QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Quantum Information
with Continuous Variables

Edited by

SAMUEL L. BRAUNSTEIN
School of Informatics,
University of Wales, Bangor. United Kingdom

and

ARUN K. PAT!
Institute of Physics, Orissa, India
and
Theoretical Physics Division,
BARe, Mumbai, India

KLUWER ACADEMIC PUBLISHERS

DORDRECHT/BOSTON/LONDON
A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN 978-90-481-6255-0

Published by Kluwer Academic Publishers,

P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Sold and distributed in North, Central and South America

by Kluwer Academic Publishers,
101 Philip Drive, Norwell, MA 02061, U.S.A.

In all other countries, sold and distributed

by Kluwer Academic Publishers,
P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

ISBN 978-90-481-6255-0 ISBN 978-94-015-1258-9 (eBook)

DOI 10.1007/978-94-015-1258-9
Printed on acid-free paper

Cover illustration:
reprinted with permission from Nature [Nature 413, 400-403 (2001)]
Copyright (2001) Macmillan Pub1ishers Ltd.
Softcover reprint of the hardcover 1st edition 2001

All Rights Reserved

© 2003 Kluwer Academic Publishers and copyright holders
as specified on appropriate pages within.
No part of this work may be feproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exc1usive use by the purchaser of the work.
 Dedicated to
Netta, Rashmi
and our parents
Contents

Preface Xl

About the Editors Xlll

Part I Quantum Computing

1. Quantum computing with qubits 3

Samuel L. Braunstein and Arun K. Pati

2. Quantum computation over continuous variables 9

Seth Lloyd and Samuel L. Braunstein

3. Error correction for continuous quantum variables 19

Samuel L. Braunstein

4. Deutsch-lozsa algorithm for continuous variables 31

Arun K. Pati and Samuel L. Braunstein

5. Hybrid quantum computing 37

Seth Lloyd

6. Efficient classical simulation of continuous variable

quantum information processes 47
Stephen D. Bartlett, Barry C. Sanders, Samuel L. Braunstein and Kae Nemoto

Part II Quantum Entanglement

7. Introduction to entanglement-based protocols 59

Samuel L. Braunstein and Arun K. Pati

8. Teleportation of continuous q~antum variables 67

SamuefL. Braunstein and H. JejJrey Kimble
Vll
viii QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

9. Experimental realization of continuous variable teleportation 77

Akira Furusawa and H. J. Kimble

10. Dense coding for continuous variables 95

Samuel L. Braunstein and H. Jeffrey Kimble

11. Multipartite Greenberger-Home-Zeilinger paradoxes

for continuous variables 105
Serge Massar and Stefano Pironio

12. Multipartite entanglement for continuous variables 111

Peter van Loock and Samuel L. Braunstein

13. Inseparability criterion for continuous variable systems 145

Lu-Ming Duan, Giza Giedke, J. Ignacio Cirac and Peter Zoller

14. Separability criterion for Gaussian states 155

Rajiah Simon

15. Distillability and entanglement purification for Gaussian states 173

Giza Giedke, Lu-Ming Duan, J. Ignacio Cirac and Peter Zoller

16. Entanglement purification via entanglement swapping 193

Steven Parker, Sugato Bose and Martin B. Plenio

17. Bound entanglement for continuous variables is a rare phenomenon 211

Pawel Horodecki, J. Ignacio Cirac and Madej Lewenstein

Part III Continuous Variable Optical-Atomic Interfacing

18. Atomic continuous variable processing and light-atoms quantum interface 231
Alex Kuzmich and Eugene S. Polzik

Part IV Limits on Quantum Information and Cryptography

19. Limitations on discrete quantum information and cryptography 269

Samuel L. Braunstein and Arun K. Pati

20. Quantum cloning with continuous variables 277

Nicolas J. Cerf

21. Quantum key distribution with continuous variables in optics 295

Timothy C. Ralph

22. Secure quantum key distribution using squeezed states 317

Daniel Gottesman and John Preskill
 Contents ix

23. Experimental demonstration of dense coding and quantum cryptography

with continuous variables 357
Kunchi Peng, Qing Pan, ling Zhang and Changde Xie

24. Quantum solitons in optical fibres:

basic requisites for experimental quantum communication 379
G. Leuchs, Ch. Silberhorn, F. Konig, P. K. Lam, A. Sizmann and N. Korolkova

Index 423
Preface

Quantum information has become a flagship of interdisciplinary research in

recent years, sweeping physicists from a variety of disciplines, as well as com-
puter scientists and mathematicians. It all started with the realization that the
principles of quantum theory open new avenues of information processing ca-
pabilities which are unavailable in the classical world. Primarily, the discovery
that quantum entanglement can be put to use has led to a watershed of activity
towards the eventual implementation of quantum computation and quantum
cryptography. This was naturally accompanied by efforts to place fundamental
quantum limits on information processing. Today, real-world applications of
quantum-information technologies seem closer and more tangible than ever.
The field of quantum information has typically concerned itself with the
manipulation of discrete systems such as quantum bits, or "qubits." How-
ever, many quantum variables, such as position, momentum or the quadrature
amplitudes of electromagnetic fields, are continuous. Quantum information
processing with continuous variables is the subject of this volume.
Initially, quantum information processing with continuous variables seemed
daunting at best, ill-defined at worst. The first real success came with the exper-
imental realization of quantum teleportation for optical fields. This was soon
followed by a flood of activity, to understand the strengths and weaknesses of
this type of quantum information and how it may be processed. The next major
breakthrough was the successful definition of a notion of universal quantum
computation over continuous variables, suggesting that such variables are as
powerful as conventional qubits for any class of computation.
In some ways continuous-variable computation may not be so different than
qubit-based computation. In particular, limitations due to finite precision make
quantum floating-point operations, like their classical counterparts, effectively
discrete. Thus we might expect a continuous-variable quantum computer to
perform no better than a discrete quantum computer. However, for some tasks
continuous-variable quantum computers are nonetheless more efficient. Indeed,
Xl
xii QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

