity check matrix
 
0 1 1 1 1 0 0
Ĥ3′ =  
def
1 0 1 1 0 1 0 .
1 1 0 1 0 0 1

b) Show that57  
1 0 0 0 0 1 1
0 1 0 0 1 0 1
Ĥ4 =  
def
 
0 0 1 0 1 1 0
0 0 0 1 1 1 1
is a parity matrix for Ham[3, 2]⊥ .

c) Show that
n o
Ham[3, 2]⊥ = b ∈ {0, 1}7 : Ĥ3 b = 0 , (−1)b1 +...+b7 = 1 .

According to Exercise 21, the quantum code words of the Steane code are
1 
|ŵ0 i = √ |0000000i + |1101001i + |1011010i + |0111100i
8 
+ |0110011i + |1100110i + |1010101i + |0001111i ,
(4.54)
1 
|ŵ1 i = √ + |1111111i + |0010110i + |0100101i + |1000011i
8 
+ |1001100i + |0011001i + |0101010i + |1110000i ,

DRAFT, October 17, 2007

57
Recall Exercise 19 a).
4.3. ERROR CORRECTING CODES 117

Exercise 22 Show for (4.54) that the following network acts as indicated:


α |0i + β |1i h s 




|0i H s h s 





|0i H s s 



|0i H s α |ŵ0 i + β |ŵ1 i .


h h h 

|0i 



h h h 

|0i 



h h h 
|0i

Exercise 23 Show that the following 10-qubit network acts as indicated:

|b1 i s |b1 i
|b2 i s |b2 i
|b3 i s |b3 i
|b4 i s |b4 i
|b5 i s |b5 i
|b6 i s |b6 i
|b7 i s |b7 i
|0i h h h h 
 E
|0i h h h h Ĥ3 b ,

|0i h h h h

where, e.g.,

s s s

.. def
.. ..
. ≡ . .

h h
h h .

The network of Exercise 23 may be used to reduce single-qubit errors of type σ̂1
or/and σ̂2 to those of type σ̂0 or/and σ̂3 :
118 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

An ideal test for the computational basis of the last 3 (ancillary) qubits
causes the 10-qubit state to collapse into some state being the direct
product of

• a (possibly only partially coherent) superposition of distortions of

the original code word by single-qubit errors, appearing in one and
the same position if of type σ̂1 or σ̂2 , and
• a base state of the ancillary 3-qubit system which corresponds ei-
ther to (0, 0, 0) , if the collapsed 7-qubit state is a distortion of the
original code word by single-qubit errors of only type σ̂0 or/and σ̂3 ,
or else corresponds to the classical error syndrome of a bit-flip in
the position, where the code word is distorted by an error of type
σ̂1 or/and σ̂2 .

If the collapsed state of the ancillary system does not correspond to

(0, 0, 0) then σ̂1 should be applied to the qubit in the position where the
code word is distorted. In any case, then, the resulting 7-qubit state will
be a (possibly only partially coherent) superposition of distortions of the
original code word by single-qubit errors of type σ̂0 or/and σ̂3 .

X
3 
Exercise 24 Show that the following 10-qubit network flips the ej 23−j -th
j=1
qubit for input of the specified type with |e1 , e2 , e3 i =
6 |0, 0, 0i and, therefore, may
be used to avoid testing the error syndrome for single-qubit errors of type σ̂1 or/and
σ̂2 :

h
h
h
h
h
h
h

|e1 i c c c s s s s |e1 i
|e2 i c s s c c s s |e2 i
|e3 i s c s c s c s |e3 i

where
def def
c ≡ ¬
ˆ s ¬
ˆ , c ≡ ¬
ˆ s ¬
ˆ .
4.3. ERROR CORRECTING CODES 119

The eventually remaining single-qubit errors of only type σ̂0 or/and σ̂3 . may be
converted into errors of type σ̂0 or/and σ̂1 by applying ÛH⊗n . Correction these errors
as just described and applying ÛH⊗n once more restores the original message.

Up to now we tacitly assumed that all devices used for error correction work
perfectly error free. Of course this is unrealistic and, actually, special care has to be
taken to prevent these devices from making things worse.

For instance, if a phase error appears for the first ancillary qubit of the error
syndrome network presented in Exercise 23 then according to Exercise this error
may propagate into all of the last four data qubits. To prevent this one could use

s
s
s
s

|0i h h h h

|0i h s

|0i h s

|0i h s

instead of
s
s
s
s

|0i h h h h

and implement in a suitable way.58

While such precautions prohibit propagation of errors of the ancillary part of
the network into the data part they do not guarantee a correct error syndrome.
Therefore the ‘measurement’ of the error syndrome should be repeated and only
used for error correction if confirmed.
In order to protect calculations against quantum noise they should be performed
directly on the encoded data.
Of course, the encoded data should be error checked sufficiently often. Especially,
the actual computation should not be started before the initial encoded state has
been checked to be free of errors.
DRAFT, October 17, 2007
58
See (Möttönen and Vartiainen, 2005, Fig. 8), in this connection.
120 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

Altogether it seems possible to implement reliable quantum computation, if

sufficient care is taken. For further details see (Preskill, 1998b; Preskill, 1998a;
Leung, 2000).

4.4 Entanglement Assisted Channels59

“Entanglement is monogamous — the more entangled Bob is with Alice, the less
entangled he can be with anyone else.”
Charles Bennett60

4.4.1 Quantum Dense Coding61

Let S = S1 ⊕ S2 ben a bipartite62 2-qubit
o system with state space H ⊗ H and
def
computational basis |ν, µi = φν ⊗ φµ . Then the so-called Bell states
ν,µ∈{0,1}

def 1
Φ+ = √ (|0, 0i + |1, 1i) ,
2
def 1
Φ− = √ (|0, 0i − |1, 1i) ,
2
(4.55)
+ def 1
Ψ = √ (|0, 1i + |1, 0i) ,
2
def 1
Ψ− = √ (|0, 1i − |1, 0i)
2
form an orthonormal basis of H ⊗ H and may be locally transformed into each
other:63  
Φ∓ = σ̂3 ⊗ 1̂ Φ± ,
 
Ψ∓ = σ̂3 ⊗ 1̂ Ψ± , (4.56)
 
Ψ+ = σ̂1 ⊗ 1̂ Φ+ .
Obviously, 2 classical bits of information may be encoded via the Bell states, e.g.:

b Φ+ ,
(0, 0) = b Φ− ,
(0, 1) = b Ψ+ ,
(1, 0) = b Φ− .
(1, 1) =
DRAFT, October 17, 2007
59
See (Lücke, 2002, Section 1.2.2) for the network models of dense coding, teleportation, and
entanglement swapping. See also (Devetak and Winter, 2003; Devetak et al., 2004) for related
protocols.
60
http://qpip-server.tcs.tifr.res.in/ qpip/HTML/Courses/Bennett/TIFR2.pdf
61
See also (Mermin, 2002).
62
See Appendix A.4.3.
63
As usual, we denote by σ̂1 , . . . , σ̂3 the Pauli operators, i.e. w.r.t. (|0i , |1i) :
     
0 1 0 −i 1 0
σ̂1 =
b , σ̂2 = b , σ̂3 = b .
1 0 +i 0 0 −1
4.4. ENTANGLEMENT ASSISTED CHANNELS 121

Thus, if Alice and Bob (situated arbitrarily far apart) initially share a pair of qubits
forming S1 ⊕ S2 in a Bell state,64 say Ψ− , then Alice may communicate 2 bits of
information by sending Bob her single qubit (system S1 ) after acting on it in an
appropriate way:

2-bit information operation by Alice new Bell state

(0,0) σ̂1 σ̂3 Φ+

(0,1) σ̂3 σ̂1 σ̂3 Φ−
(1,0) σ̂3 Ψ+
(1,1) none Ψ−

After receiving Alice’s qubit Bob just has to perform a projective measurement
w.r.t. the Bell basis65 {Φ+ , Φ− , Ψ+ , Ψ− } in order to decode the 2-bit information.
Needless to say, without entanglement Alice would not have any chance to trans-
mit more than a single bit by sending just a single qubit. Therefore, the described
procedure to communicate 2 bits by sending just 1 qubit is called quantum dense
coding (of classical information). Of course, the crucial point is that Alice has to
be given one partner of an entangled pair of qubits first. Note, however, that Alice
and Bob may store their qubits for some time in suitable quantum memory66 before
starting to communicate. Then the information carried by the sent qubit is of no
use for any potential eavesdropper.

4.4.2 Quantum Teleportation

n system S = S1 ⊕ S2 ⊕oS3 with state space H ⊗ H ⊗ H and
Consider, e.g., a 3-qubit
def
computational basis |α, β, γi = φα ⊗ φβ ⊗ φγ in the initial state67
α,β,γ∈{0,1}

1
Ψ0 = ψ ⊗ √ (φ0 ⊗ φ1 − φ1 ⊗ φ0 ) . (4.57)
2
In order to determine the effect of a projective measurement on the subsystem S1 ⊕S2
w.r.t. its Bell basis we rewrite this state in the form
Ψ0 = Φ+ ⊗ χ0 + Φ− ⊗ χ1 + Ψ+ ⊗ χ2 + Ψ− ⊗ χ3 . (4.58)
Writing
ψ = α |0i + β |1i
DRAFT, October 17, 2007
64
For the creation of photon pairs in Bell states via parametric down conversion see, e.g.,
(Gatti et al., 2003) and references given there.
65
For the implementation of such measurements see, e.g., (Paris et al., 2000; Tomita, 2000;
Kim et al., 2001).
66
For the possibility of storing optical qubits see (Gingrich et al., 2003).
67
Obviously, then, the subsystem S2 ⊕ S3 is in the Bell state corresponding to Ψ− .
122 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

and comparing
√
2ψ ⊗ √1 (φ0 ⊗ φ1 − φ1 ⊗ φ0 )
2
= α |0, 0, 1i − α |0, 1, 0i − β |1, 1, 0i + β |1, 0, 1i

with
√
2 Φ+ ⊗ χ0 + Φ− ⊗ χ1 + Ψ+ ⊗ χ2 + Ψ− ⊗ χ3
1
= |0, 0i ⊗ χ0 + |1, 1i ⊗ χ0 + |0, 0i ⊗ χ1 − |1, 1i ⊗ χ1
2 
+ |0, 1i ⊗ χ2 + |1, 0i ⊗ χ2 + |0, 1i ⊗ χ3 − |1, 0i ⊗ χ3
χ0 + χ1 χ2 + χ3 χ2 − χ3 χ0 − χ1
= |0, 0i ⊗ + |0, 1i ⊗ + |1, 0i ⊗ + |1, 1i ⊗
2 2 2 2
we get
χ0 + χ1 = +α |1i ,
χ2 + χ3 = −α |0i ,
χ2 − χ3 = +β |1i ,
χ0 − χ1 = −β |0i
and hence
χ0 = α |1i − β |0i ,
= σ̂1 σ̂3 ψ
χ1 = α |1i + β |0i ,
= σ̂1 ψ
(4.59)
χ2 = β |1i − α |0i ,
= −σ̂3 ψ
χ3 = −α |0i − β |1i ,
= −ψ .
(4.57)–(4.59) show the possibility of quantum teleportation (of quantum infor-
mation):68

If Alice and Bob (situated arbitrarily far apart) initially share a pair of
qubits forming S2 ⊕ S3 in the Bell state √12 (φ0 ⊗ φ1 − φ1 ⊗ φ0 ) then
Alice may communicate to Bob the quantum information contained in
the unknown state ψ in the following way:
Alice performs a projective measurement on S1 ⊕ S2 w.r.t. the Bell
basis {Φ+ , Φ− , Ψ+ , Ψ− } and tells Bob her result via classical commu-
nication. Bob just has to perform one of the operations σ̂1 σ̂3 , σ̂1 , σ̂3
or none — according to the outcome of Alice’s measurement — on his
DRAFT, October 17, 2007
68
See (Sanctuary et al., 2003) for critical remarks on corresponding experiments.
4.4. ENTANGLEMENT ASSISTED CHANNELS 123

qubit (system S3 ) in order to have the latter in the state ±ψ :

result of Alice’s measurement Bob’s operation

Φ+ σ̂3 σ̂1 = (σ̂1 σ̂3 )−1

Φ− σ̂1 = (σ̂1 )−1
Ψ+ σ̂3 = (σ̂3 )−1
Ψ− none

Note that the classical information sent by Alice would be of no use to an eaves-
dropper and that sending classical information avoids the decoherence problems
connected with sending qubits.

4.4.3 Entanglement Swapping

Consider a 4-qubit system S = S0 ⊕ S1 ⊕ S2 ⊕ S3 in the initial state
Ψ̂0 = Ψ− ⊗ Ψ− . (4.60)
Then the calculations of Section 4.4.2 show that
  1 
Ψ̂0 = 1̂ ⊗ 1̂ ⊗ 1̂ ⊗ σ̂1 σ̂3 |0i ⊗ Φ+ ⊗ |1i − |1i ⊗ Φ+ ⊗ |0i
2
  1 
1̂ ⊗ 1̂ ⊗ 1̂ ⊗ σ̂1 |0i ⊗ Φ− ⊗ |1i − |1i ⊗ Φ− ⊗ |0i
2
  1 
(4.61)
− 1̂ ⊗ 1̂ ⊗ 1̂ ⊗ σ̂3 |0i ⊗ Ψ+ ⊗ |1i − |1i ⊗ Ψ+ ⊗ |0i
2
1 
− |0i ⊗ Ψ− ⊗ |1i − |1i ⊗ Ψ− ⊗ |0i .
2
Now assume that
Victor has access to S0 ,
Alice has access to S1 ⊕ S2 ,
Bob has access to S3 .
Then — even though Victor, Alice, and Bob may be arbitrarily far apart, the en-
tanglement of the subsystem S0 ⊕ S1 may be swapped to the subsystem S0 ⊕ S3 in
the following way:
Alice performs a projective measurement on S1 ⊕S2 w.r.t. the Bell basis
{Φ+ , Φ− , Ψ+ , Ψ− } and tells Bob her result via classical communication.
Bob just has to perform one of the operations σ̂1 σ̂3 , σ̂1 , σ̂3 or none —
according to the outcome of Alice’s measurement — on his qubit (system
S3 ) in order to have the partial state of the subsystem S0 ⊕ S3 in the
state Ψ− .
124 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION

Thus Alice may act as an entanglement provider:

Alice prepares pairs of entangled qubits and distributes one partner of

each pair to various customers including Victor and Bob. If Victor and
Bob need to share an entangled pair they instruct Alice to perform a pro-
jective measurement on S1 ⊕ S2 w.r.t. the Bell basis {Φ+ , Φ− , Ψ+ , Ψ− }
and communicate the result to either Victor and Bob who then knows
the type of entanglement of the pair shared with Bob.

4.4.4 Quantum Cryptography69

As explained in 2.2.1 the security of the RSA encryption scheme relies on the extreme
difficulty to factorize large numbers n by classical means (and the possibility of
authentication of submitted messages). Otherwise d could be determined from n
and e . However, in view of Shor’s quantum factoring algorithm (Shor, 1994) and
the possible implementation of quantum computers such classical cryptosystems as
RSA may become insecure. Fortunately, quantum mechanics itself offers means for
secure communication exploiting the Vernam cipher70 (also called one time pad ):

Exploiting quantum mechanisms, Victor and Bob agree on a purely ran-

dom secret key c = (c1 , . . . , cn ) ∈ {0, 1}n . Then, instead of sending Bob
the pure the plaintext message c = (c1 , . . . , cn ) ∈ {0, 1}n Victor sends
him the encoded message71

e = (b1 ⊕ c1 , . . . , bn ⊕ cn )

(through some public channel) which Bob may decrypt as

b = (e1 ⊕ c1 , . . . , en ⊕ cn )

but appears purely random to all eventual eavesdroppers.

As shown by Claude Shannon (Shannon, 1949a), this cryptosystem is
absolutely secure if e is kept secret and used only once.

If Victor and Bob share enough (nearly) maximally entangled pairs of qubits72 they
may establish a secret key in the following way:
DRAFT, October 17, 2007
69
See (Bowmeester et al., 2000, Chapter 2) for a nice introduction and (Elliott et al., 2005) for
actual implementation. A commercial quantum cryptosystem is offered at: www.idquantique.com
70
Developed by Gilbert Vernam at AT&T in 1917 (first published in 1926).
71
Note that
def
b ⊕ b′ = b + b′ mod 2 ∀ b, b′ ∈ {0, 1} .

72
If the entanglement is not good enough even if it is fairly bad they may perform entanglement
distillation resulting in a smaller number of nearly perfectly entangled pairs; see Section 6.2.3.
4.4. ENTANGLEMENT ASSISTED CHANNELS 125

Via public communication they agree on an orthonormal basis (e1 , e2 , e3 )

of IR3 and on a series of joint measurements of the following type on
definite pairs:

For every tested pair the momenta of the partners are directed
parallel or antiparallel e2 and a check for linear polarization is
performed, but Victor and Bob independently and randomly
between two possibilities: Either they test whether the linear
polarization of their photon is parallel or orthogonal to e1 or
check whether the linear polarization is parallel or orthogonal
1
to e′1 = √ (e1 + e3 ).
2
As long as their choices are different their results for the corresponding
pairs are completely uncorrelated. Whenever they choose the same type
of measurement their results are (nearly) perfectly correlated. Thus
they may agree via public communication on a random secret key in the
following way:

• Alice and Bob identify those pairs for which, by chance,

they had chosen the same type of measurement and dis-
card those pairs of which at least one barter got lost and
thus did not provide a definite result.
• The remaining pairs are split into two groups.
• Bob and Alice (publicly) compare their results for the first
group in order to check perfect correlation.
• If the correlations turn out to be (nearly) perfect for the
first group then Alice and Bob agree to use their results on
the second group for a common key. Thank to the corre-
lations they need not communicate the results concerning
the second group. Therefore a possible eavesdropper has
no access to the chosen key.
• Weak deviations from perfect correlation may be error
corrected after communicating results of various parity
checks73 — of almost no use for any eavesdropper.

This cryptosystem can only be attacked by manipulating the entangled pairs before
Victor’s and Bob’s measurements. But such attack will be detected by Victor and
Bob, who can eventually discard the current key and create another one.

DRAFT, October 17, 2007

73
Alternatively, if the correlations are only marginally spoiled, one could apply standard classical
error correction74 to the appropriately encoded (and slightly disturbed) plaintext after use of the
one time pad.
126 CHAPTER 4. GENERAL ASPECTS OF QUANTUM INFORMATION
Chapter 5

Quantifying Quantum Information

In fact, the mathematical machinery we need to develop quantum infor-

mation theory is very similar to Shannon’s mathematics (typical se-
quences, random coding, . . . ); so similar as to sometimes obscure that
the conceptual context is really quite different.

(Preskill, 01, Section 5.2)

5.1 Shannon Theory for Pedestrians

For simplicity, let us consider an information source (Z, p) of the following type:
1. Letters x are randomly drawn from a finite alphabet Z = {z1 , . . . , zN } .
2. The probability for drawing the n-letter word w = (zj1 , . . . , zjn ) is1
n
Y
p(zj1 , . . . , zjn ) = p(zjν ) ∀ zj1 , . . . , zjn ∈ Z ,
ν=1

where
p(z) = probability for drawing z ∀z ∈ Z .

Consistency, of course requires

p(z1 ) + . . . + p(zN ) = 1 . (5.1)

Then, for the corresponding Shannon entropy

N
X  
def
H(Z, p) = − p(zj ) log2 p(zj ) (5.2)
j=1

DRAFT, October 17, 2007

1
Thus we assume that the probabilities for the successively drawn letters are independent.

127
128 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

one may prove2

Shannon’s noiseless coding theorem:

For arbitrarly given δ > 0 , we may associate with every n ∈ IN a set
Wδ,n of typical n-letter words3 such that:

1.
|Wδ,n | ≤ 2n(H(Z,p)+δ) ∀ n ∈ IN .

2. The probability for a drawing a n-letter word w ∈

/ Wδ,n tends to 0
for n → ∞ .

In other words:

Asymptotically, the relevant words can be indexed by ⌈n (H(Z, p) + δ)⌉

bits,4 for every fixed δ > 0 .
 
In this sense, the information gained by drawing the letter z is − log2 p(z) bits.
The average information gained by drawing a letter, correspondingly, is H(Z, p) bits.

Remarks:

1. As to be expected, we have:

H(Z, p) = 0 ⇐⇒ ∃ z0 ∈ Z : p(z0 ) = 1 . (5.3)

2. Straightforward calculation shows that5

Z = Z 1 ∪ Z2 , Z1 6= ∅ = Z1 ∩ Z2 6= Z2
 
=⇒ H(Z, p) = H {Z1 , Z2 } , p + p(Z1 ) H(Z1 , p1 ) + p(Z2 ) H(Z2 , p2 ) ,
(5.4)
DRAFT, October 17, 2007
2
See (Shannon, 1949, Appendix 3).
3
The typical words have to include essentially all those containing the letter x approximately
⌈n p(z)⌉-times for every z ∈ Z . For large n the number of such words is of the order
 
n! n H(Z,p) −N N
Q   ≈ 2 since N ! ≈ e N for large N .
Stirling
z n p(z) !

 
4
Obviously, such coding can be used for data compression, if H(Z, p) < log2 |Z| .
5
(5.4) together with (5.5) and continuity in the p(z) fixes H uniquely (Shannon, 1949, Theorem
2).
5.1. SHANNON THEORY FOR PEDESTRIANS 129

where
def X 
p(Zj ) = p(z) 

z∈Zj

∀ j ∈ {1, 2} .
def 
pj (z) = p(z)/p(Zj ) ∀ z ∈ Zj

3. H(Z, p) as a functional of p is maximal6 for constant p , i.e. for

p(z1 ) = . . . = p(zN ) = 1/N .
4.
1  
p(z) = ∀z ∈ Z =⇒ H(Z, p) = log2 |Z| . (5.5)
|Z|
5. Since  
log2 |Z| = N if |Z| = 2N ,
compression is not possible for constant p(z) (essentially all words
are typical ).

Now assume
Z =X ×Y
and define
def X
p1 (x) = p(x, y) ∀ x ∈ X , (5.6)
y∈Y
def X
p2 (y) = p(x, y) ∀ y ∈ Y , (5.7)
x∈X

p(x, y)/p2 (y)
def if p2 (y) > 0
p1 (x|y) = ∀ (x, y) ∈ Z , (5.8)
1/ |X| else

def p(x, y)/p1 (x) if p1 (x) > 0
p2 (y|x) = ∀ (x, y) ∈ Z . (5.9)
1/ |Y | else

Remark: A possible application is the following:

X = {elements drawn and sent via some channel } ,
Y = {elements received through that channel} ,
p(x, y) = joint probability for sending x and receiving y ,
p1 (x) = probability for sending x ,
p2 (y) = probability for receiving y ,
p2 (y|x) = probability for receiving y when x is sent ,
p1 (x|y) = probability for x having been sent when y is received .
DRAFT, October 17, 2007
6
The simplest way to check this this is by means of a Lagrange multiplier λ : Determine
P λ ∈ IR


and p(z1 ), . . . , p(zN ) ≥ 0 (not a priori postulating (5.1)) for which H(Z, p) + λ 1 − z∈Z p(z)
is maximal.
130 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

Then, thanks to7

a 6= b =⇒ a (ln a − ln b) > a − b ∀ a, b > 0 (5.10)
we have8
H(X × Y, p) ≤ H(X, p1 ) + H(Y, p2 ) (subadditivity) (5.11)
and
H(X × Y, p) = H(X, p1 ) + H(Y, p2 )
(5.12)
=⇒ p(x, y) = p1 (x) p2 (y) ∀ (x, y) ∈ X × Y .

Outline of proof: Replacing b by b c in (5.10) and using

ln(x)
log2 (x) = ∀x > 0,
ln(2)
we get
  a − bc
a 6= b c =⇒ a log2 (a) − log2 (b) − log2 (c) > ∀ a, b, c > 0
ln(2)
and hence
H(X, p1 ) + H(Y, p2 ) − H(X × Y, p)
  !
X X   X X  
= −  p(x, y) log2 p1 (x) − p(x, y) log2 p2 (y)
x∈X y∈Y y∈Y x∈X
X  
+ p(x, y) log2 p(x, y)
 (x,y)∈X×Y
X      
= p(x, y) log2 p(x, y) − log2 p1 (x) − log2 p2 (y)
(x,y)∈X×Y
X  .
≥ p(x, y) − p1 (x) p2 (y) ln(2)
(x,y)∈X×Y
≥ 0
with equality iff
p(x, y) − p1 (x) p2 (y) ∀ (x, y) ∈ X × Y .

Moreover, according to Shannon’s noiseless coding theorem, the conditional en-

tropy9  
def X
H1 (X|Y ) = p2 (y) H X, p1 (.|y) (5.13)
y∈Y

is the average asymptotic amount of bits of information needed in addition per Y -

part of the drawn z ∈ Z , if only these parts are known, in order to determine also
the X-parts. Accordingly, we have
H(X × Y, p) = H(Y, p2 ) + H(X|Y ) . (5.14)

DRAFT, October 17, 2007

7
Setting x = b/a , (5.10) follows from: 0 < x 6= 1 =⇒ ln x < x − 1 .
8
This corresponds to the fact that correlations between X and Y contain additional information.
9
Its quantum analogue can be negative (Horodecki et al., 2005).
5.2. ADAPTION TO QUANTUM COMMUNICATION 131

Outline of proof:

H(X|Y )
X X  
= − p2 (y) p1 (x|y) log2 p1 (x|y)
(5.13) | {z }
y∈Y x∈X
=p(x,y)/p2 (y)
X   X X  
= − p2 (y) p1 (x|y) log2 p(x, y) + p2 (y) p1 (x|y) log2 p2 (y) .
| {z }
(x,y)∈Z y∈Y x∈X
=p(x,y) | {z }
=1

Similarly we have
H(X × Y, p) = H(X, p1 ) + H(Y |X)
for X  
def
H2 (Y |X) = p1 (x) H Y, p2 (.|x) .
x∈X

Thanks to subadditivity (5.11), the mutual information

def
I(X : Y ) = H(X, p1 ) + H(Y, p2 ) − H(X × Y, p) (5.15)

is non-negative, as required by its interpretation as amount of information contained

in the correlations between X and Y . Its relation to the conditional entropy is given
by
I(X : Y ) = H(X, p1 ) − H1 (X|Y )
= H(Y, p2 ) − H2 (Y |X) (5.16)
≥ 0.

5.2 Adaption to Quantum Communication

5.2.1 Von Neumann Entropy10
The von Neumann entropy 11
def
S1 (ρ̂) = −trace (ρ̂ log2 ρ̂) ∀ ρ̂ ∈ S(H) (5.17)

(von Neumann, 1927) can be considered as a generalization of the Shannon entropy

in the following sense:
DRAFT, October 17, 2007
10
See also (Wehrl, 1978; Ohya and Petz, 1993; Petz, 2001; Ruskai, 2002).
11
One can easily prove that, at least for the (normalized) statistical operator ρ̂ of the micro-
canonical or canonical ensemble, k ln(2) S1 (ρ̂) is the usual thermodynamic entropy; see, e.g.,
(Gardiner and Zoller, 2000, Section 2.4.1). Note, however, that the von Neumann entropy —
contrary to the thermodynamic entropy — is non-extensive (not additive) for homogeneous non-
equilibrium systems. For generalizations of the von Neumann entropy as a measure of ‘mixedness’
see (Berry and Sanders, 2003) and references given there.
132 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

If n 
X 1 for α = β ,
ρ̂ = λν |φν ihφν | , hφα | φβ i =
ν=1
|{z} 0 else ,
≥0

then
S1 (ρ̂) = H(X, p1 ) (5.18)
holds for n o
def
X = |φ1 ihφ1 | , . . . , |φn ihφn | (5.19)
 
def
p1 |φν ihφν | = λν ∀ ν ∈ {1, . . . , n} . (5.20)

Warning: If {φ1 , . . . , φn } is not an orthonormal system then (5.19) and (5.20) do

not imply (5.18) in general!

Theorem 5.2.1 (Klein’s inequality12 ) For all Â, B̂ ∈ L(H) we have

0 ≤ Â 6= B̂ ≥ 0 , ker(B̂) ⊂ ker(Â)
  
=⇒ trace  ln  − ln B̂ > trace ( − B̂) .

Outline of proof: Thanks to the spectral theorem there are orthonormal systems
{φ1 , . . . , φn } and {φ′1 , . . . , φ′n } of H with
n
X n
X
 = aν |φν ihφν | , B̂ = bν |φ′ν ihφ′ν |
|{z} |{z}
ν=1 ≥0 ν=1 ≥0

for suitable a1 , . . . , bn . Then

n
X
trace (Â ln Â) = aν ln aν
ν=1

and
n
* n n
+
X X X
trace (Â ln B̂) = φα aβ |φβ ihφβ | ln bγ |φγ ihφγ | φα
α=1 β=1 γ=1
Xn
2
= aα ln(bγ ) φα | φ′γ ,
α,γ=1

hence
n n
!
X X 2
trace (Â ln Â) − trace (Â ln B̂) = aα ln aν − ln(bγ ) φα | φ′γ .
α=1 γ=1

DRAFT, October 17, 2007

12
See (Klein, 1931).
5.2. ADAPTION TO QUANTUM COMMUNICATION 133

Since, for x > 0 , ln(x) is a strictly concave function,13 the latter implies

trace (Â ln Â) − trace (Â ln B̂) (5.21)

Xn X n !
2
≥ aα ln aα − ln bγ φα | φ′γ (5.22)
α=1 γ=1
n n
!
X X 2
≥ aα − bγ φα | φ′γ (5.23)
(5.10) α=1 γ=1
Xn n
X
= aα − bγ
α=1 γ=1

= trace (Â − B̂) .

In (5.22) equality holds only if, for every α ∈ {1, . . . , n} , at most one of products
aα bγ φα | φ′γ is different from zero. Then, with suitable relabelling of the φ′γ , we
Xn
2
have bγ φα | φ′γ = bα and equality in (5.23) holds only if aα = bα for all
γ=1
α ∈ {1, . . . , n} , i.e. if  = B̂ .

Remark: As an immediate consequence of Theorem 5.2.1 we have strict

positivity of the quantum relative entropy
 
def
S(ρ̂kρ̂′ ) = trace ρ̂ (ln ρ̂ − ln ρ̂′ )

for all states ρ̂, ρ̂′ ∈ S(H) with ker(ρ̂′ ) ⊂ ker(ρ̂) and ρ̂ 6= ρ̂′ .

Corollary 5.2.2 Let H1 , H2 be Hilbert spaces and ρ̂ ∈ S(H1 ⊗ H2 ) . Then

   
S1 (ρ̂) ≤ S1 trace 2 (ρ̂) + S1 trace 1 (ρ̂) (subadditivity)

and
   
S1 (ρ̂) = S1 trace 2 (ρ̂) + S1 trace 1 (ρ̂) ⇐⇒ ρ̂ = trace 2 (ρ̂) ⊗ trace 1 (ρ̂) .

Outline of proof: Application of Klein’s inequality to

 = ρ̂ , B̂ = trace 2 (ρ̂) ⊗ trace 1 (ρ̂)

gives
0 = trace (Â − B̂)
≤ −S1 (ρ̂) − trace (Â log2 B̂)

DRAFT, October 17, 2007

13
Strictly concave functions f of x > 0 are those fulfilling
 
λ f (x1 ) + (1 − λ) f (x2 ) < f λ x1 + (1 − λ) x2 ∀ λ ∈ (0, 1) , x1 , x2 > 0 .
134 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

with equality only for ρ̂ = trace 2 (ρ̂) ⊗ trace 1 (ρ̂) . Since

trace (Â logB̂) 




= trace  log2 trace 2 (ρ̂) ⊗ 1̂ + 1̂ ⊗ log2 trace 1 (ρ̂)

 

= trace trace 2 (ρ̂) log2 trace 2 (ρ̂) + trace trace 1 (ρ̂) log2 trace 1 (ρ̂)
   
= −S1 trace 2 (ρ̂) − S1 trace 1 (ρ̂) ,

this proves the corollary.

The von Neumann entropy of a state ρ increases if the latter is changed by a

complete projective measurement operation (ignoring the results):
n o
Corollary 5.2.3 Let H be a Hilbert space, ρ̂0 ∈ S(H) , P̂1 , . . . , P̂l a set of
pairwise orthogonal projection operators on H with

P̂1 + . . . + P̂l = 1̂

and define
l
X
def
ρ̂′0 = P̂k ρ̂0 P̂k .
k=1
Then
ρ̂0 6= ρ̂′0 =⇒ S1 (ρ̂0 ) < S1 (ρ̂′0 ) .

Proof: Since
X
l 
trace (ρ̂ log2 ρ̂′ ) = trace P̂k P̂k ρ̂ log2 ρ̂′
k=1
| {z }
=1̂
X
l 
= trace P̂k ρ̂ log2 (ρ̂′ ) P̂k
k=1
X
l 
′
= trace P̂k ρ̂ P̂k log2 (ρ̂ )
[ρ̂′ ,P̂k ]− =0
k=1
= −S1 (ρ̂′ ) ,

The statement follows from Klein’s inequality (Theorem 5.2.1) applied to  =

ρ̂ , B̂ = ρ̂′ .

Warning: In general, trace preserving quantum operations may de-

crease the von Neumann entropy.14

DRAFT, October 17, 2007

14
E.g., if
K̂1 = |0ih0| K̂2 = |0ih1|
5.2. ADAPTION TO QUANTUM COMMUNICATION 135

5.2.2 Accessible Information

Assume that the ‘alphabet’ X = {ρ̂1 , . . . , ρ̂N1 } is a set of (pairwise different) states
on H . Then, by “drawing the ‘letter’ ρ̂ from X” we mean the random choice of an
individual from an ensemble in the state ρ̂ . Again, for simplicity, we assume that
for every letter ρ̂ its probability p1 (ρ̂) for being drawn is given, positive, and does
not depend on which letters have been drawn before.

The best one can do, in order to acquire information about an drawn letter,
o is
perform a POV measurement15 corresponding to some set Y = Ê1 , . . . , ÊN2 of
events Ê represented by positive bounded operators on H with16

Ê1 + . . . + ÊN2 = 1̂ .

According to quantum mechanical rules, the probability for Ê is trace (ρ̂ Ê) , if ρ̂
was drawn. Hence, the probability for ρ̂ being drawn and Ê being detected is17
def
p(ρ̂, Ê) = p1 (ρ̂) trace (ρ̂ Ê) . (5.24)

If the letter drawn is unknown, the probability for Ê is

N1
X
p2 (Ê) = p(ρ̂j , Ê)
(5.7) j=1

= trace (ρ̂0 Ê) ,

where
N1
X
def
ρ̂0 = p1 (ρ̂j )ρ̂j
j=1

is the state of the source providing the letters. Of course,18 ρ̂0 does not uniquely
determine (X, p1 ) . Nevertheless, the von Neumann entropy fulfills the Holevo
bound
N1
X
I(X : Y ) ≤ S1 (ρ̂0 ) − p1 (ρ̂j ) S1 (ρ̂j ) (5.25)
j=1

DRAFT, October 17, 2007

are the Kraus operator of the quantum operation C acting on a qubit system we have C(1̂/2) =
|0ih0| and hence  
S1 (1̂/2) > S1 C(1̂/2) = 0 .

15
If the elements of X are linearly independent, then the best result can be achieved by projective
measurement, i.e. with Ê1 , . . . , ÊN2 being projection operators (Eldar, 2003). For the importance
of considering also linearly dependent Êν ’s see, e.g., (Kaszlikowski et al., 2003).
16
Recall Corollary A.4.3, in this connection.
17
Obviously, (5.24) is consistent with (5.6).
18
Recall Corollary A.4.3.
136 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

(Nielsen and Chuang, 2001, Theorem 12.1) and the condition

∃ j1 , j2 ∈ {1, . . . , N1 } : trace (ρ̂j1 ρ̂j2 ) 6= 0 , j1 6= j2

N1
X (5.26)
=⇒ S1 (ρ̂0 ) < H(X, p1 ) + p1 (ρ̂j ) S1 (ρ̂j )
j=1

(Nielsen and Chuang, 2001, Theorem 11.10). A direct consequence of (5.25) and
(5.26) is the upper bound

∃ j1 , j2 ∈ {1, . . . , N1 } : trace (ρ̂j1 ρ̂j2 ) 6= 0 , j1 6= j2

(5.27)
=⇒ A(X, p) < H(X, p1 )

on the accessible information

def
A(X, p) = max I(X : Y ) . (5.28)
Y =POV
ˆ

Hence:

It is impossible to get full information on the letters actually drawn

unless19
ρ̂ 6= ρ̂′ =⇒ ρ̂ ρ̂′ = 0̂ ∀ ρ̂, ρ̂′ ∈ X .

Remark: Note that, for arbitrary ρ̂, ρ̂′ ∈ S(H) we have20

ρ̂ ρ̂′ = 0̂ ⇐⇒ trace (ρ̂ ρ̂′ ) = 0

⇐⇒ ρ̂ H ⊥ ρ̂′ H .

On the other hand, (5.27) (together with continuity of the entropies) implies the
bound
A(X, p) ≤ S1 (ρ̂0 ) . (5.29)

DRAFT, October 17, 2007

19
Otherwise, distinguishability of arbitrary states could be used for superluminal communication.
20
Here, positivity of the operators ρ̂, ρ̂′ is essential! For  = † ∈ L(H) the range  H of  is
also called the support of  , since

Â Ψ 6= 0 ⇐⇒ 0 6= Ψ ∈ ÂH

for such  .
5.2. ADAPTION TO QUANTUM COMMUNICATION 137

5.2.3 Distance Measures for Quantum States21

A natural distance measure for quantum states ρ̂ , ρ̂′ of a finite-dimensional22 quan-
tum system is the trace distance23
def 1
D(ρ̂, ρ̂′ ) = kρ̂ − ρ̂′ k1 , (5.30)
2
where k.k1 denotes the trace norm
q 
def +
kAk1 = trace †  (5.31)

for trace class operators  on a Hilbert space H . Especially for qubits we have
1  1 
ρ̂ = 1̂ + ρ · τ̂ , ρ̂′ = 1̂ + ρ′ · τ̂
2 2
24
and hence q
D(ρ̂, ρ̂′ ) = 14 trace + (ρ · τ̂ − ρ′ · τ̂ )2
 q 
1 +
= trace |ρ − ρ′ | 1̂
4
1
= |ρ − ρ′ | ,
2
i.e.:
For qubit states ρ̂ , ρ̂′ the trace distance is half the Euclidean distance
of the corresponding Bloch vectors ρ , ρ′ .
For general mixed states we have the following:

Lemma 5.2.4 Let H be a finite-dimensional Hilbert space and ρ̂, ρ̂′ ∈ S(H) .
Then25  
D(ρ̂, ρ̂′ ) = max trace (P̂ ρ̂) − trace (P̂ ρ̂′ ) , (5.32)
P̂ ∈P

where P denotes the set of all projection operators on H .

DRAFT, October 17, 2007
21
See also (Gilchrist et al., 2004).
22
For infinite-dimensional systems, however, the trace distance is not physically adequate
(Streater, 2003).
23
Note that 
X∞

ρ̂ = pν |Φν ihΦν | , 

 ∞
′ 1 X
ν=1
X∞ =⇒ D(ρ̂, ρ̂ ) = |pν − p′ν |
′ ′ 
 2 ν=1
ρ̂ = pν |Φν ihΦν | 
ν=1
if {Φν }ν∈IN is a MONS of H .
24
Recall that
2
|e| = 1 =⇒ (e · τ̂ ) = 1̂ ∀ e ∈ IR3 .

25
This formula holds also with P replaced by the set of all events; i.e. of all positive operators
with trace ≤ 1 .
138 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

Outline of proof: The spectral theorem tells us that there are an orthonormal basis
{φν }ν∈IN of H and real numbers λ1 , . . . , λn with
n
X
ρ̂ − ρ̂′ = λ ν φν (5.33)
ν=1

and
n
X
λν = 0 . (5.34)
ν=1

Then q n
X
+ 2
(ρ̂ − ρ̂′ ) = |λν | φν
ν=1

and, therefore,
n
1X
D(ρ̂, ρ̂′ ) = |λν |
2 ν=1
X
= λν
(5.34)
ν∈{1,...,n}

λν >0
X 
= trace λ ν φν
ν∈{1,...,n}
 λν >0
X   X 
≥ trace P̂ λ ν φν + trace P̂ λ ν φν ∀ P̂ ∈ P
ν∈{1,...,n} ν∈{1,...,n}
λν >0 λν ≤0

with equality for X

P̂ = |φν ihφν | .
ν∈{1,...,n}
λν >0

By (5.33) and linearity of the trace, this implies (5.32).

Now it is obvious that the trace distance is a metric on the set of states:26
D(ρ̂1 , ρ̂2 ) ≥ 0 ,
D(ρ̂1 , ρ̂2 ) = 0 ⇐⇒ ρ̂1 = ρ̂2 ,
D(ρ̂1 , ρ̂2 ) = D(ρ̂2 , ρ̂1 ) ,
D(ρ̂1 , ρ̂3 ) ≤ D(ρ̂1 , ρ̂2 ) + D(ρ̂2 , ρ̂3 ) .

DRAFT, October 17, 2007

26
The inequality follows from (5.32) and
   
trace (ρ̂1 ) − trace (ρ̂3 ) = trace (ρ̂1 ) − trace (ρ̂2 ) + trace (ρ̂2 ) − trace (ρ̂3 ) .
5.2. ADAPTION TO QUANTUM COMMUNICATION 139

Corollary 5.2.5 Let H1 and H2 be finite-dimensional Hilbert spaces and let

C ∈ Q(H1 , H1 ) be trace-preserving. Then Q(H1 , H1 ) is contractive w.r.t. the trace
distance, i.e.:  
D C(ρ̂), C(ρ̂′ ) ≤ D(ρ̂, ρ̂′ ) ∀ ρ̂, ρ̂′ ∈ S(H1 ) .

Outline of proof: With φ1 , . . . , φn and λ1 , . . . , λn chosen as in the proof for (5.32),

we have27
 X 
D(ρ̂, ρ̂′ ) = trace λ ν φν
ν∈{1,...,n}

!
λν >0
 X
= ¯
trace C λ ν φν
ν∈{1,...,n}

! !
λν >0
 X  X
¯
≥ trace P̂ C λ ν φν ¯
+ trace P̂ C λ ν φν
ν∈{1,...,n} ν∈{1,...,n}
λν >0 λν ≤0
  
′
= trace P̂ C(ρ̂) − ĉ(ρ̂ ) ∀ P̂ ∈ P .

This, together with (5.32), proves the theorem.