in many protocols, especially those relating to communication, they only require

linear operations together with classical feed-forward and detection. This
together with the large bandwidths naturally available to continuous (optical)
variables appears to give them the potential for a significant advantage.
Noise is the Achilles' heel of quantum computation, and continuous variables
are even more susceptible to noise than discrete variables. Since an uncount-
ably infinite number of things can go wrong with a continuous variable, error
correction protocols might be expected to require infinite redundancy. Fortu-
nately, continuous-variable error correction routines exist and require no greater
redundancy than protocols for discrete variables. With the problem of noise
potentially reduced to manageable proportions, many other hurdles persist, and
many more exciting questions remain open. Quantum information with contin-
uous variables continues to be a dynamic and exciting field of ongoing research
and development.
In this volume we present important developments in the area of quantum
information theory for continuous-variable systems from various leading re-
searchers in the field. Several introductory chapters layout some of the basics
of quantum information theory in terms of the more usual qubits. Each in-
troduction is followed by generalizations to continuous variables. In addition
to chapters on quantum computation and quantum teleportation, we have in-
cluded work on quantum dense coding, quantum error correction, some simple
attempts at generalizing quantum algorithms and technologically promising
work on quantum cryptography and quantum memory. These results apply to
any collection of continuous variables, including phonons, photons, Josephson
Junction circuits, Bose-Einstein condensates, etc. Finally, the underlying na-
ture of continuous quantum information is investigated in chapters on quantum
cloning and quantum entanglement.
At this juncture, we express our gratitude to Kluwer Academic for allowing
us to bring this work to light. We hope that this book offers the reader a
rigorous introduction to continuous-variable quantum information and some
thought-provoking snapshots of recent developments. We sincerely thank all
the authors for their contributions. A.K.P. would especially like to express his
thanks to Prof. R. K. Choudhury, Director of the Institute of Physics, in Orissa,
India, for his encouragement. Finally, both authors appreciate the support
provided by the University of Wales, Bangor, throughout this endeavor.

SAMUEL L. BRAUNSTEIN

ARUN K. PAT!
About the Editors

Samuel L. Braunstein is a Professor at the University of Wales, Bangor,

where he has taught since 1997. He is a recipient of the prestigious Royal
Society-Wolfson Research Merit Award and was awarded the honorary title of
2001 Lord Kelvin Lecturer. Before joining the University of Wales, he held a
German Humboldt Fellowship (spent at the University of Ulm). He is editor of
two books Quantum Computing and Scalable Quantum Computing and serves
on the editorial board of the journal Fortschritte der Physik. He has initiated
and is a Founding Managing Editor of Quantum Information and Computation
- the first journal dedicated specifically to this field.

Arun K. Pati is a Visiting Scientist in the Institute of Physics, Bhubaneswar,

Orissa, India since 2001. He has been a Scientific .officer in the Theoretical
Physics Division, BARC, Mumbai since 1989. His research interests include
Quantum Theory, Foundations, and Quantum Information and Computation.
He is a recipient of the Indian Physics Association NSS Memorial Award for the
year 2000 and the Indian Physical Society Award for Young Physicists for the
year 1996. He was an Associate of the Indian Academy of Science, Bangalore
from 1998-2001. Presently he is an Associate at the Center for Philosophy
and Foundation Science, New Delhi and an Honorary Research Fellow at the
University of Wales, Bangor, United Kingdom.

xiii
I

QUANTUM COMPUTING
Chapter 1

QUANTUM COMPUTING WITH QUBITS

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 I UT, United Kingdom
schmuel @sees.bangor.ac.uk

Arun K. Pati
Institute of Physics, Bhubaneswar-75IO05, Orissa, INDIA
Theoretical Physics Division, BARe, Mumbai, INDIA
akpati @iopb.res.in

Abstract We briefly give an introduction to quantum computing with qubits.

This book will introduce the reader to the area of quantum information
processing with continuous variables. However, to put it into some context
with the "conventional" approach to quantum computing with qubits we shall
give a brief introduction to its basic principles skipping all details. Briefly,
we will touch on notions of bits, qubits, quantum parallelism, and quantum
algorithms such as the Deutsch-Jozsa, the Shor factoring problem, and the
Grover quantum search.
Quantum information theory is a marriage between two scientific pillars
of modern science, namely, quantum theory and classical information theory.
Quantum theory as developed by Planck, Einstein, Schrodinger, Dirac, Heisen-
berg and many others in the early part of the last century is one of the finest
theories that explains phenomena ranging from molecules to electrons, protons,
neutrons and other microscopic particles. The mathematical theory of classi-
cal information was put forth by Shannon in the mid part of the last century.
Whatever revolution in information technology we see at present is partly due
to the ground breaking work by Shannon, Turing, Church and others.
When the ideas from information theory are carried over to quantum theory
there emerges a revolution in our ability to process information. Ultimately, the
basic ways of expressing and manipulating information require physical states
3
SoL Braunstein and A.K. Pati (eds.), Quantum ['!!ormation with Continuous Variables, 3-8.
© 2003 Kluwer Academic Publishers.
4 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

and processes. In quantum theory we know that physical processes are funda-
mentally different than those of classical physics. Therefore manipulation of
information based on quantum physical processes has also to be fundamentally
different than their classical counterparts. It was first realised by Feynman that
simulating quantum systems on a classical computer would be very inefficient
[1]. However, if one utilizes quantum systems than one can do much more. For
this reason, quantum information is distinguished from conventional classical
information.