Corollary 5.2.6 Let H be a Hilbert space and N ∈ IN . Moreover, consider

ρ̂1 , . . . , ρ̂′N ∈ S(H) and p1 , . . . , p′N ≥ 0 with
N
X N
X
pν = 1 = p′ν . (5.35)
ν=1 ν=1

Then
N N
! N N
X X 1X X
D pν ρ̂ν , p′ν ρ̂′ν ≤ |pν − p′ν | + pν D(ρ̂ν , ρ̂′ν ) . (5.36)
ν=1 ν=1 2 ν=1 ν=1

Outline of proof: By Lemma 5.2.4 there is a P̂ ∈ P with

N N
! N N
!
X X X X
′ ′ ′ ′
D pν ρ̂ν , pν ρ̂ν = trace P̂ pν ρ̂ν − P̂ pν ρ̂ν
ν=1 ν=1 ν=1 ν=1
XN    
= pν trace P̂ ρ̂ν − P̂ ρ̂′ν + (pν − p′ν ) trace (P̂ ρ̂′ν )
ν=1 | {z } | {z }
∈[0,1]
≤ D(ρ̂ν ,ρ̂′ν )
L. 5.2.4
N
X X
≤ pν D(ρ̂ν , ρ̂′ν ) + (pν − p′ν ) .
ν=1 ν∈{1,...,N }
pν −p′ν >0

DRAFT, October 17, 2007

27
Recall Footnote 18 of Chapter 4.
140 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

Together with
X N
1X
(pν − p′ν ) = |pν − p′ν |
(5.35) 2
ν∈{1,...,N } ν=1
pν −p′ν >0

this implies (5.36).

A direct consequence of (5.36) is

N N
! N
X X X
D pν ρ̂ν , pν ρ̂′ν ≤ pν D(ρ̂ν , ρ̂′ν ) . (5.37)
ν=1 ν=1 ν=1

Moreover, setting ρ̂′ν ≡ ρ̂′ in (5.37) and recalling (5.35), we get

N
! N
X X
′
D pν ρ̂ν , ρ̂ ≤ pν D(ρ̂ν , ρ̂′ ) . (5.38)
ν=1 ν=1

Another important measure for the distance of states is the Bures fidelity:28
 s 2
q q †  q q 
def +
F (ρ̂, ρ̂′ ) = trace +
ρ̂′ +
ρ̂ +
ρ̂′ +
ρ̂  .

q
Since +
|ψihψ| = |ψihψ| and hence
rq q q
+ +
|ψihψ| ρ̂′ +
|ψihψ| = +
|ψihψ| ρ̂′ |ψihψ|
q
= +
hψ | ρ̂′ ψi |ψihψ|

holds for all normalized ψ ∈ H , we have:

ρ̂ = |ψihψ| =⇒ F (ρ̂, ρ̂′ ) = hψ | ρ̂′ ψi ∀ ρ̂, ρ̂′ ∈ S(H) . (5.39)

Symmetry of the fidelity in general is not evident from its definition but follows
directly from Uhlmann’s theorem (Uhlman, 1976):
DRAFT, October 17, 2007
28
For commuting ρ̂, ρ̂′ the Bures fidelity has a simple geometrical interpretation, since
X 
ρ̂ = pν |φν ihφν | ,  !2
 Xp
Xν ′
=⇒ F (ρ̂, ρ̂ ) = pν p′ν .
ρ̂′ = p′ν |φν ihφν | 
 ν
ν

See (Chen et al., 2002) for a geometrical interpretation of the Bures fidelity of general qubit states.
5.2. ADAPTION TO QUANTUM COMMUNICATION 141

Theorem 5.2.7 Let H be a Hilbert-space and ρ̂1 , ρ̂2 ∈ S(H) . Then

F (ρ̂1 , ρ̂2 ) = max |hΨ1 | Ψ2 i|2 ,
Ψ1 ∈Tρ̂1 ,Ψ2 ∈Tρ̂2

where29 n  o
def
Tρ̂ = Ψ ∈ H ⊗ H : ρ̂ = trace 2 |ΨihΨ| ∀ ρ̂ ∈ S(H) .

Outline of proof:nLet j ∈ {1, 2}oand Ψj ∈ Tρ̂j . By the spectral theorem, there is are
(j) (j) (j) (j)
orthonormal basis φ1 , . . . , φn of H , some n′ ∈ {1, . . . , n} , and s1 , . . . , sn′ > 0
with ′
n 
X 2
ρ̂j = s(j)
ν φ(j)
ν .
ν=1
Therefore, the Schmidt-decomposition of Ψj has to be of the form
′
n
X
Ψj = s(j) (j) (j)
ν φν ⊗ ψ ν
ν=1
Xn  
p
= +
ρ̂j φ(j)
ν ⊗ ψν(j)
ν=1
n o
(j) (j)
with some orthonormal basis ψ1 , . . . , ψn of H . Considering j = 1 and j = 2
together we thus get
n′ D p
X p ED E
hΨ2 | Ψ1 i = +
ρ̂2 φ(2)
ν
+
ρ̂1 φ(1)
µ ψν(2) ψµ(1) . (5.40)
ν,µ=1

In order to rewrite the r.h.s. of (5.40) as a trace we use the unitary operators V̂ and
V̂ ′ characterized by
V̂ φ(2)
µ = φµ
(1)
∀ µ ∈ {1, . . . , n}
and D E D E
φ(2)
µ V̂ ′ φ(2)
ν = ψν(2) ψµ(1) ∀ ν, µ ∈ {1, . . . , n} .
Then
n′ D p
X p ED E
hΨ2 | Ψ1 i = +
ρ̂2 φ(2)
ν
+
ρ̂1 V̂ φ(2)
µ φ(2)
µ V̂ ′ φ(2)
ν
(5.40)
ν,µ=1
Xn′ D p p E
= φ(2)
ν
+
ρ̂2 + ρ̂1 V̂ V̂ ′ φ(2)
ν
ν=1  
√ √
= trace + ρ̂2 + ρ̂1 V̂ V̂ ′
 √ √ 
= trace V̂ V̂ ′ + ρ̂2 + ρ̂1

Applying Lemma 5.2.8, below, to the polar decomposition

r
p p  p p †  p p 
′ + + + +
V̂ V̂ ρ̂2 ρ̂1 = Û ρ̂2 + ρ̂1 +
ρ̂2 + ρ̂1 , Û unitary ,

DRAFT, October 17, 2007

29
Note that Tρ̂ is the set of all purifications of ρ̂ as introduced in Lemma A.4.8.
142 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

(see, e.g., Lemma 7.3.20 of (Lücke, eine)) we always get

 q 
2 + √ √
† √ √
2
|hΨ1 | Ψ2 i| ≤ trace +
ρ̂2 + ρ̂1 +
ρ̂2 + ρ̂1

= F (ρ̂1 , ρ̂1 ) .
(j)
Obviously, by appropriate choice of the ψν we get equality.

Lemma 5.2.8 Let H be a finite-dimensional 30 Hilbert space. Then

trace (B̂ ρ̂) ≤ B̂ ∀ ρ̂ ∈ S(H) , B̂ ∈ L(H) .

Outline of proof: By the spectral theorem, there are a an orthonormal basis

{φ1 , . . . , φn } of H and p1 , . . . , pn ≥ 0 with
n
X n
X
ρ̂ = pν |φν ihφν | , pν = 1 .
ν=1 ν=1

Then
n
X  
trace (B̂ ρ̂) = pν trace B̂ |φν ihφν |
ν=1
Xn D E
= pν φν | B̂ φν
ν=1
n
X D E
≤ pν φν | B̂ φν
ν=1 | {z }
≤kB̂ k

≤ B̂ .

Uhlmanns theorem and the definition of the Bures fidelity show:

F (ρ̂1 , ρ̂2 ) = F (ρ̂2 , ρ̂1 ) , (5.41)
F (ρ̂1 , ρ̂2 ) ∈ [0, 1] , (5.42)

1 iff ρ̂1 = ρ̂2
F (ρ̂1 , ρ̂2 ) = (5.43)
0 iff ρ̂1 ρ̂2 = 0 .
Using Uhlmanns theorem one may also show:31
q
def
A(ρ̂1 , ρ̂2 ) = arccos F (ρ̂1 , ρ̂2 ) ∈ [0, π/2] is a metric , (5.44)
 
F C(ρ̂1 ), C(ρ̂2 ) ≥ F (ρ̂1 , ρ̂2 ) , (5.45)
!
X X Xq
F pν ρ̂ν , p′ν ρ̂′ν ≥ pν p′ν F (ρ̂ν , ρ̂′ν ) . (5.46)
ν ν ν
DRAFT, October 17, 2007
30
The given may be directly extended to the infinite-dimensional case.
31
Necessary and sufficient conditions for a given set of pure states to be transformable
via a quantum operation into another given set of (not necessarily pure) states are given in
(Chefles et al., 2003, Theorem 4).
5.2. ADAPTION TO QUANTUM COMMUNICATION 143

For pure states φ, ψ :

F (P̂φ , P̂ψ ) = |hφ | ψi|

q
= 1 − D(P̂φ , P̂ψ )2 .

For general states:

q
1 − F (ρ̂1 , ρ̂2 ) ≤ D(ρ̂1 , ρ̂2 ) ≤ 1 − F (ρ̂1 , ρ̂2 )2 .

Remark: In principle, fidelity and trace distance are of equal use to

characterize the difference of states. However, usually, calculations are
easier with fidelity. Therefore only the latter will be used in the following.

Relevant for transmission of (unknown) states:

 
def
Fmin (C) = min F ρ̂, C(ρ̂)
ρ̂
 
= min F P̂ψ , C(P̂ψ ) .
(5.46) ψ

Relevant for the (approximate) realization of a gate Û as the quantum operation

C is the gate fidelity
 
def
F (Û , C) = min F Û ρ̂ Û −1 , C(ρ̂)
ρ̂
 
= min F P̂Û ψ , C(P̂Û ψ )
ψ

for which we have, e.g.,32

q q q
arccos F (Û1 Û2 , C1 ◦ C2 ) ≤ arccos F (Û1 , C1 ) + arccos F (Û1 , C1 ) .

Relevant for quantum sources producing ρ̂j with probability pj and disturbed by C
is the ensemble average fidelity
X  
def
F = pj F ρ̂j , C(ρ̂j ) .
j

DRAFT, October 17, 2007

32
Recall (5.44).
144 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION

5.2.4 Schumacher Encoding

Lemma 5.2.9 Let H be a finite-dimensional Hilbert space, ρ̂0 ∈ S(H) , and
δ > 0 . Then  
⊗n (n)
lim
n→∞
trace ρ̂ 0 Λ̂ρ̂0 ,δ = 1

and33  
(n)
2n(S1 (ρ̂0 )+δ) ≥ dim Λ̂ρ̂0 ,δ H⊗n ≥ 2n(S1 (ρ̂0 )−δ) ∀ n ∈ IN ,
(n)
where, for n ∈ IN , Λ̂ρ̂0 ,δ denotes
h
the projector onto i
the subspace of all eigenvectors
⊗n n(S1 (ρ̂0 )−δ) n(S1 (ρ̂0 )+δ)
of ρ̂0 with eigenvalues in 2 ,2 .

Lemma 5.2.10
Let H be a Hilbert space and  a self-adjoint operator on H . The for arbitrary
ψ1 , . . . , ψN ∈ H and p1 , . . . , pN > 0 we have
N N
! N
X D E 2 X X
pν ψν  ψν ≥ 2 trace  pν |ψν ihψν | − pν .
ν=1 ν=1 ν=1

Outline of proof: Apply the inequality

x2 ≥ 2 x − 1 ∀ x ∈ IR
D E
to x = ψν Â ψν .

..
.
DRAFT, October 17, 2007
33
See (Mitchison and Jozsa, 2003) in this connection.
5.2. ADAPTION TO QUANTUM COMMUNICATION 145

make a guess, Fano inequality

5.2.5 A la Nielsen/Chuang
Klein inequality (Nielsen and Chuang, 2001, Theorem 11.7):34
   
trace ρ̂ log(ρ̂) ≥ trace ρ̂ log(ρ̂′ ) ∀ ρ̂, ρ̂′ ∈ S(H) . (5.47)

 
def
Hbin (p) = H {p, 1 − p} ∀ p ∈ [0, 1] .

..
.

PPT criterion
Cat states implemented for Josephson junctions or coherent states (entangle-
ment laser)

5.2.6 Entropy35
Classically, every message can be encoded in a string of bits. But can
quantum information always be encoded in a string of qubits?!

DRAFT, October 17, 2007

34
The r.h.s of (5.47) may become −∞ . Equality holds iff ρ̂ = ρ̂′ .
35
See also (Ohya and Petz, 1993) and (Petz, 2001).
146 CHAPTER 5. QUANTIFYING QUANTUM INFORMATION
Chapter 6

Handling Entanglement1

We have seen in 4.4 that perfect entanglement may be used for implementing noiseless
quantum communication. Therefore, quantification and handling (e.g., distillation)
of entanglement is important. Here, for simplicity, we consider only bipartite sys-
tems. One might expect, then, that separability is equivalent to the existence of
corresponding (local) hidden variable models, hence to the validity of all (general-
ized) Bell inequalities.2 However, as shown in (Werner, 1989), that this is not the
case.

6.1 Detecting Entanglement

Detecting entanglement of pure states is very easy:
(
ρ̂ ∈ Spure (H1 ⊗ H2 )
ρ̂ ∈ Spure (H1 ⊗ H2 ) ∪ Ssep (H1 ⊗ H2 ) ⇐⇒  2
Lemma A.4.6 trace 2 (ρ̂) = trace 2 (ρ̂) .
TheoremA.4.7

It is mixedness which can make detection of entanglement a very hard problem

(Gurvits, 2002).

6.1.1 Entanglement Witnesses and Non-Completely Posi-

tive Mappings
Lemma 6.1.1 A state ρ̂ of the bipartite system S with state space H = H1 ⊗ H2
is non-separable 3 iff it possesses an entanglement witness, i.e. an operator Ŵ ∈
L(H1 ⊗ H2 ) fulfilling the following two properties: 4
DRAFT, October 17, 2007
1
See also (Horodecki et al., 2001) and references given there.
2
See (Werner and Wolf, 201; Collins and Gisin, 2003) in this connection.
3
Recall Definition A.4.4.
4
The second property guarantees that Ŵ is Hermitian — thanks to the polarization identity
3
1X α
h(x1 , x1 ) = (−i) |x1 + iα x2 ihx1 + iα x2 | ,
4 α=0
valid for every mapping h that is linear in the second and conjugate linear in the first argument;
especially for h(φ, ψ) = |φihψ| .

147
148 CHAPTER 6. HANDLING ENTANGLEMENT

1.
trace (ρ̂ Ŵ ) < 0 .

2.
ρ̂′ separable =⇒ trace (ρ̂ Ŵ ) ≥ 0 ∀ ρ̂′ ∈ S(H1 ⊗ H2 ) .

Outline of proof: The statement follows from the known fact5 that for every point
X outside a convex set K there is a hyperplane separating X from K .

n o
(j)
Lemma 6.1.2 (Jamiolkowski) For j ∈ {1, 2} , let φ1 , . . . , φ(j)
nj be a MONS of
the Hilbert space Hj . Then for every Ŵ ∈ L(H1 ⊗ H2 ) there is a unique linear
mapping LŴ : L(H1 ) −→ L(H2 ) with

Ŵ = n1 (1 ⊗ LŴ ) (P̂H+1 ) , (6.1)

where +*
n1 n1
def 1 X X
P̂H+1 = φ(1) (1)
ν ⊗ φν φ(1) (1)
µ ⊗ φµ . (6.2)
n1 ν=1 µ=1

LŴ fulfills
 ED  n2
X D  E ED
LŴ φ(1)
ν1 φ(1)
µ1 = φ(1) (2) (1) (2)
ν1 ⊗ φν2 | Ŵ φµ1 ⊗ φµ2 φ(2)
ν2 φ(2)
µ2 (6.3)
ν2 ,µ2 =1

for all ν1 , ν2 ∈ {1, . . . , n1 } and:

LŴ positive ⇐⇒ trace (Ŵ ρ̂) ≥ 0 ∀ ρ̂ ∈ Ssep (H1 ⊗ H2 ) , (6.4)
where6
def
Ssep (H1 ⊗ H2 ) = {ρ̂ ∈ S(H1 ⊗ H2 ) : ρ̂ separable} . (6.5)

Outline of proof: Defining LŴ by (6.3) plus linear continuation gives

n1
X n2
X D  E ED
Ŵ = φ(1)
ν1 ⊗ φ (2)
ν2 | Ŵ φ (1)
µ1 ⊗ φ (2)
µ2 φ (1)
ν1 ⊗ φ (2)
ν2 φ(1) (2)
µ1 ⊗ φµ2
ν1 ,µ1 =1 ν2 ,µ2 =1
X n1  ED   ED 
= φ(1)
ν1 φ(1)
µ1 ⊗ LŴ φ(1)
ν1 φ(1)
µ1
(6.3)
ν1 ,µ1 =1 !

n1
X ED
= 1 ⊗ LŴ φ(1)
ν1 ⊗ φ(1)
ν1 φ(1) (1)
µ1 ⊗ φµ1
lin. cont.
ν1 ,µ1 =1

DRAFT, October 17, 2007

5
See, e.g., (Neumark, 1959, $ 1, No. 9) and (Robertson and Robertson, 1967, Kap. 1, Satz 8).
6
Note that
Ssep (H1 ⊗ H2 ) = S(H1 ) ⊗alg S(H2 ) .
Theorem A.4.5
6.1. DETECTING ENTANGLEMENT 149

and hence (6.1). Conversely, (6.1) gives

 ED  D X n1  ED   ED  E
LŴ φ(1)
ν1
(1)
φµ1 = φ(1)
ν1 φ
(1)
′
ν1 φ
(1)
′
µ1 ⊗ L Ŵ φ
(1)
′
ν1 φ
(1)
′
µ1 φ (1)
µ1
D ν1 ,µ1 =1 E

+
= φ(1)
ν1 dim(H1 ) 1 ⊗ LŴ (P̂H1 ) φµ1
(1)
lin.
D E
= φ(1)
ν1 Ŵ φµ1
(1)
,
(6.1)

i.e. (6.3). Since

! 
N
X N
X  
(1) (2) (1) (2)
trace Ŵ ρ̂k ⊗ ρ̂k = dim(H1 ) trace ρ̂k ⊗ LŴ ρ̂k ,
Jami
k=1 k=1

positivity of LŴ implies nonnegativity of trace (Ŵ ρ̂) for separable7 ρ̂ . Conversely
the latter implies positivity of LŴ , since
D  ED  E D E
φ(2) LŴ φ(1) φ(1) φ(2) = Ψφ(1) ,φ(2) Ŵ Ψφ(1) ,φ(2) ∀ φ(1) ∈ H1 , φ(2) ∈ H2 ,
(6.3)

where  
dim(H1 ) D E
def
X
Ψφ(1) ,φ(2) =  φ(1) φ(1)
ν1 φ(1)
ν1
 ⊗ φ(2) .
ν1 =1

n o
(j)
Corollary 6.1.3 For j ∈ {1, 2} , let φ1 , . . . , φ(j)
nj be a MONS of the Hilbert
space Hj . Then for every Ŵ ∈ L(H1 ⊗ H2 ) we have8
 
trace (ρ̂ Ŵ ) < 0 =⇒ 1 ⊗ L†Ŵ (ρ̂) 6≥ 0 ∀ ρ̂ ∈ S(H1 ⊗ H2 ) (6.6)

and

trace (ρ̂ Ŵ ) ≥ 0 ∀ ρ̂ ∈ Ssep (H1 ⊗ H2 ) ⇐⇒ L†Ŵ (ρ̂2 ) ≥ 0 ∀ ρ̂2 ∈ S(H2 ) , (6.7)

where L†Ŵ denotes the Hilbert-Schmidt adjoint of the linear mapping LŴ that is
characterized by (6.3), i.e.:
   
trace L†Ŵ (Â2 ) Â1 = trace Â2 LŴ (Â1 ) ∀ Â1 ∈ L(H1 ) , Â2 ∈ L(H2 ) .

DRAFT, October 17, 2007

7
Recall Footnote 6.
8
Positivity of  ∈ L(H) is easily to checked:
dim(H)
X
 ≥ 0 ⇐⇒ det( − x 1̂) = |{z}
c (−1)ν cν xν .
|{z}
∈IR ν=1 ≥0
150 CHAPTER 6. HANDLING ENTANGLEMENT

Outline of proof: (6.6) follows from

 
1
+
trace (ρ̂ Ŵ ) = trace ρ̂ 1 ⊗ LŴ (P̂H )
n1 (6.1) 1
  
† +
= trace 1 ⊗ LŴ (ρ̂) P̂H1

and (6.7) from (6.4).

We may conclude:

 
1. If Ŵ is an entanglement witness for ρ̂ then 1 ⊗ L†Ŵ (ρ̂) 6≥ 0 and, there-
fore, the positive map L†Ŵ cannot be completely positive.

2. For every positive mapping L′ : H2 −→ H1 we have:9

(1 ⊗ L′ )(ρ̂) 6≥ 0 =⇒ ρ̂ ∈
/ Ssep (H1 ⊗ H2 ) .

3. A state ρ̂ ∈ S(H1 ⊗ H2 ) is separable if and only if (1 ⊗ L′ )(ρ̂) is positive

for all positive mappings L′ : H2 −→ H1 .

Remark: In general, for detecting entanglement, L†Ŵ is more useful

than Ŵ since
i.g.
L†Ŵ (ρ̂) 6≥ 0 6=⇒ trace (Ŵ ρ̂) < 0 .
Ŵ , on the other hand, allows to detect entanglement by local measure-
ments (Gühne et al., 2002).

6.1.2 Examples
Lemma 6.1.4 The flip operator of the bipartite system S with state space H⊗H ,
i.e. the linear Operator F̂ on H ⊗ H characterized by
def
F̂ (ψ1 ⊗ ψ2 ) = ψ2 ⊗ ψ1 ∀ ψ1 , ψ2 ∈ H , (6.8)
has the following properties:10
1.
F̂ = F̂ † , F̂ 2 = 1̂ .

DRAFT, October 17, 2007

9
Of course this statement is relevant for non-completely positive L′ , only.
10
Recall Definition A.4.4.
6.1. DETECTING ENTANGLEMENT 151

2.
F̂ = P̂+ − P̂− ,
where the
1 
def
1̂ ± F̂
P̂± =
2
are the projectors onto the symmetric resp. anti-symmetric pure states of S :
Ψ ∈ P̂± (H ⊗ H) ⇐⇒ F̂ Ψ = ±Ψ̂ ∀Ψ ∈ H ⊗ H.
3. For all β ∈ IR :
1̂ + β F̂ ≥ 0 ⇐⇒ β ∈ [−1, +1] .
4.
trace (F̂ ) = − dim(H) .

5. For all ρ̂ ∈ S(H ⊗ H) :

ρ̂ separable =⇒ trace (F̂ ρ̂) ≥ 0 .

Outline of proof: The first four statements are more or less obvious and the last
one follows from
    
trace F̂ |ψ1 ihψ1 | ⊗ |ψ2 ihψ2 | = trace F̂ |ψ1 ⊗ ψ2 ihψ1 ⊗ ψ2 |
 
= trace |ψ2 ⊗ ψ1 ihψ1 ⊗ ψ2 |

= hψ1 ⊗ ψ2 | ψ2 ⊗ ψ1 i
2
= |hψ1 | ψ2 i| .

Consequence: The flip operator F̂ is an entanglement witness for the

mixed Werner states 11 ρ̂W (β) with β ∈ [−1, −1/ dim(H)] , where
def 1̂ + β F̂
ρ̂W (β) = ∀ β ∈ [−1, +1] . (6.9)
trace (1̂ + β F̂ )

For H1 = H2 = H the positive linear mapping (6.3) associated

n with the o flip operator
12 (1) (1)
on H ⊗ H is the transposition w.r.t. {φ1 , . . . , φn } = φ1 , . . . , φn1 :
   
LF̂ |φν1 ihφν2 | = T |φν1 ihφν2 |
def
= |φν2 ihφν1 | (6.10)
 
= L′F̂ |φν1 ihφν2 | ∀ ν1 , ν2 ∈ {1, . . . , n} .
DRAFT, October 17, 2007
11
Compare Footnote 27 of Chapter 4.
12
The transposition was considered at the end of 4.2.1.
152 CHAPTER 6. HANDLING ENTANGLEMENT

While the flip operator cannot be an entanglement witness for, e.g., the F̂ -invariant
pure states,13 a two-qubit state ρ̂ is separable iff its partial transpose (1 ⊗ T) (ρ̂) is
positive. Slightly more generally we have:
n o
(j)
Theorem 6.1.5 For j ∈ {1, 2} , let φ1 , . . . , φ(j)nj be a MONS of the Hilbert
space Hj . If n1 + n2 ∈ {4, 5} then an arbitrarily given state
n1
X n2
X ED
ρ̂ = ρν1 ν2 ,µ1 µ2 φ(1) (2)
ν1 ⊗ φν2 φ(1) (2)
µ1 ⊗ φµ2 ∈ S(H1 ⊗ H2 )
ν1 ,µ=1 ν2 ,µ2 =1

is separable iff it has a positive partial transpose:14

n1
X n2
X ED
ρν1 ν2 ,µ1 µ2 φ(1) (2)
ν1 ⊗ φµ2 φ(1) (2)
µ1 ⊗ φν2 ≥ 0 .
ν1 ,µ=1 ν2 ,µ2 =1

Proof:15 See (Horodecki et al., 1996, Theorem 3).

n o
(1)
Another example for H1 = H2 = H and {φ1 , . . . , φn } = φ1 , . . . , φ(1)
n1 =
n o
(2)
φ1 , . . . , φ(2)
n1 is

Ŵ = 1̂H⊗H − n P̂H+
n 
X   
= |φν ihφµ | ⊗ δνµ 1̂ − |φν ihφµ |
ν, µ=1
 
+
= m 1 ⊗ LD
H (P̂H ) ,

where
LŴ (B̂) = LD
H (B̂) (6.11)
(6.1)
def
= trace (B̂) − B̂ ∀ B̂ ∈ L(H) , (6.12)
Obviously, LD
H is positive and
 
1 ⊗ LD
H (ρ̂) = trace 2 (ρ̂) ⊗ 1̂ − ρ̂ ∀ ρ̂ ∈ L(H) .

Therefore,16
trace 2 (ρ̂) ⊗ 1̂ 6≥ ρ̂ =⇒ ρ̂ ∈
/ Ssep (H ⊗ H) . (6.13)

DRAFT, October 17, 2007

13
See, however, (Horodecki and Horodecki, 1996).
14
Positivity of the partial
n transpose, contrary
o to the partial transpose itself, does not depend on
(2) (2)
the choice for the MONS φ1 , . . . , φn2 .
15
The essential point is that — provided n1 + n2 ∈ {4, 5} — every positive mapping L′ :
L(H2 ) −→ L(H1 ) is decomposable; see (Labuschagne et al., 2003) and references given there.
For entangled PPT states in higher dimensions see (Ha et al., 2003).
16
See (Hiroshima, 2003) in this connection.
6.1. DETECTING ENTANGLEMENT 153

6.1.3 Other Criteria17

Up to now, in view of Corollary 6.1.3, we considered the entanglement criterion

(1 ⊗ |{z}
L )(ρ̂) 6≥ 0 =⇒ ρ̂ ∈
/ Ss (H1 ⊗ H1 ) . (6.14)
≥0

only for positive maps L : L(H2 ) −→ L(H1 ) . But, of course, (6.14) also holds for
positive maps L : L(H2 ) −→ L(H2 ) . Typical examples are

1. L = T :

Every separable state ρ̂ ∈ S(H1 ⊗ H2 ) is PPT, i.e. has a positive

partial transpose (1 ⊗ T)(ρ̂) — for every choice of MONS.

2. L = LD
H :
ρ̂ ∈ Ssep (H1 ⊗ H2 ) =⇒ trace 2 (ρ̂) ⊗ 1̂ ≥ ρ̂ . (6.15)

Remarks:

1. Since18
 †
(1 ⊗ T)(ρ̂) = (1 ⊗ T)(ρ̂) ∀ ρ̂ ∈ S(H1 ⊗ H2 )

and  
trace (1 ⊗ T)(ρ̂) = 1 ∀ ρ̂ ∈ S(H1 ⊗ H2 ) ,
we have19

(1 ⊗ T)(ρ̂) 6≥ 0 ⇐⇒ k(1 ⊗ T)(ρ̂)k1 > 1 ∀ ρ̂ ∈ S(H1 ⊗ H2 ) ,

as exploited in (Vidal and Werner, 2002).

2. Since20
T (τ̂ ν ) = (−1)δν2 τ̂ ν ∀ ν ∈ {0, . . . , 3} ,
we have for two-qubit states ρ̂ :21
 
(1 ⊗ T)(ρ̂) = ρ̂ ⇐⇒ trace (τ̂ µ ⊗ τ̂ 2 ) ρ̂ = 0 ∀ µ ∈ {0, . . . , 3} .

DRAFT, October 17, 2007

17
See also (Doherty et al., 2003)
18
Note that  †
(1 ⊗ T)(† ) = (1 ⊗ T)(Â) ∀  ∈ L(H1 ⊗ H2 ) .

19
Recall (5.31).
20
Recall (A.22)
21
Compare (Altafini, 2003, Sect. II.B., Corollary 1).
154 CHAPTER 6. HANDLING ENTANGLEMENT

3. For all ρ̂ ∈ S(H1 ⊗ H2 ) :

λ Ψ 6= 0
(1 ⊗ T)(ρ̂) Ψ = |{z}
(  <0 
(1 ⊗ T) |ΨihΨ| entanglement witness for ρ̂ ,
=⇒
Ψ non-separable .

Let us list some other sufficient criteria for entanglement:

1. Range criterion:22
 ∗  
ψ (1) ⊗ ψ (2) ∈
/ range (1 ⊗ T)(ρ̂) ∀ ψ (1) ⊗ ψ (2) ∈ range(ρ̂)
(6.16)
=⇒ ρ̂ ∈
/ Ssep (H1 ⊗ H2 ) ,

where n2 D
  X E
(2) ∗
ψ = ψ (2) | φ(2)
ν φ(2)
ν
(4.9) ν=1
n o
(2)
depends on the MONS φ1 , . . . , φ(2)
n2 of H2 .

Outline of proof: By Theorem A.4.5 (and the spectral theorem), every ρ̂ ∈

Ssep (H1 ⊗ H2 ) can be written in the form
N
X ED
(1) (2) (1) (2)
ρ̂ = ψk ⊗ ψk ψk ⊗ ψk .
k=1

Then Lemma A.4.2 implies

(1) (2)
ψk ⊗ ψk ∈ range(ρ̂) ∀ k ∈ {1, . . . , N }
and, since23
N
X  ∗ ED  ∗
(1) (2) (1) (2)
(1 ⊗ T)(ρ̂) = ψk ⊗ ψk ψk ⊗ ψk ,
k=1

also  ∗  
(1) (2)
ψk ⊗ ψk ∈ range (1 ⊗ T)(ρ̂) ∀ k ∈ {1, . . . , N } .

2. Entropic inequalities:24
)
ρ̂ ∈ Ssep (H1 ⊗ H2 )  
=⇒ Sα trace j (ρ̂) ≤ Sα (ρ̂) ∀ j ∈ {1, 2} , (6.17)
α ∈ {0, 1, 2, ∞}

DRAFT, October 17, 2007

22
This criterion especially implies that every state ρ̂ ∈ S(H1 ⊗ H2 ) containing no product state
in its range must be entangled. For a non-separable PPT state on C2 ⊗ C4 containing product
states in its range and fulfilling the range criterion see (Horodecki et al., 2001, Eqn. (27)).
23
Recall (4.14).
24
See (Vollbrecht and Wolf, 2002, Sect. III) and references given there.
6.1. DETECTING ENTANGLEMENT 155

(more information for the total state than for the partial states), where25
 
def
log2 trace (ρ̂α )
Sα (ρ̂) = ∀ α ∈ (0, 1) ∪ (1, ∞) (6.18)
1−α
and26
def
Sk (ρ̂) = lim Sα (ρ̂) ∀ k ∈ {0, ∞} .
α→k

3. Greatest cross norm criterion (Rudolph, 2000, Thm. 5):

ρ̂ ∈ Ssep (H1 ⊗ H2 ) ⇐⇒ kρ̂kγ = 1 , (6.19)

where
X
N N
X 
def (1) (2) (1) (2)
 = inf Âk Âk : Âk ⊗ Âk =  , N ∈ IN (6.20)
γ 1 1 | {z } | {z }
k=1 k=1
∈L(H1 ) ∈L(H2 )

for  ∈ L(H1 ⊗ H2 ) .
4. Computable cross norm criterion27 (Rudolph, 2002, Prop. 19):

ρ̂ ∈ Ssep (H ⊗ H) =⇒ kA(ρ̂)k ≤ 1 , (6.21)

 
where the linear mapping A : L(H ⊗ H) −→ L L(H) is characterized by
 
A |φν1 ⊗ φν2 ihφµ1 ⊗ φµ2 | Â
def
D E (6.22)
= φν2 Â φµ2 |φν1 ihφµ1 | ∀ ν1 , ν2 , µ1 , µ2 ∈ {1, . . . , n} , Â ∈ L(H)

w.r.t. the fundamental MONS {φ1 , . . . , φn } of H .

5. Reduction criterion28 (Horodecki and Horodecki, 1999):
(
trace 2 (ρ̂) ⊗ 1̂ − ρ̂ ≥ 0 ,
ρ̂ ∈ Ssep (H1 ⊗ H2 ) ⇐⇒ (6.23)
1̂ ⊗ trace 1 (ρ̂) − ρ̂ ≥ 0 .

DRAFT, October 17, 2007

25
The normalization factor 1/1 − α guarantees
 
max Sα (ρ̂) = log2 dim(H)
ρ̂∈S(H)

for the so-called Renỳi quantum entropies Sα ; see also (Lavenda and Dunning-Davies, 2003).
26
Note also that
S1 (ρ̂) = lim Sα (ρ̂) .
1<α→1

27
Note that for dim(H) ≥ dim(H1 ), dim(H2 ) there is a canonical mapping of S(H1 ⊗ H2 ) into
S(H ⊗ H) respecting separability. Therefore, it is sufficient to consider the case H1 = H2 .
28
See (Hiroshima, 2003), in this connection.
156 CHAPTER 6. HANDLING ENTANGLEMENT

6.2 Local Operations and Classical Communica-

tion (LOCC)
6.2.1 General Aspects
By local operations and classical communication (LOCC) state transforma-
tions of the form29
N
X  
′ (1) (2)
ρ̂ 7−→ ρ̂ = LLOCC (ρ̂) = pk Ck ⊗ Ck (ρ̂) (6.24)
k=1

can be implemented on a bipartite system with state space H1 ⊗ H2 , where

N
X
N ∈ IN p =1
k
|{z}
j=1
>0

(1) (2)
and the Ck resp. Ck are trace preserving elements of Q(H1 , H1 ) resp. Q(H2 , H2 ) .
Obviously,

ψ (1) ∈ H1 , ψ (2) ∈ H2   
=⇒ SLOCC ψ (1) ⊗ ψ (2) = Ssep (H1 ⊗ H2 ) , (6.25)
ψ (1) = ψ (2) = 1 

where
def
SLOCC (ρ̂) = {states ρ̂′ of the form (6.24)} .

In order to characterize the possible transformations of pure states by LOCC we

need the following definition:

Definition 6.2.1 Let Ĥ, Ĥ ′ be Hermitian operators on the Hilbert space H with
spectral decompositions
n
X n
X
Ĥ = Eν |ψν ihψν | , Ĥ ′ = Eν′ |ψν′ ihψν′ | .
ν=1 ν=1

Then we write Ĥ  Ĥ ′ iff 30 there p1 , . . . , pn ≥ 0 and permutations π1 , . . . , πn ∈ Sn

with n n
X X
Eµ = pν Eπ′ ν (µ) ∀ µ ∈ {1, . . . , n} , pν = 1 .
ν=1 ν=1

DRAFT, October 17, 2007

29
Obviously, the LOCC quantum operations (6.24) form a group.
30
Obviously,  does not depend on the choice of orthonormal eigenbases for Ĥ, Ĥ ′ and gives a
semi-ordering of S(H) .
6.2. LOCAL OPERATIONS AND CLASSICAL COMMUNICATION 157

Theorem 6.2.2 Let Ĥ, Ĥ ′ be Hermitian operators on the Hilbert space H . Then
Ĥ  Ĥ ′ iff there are p1 , . . . , pn ≥ 0 and unitary Operators Û1 , . . . , Ûn on H with
n
X n
X
Ĥ = pν Ûν Ĥ ′ Ûν† , pν = 1 .
ν=1 ν=1

Proof: See (Nielsen and Chuang, 2001, Theorem 12.13).

For pure states and H1 = H2 , concerning LOCC, nothing is lost if the operations
on one side are restricted to be unitary:

Lemma 6.2.3 Let H be a Hilbert space, 0 6= Ψ ∈ H ⊗ H and M̂ ∈ L(H) . Then

there are N̂ , Û ∈ L(H) with
   
1̂ ⊗ M̂ Ψ = N̂ ⊗ Û Ψ , Û † = Û −1 .

Outline of proof: According ton Theorem A.4.7 o there

n are non-negativeo numbers
(1) (1) (2) (2)
s1 , . . . , sn and orthonormal bases φ1 , . . . , φn and φ1 , . . . , φn of H with
n
X def
Ψ= sα φ(1) (2)
α ⊗ φα , n = dim(H) .
α=1

Then, writing
n
X ED
M̂ = Mνµ φ(2)
ν φ(2)
µ
ν,µ=1

and defining
n
X ED
def
N̂ ′ = Mνµ φ(1)
ν φ(1)
µ
ν,µ=1
we get    
Ŝ N̂ ′ ⊗ 1̂ Ψ = 1̂ ⊗ M̂ Ψ ,

where Ŝ denotes the linear swap operator characterized by

def
Ŝ φ ⊗ ψ = ψ ⊗ φ ∀ φ, ψ ∈ H .
   
Therefore N̂ ′ ⊗ 1̂ Ψ and 1̂ ⊗ M̂ Ψ have the same Schmidt coefficients, i.e. there
are unitary Û , V̂ ∈ L(H) with
    
1̂ ⊗ M̂ Ψ = V̂ ⊗ Û N̂ ′ ⊗ 1̂ Ψ .

def
Defining N̂ = V̂ N̂ ′ we get the statement of the lemma.

 
(1) (2)
The state LLOCC (ρ̂) in (6.24) can be pure only if it coincides with Ck ⊗ Ck (ρ̂)
for all k ∈ {1, . . . , N } with pk 6= 0 . This, together with Theorem 6.2.2 allows us to
prove the following:
158 CHAPTER 6. HANDLING ENTANGLEMENT

Theorem 6.2.4 Let H be a Hilbert space. Then

ρ̂′ ∈ SLOCC (ρ̂) ⇐⇒ trace 2 (ρ̂)  trace 2 (ρ̂′ ) ∀ ρ̂, ρ̂′ ∈ Spure (H ⊗ H) .

Outline of proof: Assume that (6.24) holds. Then, by Lemma 6.2.3, LLOCC can
be chosen such that
XN′    
′
ρ̂ = M̂j ⊗ Ûj ρ̂ M̂j† ⊗ Ûj†
j=1 | {z }
∝ρ̂′

holds with N ′ ∈ IN and M̂j , Ûj ∈ L(H) fulfilling

′
N
X
M̂j† M̂j = 1̂ , Ûj = Ûj† ∀ j ∈ {1, . . . , N ′ } . (6.26)
j=1

This implies
M̂j trace 2 (ρ̂) M̂j† = p′j trace 2 (ρ̂′ ) ∀ j ∈ {1, . . . , N ′ } , (6.27)
where  
def
p′j = trace M̂j trace 2 (ρ̂) M̂j† ∀ j ∈ {1, . . . , N ′ } . (6.28)
Using the polar decomposition
r † p
p p
trace 2 (ρ̂) M̂j† trace 2 (ρ̂) M̂j† trace 2 (ρ̂) M̂j†
+ + + +
= Ûj
q
= Ûj +
M̂j trace 2 (ρ̂) M̂j†
q
= Ûj +
p′j trace 2 (ρ̂′ )
(6.27)

and multiplying from the right with its adjoint gives

p p
+
trace 2 (ρ̂) M̂j† M̂j + trace 2 (ρ̂) = p′j Ûj trace 2 (ρ̂′ ) Ûj† ∀ j ∈ {1, . . . , N ′ } .
Finally, summing over j gives
′
N
X
trace 2 (ρ̂) = p′j Ûj trace 2 (ρ̂′ ) Ûj† (6.29)
(6.26)
j=1

and hence trace 2 (ρ̂)  trace 2 (ρ̂′ ) , by Theorem 6.2.2, since (6.26) and (6.28) imply
′
N
X
p′j = 1 . (6.30)
j=1
|{z}
≥0

Conversely, if ρ̂1  ρ̂′1 , where

def def
ρ̂1 = trace 2 (ρ̂1 ) , ρ̂′1 = trace 2 (ρ̂′1 ) ,
then Theorem 6.2.2 implies (6.29) for suitable unitary Û1 , . . . , ÛN ′ ∈ L(H) and
p′1 , . . . , p′N ′ > 0 fulfilling (6.30). Then, if we define
q q
−1
def 1̂ − P̂ρ̂ H
p′j ρ̂′1 Ûj† ∀ j ∈ {1, . . . , N ′ }
+
M̂j = +
ρ̂1 /\ ρ̂ H P̂ρ̂ H + q
p′j
6.2. LOCAL OPERATIONS AND CLASSICAL COMMUNICATION 159

(6.29) and (6.30) imply

′
N
X
M̂j† M̂j = 1̂ (6.31)
j=1

and
   
trace 2 M̂j ⊗ 1̂ ρ̂ M̂j† ⊗ 1̂ = p′j trace 2 (ρ̂′ ) ∀ j ∈ {1, . . . , N ′ } , (6.32)

Since
 (    
trace 2 (ρ̂′′ ) = trace 2 (ρ̂′ ) 1̂ ⊗ Û ρ̂′′ 1̂ ⊗ Û † = ρ̂′
=⇒
ρ̂′ , ρ̂′′ ∈ Spure (H ⊗ H) TheoremA.4.7 for some unitary Û ∈ L(H) ,

(6.32) shows that there are unitary V̂1 , . . . V̂N ′ ∈ L(H) with
   
M̂j ⊗ V̂j ρ̂ M̂j† ⊗ V̂j† = p′j ρ̂′ ∀ j ∈ {1, . . . , N ′ } .

This, together with (6.30) and (6.31), shows that ρ̂ can be transformed into ρ̂′ by
LOCC.