1. QUANTUM COMPUTATION
The physics of information and computation are intimately related. Informa-
tion is encoded in the state of a physical system, whereas computation involves
the processing of this information through actions on the physical system. This
processing must obey physical law. Therefore, the study of information and
computation are linked through the study of underlying physical processes. If
the physical processes obey the rules of classical physics, the corresponding
computation is dubbed "classical." If on the other hand, the underlying pro-
cesses are subjected to quantum mechanical rules, the resulting computation
will be called "quantum computation." The logic that lies at the heart of conven-
tional computers and quantum computers is therefore fundamentally different.
Quantum computation is a particular way of processing information which uti-
lizes the principles of the linearity of quantum mechanics, out of which comes
quantum superposition, quantum entanglement and quantum parallelism. This
was first suggested by Deutsch [2].
In a conventional computer information is stored as binary digits (bits)
usually (logically) labeled 0 and 1. To represent a bit, one may use any
physical system that one likes provided it allows two distinct states. Two
bits of information can be stored in any system allowing 22 = 4 possible
distinct states. Similarly, n bits of information may be represented in any
system capable of providing 2n distinct states. In each case, there is only one
configuration at a time of the logical bits (e.g., 110 for 3 bits). Information
stored in these binary digits can be manipulated using elementary logic gates
that obey Boolean algebra. For example, in a conventional classical computer
one can manipulate information using sequence of logical operations such
as AND, OR, NOT, and XOR gates. These gates may be built into circuits
constructing any possible Boolean functions [3].

1.1 QUBITS
Let us represent a bit, 0 or 1, by saying that the spin of a neutron is up
or down, or we could sayan atom is in ground or in an excited state, or a
photon is horizontally or vertically polarized. All these systems are examples
 Quantum computing with qubits 5

of two-state quantum systems because the two states are orthogonal and hence
logically distinct from each other. When a quantum system is in a given basis
state it may be said to carry classical information. However, quantum theory
also allows states which are in linear superpositions of these basis states. If
we use the logical label for each of the basis states then the most general pure
quantum state for a "two-state" system is given by

11/1} = aiD} + ,611} , (1.1)

for some complex numbers a and,6 which satisfy lal 2 + 1,61 2 = 1. According
to Dirac a quantum state 1/1 is denoted by a "ket" 11/1} E 1-£ = ((J (a two-
dimensional Hilbert space). Such a system contains a quantum bit or "qubit"
of information [4].
A single qubit allows two inputs to be stored (and possibly processed) si-
multaneously. As we add more qubits the number of simultaneous possibilities
grows very quickly. For example, for two qubits, which have logically distinct
states labeled by 00,01,10 and 11 the most general state may be written as a
superposition of these four possibilities

aIOO} + ,6101 + c51 10} + ,Ill} E 1-£ = ~ . (1.2)

For n qubits with 2n possible distinct logical basis states the most general pure
state will take the form of a simultaneous superposition of this exponential
number of possibilities, such as

(1.3)

This ability for quantum states to simultaneously represent (and process) an

exponential number of "logical states" demonstrates the fundamental difference
between classical and quantum computers.
In the case of a composite system of two or more qubits (or any kind of
subsystems) as seen in Eq, (1.3), the result of this superposition can give
rise to quantum entanglement (inter-twinedness). If a pure composite state
cannot be written as a product of individual states for each subsystem then it
is an entangled state. The word entanglement neatly encapsulates the novel
non-classical correlations accessible to quantum systems, yet which cannot be
described within any local realistic model since they allow for the violation of
Bell inequalitiesinequality,Bell [5]. What this means is that one cannot attribute
definite properties to the individual subsystems of an entangled state. Charlie
Bennett's metaphor for a pair of entangled particles is that they are to be likened
to a pair of lovers. If a pair of particles are entangled, then a measurement
of one, will, in some sense, instantaneously affect the other no matter how far
apart they lie.
6 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

2. QUANTUM PARALLELISM AND ALGORITHMS

It is possible to design new types of logic gates, generalizing those that
work for classical bits, that act on qubits. These quantum gates may also be
combined into circuits in such a way as to allow the most general (unitary)
transformations between quantum states. We may think of these circuits as
performing particular quantum algorithms for a given input. Because more
general transformations than simple Boolean functions are available, quantum
superpositions may be created using these quantum circuits. This allows for
the ability of quantum computers to perform many computational tasks in par-
allel. Further, because the potential amount of parallelism grows exponentially
with the number of qubits, this feature cannot be efficiently simulated in any
conventional computer, no matter how parallel its architecture.
To illustrate quantum parallelism, imagine that we had access to a black
box that computes a given function f(x) from an input x of n qubits (x = 0,
1, ... , 2n). Quantum mechanically because one can create a superposition
of all inputs, one could perform all N = 2n function evaluations in a single
go. Classically, this would require the N separate function evaluations. In
particular, one can start by preparing a "register" of n qubits in an equal
superposition of all possible bit strings given by

1 N-l

.;N ~ Ix). (1.4)

Further, we suppose that the function f is evaluated by some unitary operation

Uf acting via [2]

Uf : Ix)IO) -t Ix)lf(x)) , (1.5)

on any given input x, with the input located in the first register and the output
stored in a second register. Then by the linearity of quantum mechanics, the
action of Uf on the equal superposition of the input register plus an extra output
register will produce

1 N-l 1 N-l
Uf : .;N ~ Ix)IO) -t .;N ~ Ix)lf(x)) . (1.6)

That is, all possible function evaluations have been done in a single step. This
massive parallelism comes for free if one could ever build a quantum computer.
Unfortunately, we are unable to extract all this information in a single ob-
servation of the resulting quantum state. Instead, we will only be able to
probabilistically extract the information encoded in a single "branch" or term
in this superposition. Naively then the quantum parallelism is all lost at the
 Quantum computing with qubits 7

stage of measurement. Fortunately, this need not be the case. Instead, by

performing a further unitary transformation before the final measurement one
may extract some information of this parallel information. Thus, typically
algorithms performed on a quantum computer are probabilistic in nature.
Although a quantum computer cannot solve any problem a conventional
computer fails at, there are a number of problems that they can solve far more
efficiently. Efficiency of a computation is measured by the computational
complexity which, roughly speaking, is how many steps are required to solve a
given problem as a function of the size of the input (specifying the problem).
For example, if an input is specified by a single number N, then the size
of the input corresponds to the length of this number L = log N. If the
computation runs in a number of steps which is a polynomial in L, then the
problem is considered tractable. If the number of steps grows exponentially,
the problem is considered hard. In recent years there have been three important
quantum computational algorithms discovered: the Deutsch-Jozsa algorithm,
Shor's algorithm and Grover's algorithm. (For details of these algorithms see
Ref. [9].)