Lemma 6.2.5 Let Ĥ, Ĥ ′ be Hermitian operators on the Hilbert space H with
spectral decompositions
n
X n
X
Ĥ = Eν |ψν ihψν | , Ĥ ′ = Eν′ |ψν′ ihψν′ | .
ν=1 ν=1

Then Ĥ  Ĥ ′ iff the conditions

n
X n
X
Eν = Eν′ (6.33)
ν=1 ν=1

and ′ ′
n
X n
X
′
max Eπ(ν) ≤ max Eπ(ν) ∀ n′ ∈ {1, . . . , n − 1} (6.34)
π∈Sn π∈Sn
ν=1 ν=1
are fulfilled.

Proof: See (Nielsen and Chuang, 2001, Proposition 12.11).

Remarks:
1. Obviously,
1
1̂  ρ̂1 ∀ ρ̂1 ∈ S(H) .
dim H
2. According to Theorem 6.2.4, therefore,31
 
Spure (H ⊗ H) ⊂ SLOCC P̂H+ .
DRAFT, October 17, 2007
31
Recall (6.2).
160 CHAPTER 6. HANDLING ENTANGLEMENT

3. For qubit states ρ̂, ρ̂′ we always have either ρ̂  ρ̂′ or ρ̂′  ρ̂ or both.
4. However, for higher dimensional H neither ρ̂  ρ̂′ nor ρ̂′  ρ̂ need
be true for ρ̂, ρ̂′ ∈ S(H) as application of Lemma 6.2.5 to, e.g., the
case
1 1
ρ̂ = (7 φ1 + 7 φ2 + φ3 ) , ρ̂ = (11 φ1 + 2 φ2 + 2 φ3 )
15 15
shows if {φ1 , φ2 , φ3 } is a MONS of H .

Note that LOCC would be much more powerful if intermediate use of nonlocally
entangled ancillary pairs, restoring their original states, could be made:
n o
(j)
For j = 1 resp. j = 2 let φ1 , . . . , φ(j)
nj be a MONS of the Hilbert
space Hj , let
n1
X n1
X
Ψ= λνµ φ(1) (1)
ν ⊗ φµ , Ψ′ = λ′νµ φ(1) (1)
ν ⊗ φµ
νµ=1 ν,µ1

be pure states of the bipartite system with state space H1 ⊗ H1 and let
n1 X
X n2    
(2)
Φ= λνµ λanc (1) (2)
αβ φν ⊗ φα ⊗ φ(1)
µ ⊗ φβ ,
ν1 α,β=1
Xn1 Xn2    
(2)
Φ′ = λ′νµ λanc (1) (2)
αβ φν ⊗ φα ⊗ φ(1)
µ ⊗ φβ
ν1 α,β=1

be pure states of the bipartite system with state space (H1 ⊗ H2 ) ⊗

(H1 ⊗ H2 ). Then it may happen that32

|Ψi 6 |Ψ′ i but |Φi  |Φ′ i

n2
X (2)
— an effect of the ancillary system in the state λanc (2)
αβ φα ⊗ φβ that
α,β=1
is called entanglement catalysis.

Finally, given ρ̂, ρ̂′ ∈ Spure (H ⊗ H) , let us note that ρ̂ can be transformed into ρ̂′ by
local operations without communication iff

trace 2 (ρ̂′ ) ∝ K̂ trace 2 (ρ̂) K̂ † , K̂ † K̂ ≤ 1̂ (6.35)

holds for some K̂ ∈ L(H) . Let

n
X n
X
trace 2 (ρ̂) = pν |ψν ihψν | , trace 2 (ρ̂′ ) = p′ν |ψν′ ihψν′ |
ν=1 ν=1
DRAFT, October 17, 2007
32
See (Nielsen and Chuang, 2001, Exercise 12.21) for an explicit example with (n1 , n2 ) = (4, 2) .
6.3. QUANTIFICATION OF ENTANGLEMENT 161

be spectral decompositions of ρ̂, ρ̂′ . Then, exploiting unitary transformations and

the polar decomposition of K̂ one can show that (6.35) is equivalent to existence of
real numbers k1 , . . . , kn and a permutation π ∈ Sn with:
n
X
(kν )2 ≤ 1 , p′ν = (kν )2 pπ(ν) ∀ ν ∈ {1, . . . , n} .
ν=1

6.2.2 Entanglement Dilution

..
.

6.2.3 Entanglement Distillation

See, e.g., (Bowmeester et al., 2000, Section 8.4) and (Devetak and Winter, 2005).

6.3 Quantification of Entanglement33

For pure states |ΨihΨ| of a bipartite system S with state space H = H1 ⊗ H2
there is a generally accepted measure of entanglement, namely the entropy of
entanglement 34
    
def
Epure |ΨihΨ| = S1 trace 2 |ΨihΨ| (6.36)
  
= S1 trace 1 |ΨihΨ| ,
Th. A.4.7
i.e. the von Neumann entropy of the partial states. Since
 ′
 ′
n n
X √ X
Epure  pν φ(1)
ν ⊗ φ(2)
ν
 =− p log2 (pν )
ν
|{z}
ν=1 ν=1
>0
 
holds for Schmidt decompositions, Epure |ΨihΨ| becomes maximal35 for
n o 1
n′ = min dim(H1 ), dim(H2 ) , pν = ∀ ν ∈ {1, . . . , n′ } .
n′
DRAFT, October 17, 2007
33
See also (Řeháček and Hradil, 2002).
34
The unit of entanglement is one ebit.
35
Recall Remark 3 in the beginning of Section 5.1.
162 CHAPTER 6. HANDLING ENTANGLEMENT

Hence    n o
max E |ΨihΨ| = log2 min dim(H1 ), dim(H2 ) (6.37)
Ψ∈H1 ⊗H2
kΨk=1

holds for E = Epure .

Strict requirements for every entanglement measure E on S(H1 ⊗ H2 ) :

1.
E(ρ̂) ≥ 0 ∀ ρ̂ ∈ S(H1 ⊗ H2 ) .

2.
E(ρ̂) = 0 ∀ ρ̂ ∈ Ssep (H1 ⊗ H2 ) .

3.
E(ρ̂) = 0 =⇒ ρ̂ ∈ Ssep (H1 ⊗ H2 ) ∀ ρ̂ ∈ S(H1 ⊗ H2 )
(could be relexed by cosideration of additional entanglement measures).

4.
ρ̂′ ∈ SLOCC (ρ̂) =⇒ E(ρ̂′ ) ≤ E(ρ̂) ∀ ρ̂, ρ′ ∈ S(H1 ⊗ H2 ) .

5. For all pure states |Ψ1 ihΨ1 | , |Ψ2 ihΨ2 | ∈ S(H1 ⊗ H2 ) :

       
Epure |Ψ1 ihΨ1 | < Epure |Ψ2 ihΨ2 | =⇒ E |Ψ1 ihΨ1 | < E |Ψ2 ihΨ2 | .

Desirable for entanglement measures E on S(H1 ⊗ H2 ) :

1.
E(ρ̂) ≥ E(ρ̂′ ) =⇒ ρ̂′ ∈ SLOCC (ρ̂) ∀ρ̂, ρ̂′ ∈ S(H1 ⊗ H2 )
(not possible).

2. Continuity

3. Additivity

4. Subadditivity

5. Convexity

Standard entanglement measures:36

DRAFT, October 17, 2007
36
There cannot be a unique entanglement measure (Morikoshi et al., 2003).
6.3. QUANTIFICATION OF ENTANGLEMENT 163

1. Entanglement of formation37 is the convex continuation

P  
def
EF (ρ̂) = P inf ν pν S1 trace 2 (ρ̂ν )
ρ̂= pν ρ̂ν
ν |{z} |{z}
0 pure

of the entropy of entanglement (6.36) to all of S(H1 ⊗ H2 ) .

Additivity of the entanglement of formation is generally conjectured but not
yet proved.
2.
Squashed Entanglement (Christandl and Winter, 2004)

• In general, Werner states are not maximally entangled, i.e. their entangle-
ment of formation (tangle) is not maximal for given a fixed degree of mixedness
(linear entropy) (White et al., 2001).
• Compare with (Eckert et al., 2002). (see also quant-ph/0210107)
• What about local superselection rules? (Verstraete and Cirac, 2003;
Bartlett and Wiseman, 2003)
• Exploit the notion of truncated expectation values. (Lee et al., 2003)
• What is the generalization of the latter for mixed states?
• Existiert ein Abstandsmaß a la (Lee et al., 2003)?
• Warum verwendet man nicht den Hilbert-Schmidt-Abstand von
Zuständen, der sich leicht mithilfe von Erwartungswerten von (Pro-
dukten von) Pauli-Operatoren ausdrücken läßt?
• Besteht ein Zusammenhang mit (Lee et al., 2003)? Der Abstand
zur Menge der separablen Zustände sollte doch ein Maß für Ver-
schränktheit sein...
Remarks:
1. For (dim(H1 ), dim(H1 )) either (2,2) or (3,2) PPT w.r.t. the second
factor is necessary and sufficient for separability (Horodecki et al., 1996).
2. Otherwise states with bound entanglement, i.e. entangled PPT
states, exist (Horodecki et al., 1996, Appendix).
3. In the 2-qubit case every entangled mixed state can be represented
as a convex combination of a separable (in general mixed) state
with a pure entangled state (Lewenstein and Sanpera, 1998). The
representation with minimal norm of the pure state is unique.
4. Also this shows that, for the 2-qubit case, separability is equivalent
to PPT.
DRAFT, October 17, 2007
37
For pairs of qubits the entanglement of formation coincides with the concurrence
(Wootters, 2001, Section 3.1).
164 CHAPTER 6. HANDLING ENTANGLEMENT
Appendix A

A.1 Turing’s Halting Problem

The halting problem is the following:

There is no algorithm by which one may decide for every program and
every finite input to the program whether the program will halt or loop
forever.

The proof given by Turing is essentially as follows:

Since every program may be encoded into a finite sequence (b1 , . . . , bn )

of bits the programs may be indexed by the corresponding numbers
Pn n−ν
ν=1 bν 2 . The same holds for all finite inputs. Now assume that
there is an algorithm telling us for all (j, k) ∈ ZZ2+ whether program j
will halt on input k . Then this algorithm may be used to write a program
P with the following property:

For every j ∈ ZZ+ , program P will hold on input j iff program

j will not.

Obviously, program P is different from program j for every j ∈ ZZ+ — a

contradiction.

A heuristic explanation is the following:

There are uncountably many possibilities for infinite loops which, there-
fore, cannot be checked in a systematic way. But we cannot be sure
whether a given program will halt on a given input or not unless

• an infinite loop is found by chance or

• the program was actually tested on the input and found to halt.

Let us finally note that the halting problem is a solution to Hilbert’s 23rd
problem (see http://aleph0.clarku.edu/ djoyce/hilbert/).

165
166 APPENDIX A.

A.2 Some Remarks on Quantum Teleportation

Even though quantum teleportation, described in 1.2.2, seems to indicate some kind
of quantum nonlocality, there is a naive ‘explanation’ relying on locality and some
kind of realism:
There is a set of four compatible relations, corresponding to the Bell
states, between a pair of qubits. These correlations are so strong that
the state (predicting ensemble averages) of the second qubit is fixed by
that of the first qubit and the Bell relation (considered as an element of
reality). Therefore, Alice need only inform Bob about the Bell relation
of qubit 1 to some qubit 2 with known Bell relation to Bob’s qubit
3 in order to enable Bob to transform qubit 3 into the unknown state
of qubit 1. If qubits 1 and 2 are accessible to Alice and far apart from
Bob, then Alice can access this information without influencing Bob’s
qubit (thanks to locality) , although disturbing qubits 1 and 2 in an
uncontrollable way.
Strictly speaking, of course, this picture is inconsistent:
Two qubits may be in a factorized 2-qubit state with factors meeting
none of the Bell relations. Nevertheless we claim that one of the Bell
relations is an element of reality which we find by measuring w.r.t. to
the Bell basis.
This inconsistency, however, is typical for our talking about quantum systems:
Even when we know the state Φ of a system (since its preparation is well
specified), we may ask for the probability |hΨ | Φi|2 to find it in another
state Ψ .
Concerning the Bell relation the situation is even less disturbing:
The 1-qubit states give no information about the actual relation between
the partners of the individual pairs. Selection into subensembles corre-
sponding to the 4 Bell relations has to be expected to change the partial
1-qubit states – even from a classical point of view.

In order to test for the Bell relations it seems necessary to get the qubits into
contact1 — not necessarily into interaction (Resch et al., 2002; Hofmann and Takeuchi, 2002).
This way they loose their identity – a natural reason for the change of the total (in-
ternal) state through measurement.

It seems that the Bell relations may be taken as elements of reality, but they
can be applied to only one (freely chosen) set of 1-particle ‘properties’.2
DRAFT, October 17, 2007
1
See (Lloyd, 2000), however.
2
More generally, see (Griffiths, 2002).
A.3. QUANTUM PHASE ESTIMATION AND ORDER FINDING 167

A.3 Quantum Phase Estimation and Order Find-

ing
Exercise 25 Show that the following quantum network acts as indicated if Ψ̆ϕ is
a n-qubit eigenstate of the unitary Operator Û with eigenvalue ei ϕ :

m

|0i H ··· s √1 |0i + ei 2 ϕ
|1i
2
.. .. .. .. ..
. . . . .  
1
|0i H s ··· √1 |0i + ei 2 ϕ
|1i
2  
0
|0i H s ··· √1 |0i + ei 2 ϕ
|1i
2
 ··· 
 
Ψ̆ϕ .. 0 .. 1 .. m .. Ψ̆ϕ .
 . Û 2 . Û 2 . Û 2 . 
···

First, let us consider the case

I(b)
ϕ = 2π for some b ∈ {0, 1}m+1 . (A.1)
2m+1
Since  
m+1
X
i 2(m+1)−ν ϕ (m+1)−ν−µ 
(A.1) =⇒ e = exp i 2π bµ 2 ,
µ=(m+1)−ν+1

(2.22) tells us that

m+1
!
O 
−1 − m+1 i 2(m+1)−ν ϕ
(A.1) =⇒ |bi = F̂ 2 2 |0i + e |1i . (A.2)
ν=1

Therefore, the phase ϕ considered in Exercise 25 can be determined by applying the

inverse quantum Fourier transform to the state
m+1
O 
def m+1 (m+1)−ν ϕ
Φ̆ϕ = 2− 2 |0i + ei 2 |1i
ν=1
2m+1
(A.3)
X−1
− m+1
= 2 2 ei ϕ j |jim+1
j=0

of the first m + 1 qubits of the output produced by the described network and
measuring the result — if (A.1) holds exactly and everything works perfectly.
168 APPENDIX A.

If ϕ ∈ [0, 2π) is not of the form (A.1), we have

2m+1
X−1
1 2π
F̂ −1
Φ̆ϕ = m+1
ei ϕ j e−i k 2m+1 j |kim+1
(A.3),(2.18) 2 j,k=0

1 X−1 
2m+1
i2π ( 2π
ϕ
j
= e − k
2m+1
) |kim+1
2m+1 j,k=0

1 − ei2π(2 2π −k)
m+1 ϕ
2m+1
X−1
1
= |kim+1
1 − ei2π( 2π − 2m+1 )
ϕ
2m+1 k=0
k

2m+1
X−1
m+1
1 1 − ei 2 ϕ
= |kim+1 . (A.4)
1 − ei2π( 2π − 2m+1 )
ϕ
2m+1 k=0
k

Then (A.4) implies that the probability p(k) for finding |kim+1 when testing F̂ −1 Φ̆ϕ
fulfills the inequality
1 −2
1 − ei2π( 2π − 2m+1 )
ϕ k
p(k) ≤ . (A.5)
22m
Let us define  
defϕ
D(k) = min 2m+1− ν − k ∀ k ∈ ZZ .
ν∈ZZ 2π
Then, given d ∈ {3, . . . , 2m − 1} , the probability pd for getting any state |kim+1
with D(k) > d when testing F̂ −1 Φ̆ϕ fulfills the inequality
1
pd ≤ . (A.6)
2 (d − 2)


Proof: Choosing k0 ∈ 0, . . . , 2m+1 − 1 such that
def ϕ
∆ = 2m+1 − k0 ∈ (0, 1)
2π
we get
X
pd = p(k)
k∈{0,...,2m+1 −1}
D(k)>d

1 X ∆ k−k0 −2
≤ 1 − ei2π 2m+1 e−i2π 2m+1
(A.5) 22m
k∈{0,...,2m+1 −1}
minν∈ZZ |k0 −k−ν 2m+1 |≥d

1 X ∆ j −2
= 1 − ei2π 2m+1 e−i2π 2m+1
22m
j∈{−2m +1,...,2m }
minν∈ZZ |j−ν 2m+1 |≥d

1 X ∆ j −2
≤ 1 − ei2π 2m+1 e−i2π 2m+1 .
22m
j∈{−2m +1,...,2m }
j ∈{−d+1,...,d−1}
/
A.3. QUANTUM PHASE ESTIMATION AND ORDER FINDING 169

By (A.6), therefore3

X −2
1 2 ∆−j
pd ≤ 2π m+1
22m π 2
j∈{−2m +1,...,2m }
j ∈{−d+1,...,d−1}
/
 
−d 2m
1 X −2
X −2
= (∆ − j) + (∆ − j) 
4 j=−2m +1
j=d
 m

−d 2
1 X X −2
≤ j −2 + (1 − j) 
4 j=−2m +1
j=d
m
2X −1
1
≤ j −2
2
j=d−1
Z ∞
1 dx
≤
2 d−2 x2
1
= .
2 (d − 2)

(A.6) tells us that, with probability ≥ 1 − (d − 2)−1 /2 , testing F̂ −1 Φ̆ϕ w.r.t. the
computational (m + 1)-qubit base gives a state |bi for which

I(b) d
min ϕ − 2π ν − 2π ≤ 2π .
ν∈ZZ 2m+1 2m+1

Now, let x and N be arbitrarily given coprime positive integers. Then the order
finding problem is to determine
n o
def ′
r = min r′ ∈ IN : xr = 1 mod N ,

the order of x modulo N .

Defining n o
def
L = min l ∈ IN : N ≤ 2l
and
def
[j mod N ] = min {k ∈ ZZ+ : k = j mod N } ∀ j ∈ ZZ ,
DRAFT, October 17, 2007
3
Note that
1 θ |θ| |θ|
1 − eiθ = sin ≥ √ ≥ ∀ θ ∈ [−π, +π] ,
2 2 2 2 π
since  
d θ θ 1 θ 1
sin − √ = cos − √ ≥ 0 ∀ θ ∈ [0, +π] .
dθ 2 2 2 2 2 2 2
170 APPENDIX A.

we have4 ( )
h iE
j
r= x mod N : j ∈ IN
L

and Dh i h iE
xj mod N xk mod N = δjk ∀ j, k ∈ {0, 1, . . . , r − 1} .
L
Therefore the states
r−1 h iE
def 1 X −i 2π s j
Ψs = √ e r xj mod N ∀ s ∈ {0, . . . , r − 1} (A.7)
r j=0 L

are normalized and, thanks to (2.17), fulfill the equation

1 r−1
X s
h iE
√ e+i 2π r j Ψs = xj mod N ∀ j ∈ {0, . . . , r − 1} .
r s=0 L

Especially for j = 0 the latter gives

1 r−1
X
|1iL = √ Ψs . (A.8)
r s=0

We do not yet know the states Ψs explicitly since we do not yet know r . But we
know that these states exist and have the nice property that they are eigenstates of
the unitary5 L-qubit operator Û characterized by
 n o
def |[x y mod N ]iL if y ∈ {0, . . . , N − 1}
Û |yiL = ∀y ∈ 0, . . . , 2L − 1 .
|yi else
More precisely, we have
s
Û Ψs = ei 2π r Ψs ∀ s ∈ {0, . . . , r − 1} .

Therefore, replacing n by L and Ψ̆ϕ by |1iL , in Exercise 25 we get the total output
state6
1 r−1
X
√ Φ̆2π rs ⊗ Ψs .
r s=0
Since the Ψs form an orthonormal subset of the L-qubit state space, we may assume
that the partial state of the system of the first m + 1 qubits is one of the vector
states Φ̆2π rs with equal probability and the results concerning phase estimation show:

DRAFT, October 17, 2007

4
Recall (1.12).
5
Note that, because gcd(x, N ) = 1 ,

y1 6= y2 mod N =⇒ x y1 6= x y2 mod N .

6
Recall (A.3).
A.4. FINITE-DIMENSIONAL QUANTUM KINEMATICS 171

When F̂ −1 Φ̆2π rs is tested w.r.t. the computational (m + 1)-qubit basis

the probability for getting a state |kim+1 with
s k d
− m+1 ≤ m+1 ,
r 2 2
1
for d ≥ 3 , is not less than 1 − .
2 (d − 2)
Choosing d and 2m+1 /d sufficiently large we obtain, this way, an excellent approxi-
mation k/2m+1 of s/r for some random s ∈ {0, . . . , r − 1} .

A.4 Finite-Dimensional Quantum Kinematics

A.4.1 General Description
• The state space of a finite-dimensional quantum system is a finite-dimensional
complex Euclidean space H the inner product of which we denote by h. | .i .
• The vectors ψ ∈ H with norm 1 correspond to pure states.7
• In the interaction picture, used here, the state vectors ψ do not depend on
time as long as the states are not disturbed by additional interaction (e.g. with
some ‘measurement’ apparatus).
• The state vectors label equivalence classes of preparation procedures for en-
sembles of individual systems of the considered type.
• Preparation procedures are called equivalent if the ensembles they provide
cannot be distinguished by the statistical outcome of measurements.
• Important measurements performable (in principle) on individual elements of
an ensemble are projective measurements:
The individual drawn from an ensemble with state vector
n
X
ψ= hφν | ψi φν
ν=1

will be forced to a transition (if necessary) into one of the φν -

ensembles.8 According to the rules of quantum theory we have9

|hφν | ψi|2 = probability for the transition ψ 7→ φν (A.9)

for all ν ∈ {1, . . . , n} .
DRAFT, October 17, 2007
7
We assume that there are no superselection rules.
8
I.e., an ensemble formed by a (sufficiently ) large number of individuals for which φν is ‘mea-
sured’ actually corresponds (sufficiently well) to φν , unless additional perturbations have appeared.
9
Therefore, hφn u | ψi is called probability amplitude for the transition ψ 7→ φν .
172 APPENDIX A.

• Observables are operators  ∈ L(H) of the form

n
X
 = a |φν ihφν | ,
ν
|{z}
{φ1 , . . . , φn } an orthonormal basis of H , (A.10)
ν=1
∈IR

with the following interpretation:

In a state with state vector φν the physical entity A (corresponding

to Â) has the definite value aν .

This together with (A.9) implies:10

(
  expectation value for A
trace |ψihψ| Â = (A.11)
in a state with state vector ψ .

• Individuals which are only known to be members of an ensemble with state

N
X
vector ψj with probability λj for j ∈ {1, . . . , N } , λj = 1 , form an ensemble
j=1
to be described by the density matrix 11
N
X
ρ̂ = λj ψj , trace (ρ̂) = 1 , (A.12)
j=1
|{z} |{z}
≥0 normalized

in the sense that

trace (ρ̂ Â) = expectation value for  . (A.13)

• In this sense, the set of all states corresponds to

n o
def
S(H) = Â ∈ L(H) : Â ≥ 0 , trace (Â) = 1 .

States12 with ρ̂2 6= ρ̂ are called mixed.

DRAFT, October 17, 2007

10
Note that
  Xn D E
2
trace |ψihψ| Â = aν |hφν | ψi| = ψ Â ψ .
ν=1

11
Such ensembles arise, e.g., from projective measurements on pure states if the individuals are
not selected according to the ‘measurement’ results.
12
From now on we identify states with their density matrices. Note that

ρ̂ = |ψihψ| for some ψ ∈ H ⇐⇒ ρ̂2 = ρ̂ .

A.4. FINITE-DIMENSIONAL QUANTUM KINEMATICS 173

Lemma A.4.1 Let ρ̂1 , ρ̂2 ∈ S(H) . Then

0 ≤ trace (ρ̂1 ρ̂2 ) ≤ 1
and
trace (ρ̂1 ρ̂2 ) = 1 ⇐⇒ ∃ ψ ∈ H : ρ̂1 = ρ̂2 = |ψihψ| .

Outline of proof: Thanks to the spectral theorem there are an orthonormal basis
{φ1 , . . . , φn } of H and λ1 , . . . , λn ≥ 0 with
n
X
ρ̂1 = λν |φν ihφν |
ν=1

and hence
n
X
trace (ρ̂1 ρ̂2 ) = λν hφµ | φν ihφν | ρ̂2 φµ i
ν,µ=1
Xn
= λν hφν | ρ̂2 φν i .
ν=1
Therefore, 0 ≤ trace (ρ̂1 ρ̂2 ) ≤ 1 and
 
trace (ρ̂1 ρ̂2 ) = 1 ⇐⇒ 0 < λν < 1 =⇒ hφν | ρ̂2 φν i = 1 ∀ µ ∈ {1, . . . , n}
⇐⇒ ∃ ν0 ∈ {1, . . . , n} : ρ̂1 = ρ̂2 = |φν0 ihφν0 | .

Lemma A.4.2 Let13 ψ1 , . . . , ψN ∈ H . Then

N
! N
!
X [
|ψk ihψk | H = span {ψk } . (A.14)
k=1 k=1

Outline of proof: Thanks to the spectral theorem there is an ONS {φ1 , . . . , φn′ } ⊂
H with
XN n′
X
|ψk ihψk | = λν |φν ihφν | (A.15)
k=1 ν=1
for suitable λ1 , . . . , λn′ > 0 and, consequently,
N
! N
!
X [
|ψk ihψk | H = span {φk } . (A.16)
k=1 k=1

Then
N
X N
X
2
|hχ | ψk i| = hχ | ψk ihψk | χi
k=1 k=1
Xn′
= λν hχ | φν ihφν | χi
(A.15) ν=1
Xn′
2
= λν |hχ | φν i| ∀χ ∈ H
ν=1

DRAFT, October 17, 2007

13
If not stated otherwise, by H , H1 , , H2 we denote arbitrarily given finite-dimensional complex
Euclidean vector spaces.
174 APPENDIX A.

and, consequently,

χ ⊥ ψk ∀ k ∈ {1, . . . , N } ⇐⇒ χ ⊥ φν ∀ ν ∈ {1, . . . , n} ,

i.e. !  
′
N
[ n
[
span {ψk } = span  {φν } .
k=1 ν=1

The latter together with (A.16) implies (A.14).

Corollary A.4.3 Let ψ1 , . . . , ψN , ψ1′ , . . . , ψN

′
∈ H . Then
N
X N
X
|ψk ihψk | = |ψk′ ihψk′ | . (A.17)
k=1 k=1
 
holds iff there is a unitary N × N -matrix Uj k with14

N
X
ψk = Uk j ψj′ ∀ k ∈ {1, . . . , N } . (A.18)
j=1

Proof: Assume that (A.17) holds and, as in the proof of Lemma A.4.2, let us choose
an orthonormal basis {φ1 , . . . , φn } ⊂ H and λ1 , . . . , λn′ > 0 for which (A.15) holds.
Then Lemma A.4.2 implies
!  ′ 
[N [n
span {ψk } = span  {φν } .
k=1 ν=1

Especially, the ψk ’s can be written as linear combinations

′
n
X p
ψk = ckν +
λ ν φν
ν=1

of φν ’s. Then we have

′
n
X n′ 
N X
X ∗
λν |φν ihφν | = ckµ ckν λν |φν ihφµ | .
ν=1 (A.15)
k=1 ν,µ=1

Since the |φν ihφµ | form an ONS w.r.t. the Hilbert-Schmidt scalar product
D E
def
 B̂ = trace († B̂) ∀ Â, B̂ ∈ S(H) , (A.19)

this implies
N
X
∗
ckν ckµ = δνµ ∀ µ, ν ∈ {1, . . . , n′ } ,
k=1

DRAFT, October 17, 2007

 
14
In general, of course, the Uj k are not fixed by (A.18)
A.4. FINITE-DIMENSIONAL QUANTUM KINEMATICS 175

i.e. the  
c1ν
 
cν =  ...  , ν ∈ {1, . . . , n′ } ,
cN ν
form an orthonormal system in CN .
Extending this to an orthonormal basis of CN
we get a unitary N × N -matrix ckν with

N
X p
ψk = ckν +
λ ν φν ∀ k ∈ {1, . . . , N } , (A.20)
ν=1

where
def
λν = 0 for ν > n′ .
ν

Similarly, we get a unitary N × N -matrix c′k with

N
X p
ν
ψk′ = c′k +
λ ν φν ∀ k ∈ {1, . . . , N }
ν=1

and hence
p N
X ν
∗
+
λ ν φν = c′l ψl′ ∀ ν ∈ {1, . . . , N } . (A.21)
l=1

Combining (A.20) with (A.21) we get (A.18) for the unitary matrix with components
N
X
def ν
∗
Uk l = ckν c′l ∀ k, l ∈ {1, . . . , N } .
ν=1

Conversely, (A.17) follows from (A.18) by straightforward calculation.

A.4.2 Qubits
Qubits are 2-dimensional quantum systems for which some orthonormal computa-
tional basis {|0i , |1i} of their state space H is chosen. According to the standard
convention     
α 
ψ= w.r.t. |0i , |1i 
β ∀ α, β ∈ C
def 

⇐⇒ ψ = α |0i + β |1i
we have    
1 0
|0i = , |1i = .
0 1
 
A11 A12
All  ∈ L(H) may be identified with their matrix w.r.t. the computa-
A21 A22
tional basis:
      
α A11 A12 α
ψ= =⇒ Â ψ = w.r.t. |0i , |1i .
β A21 A22 β
176 APPENDIX A.
n √ √ o
Here an orthonormal basis w.r.t. the scalar product (A.19) is τ̂ 0 / 2, . . . , τ̂ 3 / 2 ,
where
       
0 1 0 1 0 1 2 0 −i 3 1 0
τ̂ = τ̂ = τ̂ = τ̂ = (A.22)
0 1 1 0 i 0 0 −1
are the well-known Pauli matrices. Therefore we have
3
1X
 = trace (Âτ̂ ν ) τ̂ ν ∀  ∈ L(H) ,
2 ν=0
especially
1 
ρ̂ = 1̂ + ρ · τ̂ ∀ ρ̂ ∈ S(H) ,
2
where the vector15
def
ρ = trace (ρ̂ τ̂ )
fulfills16
|ρ| ≤ 1 , ρ̂2 = ρ̂ ⇐⇒ |ρ| = 1 .
Note that the components of ρ are just expectation values of observables which are
sufficient for quantum state tomography.

Remark: Since
(eϑ,ϕ · τ̂ ) χϑ,ϕ = χϑ,ϕ
holds for    
sin ϑ cos ϕ ϑ −i ϕ

def def 
e cos  2

eϑ,ϕ =  sin ϑ sin ϕ  ,χϑ,ϕ =  1

ϕ
+i 2 ϑ,
cos ϑ e sin
1
every pure state corresponds to a definite spin orientation in the spin- 12 case — where
h̄
2 τ̂ is the spin vector observable .

A.4.3 Bipartite Systems

n o
(j)
For j = 1, 2 let φ1 , . . . , φ(j)
nj be an orthonormal basis of the state space Hj
of a quantum mechanical system Sj . If S1 and S2 are distinguishable then the
bipartite system S composed of these two has the state space H = H1 ⊗ H2 with
orthonormal basis
n o
φ(1) (2)
ν ⊗ φµ : ν ∈ {1, . . . , n1 } , µ ∈ {1, . . . , n2 } .

DRAFT, October 17, 2007

15
Usually, the vector of coherence ρ is called Bloch vector when associated with electron spin
and Stokes vector when associated with photon polarization. More generally see (Altafini, 2003).
16
Note that
1 2

1 = trace (ρ̂) ≥ trace (ρ̂2 ) = 1 + |ρ| .
2
A.4. FINITE-DIMENSIONAL QUANTUM KINEMATICS 177

Extending ⊗ to a bilinear mapping of H1 × H2 into H1 ⊗ H2 we get

 
L(H1 ⊗ H2 ) = span Â1 ⊗ Â2 : Â1 ∈ L(H1 ) , Â2 ∈ L(H2 ) ,

where the linear operators Â1 ⊗ Â2 are fixed by

        
def
Â1 ⊗ Â2 ∀ ψ (1) , ψ (2) ∈ H1 ⊗ H2 .
ψ (1) ⊗ ψ (2) = Â1 ψ (1) ⊗ Â2 ψ (2)
(A.23)
If Â1 resp. Â2 is the observable of S1 resp. S1 corresponding to the physical entity A1
resp A2 then Â1 ⊗ Â2 is the observable corresponding to the physical entity A1 A2 .

If one is only interested in the subsystem S1 of the total system S in state ρ̂ then
it is sufficient to know its partial state 17
n2 D
X E
def def
ρ̂1 = trace 2 (ρ̂) = φ(2) (2)
µ ρ̂ φµ , (A.24)
µ=1

since:   
trace ρ̂ Â1 ⊗ 1̂ = trace (ρ̂1 Â1 ) ∀ Â1 ∈ L(H1 ) .
Note that
i.g.
ρ̂ pure 6=⇒ ρ̂1 pure .

Example: If S1 and S2 are qubit systems then for the the Bell state
       
def 1 1 0 0 1
Ψ− = √ ⊗ − ⊗
2 0 1 1 0
we have  ED 
1
trace 2 Ψ− 1̂ , Ψ− =
2
i.e. the partial states give no information at all:
 ED   
1
trace Ψ− Ψ− |ψihψ| ⊗ 1̂ = kψk2 ∀ ψ ∈ C2 .
2
But there are strong (non-classical) correlations between the subsystems,
since
1
Ψ− = √ (ψ ⊗ ψ⊥ − ψ⊥ ⊗ ψ)
2
DRAFT, October 17, 2007
17
The so-called partial trace trace 2 w.r.t. the second factor is the linear mapping of L(H1 ⊗H2 )
into L(H1 ) characterized by
 
trace 2 |ψ1 ⊗ ψ2 ihψ1′ ⊗ ψ2′ | = hψ2′ | ψ2 i |ψ1 ihψ1′ | ∀ ψ1 , ψ1′ ∈ H1 , ψ2 , ψ2′ ∈ H2 .
| {z }
 
=trace |ψ2 ihψ2′ |

The partial trace w.r.t. the first factor, written trace 1 , is defined similarly.
178 APPENDIX A.

for all normalized ψ ∈ C2 , where

   
α def −β ∗
= for all α, β ∈ C ,
β ⊥ α∗

and hence
 ED     1
trace Ψ −
Ψ −
|ψihψ| ⊗ |φihφ| = |hφ | ψ⊥ i|2
2
holds for all normalized ψ, φ ∈ C2 .

Definition A.4.4 A state ρ̂ of the bipartite system S with state space H1 ⊗ H2 is

called separable iff there is a sequence {ρ̂N }N ∈IN of states of the form
N
X
ρ̂N = λ
ν
|{z} ν ρ̂(1) ⊗ ρ̂(2)
ν (A.25)
ν=1
|{z} |{z}
≥0 ∈S(H1 ) ∈S(H2 )

with 18
lim kρ̂ − ρ̂N k1 = 0 .
N →∞

Theorem A.4.5 (Horodecki) Let S be a bipartite system with state space H1 ⊗

H2 . Then the separable states of S are exactly those of the form (A.25) with

N ≤ (dim(H1 ⊗ H2 ))2 .

Proof: See (Horodecki, 1997).

Lemma A.4.6 Let Ψ be a normalized vector in H1 ⊗ H2 . Then ρ̂ = |ΨihΨ| is

separable iff
Ψ = φ(1) ⊗ φ(2) (A.26)
holds for some normed φ(1) ∈ H1 and φ(2) ∈ H2 .

Outline of proof: Let ρ̂ = ρ̂2 be separable, hence of the form

N
X N
X
ρ̂ = λν ρ̂(1)
ν ⊗ ρ̂(2)
ν , λν = 1 .
|{z} |{z} |{z}
ν=1 >0 ν=1
∈S(H) ∈S(H)

DRAFT, October 17, 2007

18
Recall (5.31) and Footnote 22 of Chapter 5.
A.4. FINITE-DIMENSIONAL QUANTUM KINEMATICS 179

Then, by Lemma A.4.1,

1 = trace (ρ̂2 )
XN
= λν λµ trace H1 (ρ̂(1) (1) (2) (2)
ν ρ̂µ ) trace H2 (ρ̂ν ρ̂µ )
ν,µ=1
| {z } | {z }
∈[0,1] ∈[0,1]

and, therefore,
trace H1 (ρ̂(1) (1) (2) (2)
ν ρ̂µ ) = trace H2 (ρ̂ν ρ̂µ ) = 1 ∀ ν, µ ∈ {1, . . . , N } .

By by Lemma A.4.1, again, the latter is equivalent to the existence of normed φ(1) ∈
H1 and φ(2) ∈ H2 with
ED
ρ̂(j)
ν = φ
(j)
φ(j) ∀ ν ∈ {1, . . . , N } , j ∈ {1, 2}

and hence (A.26).

Conversely, (A.26) together with
ED
def
ρ̂(j) = φ(j) φ(j) ∀ j ∈ {1, 2}

implies
|ΨihΨ| = ρ̂(1) ⊗ ρ̂(2)
and hence separability of |ΨihΨ| .

Theorem A.4.7 Every vector state on H1 ⊗ H2 allows a Schmidt decomposi-

tion,19 i.e. for every Ψ ∈n H1 ⊗ H2 there o are a
n unique Schmidt o number n′ ∈ IN
(1) (1) (2) (2)
and orthonormal subsets φ1 , . . . , φn′ resp. φ1 , . . . , φn′ of H1 resp. H2 with
n ′
X
Ψ= sν
|{z}
φ(1) (2)
ν ⊗ φν (A.27)
ν=1 >0

for suitable Schmidt

 coefficients s1 , . . . , sn′ . If the eigenvalues of the partial
state trace 2 |ΨihΨ| are non-degenerated then the Schmidt decomposition of Ψ is
unique.

Outline of proof: Consider anyn Ψ ∈ H1 ⊗ H o 2 . Thanks to the spectral theorem

(1) (1)
there are an orthonormal basis φ1 , . . . , φn1 of H1 , a positive integer n′ ≤ n1 ,
and s1 , . . . , sn′ > 0 with
  Xn1 ED
2 def
trace 2 |ΨihΨ| = (sν ) φ(1)
ν φ(1)
ν , sν = 0 ∀ ν > n′ . (A.28)
ν=1

Then there are ψ1 , . . . , ψn1 ∈ H2 with

n1
X
Ψ= φ(1)
ν ⊗ ψν (A.29)
ν=1

DRAFT, October 17, 2007

19
For the generalization to n-partite systems see (Carteret et al., 2000)
180 APPENDIX A.

and, therefore,
  n1
X ED
trace 2 |ΨihΨ| = hψµ | ψν i φ(1)
ν φ(1)
µ .
ν,µ=1

The latter together with (A.28) implies

2
hψµ | ψν i = (sν ) δνµ ∀ ν, µ ∈ {1, . . . , n1 } .

Thus, with
ψν
def
φ(2)
ν = ∀ ν ∈ {1, . . . , n′ } ,
sν
(A.29) becomes equivalent to (A.27). Since, conversely, (A.27) implies (A.28) the
stated uniqueness properties are obvious.

Remarks:
1. Note that the Bell states (4.55) have Schmidt number 2 and
hence, by Lemma A.4.6, be separable.
2. The vector state
1 
|0, 0i + |0, 1i + |1, 0i + |1, 1i ,
2
   
however, is separable since equal to Ĥ |0i ⊗ Ĥ |0i , where
   
def 1 def 1
Ĥ |0i = √ |0i + |1i , Ĥ |1i = √ |0i − |1i
2 2
characterizes the unitary Hadamard operator, strongly used in
quantum computing.
3. For mixed states ρ̂ ∈ S(H1 ⊗ H2 ) there is a Schmidt-like decom-
position of the Form
n ′
X
ρ̂ = s ν
|{z}
Â(1) (2)
ν ⊗ Âν
ν=1
>0

with n′ ≤ (dim H1 )2 , (dim H2 )2 . According to (Herbut, 2002, Corol-

lary 1) the operators Â(j)
ν may be chosen Hermitian and such that
 
ν 6= µ =⇒ trace Â(j) (j)
ν µ = 0.

However, in general, they cannot be positive.

Pure states are non-separable iff their partial states are mixed.20 Therefore, the
correlations in non-separable pure states are non-classical.21 Obviously, the partial
DRAFT, October 17, 2007
20
Usually, non-separable states are called entangled; see (Verstraete and Cirac, 2003), however.
21
In a way, also the mixed non-separable states are non-classically correlated (Werner, 1989).
A.4. FINITE-DIMENSIONAL QUANTUM KINEMATICS 181

transpose
 
  n1
X n2
X ED
1 ⊗ T̄  λν1 µ1 ν2 µ2 φ(1) (2)
ν1 ⊗ φν2 φ(1) (2) 
µ1 ⊗ φµ2
ν1 ,µ1 =1 ν2 ,µ2 =1
n2
X ED
= λν1 µ1 ν2 µ2 φ(1) (2)
ν1 ⊗ φµ2 φ(1) (2)
µ1 ⊗ φν2
(4.13) ν2 ,µ2 =1

n o
(2)
w.r.t. the orthonormal basis φ1 , . . . , φ(2)
n2 of H2 must be positive for separable
states: !
N T2 N
X X
(1 ⊗ T) λν ρ̂(1)
ν ⊗ ρ̂(2)
ν = λν ρ̂(1) (2)
ν ⊗ T(ρ̂ν ) ≥ 0 .
ν=1 ν=1

Therefore, the mixed Werner states Ŵλ with λ > 1/3 , considered at the end of
Section 4.2.1, are non-separable.22

Lemma A.4.8 For every state ρ̂ ∈ H there is a purification in H ⊗ H , i.e. a

normalized vector Ψ ∈ H ⊗ H with
 
ρ̂ = trace 2 |ΨihΨ| . (A.30)

Outline
n of proof:
o Thanks to the spectral theorem there are an orthonormal basis
(1) (1)
φ1 , . . . , φn of H and λ1 , . . . , λn ≥ 0 with

n
X ED n
X
ρ̂ = λν φ(1)
ν φ(1)
ν , λν = 1 . (A.31)
ν=1 ν=1

For the normalized vector23

n p
X
def
Ψ = +
λν φ(1) (1)
ν ⊗ φν
ν=1

then, we get (A.30).

DRAFT, October 17, 2007

22
A decomposition of the 2-qubit Werner states, taking the form (A.25) for λ ≤ 1/3 , is
presented on page 6 of: http://www.physik.uni-augsburg.de/weh-school/bruss.pdf
23
For the set of all suitable Ψ see (Kuah and Sudarshan, 2003, Lemma 1).
182 APPENDIX A.
Bibliography

Aaronson, S. and Gottesman, D. (2004). Improved simulation of stabilizer circuits.