• Deutsch-Jozsa algorithm: Here one aims to determine some "global"

information about a binary function f (x). In particular, we are given a
promise that either the function yields a constant (0 or 1) for all inputs,
or the function is balanced, with exactly half the outputs O. We wish
to determine whether the function is constant or balanced. For inputs
consisting of n bits, x E {O, 1,2, ... ,2n - I}, a conventional computer
would take 0(2n) steps to decide. But on a quantum computer, the
Deutsch-Jozsa algorithm can tell us the answer in a single evaluation of
the function [6]! Thus, to determine the global nature of such a function
precisely could be achieved with an exponential speed-up in a quantum
computer (however, this exponential gap vanishes if we allow classically
probabilistic algorithms).
• Shor's algorithm: Here one aims to factorize a composite number x.
In general this is believed to be computationally intractable. On a con-
ventional computer the currently best known algorithm takes essentially
an exponential number of steps 0(exp[2Ll/3(logL)2/3]), L = logx.
Shor discovered that a quantum computer can do the job in a polynomial
number 0(L3) [7]. For example, to factor a number with 250 deci-
mal digits, an estimate of around 800,000 years has been suggested as
required for a large network of currently available classical computers.
By contrast, on a quantum computer, this could be done in seconds to
hours depending on the computer's clock speed. Shor's landmark work
generated wide interest in the field for physicists, computer scientists and
mathematicians alike.
8 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

• Grover's algorithm: In this algorithm one aims to find one particular

item from a large unsorted (virtual) database containing N items. Classi-
cally, one needs to query O(N) times from the database in order to find a
specific marked item. Quantum mechanically, on the other hand, one can
complete the search in O(..fN) steps [8]. More precisely, one models
a database by a virtual table or function f(x), x E {O, 1, ... ,N - I}.
We are told that f(x) = 1 for a single x = y and 0 for all other values.
The problem is to find this y. Grover proved that quantum mechan-
ics helps to find the desired item from the virtual database in O(..fN)
steps. This corresponds to a square-root speed-up. Since unstructured
searches are so ubiquitous throughout computation, Grover's algorithm
has a potentially important role.

These discoveries herald radically new ways of thinking about computation,

information, and programming in general. It is worth mentioning that all
three of these algorithms have been implemented (though some might say
merely simulated) on primitive quantum computers. There have been various
proposals for building a quantum computer but a full scale machine appear
to be far off. The experimental proposals include isolating and manipulating
qubits in ion traps, solid state based devices such as SQUIDS, quantum dots,
NMR techniques and many more (see for example Ref. [10]). In part I of
this book we shadow many of the developments found in qubit-based quantum
computers with analogous work based on quantum continuous variables.

References
[1] R. Feynman, Found. of Physics 16, 507 (1986).
[2] D. Deutsch, Proc. R. Soc. London. A, 400, 97 (1985).
[3] G. Boole, An Investigation of The Laws of Thought reprinted as (Dover,
New York, 1958).
[4] B. Schumacher, Phys. Rev. A 51,2738 (1995).
[5] J. Bell, Speakables and Unspeakables in Quantum Theory (Oxford Uni-
versity Press).
[6] D. Deutsch and R. Jozsa, Proc. R. Society (London) A 439,553 (1992).
[7] P. Shor, in Proc 35th Annual Symp. on Found. of Compo Sci. (IEEE
Computer Society Press, 1994).
[8] L. K. Grover, Phys. Rev. Lett. 79, 325 (1997).
[9] M. A. Nielsen and 1. L. Chuang, Quantum Computation and Quantum
Information (Cambridge University Press, Cambridge, England, 2000).
[10] H. K. Lo and S. L. Braunstein (Eds), Scalable Quantum Computers
(Wiley-VCH Verlag, Berlin, 2001).
Chapter 2

QUANTUM COMPUTATION
OVER CONTINUOUS VARIABLES*

Seth Lloyd
MIT Department of Mechanical Engineering
MIT 3-160, Cambridge, Mass. 02139, USA
[email protected]

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 I UT, United Kingdom
[email protected]

Abstract This paper provides necessary and sufficient conditions for constructing a uni-
versal quantum computer over continuous variables. As an example, it is shown
how a universal quantum computer for the amplitudes of the electromagnetic
field might be constructed using simple linear devices such as beam splitters and
phase shifters, together with squeezers and nonlinear devices such as Kerr-effect
fibers and atoms in optical cavities. Such a device could in principle perform
"quantum floating point" computations. Problems of noise, finite precision, and
error correction are discussed.

Quantum computation has traditionally concerned itself with the manipu-

lation of discrete systems such as quantum bits, or "qubits" [1, 2]. Many
quantum variables, such as position and momentum or the amplitudes of elec-
tromagnetic fields, are continuous. Although noise and finite precision make
precise manipulations of continuous variables intrinsically more difficult than
the manipulation of discrete variables, because of the recent developments in
quantum error correction [3, 4, 5] and quantum teleportation [6, 7] of contin-
uous quantum variables, it is worthwhile addressing the question of quantum
computation over continuous variables.

·S. Lloyd and S. L. Braunstein, Physical Review Letters 82,1784-1787 (1999).

Copyright (1999) by the American Physical Society.