Phys. Rev. A, 70:052328. quant-ph/0406196. 33

Aho, A. V. and Svore, K. M. (2003). Compiling quantum circuits using the Palin-
drome transform. quant-ph/0311008, (1–17). 31

Alber, G., Beth, T., Horodecki, M., Horodecki, P., Horodecki, R., Rötteler, M.,
Weinfurter, H., Werner, R., and Zeilinger, A. (2001). Quantum Information,
volume 173 of Springer Tracts in Modern Physics. Springer. An Introduction
to Basic Theoretical Concepts and Experiments. 3

Alicki, R. (2003). Controlled quantum open systems. Lecture Notes in Physics,

622:121–139. quant-ph/0302132. 103

Altafini, C. (2003). Tensor of coherences parameterization of multiqubit density

operators for entanglement characterization. Phys. Rev. A, 69:012311. quant-
ph/0308019. 153, 176

Ambainis, A. (2005). Quantum search algorithms. quant-ph/0504012, pages 1–12.

36

Aniello, P., Man’ko, V., Marmo, G., A., Porzio, , and Zaccaria, S. S. F. (2003).
Trapped ions interacting with laser fields: a perturbative analysis without ro-
tating wave approximation. quant-ph/0301138, pages 1–26. 82

Audretsch, J., editor (2002). Verschränkte Welt. WILEY-VCH, Weinheim. Faszi-

nation der Quanten. 3

Baladi, V. and Vallee, B. (2003). Euclidean algorithms are Gaussian. mp arc/03-

474, pages 1–45. 46

Barenco, A., Bennet, C., Cleve, R., DiVincenzo, D., Margolus, N., Shor, P., Sleator,
T., Smolin, J., and Weinfurter, H. (1995). Elementary gates for quantum com-
putation. Phys. Rev. A, 52:3457–3467. quant-ph/9503016. 29

Bartlett, S. D., Diamanti, E., Sanders, B. C., and Yamamoto, Y. (2002). Photon
counting schemes and performance of non-deterministic nonlinear gates in linear
optics. quant-ph/0204073, pages 1–9. 19, 55

183
184 BIBLIOGRAPHY

Bartlett, S. D. and Wiseman, H. M. (2003). Entanglement in the presence of super-

selection rules. Phys. Rev. Lett., 91:097903. quant-ph/0303140. 91, 163

Bennet, C. H., Brassard, G., Crépeau, C., Jozsa, R., Peres, A., and Wootters, W. K.
(1993). Teleporting an unknown quantum state via dual classical and Einstein-
Podolsky-Rosen channels. Phys. Rev. Lett., 70:1895–1899. 26

Berry, D. W. and Sanders, B. C. (2003). Bounds on general entropy measures. J.

Phys. A, 36:12255. quant-ph/0305059. 131

Bertlmann, R. A. and Zeilinger, A., editors (2002). Quantum [Un]speakables.

Springer. From Bell to Quantum Information. 3

Bettelli, S., Serafini, L., and Calarco, T. (2001). Toward an architecture for quantum
programming. cs.PL/0103009, pages 1–23.
(http://sra.itc.it/people/serafini/qlang/). 19

Bowmeester, D., Ekert, A., and Zeilinger, A., editors (2000). The Physics of Quan-
tum Information. Springer-Verlag. 3, 26, 124, 161

Boyer, M., Brassard, G., Høyer, P., and Tapp, A. (1998). Tight bounds on quantum
searching. Fortschr. Phys., 46(493–506). quant-ph/9605034. 35, 36

Brezinski, C. (1991). History of Continued Fractions and Padé Approximants, vol-

ume 12 of Springer Series in Computational Mathematics. Springer-Verlag.
46

Brigham, E. O. (1974). The Fast Fourier Transform. Prentice-Hall, Inc., Englewood

Cliffs, NJ. 43

Browne, D. E. and Rudolph, T. (2004). Resource-efficient linear optical quantum

computation. quant-ph/0405157, pages 1–5. 68

Brukner, C., Pan, J.-W., Simon, C., Weihs, G., and Zeilinger, A. (2003). Proba-
bilistic instantaneous quantum computation. Phys. Rev. A, 67:034304. quant-
ph/0109022. 19, 61

Brukner, C., Zukowski, M., and Zeilinger, A. (2001). The essence of entanglement.
quant-ph/0106119, pages 1–10. 24

Bruß, D. (2003). Quanteninformation. Fischer Taschenbuch Verlag. 3

Bub, J. (2001). Maxwell’s demon and the thermodynamics of computation. Studies

in History and Philosophy of Modern Physics, 32:569–579. quant-ph/0203017.
15

Buscemi, F., D’Ariano, G. M., and Sacchi, M. F. (2003). Physical realizations of

quantum operations. Phys. Rev. A, 68:042113. quant-ph/0305180. 94
BIBLIOGRAPHY 185

Bužek, V. and Šašura, M. (2002). Cold trapped ions as quantum information pro-
cessors. J. Mod. Opt. quant-ph/0112041. 71

Carteret, H. A., Higuchi, A., and Sudbery, A. (2000). Multipartite generalisa-

tion of the Schmidt decomposition. J. Mathem. Phys., 41:7932–7939. quant-
ph/0006125. 179

Cerf, N. J., Adami, C., and Kwiat, P. G. (1998). Optical simulation of quantum
logic. Phys. Rev. A, 57:R1477–R1480. quant-ph/9706022. 56

Chefles, A., Jozsa, R., and Winter, A. (2003). On the existence of physical trans-
formations between sets of quantum states. quant-ph/0307227, pages 1–8. 142

Chen, J.-L., Fu, L., Ungar, A. A., and Zhao, X.-G. (2002). Geometric observation
for the Bures fidelity between two states of a qubit. Phys. Rev. A, 65:024303.
quant-ph/0112169. 140

Chen, Q., Cheng, J., Wang, K.-L., and Du, J. (2005). Efficient construction of 2-D
cluster states with probabilistic quantum gates. quant-ph/0507066, pages 1–4.
70

Cheng, K.-W. and Tseng, C.-C. (2002). Quantum plain and carry look-ahead adders.
quant-ph/0206028, pages 1–16. 16

Childs, A. M., Leung, D. W., and Nielsen, M. A. (2005). Unified derivations of

measurement-based schemes for quantum computation. Phys. Rev. A, page
032318. quant-ph/0404132. 68

Christandl, M. and Winter, A. (2004). “Squashed entanglement”- An additive en-

tanglement measure. J. Mathem. Phys., 45:829–840. quant-ph/0308088. 163

Cleve, R., Ekert, A., Macchiavello, C., and Mosca, M. (1998). Quantum algorithms
revisited. Proc. R. Soc. Lond. A, 454:339–354. quant-ph/9708016. 23

Collins, D. and Gisin, N. (2003). A relevant two qubit Bell inequality inequivalent
to the CHSH inequality. quant-ph/0306129, pages 1–3. 147

Coppersmith, D. (1994). An approximate Fourier transform useful in quantum

factoring. quant-ph/0201067, pages 1–8. IBM Research RC 19642. 44

Cover, T. M. and Thomas, J. A. (1991). Elements of Information Theory. Wiley

Series in Telecommunications. Joh Wiley & Sons, Inc.

D’Alessandro, D. (2001). Optimal evaluation of generalized Euler angles with ap-

plications to classical and quantum control. quant-ph/0110120, pages 1–13.
29

Davies, E. B. (1976). Quantum Theory of Open Systems. Academic Press. 95, 104
186 BIBLIOGRAPHY

Deutsch, D. and Ekert, A. (1998). Quantum computation. Physics World, 11:47–52.

36
Devetak, I., Harrow, A. W., and Winter, A. (2004). A family of quantum protocols.
Phys. Rev. Lett., 93:230504. quant-ph/0308044. 120
Devetak, I. and Winter, A. (2003). Distilling common randomness from bipartite
quantum states. quant-ph/0304196, pages 1–22. 120
Devetak, I. and Winter, A. (2005). Distillation of secret key and entanglement from
quantum states. Proc. R. Soc. Lond. A, 461:207–235. quant-ph/0306078. 161
Dı́az-Caro, A. (2005). On the teleportation of n-qubit states. quant-ph/0505009,
pages 1–5. 26
Dieks, D. (1982). Communication by EPR devices. Physics Letters A, pages 271–
272. 91
Diţă, P. (2001). Factorization of unitary matrices. math-ph/0103005, pages 1–12.
28
DiVincenzo, D. P. (2000). The physical implementation of quantum computation.
Fortschr. Phys., 48:771–783. quant-ph/9705009. 51
Doherty, A. C., Parrilo, P. A., and Spedalieri, F. M. (2003). A complete family of
separability criteria. Phys. Rev. A, 69:022308. quant-ph/0308032. 153
Dong, D.-Y., Chen, C.-L., Zhang, C.-B., and Chen, Z.-H. (2005). Quantum robot:
Structure, algorithms and applications. quant-ph/0506155, pages 1–19. 36
Dowling, J. P., Franson, J. D., Lee, H., and Milburn, G. J. (2004). Towards linear
optical quantum computers. quant-ph/0402090, pages 1–9. 60
Draper, T. G. (2000). Addition on a quantum computer. quant-ph/0008033, pages
1–8. 16
Eckert, K., Schliemann, J., Bruß, D., and Lewenstein, M. (2002). Quantum correla-
tions in systems of indistinguishable particles. Annals of Physics, 299:88–127.
quant-ph/0203060. 18, 163
Ekert, A., Hayden, P., and Inamori, H. (2000). Basic concepts in quantum compu-
tation. quant-ph/0011013, pages 1–37. Lectures given at les Houches Summer
School on ”Coherent Matter Waves”, July-August 1999. 3
Eldar, Y. C. (2003). Von Neumann measurement is optimal for detecting linearly
independent mixed quantum states. quant-ph/0304077, pages 1–4. 135
Elliott, C., Colvin, A., Pearson, D., Pikal, O., Schlafer, J., and Yeh, H. (2005).
Current status of the DARPA quantum network. quant-ph/0503058, pages 1–
12. 124
BIBLIOGRAPHY 187

Fattal, D., Diamanti, E., Inoue, K., and Yamamoto, Y. (2003). Quantum telepor-
tation with a quantum dot single photon source. quant-ph/0307105, pages 1–4.
66

Feynman, R. P. (1982). Simulating physics with computers. Int. J. Theoret. Phys.,

21:467–488.

Feynman, R. P. (1985). Quantum mechanical computers. Optics News, February:11–

20.

Galindo, A. and Martı́n-Delgado, M. A. (2002). Information and computation: Clas-

sical and quantum aspects. Rev. Mod. Phys., 74:347–423. quant-ph/0112105.

Gardiner, C. W. and Zoller, P. (2000). Quantum Noise, volume 56 of Springer

Series in Synergetics. Springer, 2. edition. A Handbook of Markovian and
Non-Markovian Quantum Stochastic Methods with Applications to Quantum
Optics. 131

Gasparoni, S., Pan, J.-W., Walther, P., Rudolph, T., and Zeilinger, A. (2004). Re-
alization of a photonic CNOT gate sufficient for quantum computation. quant-
ph/0404107, pages 1–4. 71

Gatti, A., Zambrini, R., Miguel, M. S., and Lugiato, L. A. (2003). Multi-photon,
multi-mode polarization entanglement in parametric down-conversion. quant-
ph/0306133, pages 1–22. 121

Ghosh, P. K. (1995). Ion Traps. Clarendon Press, Oxford. 71, 75

Giacomini, S., Sciarrino, F., Lombardi, E., and Martini, F. D. (2002). “Active”
teleportation of a quantum bit. Phys. Rev. A, 66:030302. quant-ph/0204158.
26

Gilchrist, A., Langford, N. K., and Nielsen, M. A. (2004). Distance measures to

compare real and ideal quantum processes. quant-ph/0408063, pages 1–15. 137

Gilchrist, A. and Milburn, G. J. (2002). Conditional phase shifts using trapped

atoms. quant-ph/0208157, pages 1–6. 60

Gingrich, R. M., Ko, P., Lee, H., Vatan, F., and Dowling, J. P. (2003). An all linear
optical quantum memory based on quantum error correction. Phys. Rev. Lett.,
91:217901. quant-ph/0306098. 121

Gisin, N. and Gisin, B. (2002). A local variable model for entanglement swapping
exploiting the detection loophole. quant-ph/0201077, pages 1–5. 26

Gottesman, D. and Chuang, I. L. (1999). Demonstrating the viability of universal

quantum computation using teleportation and single-qubit operations. Nature,
402:390–393. quant-ph/9908010. 61
188 BIBLIOGRAPHY

Griffiths, R. B. (2002). The nature and location of quantum information. Phys.

Rev. A, 66:012311. quant-ph/0203058. 166

Grover, L. K. (1996). A fast quantum mechanical algorithm for database search. In

Proceedings of the 28th Annual ACM Symposium on the Theory of Computing,
Philadelphia (1996), pages 212–219. 34

Grover, L. K. and Radhakrishnan, J. (2004). Is partial quantum search of a database

any easier? quant-ph/0407122, pages 1–9. 36

Gruska, J. (1999). Quantum Computing. Advanced Topics in Computer Science

Series. McGraw-Hill.

Gu, M. and Weedbrook, C. (2005). Quantum passwords. quant-ph/0506255, pages

1–5. 92

Gühne, O., Hyllus, P., Bruß, D., Ekert, A., Lewenstein, M., Macchiavello, C., and
Sanpera, A. (2002). Detection of entanglement with few local measurements.
Phys. Rev. A, 66:062305. quant-ph/020508. 150

Gulde, S., Riebe, M., Lancaster, G. P. T., Becher, C., Eschner, J., Häffner, H.,
Schmidt-Kaler, F., Chuang, I. L., and Blatt, R. (2003). Implementation of the
Deutsch-Jozsa algorithm on an ion-trap quantum computer. Nature, 421:48–50.
23

Gurvits, L. (2002). Quantum matching theory (with new complexity theoretic, com-
binatorial and topological insights on the nature of the quantum entanglement).
quant-ph/0201022, pages 1–10. 147

Ha, K.-C., Kye, S.-H., and Park, Y. S. (2003). Entangled states with positive
partial transpose arising from indecomposable positive linear maps. Phys. Lett.
A, 313:163–174. quant-ph/0305005. 152

Haderka, O., Hamar, M., and Perina, J. (2003). Experimental multi-photon-

resolving detector using a single avalanche photodiode. quant-ph/0302154,
pages 1–11. 55

Hamming, R. W. (1950). Error detecting and error correcting codes. The Bell System
Technical Journal, 26:147–160. http://www.engelschall.com/u/sb/hamming/.
108

Hayden, P., Jozsa, R., Petz, D., and Winter, A. (2004). Structure of states which
satisfy strong subadditivity of quantum entropy with equality. Commun. Math.
Phys., pages 359–374. quant-ph/0304007.

Heiss, D., editor (2001). Fundamentals of Quantum Information. Springer. Quantum

Computation, Communication, Decoherence and All That.
BIBLIOGRAPHY 189

Herbert, N. (1982). FLASH—A superluminal communicator based upon a new kind

of quantum measurement. Foundations of Physics, 12:1171–1181. 92

Herbut, F. (2002). Hermitian Schmidt decomposition and twin observables of bi-

partite mixed states. J. Phys. A, 35:1691–1708. quant-ph/0305181. 180

Hiroshima, T. (2003). Majorization criterion for distillability of a bipartite quantum

state. Phys. Rev. Lett., 91:057902. quant-ph/0303057. 152, 155

Hirvensalo, M. (2001). Quantum Computing. Natural Computing Series. Springer-

Verlag.

Hofmann, H. F. and Takeuchi, S. (2002). Quantum phase gate for photonic qubits
using only beab splitters and post-selection. Phys. Rev. A, 66:024308. quant-
ph/0204045. 60, 166

Horodecki, M. and Horodecki, P. (1999). Reduction criterion of separability and lim-

its for a class of protocols of entanglement distillation. Phys. Rev. A, 59:4206–
4216. quant-ph/9708015. 155

Horodecki, M., Horodecki, P., and Horodecki, R. (1996). Separability of mixed

states: Necessary and sufficient conditions. Phys. Lett. A, 223:1–8. quant-
ph/9605038. 152, 163

Horodecki, M., Horodecki, P., and Horodecki, R. (2001). Mixed-state entanglement

and quantum communication. Springer Tracts in Modern Physics, 173:151–195.
quant-ph/0109124. 147, 154

Horodecki, M., Oppenheim, J., and Winter, A. (2005). Quantum information can
be negative. quant-ph/0505062, pages 1–8. 130

Horodecki, P. (1997). Separability criterion and inseparable mixed states with pos-
itive partial transposition. Phys.Lett. A, 232:333–339. quant-ph/9703004. 178

Horodecki, R. and Horodecki, M. (1996). Information-theoretic aspects of quantum

inseparability of mixed states. Phys. Rev. A, 54:1838–1843. quant-ph/9607007.
152

James, D. F. V. (1998). Quantum dynamics of cold trapped ions with application

to quantum computation. App. Phys. B, 66:181. 77

James, D. F. V. (2000). Quantum computation with hot and cold ions: An as-
sessment of proposed schemes. Fortschritte der Physik, 48:811–821. quant-
ph/0003122. 71

Kahn, D. (1967). The Codebreakers: The Story of Secret Writing. Macmillan, New
York. 37
190 BIBLIOGRAPHY

Kaszlikowski, D., Gopinathan, A., Liang, Y. C., Kwek, L. C., and Englert, B.-G.
(2003). How well can you know the edge of a quantum pyramid? quant-
ph/0307086, pages 1–3. 135

Kaye, P. and Mosca, M. (2004). Quantum networks for generating arbitrary quan-
tum states. quant-ph/0407102, pages 1–3. 68

Keyl, M. and Werner, R. F. (2002). How to correct small quantum errors. Lecture
Notes in Physics, 611:263. quant-ph/0206086. 115

Kim, Y.-H., Kulik, S. P., and Shih, Y. (2001). Quantum teleportation with a com-
plete Bell state measurement. Phys. Rev. Lett., 86:1370. quant-ph/0010046.
62, 121

Klein, O. (1931). Zur Quantenmechanischen Begründung des zweiten Hauptsatzes

der Wärmelehre. Z. Physik, 72:767–775. 132

Knill, E., Laflamme, R., and Milburn, G. (2000). Efficient linear optics quantum
computation. Nature, 409:46. quant-ph/0006088. 63

Knill, E., Laflamme, R., and Milburn, G. J. (2001). A scheme for efficient quantum
computation with linear optics. Nature, 409:46–52. 51, 60, 63, 67

Knill, E., Laflamme, R., and Viola, L. (1999). Theory of quantum error correction
for general noise. quant-ph/9908066, pages 1–6. 106

Korepin, V. E. and Grover, L. K. (2005). Simple algorithm for partial quantum

search. quant-ph/0504157, pages 1–3. 36

Kuah, A. and Sudarshan, E. (2003). Manifold of density matrices. quant-

ph/0307218, pages 1–5. 181

Kuah, A. and Sudarshan, E. C. G. (2005). Extension maps. quant-ph/0503119,

pages 1–6. 95, 97

Kwiat, P. G., Mitchell, J. R., Schwindt, P. D. D., and White, A. G. (2000).

Grover’s search algorithm: An optical approach. J.Mod.Opt., 47:257–266.
quant-ph/9905086. 56

Labuschagne, L. E., Majewski, W. A., and Marciniak, M. (2003). On k-

decomposability of positive maps. math-ph/0306017, pages 1–23. 152

Landauer, R. (1961). Irreversibility and heat generation in the computing process.

IBM J. Res. Dev., 5:183–191. 15

Landauer, R. (1998). Energy needed to send a bit. Proc. R. Soc. Lond. A, 454:305–
311. 15
BIBLIOGRAPHY 191

Lavenda, B. H. and Dunning-Davies, J. (2003). Additive entropies of degree-q and

the Tsallis entropy. physics/0310117, pages 1–13. 155
Lee, H., Oh, S. D., and Ahn, D. (2003). Entanglement measure for any quantum
states. quant-ph/0306127, pages 1–6. 163
Lee, J.-S., Chung, Y., Kim, J., and Lee, S. (1999). A practical method of construct-
ing quantum combinational logic circuits. quant-ph, 9911053:1–6. 14
Leung, D. (2000). Towards robust quantum computation. PhD thesis, Stanford
University. cs.CC/0012017. 120
Lewenstein, M. and Sanpera, A. (1998). Separability and entanglement of composite
quantum systems. Phys. Rev. Lett., 80:2261–2264. quant-ph/9707043. 163
Lim, Y. L., Beige, A., and Kwek, L. C. (2004). Repeat-until-success quantum
computing. quant-ph/0408043, pages 1–5. 68
Lindner, H., Brayer, H., and Lehmann, C. (1999). Taschenbuch der Elektrotechnik
und Elektronik. Fachbuchverlag Leipzig imCarl Hanser Verlag. 14
Lloyd, S. (2000). Computation without interaction. quant-ph/0004010, pages 1–8.
166
Long, G.-L. and Sun, Y. (2001). Efficient scheme for initializing a quantum reg-
ister with an arbitrary superposed state. Phys. Rev. A, 64:014303. quant-
ph/0104030. 18
Lücke, W. (1996). Axiomatic quantum theory. Acta Phys. Pol., 27:2357–2385. 103
Lücke, W. (Clausthal, SS 2002). Quantum computers. 98, 120
Lücke, W. (edyn). Elektrodynamik .
http://www.wolfgang-luecke.de/skripten/edyn.pdf. 52
Lücke, W. (eine). Ergänzungen zu “Mathematische Methoden der Physik”.
http://www.wolfgang-luecke.de/skripten/eine.pdf. 29, 92, 101, 142
Lücke, W. (mech). Klassische Mechanik .
http://www.wolfgang-luecke.de/skripten/mech.pdf. 29
Lücke, W. (musi). Theorie zur Physik der Musikinstrumente.
http://www.wolfgang-luecke.de/skripten/musi.html. 41
Lücke, W. (nlqo). Introduction to photonics.
http://www.wolfgang-luecke.de/skripten/nlqo.pdf. 19, 58, 59, 62, 64, 81,
82, 84, 93
Lücke, W. (qft). Particles and fields.
http://www.wolfgang-luecke.de/skripten/qft.pdf. 20
192 BIBLIOGRAPHY

Lücke, W. (tdst). Thermodynamik und Statistik .

http://www.wolfgang-luecke.de/skripten/tdst.pdf. 20

MacWilliams, F. J. and Sloane, N. J. A. (1998). The theory of error-correcting codes.