9
10 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

At first it might seem that quantum computation over continuous variables

is an ill-defined concept. First consider quantum computation over discrete
variables. A universal quantum computer over discrete variables such as qubits
can be defined to be a device that can by local operations perform any desired
unitary transformation over those variables [1,2,8]. More precisely, a universal
quantum computer applies "local" operations that effect only a few variables at
a time (such operations are called quantum logic gates): by repeated application
of such local operations it can effect any unitary transformation over a finite
number of those variables to any desired degree of precision. Now consider
the continuous case. Since an arbitrary unitary transformation over even a
single continuous variable requires an infinite number of parameters to define,
it typically cannot be approximated by any finite number of continuous quan-
tum operations such as, for example, the application of beam splitters, phase
shifters, squeezers, and nonlinear devices to modes of the electromagnetic field.
It is possible, however, to define a notion of universal quantum computation
over continuous variables for various subclasses of transformations, such as
those that correspond to Hamiltonians that are polynomial functions of the op-
erators corresponding to the continuous variables: A set of continuous quantum
operations will be termed universal for a particular set of transformations if one
can by a finite number of applications of the operations approach arbitrarily
closely to any transformation in the set.
This paper provides necessary and sufficient conditions for universal quan-
tum computation over continuous variables for transformations that are poly-
nomial in those variables. Such a continuous quantum computer is shown
to be capable in principle of performing arithmetical manipulations of con-
tinuous variables in a "quantum floating point" computation. In principle, a
continuous quantum computer could perform tasks that a discrete quantum
computer cannot. In practice, noise and finite precision make quantum float-
ing point operations, like their classical counterparts, effectively discrete. A
quantum computer that uses continuous variables cannot therefore perform a
task that a discrete quantum computer cannot. However, continuous quan-
tum computers may still be able to perform some tasks more efficiently than
their discrete counterparts. The results derived here apply to any collection of
continuous variables, including phonons, photons, Josephson Junction circuits,
Bose-Einstein condensates, etc. To be concrete, as results are derived they will
be expressed both in terms of abstract continuous variables and in the familiar
context of quadrature amplitudes of the electromagnetic field.
Consider a single continuous variable corresponding to an operator X. Let P
be the conjugate variable: [X, P] = i. For example, X and P could correspond
to quadrature amplitudes of a mode of the electromagnetic field (the quadrature
amplitudes are the real and imaginary parts of the complex electric field).
First investigate the problem of constructing Hamiltonians that correspond to
 Quantum computation over continuous variables 11

arbitrary polynomials of X and P. It is clearly necessary that one be able to

apply the Hamiltonians ±X and ±P themselves. In the Heisenberg picture,
applying a Hamiltonian H gives a time evolution for operators A = i[H, A],
so that A(t) = eiHt A( 0) e- iHt . Accordingly, applying the Hamiltonian X
for time t takes X --+ X, P --+ P - t, and applying P for time t takes
X --+ X + t, P --+ P: the Hamiltonians X and P have the effect of shifting
the conjugate variable by a constant. In the case of the electromagnetic field,
these Hamiltonians correspond to linear displacements or translations of the
quadrature amplitudes.
A simple geometric construction allows one to determine what Hamiltonian
transformations can be constructed by the repeated application of operations
from some set. Apply the Hamiltonian A for time 8t, followed by B, -A, - B,
each for the same time. Since

in the limit that 8t --+ 0, the result is the same as if one had applied the Hamil-
tonian irA, B] for time 8t 2 . In general, if one can apply a set of Hamiltonians
{±Hi}, one can construct any Hamiltonian that is a linear combination of
Hamiltonians of the form ±i[Hi, Hj], ±[Hi' [Hj, Hk]], etc. [9,10,11,12,13],
and no other Hamiltonians. That is, one can construct the Hamiltonians in the
algebra generated from the original set by commutation. This key point, orig-
inally derived in the context of quantum control and discrete quantum logic,
makes it relatively straightforward to determine the set of Hamiltonians that
can be constructed from simpler operations.
Now apply this result to the continuous variables introduced above. Since
[X, P] = i, the application of the translations ±X and ±P for short periods
of time clearly allows the construction of any Hamiltonian aX + bP + c that
is linear in X and P; this is all that it allows. To construct more complicated
Hamiltonians one must also be able to perform operations that are higher order
polynomials in X and P. Suppose now that one can apply the quadratic
Hamiltonian H = (X2 + p 2)/2. Since P = i[H, P] = X, X = i[H, X] =
- P, application of this Hamiltonian for time t takes X --+ cos tX - sin tP,
P --+ cos tP + sin tX. If X and P are quadrature amplitudes of a mode of
the electromagnetic field, then H is just the Hamiltonian of the mode (with
frequency w = 1) and corresponds to a phase shifter. Hamiltonians of this
form can be enacted by letting the system evolve on its own or by inserting
artificial phase delays. Note that since eiHt is periodic with period 1/47r, one
can effectively apply - H for a time 8t by applying H for a time 47r - 8t. The
simple commutation relations between H, X and P imply that the addition
of ±H to the set of operations that can be applied allows the construction of
Hamiltonians of the form aH + bX + cP + d.
12 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Suppose that in addition to translations and phase shifts one can a{lply
the quadratic Hamiltonian ±S = ±(XP + PX)/2. S has the effect X =
i[S, X] = X, P = i[S,P] = -P, i.e., applying +S takes X -t etX,
P -t e- t P: S "stretches" X and "squeezes" P by some amount. Similarly
-S squeezes X and stretches P. In the case of the electromagnetic field,
S corresponds to a squeezer operating in the linear regime. It can easily be
verified that [H, S] = i(X2 - p2). Looking at the algebra generated from
X, P, H and S by commutation, one sees that translations, phase shifts, and
squeezers allow the construction of any Hamiltonian that is quadratic in X and
P, and of no Hamiltonian of higher order.
To construct higher order Hamiltonians, nonlinear operations are required.
One such operation is the "Kerr" Hamiltonian H2 = (X2 + p2)2, corre-
sponding to a X3 process in nonlinear optics. This higher order Hamilto-
nian has the key feature that whereas commuting the previous Hamiltoni-
ans, X, P, H, S with some polynomial in X and P resulted in a polyno-
mial with the same or lower order, commuting H2 with a polynomial in X
and P typically increases its order. By evaluating a few commutators, e.g.,
[H2, X] = i(X2p+PX 2 +2p3)/2, [H 2,P] = -i(P2X +XP2 +2X 3)/2,
[X, [H2, S]] = p3, [P, [H2, S]] = X 3 one sees that the algebra generated
by X, P, H, Sand H2 by commutation includes all third order polynomials
in X and P. A simple inductive proof now shows that one can construct
Hamiltonians that are arbitrary Hermitian polynomials in any order of X and
P. Suppose that one can construct any polynomial of order M or less, where
M is of degree at least 3. Then since [P 3, pm xn] = ip m+2X n- 1 + lower
order terms, and [X 3, pm xn] = ipm-l xn+2+ lower order terms, one can
by judicious commutation of X 3 and p 3 with monomials of order M construct
any monomial of order M + 1. Since any polynomial of order M + 1 can
be constructed from monomials of order M + 1 and lower, by applying linear
operations and a single nonlinear operation a finite number of times one can
construct polynomials of arbitrary order in X and P to any desired degree of
accuracy. Comparison with similar results for the discrete case [14] shows that
the number of operations required grows as a small polynomial in the order of
the polynomial to be created, the accuracy to which that polynomial is to be
enacted, and the time over which it is to be applied.
The use of the Kerr Hamiltonian H2 was not essential: any higher order
Hamiltonian will do the trick. Note that commutation of a polynomial in X
and P with X and P themselves (which have order 1) always reduces the order
of the polynomial by at least 1, commutation with H and S (which have order
2) never increases the order, and commutation with a polynomial of order 3
or higher typically increases the order by at least 1. Judicious commutation
of X, P, Hand S with an applied Hamiltonian of order 3 or higher therefore
 Quantum computation over continuous variables 13

allows the construction of arbitrary Hermitian polynomials of any order in X

andP.
The above set of results shows that simple linear operations, together with a
single nonlinear operation, allow one to construct arbitrary polynomial Hamil-
tonian transformations of a single quantum variable. Let us now tum to more
than one variable, e.g., the case of an interferometer in which many modes of
the electromagnetic field interact. Suppose now that there are many variables,
{Xi, Pd, on each of which the simple single-variable operations described
above can be performed. Now let the variables interact with each other. For
simplicity, we assume that we can apply interaction Hamiltonians of the form
±Bij = ±(PiXj - XiPj): a more complicated interaction Hamiltonian can
always be used to generate interactions of this form by combining it with single-
variable operations. Since Xi = i[Bij,Xi] = Xj Xj = i[Bij,Xj] = -Xi,
Pi = i[Bij, Pi] = Pj, Pj = i[Bij, Pj ] = -Pi, this operation has the effect
of taking Ai -+ cos tA + sin tAj, Aj -+ cos tAj - sin tAi' for Ai = Xi, Pi,
A j = Xj, Pj. For the electromagnetic field, Bij functions as a beam splitter,
linearly mixing together the two modes i and j. By repeatedly taking commu-
tators of Bij with polynomials in Xi, Pi, for different i, it can be easily seen by
the same algebraic arguments as above that it is possible to build up arbitrary
Hermitian polynomials in {Xi, Pi}.
This concludes the derivation of the main result: simple linear operations on
continuous variables, together with any nonlinear operation and any interaction
between variables suffice to enact to an arbitrary degree of accuracy Hamilto-
nian operators that are arbitrary polynomials of the set of continuous variables.
In the case of modes of the electromagnetic field, linear operations such as
translations, phase shifts, squeezers, and beam splitters, combined with some
nonlinear operation such as a Kerr nonlinearity, allow one to perform arbitrary
polynomial transformations on those modes. Note that in contrast to the case
of qubits, in which a nonlinear coupling between qubits is required to perform
universal quantum computation, in the continuous case only single variable
nonlinearities are required, along with linear couplings between the variables.
In analog with information over classical continuous variables, which is mea-
sured in units of "nats" (l nat = log2e bits), the unit of continuous quantum in-
formation will be called the "qunat." Two continuous variables in the pure state
l~h2 possess -trpIlnpI qunats of entanglement, where PI = tr21~h2(~I.
For two squeezed vacua (squeezed by an amount e- r ) entangled using a beam
splitter as in Refs. [5-7] the entropy so computed from the approximate EPR
state is given by

S(p) = (1 + n) In(1 + n) - nlnn qunats (2.2)

with n = er sinh r. For example, e2r = 10 gives 10 dB of squeezing in power,

corresponding to r = 1.15129. By Eq. (2.2), two continuous variables entan-
14 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