North-Holland, New York.

Mahler, G. and Weberruß, V. A. (1995). Quantum Networks. Springer-Verlag.

Dynamics of Open Nanostructures.

Mermin, N. D. (2002). Deconstructing dense coding. Phys. Rev. A, 66:032308.

quant-ph/0204107. 24, 25, 120

Mermin, N. D. (2004). Copenhagen computation: How I learned to stop worrying

and love Bohr. IBM Journal of Research and Development, 48:53–62. quant-
ph/0305088.

Mershin, A., Nanopoulos, D. V., and Skoulakis, E. M. C. (2000). Quantum brain?

quant-ph, 0007088:1–37.

Mertens, S. (2002). Computational complexity for physicists. Computing in Science

& Engineering, 4:31–41. cond-mat/0012185. 31

Metcalf, H. J. and van der Straten, P. (1999). Laser Cooling and Trapping. Springer-
Verlag. 88

Milburn, G. J. (1996). Quantum Technology. Frontiers of science. Allen & Unwin.

Mitchison, G. and Jozsa, R. (2001). Counterfactual computation. Proc. Roy. Soc.

Lond., A457:1175–1194. quant-ph/9907007. 19

Mitchison, G. and Jozsa, R. (2003). Towards a geometrical interpretation of quan-

tum information compression. quant-ph/0309177, pages 1–11. 144

Mizrahi, S. S. and Dodonov, V. V. (2002). Creating quanta with ‘annihilation’

operator. quant-ph/0207035, pages 1–8. 53

Morikoshi, F., Santos, M. F., and Vedral, V. (2003). Accessibility of physical states
and non-uniqueness of entanglement measure. J. Phys. A, 37:5887. quant-
ph/0306032. 162

Möttönen, M. and Vartiainen, J. J. (2005). Decompositions of general quantum

gates. quant-ph/0504100, pages 1–25. 119

Munro, W., Nemoto, K., Spiller, T., Barrett, S., Kok, P., and Beausoleil, R. (2005a).
Efficient optical quantum information processing. quant-ph/0506116, pages 1–
10. 59

Munro, W. J., Nemoto, K., and Spiller, T. P. (2005b). Weak nonlinearities: A new
route to optical quantum computation. quant-ph/0507084, pages 1–7. 59
BIBLIOGRAPHY 193

Neumark, M. A. (1959). Normierte Algebren. VEB Deutscher Verlag der Wis-

senschaften, Berlin. 148

Nielsen, M. A. (2004). Optical quantum computation using cluster states. Phys.

Rev. Lett., 93:040503. quant-ph/0402005. 63, 70, 71

Nielsen, M. A. (2005). Cluster-state quantum computation. quant-ph/0504097,

pages 1–15. 68

Nielsen, M. A. and Chuang, I. L. (1997). Programmable quantum gate arrays. Phys.

Rev. Lett., pages 321–324. quant-ph/0703032. 70

Nielsen, M. A. and Chuang, I. L. (2001). Quantum Computation and Quantum

Information. Cambridge University Press. 3, 39, 40, 47, 72, 91, 136, 145, 157,
159, 160

Nielsen, M. A. and Dawson, C. M. (2004). Fault-tolerant quantum computation

with cluster states. quant-ph/0405134, pages 1–31. 70

Nussbaumer, H. J. (1982). Fast Fourier Transform and Convolution Algorithms,

volume 2 of Springer Series in Information Sciences. Springer-Verlag. 43

Ohya, M. and Petz, D. (1993). Quantum Entropy and Its Use. Springer-Verlag. 131,
145

Ottaviani, C., Rebic, S., Vitali, D., and Tombesi, P. (2005). Quantum phase gate
operation based on nonlinear optics: Full quantum analysis. quant-ph/0507137,
pages 1–4. 59

Paris, M., Plenio, M., Jonathan, D., Bose, S., and D’Ariano, G. (2000). Optical
Bell measurement by Fock filtering. Physics Letters A, 273:153–158. quant-
ph/9911036. 121

Parker, M. C. and Walker, S. D. (2003). Information transfer and Landauer’s prin-

ciple. physics/0310116, pages 1–5. 15

Patel, A. (2001). Quantum algorithms and the genetic code. Pramana, 56:367–381.
quant-ph/0002037.

Peres, A. (2002). How the no-cloning theorem got its name. quant-ph/0205076,
pages 1–4. 18, 91

Perron, O. (1954). Die Lehre von den Kettenbrüchen, volume I: Elementare Ket-
tenbrüche. B.G. Teubner Verlagsgesellschaft, Stuttgart. 46

Perron, O. (1957). Die Lehre von den Kettenbrüchen, volume I: Analytisch-

funktionentheoretische Elementare Kettenbrüche. B.G. Teubner Verlagsge-
sellschaft, Stuttgart. 46
194 BIBLIOGRAPHY

Petz, D. (2001). Entropy, von Neumann and the von Neumann entropy. math-
ph/0102013, pages 1–10. 131, 145

Pittman, T. B. and Franson, J. D. (2002). Cyclical quantum memory for photonic

qubits. Phys. Rev. A, 66:062302. 63

Plenio, M. B. and Vitelli, V. (2001). The physics of forgetting: Landauer’s erasure

principle and information theory. Contemporary Physics, 42:25 –60. quant-
ph/0103108. 15

Pless, V. (1989). Introduction to the Theory of Error-Correcting Codes. John Wiley

& Sons. 108

Preskill, J. (1998a). Fault-tolerant quantum computation. In Lo, H.-K., Popescu, S.,

and Spiller, T. P., editors, Introduction to Quantum Computation and Informa-
tion, pages 213–269. World Scientific. ISBN 981-02-339-X. quant-ph/9712048.
120

Preskill, J. (1998b). Reliable quantum computers. Proc. R. Soc. Lond. A, 454:385–

410. quant-ph/9705031. 120

Preskill, J. (2000–01). Physics 219/Computer Science 219 Quantum Computation

(Formerly Physics 229). (http://theory.caltech.edu/∼preskill/ph229/#lecture).
3, 115, 127

Pütz, J., editor (1971). Einführung in die Elektronik. Bücher des Wissens. Fischer
Taschenbuch Verlag GmBH, Frankfurt am Main. 11

Ralph, T. C., White, A. G., Munro, W. J., and Milburn, G. J. (2002). Simple
scheme for efficient linear optics quantum gates. Phys. Rev., A65:012314. quant-
ph/0108049. 19, 58, 60

Raussendorf, R. (2003). Measurement-based quantum computation with cluster

states. Dissertation, Ludwig-Maximilians Universität, München. 68

Raussendorf, R. (2005). Quantum computation via translation-invariant operations

on a chain of qubits. quant-ph/0505122, pages 1–6. 68

Raussendorf, R. and Briegel, H. J. (2001). A one-way quantum computer. Phys.

Rev. Lett., 86:5188–5191. quant-ph/0010033. 68

Raussendorf, R., Browne, D. E., and Briegel, H. J. (2002). The one-way quan-
tum computer - a non-network model of quantum computation. J. Mod. Opt.,
49:1299–1306. quant-ph/0108118. 19

Reck, M., Zeilinger, A., Bernstein, H. J., and Bertani, P. (1994). Experimental
realization of any discrete unitary operator. Phys. Rev. Lett., 73:58–61. 28
BIBLIOGRAPHY 195

Řeháček, J. and Hradil, Z. (2002). Quantification of entanglement by means of

convergent iterations. Phys. Rev. Lett., 90:127904. quant-ph/0205071. 161
Resch, K. J., Lundeen, J. S., and Steinberg, A. M. (2002). Practical creation and
detection of polarization Bell states using parametric down-conversion. quant-
ph/0204034, (1–13). 166
Rivest, R. L., Shamir, A., and Adleman, L. (1978). A method for obtaining dig-
ital signatures and public-key cryptosystems. Communications of the ACM,
21(2):120–126. hppt://theory.lcs.mit.edu/∼cis/pubs/rivest/rsapaper.ps. 37
Robertson, A. and Robertson, W. (1967). Topologische Vektorräume, volume
164/164a. Bibliographisches Institut · Mannheim. 148
Rosé, H., Aßelmeyer-Maluga, T., Kolbe, M., Niehörster, F., and Schramm, A.
(2004). The Fraunhofer quantum computing portal - www.qc.fraunhofer.de
- A web-based simulator of quantum computing processes. quant-ph/0406089,
pages 1–7. 48
Rosenberg, D., Lita, A. E., Miller, A. J., and Nam, S. W. (2005). Noise-free high-
efficiency photon-number-resolving detectors. Phys. Rev. A, 71:061803. quant-
ph/0506175. 57
Rudolph, O. (2000). A separability criterion for density operators. J. Phys. A,
33:3951–3955. quant-ph/0002026. 155
Rudolph, O. (2002). Further results on the cross norm criterion for separability.
quant-ph/0202121, pages 1–19. 155
Rudolph, T. and Pan, J.-W. (2001). A simple gate for linear optics quantum com-
puting. quant-ph/0108056, pages 1–1. 60
Rudolph, T. and Virmani, S. S. (2005). A relational quantum computer using only
two-qubit total spin measurement and an initial supply of highly mixed single
qubit states. quant-ph/0503151, pages 1–5. 68
Ruskai, M. B. (2002). Inequalities for quantum entropy: A review with conditions
for equality. J. Math. Phys., 43:4358–4375. quant-ph/0205064. 131
Sanaka, K., Jennewein, T., Pan, J.-W., Resch, K., and Zeilinger, A. (2003). Exper-
imental nonlinear sign shift for linear optics quantum computation. Phys. Rev.
Lett., 92:017902. quant-ph/0308134. 58
Sanctuary, B. C., Presse, S., Lazzara, T. D., Henderson, E. J., and Hourani, R. F.
(2003). Interpretation of “non-local” experiments using disentanglement. quant-
ph/0308026, pages 1–14. 122
Schlingemann, D. (2001). Stabilizer codes can be realized as graph codes. quant-
ph/0111080, (1–7). 115
196 BIBLIOGRAPHY

Schlingemann, D. and Werner, R. F. (2000). Quantum error-correcting codes asso-

ciated with graphs. Phys. Rev. A, 65:012308. quant-ph/0012111. 115

Schneier, B. (1996). Applied Cryptography. John Wiley & Sons, Inc. 37

Schroeder, M. R. (1997). Number Theory in Science and Communication, volume 7

of Springer Series in Information Sciences. Springer-Verlag. 39

Shabani, A. and Lidar, D. A. (2006). Quantum error correction beyond completely

positive maps. quant-ph/0610028, pages 1–5. 94

Shannon, C. (1949a). Communication theory of secrecy systems. Bell Syst. Tech.

J., 28:656–715. 124

Shannon, C. E. (1949b). The mathematical theory of communication. University of

Illinois Press. 3, 11, 91, 128

Shende, V. V., Bullock, S. S., and Markov, I. L. (2004a). A practical top-down

approach to quantum circuit synthesis. quant-ph/0406176, pages 1–2. 16

Shende, V. V., Markov, I. L., and Bullock, S. S. (2004b). On universal gate libraries
and generic minimal two-qubit quantum circuits. Phys. Rev. A, 69:062321.
quant-ph/0308033. 31

Shende, V. V., Prasad, A. K., Markov, I. L., and Hayes, J. P. (2003). Reversible
logic circuit synthesis. IEEE Trans. on CAD, 22:710–722. quant-ph/0207001.
14, 16

Shor, P. (1994). Algorithms for quantum computation: discrete logarithms and

factoring. In Proceedings 35th Annual Symposium on Foundations of Computer
Science, Santa Fe, NM, USA, 20–22 Nov. 1994, pages 124–134. IEEE Comput.
Soc. Press. 40, 124

Shor, P. W. (1997). Polynomial-time algorithms for prime factorization and discrete

logarithms on a quantum computer. SIAM J. Sci. Statist. Comput., 26:1484–
1509. quant-ph/9508027. 45

Shor, P. W. (2000). Introduction to quantum algorithms. quant-ph/0005003. Course

at the January 2000 AMS meeting. 33

Singh, S. (2002). Code Book: The Science of Secrecy from Ancient Egypt to Quantum
Cryptography. Random House Children’s Publishing. 37

Sleator, T. and Weinfurter, H. (1994). Realizable universal quantum logic gates.

Phys. Rev. Lett., 74:4087–4090. 31

Streater, R. F. (2003). Duality in quantum information geometry. math-ph/0308037,

pages 1–7. 137
BIBLIOGRAPHY 197

Tarasov, V. E. (2002). Quantum computations by quantum operations on mixed

states. quant-ph/0201033, pages 1–11. 19
Toffoli, T. (1980a). Reversible computing. In Goos, G. and Hartmanis, J., editors,
Automata, Languages and Programming, pages 632–644. Springer-Verlag. 15
Toffoli, T. (1980b). Reversible computing. Technical Memo MIT/LCS/TM-151,
MIT Lab. for Comp. Sci. (http://pm1.bu.edu/∼tt/publ.html). 16
Tomita, A. (2000). Complete Bell state measurement with controlled photon ab-
sorption and quantum interference. quant-ph/0006093, pages 1–4. 121
Tsai, I. M. and Kuo, S. Y. (2001). A systematic algorithm for quantum boolean
circuits construction. quant-ph/0104037, pages 1–12. 16
Tucci, R. R. (2004). QC Paulinesia. quant-ph/0407215, pages 1–45. See also:
www.ar-tiste.com/PaulinesiaVer1.pdf. 14
Tulsi, T., Grover, L., and Patel, A. (2005). A new algorithm for directed quantum
search. quant-ph/0505007, pages 1–12. 36
Uhlman, A. (1976). The ‘transition probability’ in the state space of a ∗ -algebra.
Rep. Math. Phys., 9:273–279. 140
Varnava, M., Browne, D. E., and Rudolph, T. (2005). Loss tolerant one-way quan-
tum computation – a horticultural approach. quant-ph/0507036, pages 1–6.
68
Vatan, F. and Williams, C. (2004). Optimal quantum circuits for general two-qubit
gates. Phys. Rev. A, 69:032315. quant-ph/0308006. 31
Vedral, V., Barenco, A., and Ekert, A. (1996). Quantum networks for elementary
arithmetic operations. Phys. Rev. A, 54:147–153. quant-ph/9511018. 16, 40
Verstraete, F. and Cirac, J. (2003). Quantum–nonlocality in the presence of su-
perselection rules and some applications. Phys. Rev. Lett., 91:10404. quant-
ph/0302039. 91, 163, 180
Vidal, G. and Werner, R. F. (2002). A computable measure of entanglement. Phys.
Rev. A, 65:032314. quant-ph/0102117. 153
Vollbrecht, K. G. H. and Wolf, M. M. (2002). Conditional entropies and their relation
to entanglement criteria. J. Mathem. Phys., 43:4299. quant-ph/0202058. 154
von Neumann, J. (1927). Thermodynamik quantenmechanischer Gesamtheiten.
Nachr. der Gesellschaft der Wiss. Gött., pages 273–291. 131
Waks, E., Inoue, K., Diamanti, E., and Yamamoto, Y. (2003). High efficiency pho-
ton number detection for quantum information processing. quant-ph/0308054,
pages 1–10. 55
198 BIBLIOGRAPHY

Walther, P., Resch, K., Rudolph, T., Schenck, E., Weinfurter, H., Vedral, V., As-
pelmeyer, M., and Zeilinger, A. (2005). Experimental one-way quantum com-
puting. Nature, 434:169–176. quant-ph/0503126. 36, 68

Wehrl, A. (1978). General properties of entropy. Rev. Modern Phys., 50:221–260.

131

Werner, R. (2001). Quantum information theory — an invitation. Springer Tracts

in Modern Physics, 173:14–57. quant-ph/0101061. 18, 92, 94

Werner, R. F. (1989). Quantum states with Einstein-Podolsky-Rosen correlations

admitting a hidden-variable model. Phys. Rev. A, 40:4277–4281. 100, 147, 180

Werner, R. F. and Wolf, M. M. (201). All multipartite Bell correlation inequalities

for two dichotomic observables per site. quant-ph/0102024, pages 1–11. 147

White, A. G., James, D. F. V., Munro, W. J., and Kwiat, P. G. (2001). Exploring
Hilbert space: accurate characterisation of quantum information. Phys. Rev.
A, 64:R030302. quant-ph/0108088. 163

Wootters, W. K. (2001). Entanglement of formation and concurrence. Quantum

Information and Computation, 1:27–44. 163

Wootters, W. K. and Zurek, W. H. (1982). A single quantum cannot be cloned.

Nature, 299:802–803. 18, 91

Wunderlich, C. and Balzer, C. (2003). Quantum measurements and new concepts

for experiments with trapped ions. Advances in Atomic, Molecular, and Optical
Physics, 49:295–376. quant-ph/0305129. 71

Younes, A. and Miller, J. (2003). Automated method for building CNOT based
quantum circuits for Boolean functions. quant-ph/0304099, pages 1–18. 16

Zhao, Z., Zhang, A.-N., Chen, Y.-A., Zhang, H., Du, J.-F., Yang, T., and Pan, J.-
W. (2005). Experimental demonstration of a non-destructive controlled-NOT
quantum gate for two independent photon-qubits. Phys. Rev. Lett., 94:030501.
quant-ph/0404129. 71
Index

adder, 16 word, 108

coding
balanced, 22 quantum dense, 24
Bell coherence
measurement, 25, 26 vector of, 176
network, 24, 26 coherent superposition, 18, 23
states, 24 complexity
Bernstein-Vazirani computational, 31
oracle, 23 computation
binary digit, 11 -al basis, 18–20
bit, 11 classical, 15, 17, 22
qu-, 18 reversible, 15, 16
Bloch complexity, 31
vector, 176 concurrence, 163
Campbell-Hausdorff formula, 64 CSS codes, 114
circuit decoherence, 104
classical logic, 12, 15 dense coding, 24
computational, 11 Deutsch-Jozsa
equivalent, 11 oracle, 22, 23
graph, 12 problem, 22
logic, 14 dual code, 109
classical
computation, 15, 17, 22 ebit, 161
network, 18 entangled, 24
logic circuit, 12, 15 pure state, 24, 26, 103
reversible computation, 15, 16 entanglemen
reversible network, 15 catalysis, 160
closed quantum system, 103 entanglement
code of formation, 163
Calderbank-Shor-Steane, 114 swapping, 26
linear classical witness, 147, 151
dual, 109 entropic inequalities, 154
quantum, 110 entropy
words, 110 Renỳi, 155
stabilizer, 113 equivalent
Steane, 116 classical network, 11

199
200 INDEX

preparation procedures, 171 optimization, 16

quantum network, 21 quantum
error correction, 18 equivalent, 21
Euler angles, 29 quantum computational, 18
reversible, 15
fast Fourier, 43 teleportation, 25
flip
operator, 100 open quantum system, 104
Fourier transform operator
fast, 43 positive
Fredkin gate, 14 support, 136
function oracle
balanced, 22 Bernstein-Vazirani, 23
decision, 14 Deutsch-Jozsa, 22, 23
order
gate, 11 of an integer modulo N , 169
Fredkin, 14
non-deterministic, 19 Partial transpose, 152
quantum, 19, 20 Pauli
universal, 27 group, 111
reversible, 13 Pauli matrices, 104
Toffoli, 14 polarization identity, 147
universal, 15, 16 post selection, 26
graph, 12 problem
Deutsch-Jozsa, 22
halting problem, 11, 165 halting, 165
Hamming projective measurement, 171
distance, 108
quantum
interaction picture, 171 closed system, 103
LOCC, 156 code, 110
logic circuit, 14 code words, 110
computation
measurement, 18, 19 network, 18
Bell, 25, 26 result, 19
projective, 171 dense coding, 24
mixture, 104 gate, 19, 20
measurement, 18
network network
adder, 16 equivalent, 21
Bell, 24, 26 open system, 104
classical parallelism, 18
2-bit, 17 state
computational, 18 entangled, 24, 26, 103
reversible, 15 tomography, 176
INDEX 201

state collapse, 18 norm, 137

teleportation, 25, 26 partial, 177
wire, 19 transposition, 100, 151
qubit, 18 partial, 152

Range criterion, 154 universal

register, 11 gate, 15, 16
n-qubit, 19 quantum gate, 27
state space, 18
Renỳi entropies, 155 Werner states, 151
result wire, 15
of a quantum computation, 19 quantum, 19
reversible
classical network, 15

Sleator-Weinfurter construction,
31
stabilizer, 113
codes, 113
state
Bell, 24
mixed, 104
pure, 171
quantum
entangled, 24, 26, 103
separable, 151, 178
Werner, 151
state space, 171
Steane code, 116
Stokes
vector, 176
superoperator, 96
superposition
coherent, 23

teleportation, 26
network, 25
of entanglement, 26
tensor product
formalism of quantum mechanics,
19
Toffoli gate, 14
tomography
quantum state, 176
trace