gled using a 10 dB squeezer then possess 2.60777 qunats of shared, continuous

quantum information, equivalent to 3.76221 qubits of discrete quantum infor-
mation. This is comparable to the degree of entanglement currently available
using ion-trap quantum computers.
Quantum computation over continuous variables can be thought of as the sys-
tematic creation and manipulation of qunats. Universal quantum computation
for polynomial transformations of continuous variables effectively allows one to
perform quantum floating point manipulations on those variables. For example,
it is clearly possible using linear operations alone to take the inputs Xl, X 2 and
to map them to Xl, aX l + bX2 + c. Similarly, application ofthe three-variable
Hamiltonian X 1X 2P3 takes Xl -T Xl, X 2 -T X 2, X3 -T X3 + X1X2t: that
is, this operation allows one to multiply Xl and X 2 and place the result in the
"register" X 3 . A wide variety of quantum floating point operations are possible.
Any polynomial transformation of the continuous variables is clearly possible,
as is any transformation that can be infinitesimally represented by a conver-
gent power series. Just as classical computation over continuous variables in
principle allows one to solve problems more rapidly than is possible digitally
[14], it is interesting to speculate that quantum computation over continuous
variables might in principle allow the solution of problems more rapidly than
is possible using a "conventional," discrete quantum computer. Continuous
variable computation has its own set of problems that might be sped up by the
application of continuous quantum computation: For example, such a continu-
ous quantum computer might be used to investigate continuous NP-complete
problems such as the four-feasibility problem, that is, the problem of deciding
whether or not a real degree 4 polynomial in n variables has a zero [15]. In
practice, of course, due to finite precision a continuous quantum computer will
effectively be able to solve the same set of problems that a conventional discrete
quantum computer can, although it may be able to perform some operations
more efficiently.
The ability to create and manipulate qunats depends crucially on the strength
of squeezing and of the nonlinearities that one can apply. 10 dB squeezers
(6 dB after attenuation in the measurement apparatus) currently exist [16].
High Q cavity quantum electrodynamics can supply a strong Kerr effect in
a relatively lossless context, and quantum logic gates constructed for qubits
could be used to provide the nonlinearity for continuous quantum variables as
well [17]. Here the fact that only single-mode nonlinearities are required for
universal quantum computation simplifies the problem of effecting continuous
quantum logic. Nonetheless, the difficulty of performing repeated nonlinear
operations in a coherent and loss-free manner is likely to limit the possibilities
for quantum computation over the amplitudes of the electromagnetic field.
Vibrational modes of ions in traps or excitations of a Bose-Eistein condensate
 Quantum computation over continuous variables 15

might provide the long-lived, lossless states required for quantum computation
over continuous variables.
Noise poses a difficult problem for quantum computation [18, 19,20], and
continuous variables are more susceptible to noise than discrete variables.
Since an uncountably infinite number of things can go wrong with a contin-
uous variable, it might at first seem that continuous error correction routines
would require infinite redundancy. In fact, continuous quantum error correc-
tion routines exist and require no greater redundancy than conventional routines
[3, 4, 5]. Such routines are capable of correcting for noise and decoherence
in principle: In practice, measurement noise, losses, and the lack of perfect
squeezing will lead to imperfect error correction [5]. Surprisingly, continuous
quantum error correction routines are in some sense easier to enact than discrete
quantum error correction routines, in that the continuous routines can be imple-
mented using only linear operations together with classical feedback [5]. The
relative simplicity of such routines suggests that robust, fault-tolerant quantum
computation may in principle be possible for continuous quantum variables
as well as for qubits. (A scheme for quantum computation is fault-tolerant
if quantum computations can be carried out even in the presence of noise and
errors [21, 22]. A fault-tolerant scheme that allows for arbitrarily long quantum
computations to be carried out is said to be robust [23].) If this is indeed the
case then quantum computation over continuous variables, despite its intrinsic
difficulties, may be an experimentally viable form of quantum information pro-
cessing. Continuous variables might be used to simulate continuous quantum
systems such as quantum field theories. Even in the absence of fault tolerance,
the large bandwidths available to continuous quantum computation make it
potentially useful for quantum communications and cryptography [24].

S.L. would like to thank H. Haus and H. J. Kimble for useful discussions. This
work was supported by DARPA under the QUIC initiative.

References
[1] D. DiVincenzo, Science 270,255 (1995).
[2] S. Lloyd, Sci. Am. 273, 140 (1995).
[3] S. Lloyd and J. J.-E. Slotine, Phys. Rev. Lett. 80,4088 (1998).
[4] S. L. Braunstein, Phys. Rev. Lett. 80, 4084 (1998).
[5] S. L. Braunstein, Nature (London) 394, 47 (1998).
[6] S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998).
[7] A. Furusawa, et at, Science 282, 706 (1998).
[8] This definition of quantum computation corresponds to the normal "cir-
cuit" definition of quantum computation as in, e.g., D. Deutsch, Proc.
16 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

R. Soc. London A, 425, 73 (1989); A. C.-c. Yao, in Proceedings of the

36th Annual Symposium on Foundations of Computer Science, edited by
S. Goldwasser, (IEEE Computer Society, Los Alamitos, CA, 1995), pp.
352-361. The works of M. Reck et aI., Phys. Rev. Lett. 73, 58 (1994),
and N. J. Cerf, C. Adami, and P. G. Kwiat, Phys. Rev. A 57, R1477
(1998), showing how to perform arbitrary unitary operators using only
linear devices such as beam splitters, though of considerable interest and
potential practical importance, does not constitute quantum computation
by the usual definition. Reck et al. and Cerf et al. propose performing
arbitrary unitary operations on N variables not by acting on the variables
themselves but by expanding the information in the variables into an inter-
ferometer with O(2N) arms and acting in this exponentially larger space.
Local operations on the original variables correspond to highly nonlocal
operations in this "unary" representation: To flip a single bit requires one
to act on half [O(2N-l)] of the arms of the interferometer. Actually to
perform quantum computation on qubits using an interferometer requires
nonlinear operations as detailed in Y. Yamamoto, M. Kitagawa, and K.
Igeta, in Proceedings of the 3rd Asia-Pacific Physics Conference, edited
by Y. W. Chan, A. F. Leung, C. N. Yang, K. Young (World Scientific,
Singapore, 1988), pp. 779-799; G. J. Milburn, Phys. Rev. Lett. 622124
(1989).
[9] G. M. Huang, T. J. Tam, J. W. Clark, J. Math. Phys. (N.Y.) 24,2608-2618
(1983).
[10] Differential Geometric Control Theory, edited by R. W. Brockett, R. S.
Millman and H. J. Sussman, (Birkhauser, Boston, 1983); Nonholonomic
Motion Planning, edited by Z. Li and J. F. Canney (Kluwer Academic,
Boston, 1993).
[11] v. Ramakrishna, M. V. Salapaka, M. Dahleh, H. Rabitz and A. Peirce,
Phys. Rev. A 51, 960-966 (1995).
[12] S. Lloyd, Phys. Rev. Lett. 75, 346-349 (1995).
[13] D. Deutsch, A. Barenco and A. Ekert, Proc. R. Soc. London A 449,
669-677 (1995).
[14] S. Lloyd, Science 273, 1073 (1996).
[15] L. Blum, M. Shub and S. Smale, Bull. Am. Math. Soc. 21,1-46 (1989).
[16] L. A. Wu et al., Phys. Rev. Lett. 57, 2520 (1986).
[17] Q. A. Turchette, et ai, Phys. Rev. Lett. 75, 4710-4713 (1995).
[18] R. Landauer, Nature (London) 335, 779-784 (1988).
[19] R. Landauer, Phys. Lett. A 217, 188-193 (1996).
[20] R. Landauer, Phil. Trans. R. Soc. London A 335,367-376 (1995).
 Quantum computation over continuous variables 17

[21] P. Shor, Proceedings of the 37th Annual Symposium on the Foundations of

Computer Science (IEEE Computer Society Press, Los Alamitos, 1996),
pp.56-65.
[22] D. P. DiVincenzo and P. W. Shor, Phys. Rev. Lett. 77, 3260-3263 (1996).
[23] R. Laflamme, M. Knill and W.H. Zurek, Science 279, 342 (1998); D.
Aharanov and Ben-Or, quant-ph; J. Preskill, Proc. R. Soc. London A 454,
385 (1998).
[24] C. H. Bennett and G. Brassard, in Proceedings of the IEEE International
Conference on Computers, Systems, and Signal Processing, Bangalore,
India, (IEEE Press, New York, 1984), pp. 175-179. A. K. Ekert, et ai,
Phys. Rev. Lett. 69, 1293 (1992); P. D. Townsend, J. G. Rarity and P.
R. Tapster, Electronics Letters 29, 1291 (1993); R. J. Hughes, et at, in
Advances in Cryptology: Proceedings of Crypto 96, (Springer-Verlag,
New York, 1997), pp. 329-343; A. Muller et al. Appl. Phys. Lett. 70, 793
(1997).
Chapter 3

ERROR CORRECTION FOR

CONTINUOUS QUANTUM VARIABLES*

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 1 UT, United Kingdom
[email protected]

Abstract We propose an error correction coding algorithm for continuous quantum vari-
ables. We use this algorithm to construct a highly efficient 5-wave-packet code
which can correct arbitrary single wave-packet errors. We show that this class of
continuous variable codes is robust against imprecision in the error syndromes.
A potential implementation of the scheme is presented.

Quantum computers hold the promise for efficiently factoring large integers
[1]. However, to do this beyond a most modest scale they will require quantum
error correction [2]. The theory of quantum error correction is already well
studied in two-level or spin-! systems (in terms of qubits or quantum bits)
[2, 3, 4, 5, 6, 7]. Some of these results have been generalized to higher-spin
systems [8, 9, 10, 11]. This work applies to discrete systems like the hyperfine
levels in ions but is not suitable for systems with continuous spectra, such as
unbound wave packets. Simultaneously with this paper, Lloyd and Slotine
present the first treatment of a quantum error correction code for continuous
quantum variables [12], demonstrating a 9-wave-packet code in analogy with
Shor's 9-qubit coding scheme [2]. Such codes hold exciting prospects for
the complete manipulation of quantum systems, including both discrete and
continuous degrees of freedom, in the presence of inevitable noise [13].
In this Letter we consider a highly efficient and compact error correction cod-
ing algorithm for continuous quantum variables. As an example, we construct
a 5-wave-packet code which can correct arbitrary single-wave-packet errors.
We show that such continuous variable codes are robust against imprecision in

* s. L. Braunstein, Physical Review Letters 80, 4084-4087 (1998).

Copyright (1998) by the American Physical Society.

19
20 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

the error syndromes and discuss potential implementation of the scheme. This
paper is restricted to one-dimensional wave-packets which might represent the
}\lave function of a nonrelativistic one-dimensional particle or the state of a
single polarization of a transverse mode of electromagnetic radiation. We shall
henceforth refer to such descriptions by the generic term wave packets [14].
Rather than starting from scratch we shall use some of the theory that has
already been given for error correction on qubits. In particular, Steane has
noted that the Hadamard transform,

(3.1)

maps phase flips into bit flips and can therefore be used to form a class of
quantum error correction codes that consist of a pair of classical codes, one for
each type of "flip" [3]. This mapping between phase and amplitude bases is
achieved with a rotation about the y-axis by 7f /2 radians in the Bloch sphere
representation of the state. In analogy, the position and momentum bases of a
continuous quantum state may be transformed into each other by 7f /2 rotations
in phase-space. This transition is implemented by substituting the Hadamard
rotation in the Bloch sphere by a Fourier transform between position and
momentum in phase-space. This suggests that we could develop the analogous
quantum error correction codes for continuous systems [15].
We shall find it convenient to use a units-free notation where

position = x x (scale length)

momentum = p / (scale length) , (3.2)

where x is a scaled length, p is a scaled momentum and we have taken Ii = ~.

(We henceforth drop the modifier "scaled.") The position basis eigenstates Ix)
are normalized according to (x'I x) = 8(x' - x) with the momentum basis given
by

(3.3)

To avoid confusion we shall work in the position basis throughout and so define
the Fourier transform as an active operation on a state by

(3.4)

where both x and yare variables in the position basis. Note that Eqs. (3.3)
and (3.4) correspond to a change of representation and a physical change of the
state, respectively.
 Error correction/or continuous quantum variables 21

In addition to the Fourier transform we shall require an analog to the bit-

wise exclusive-OR (XOR) gate for continuous variables. The XOR gate has
many interpretations including controlled-NOT gate, addition modulo 2 and
parity associated with it. Of these interpretations the natural generalization to
continuous variables is addition without a cyclic condition, which maps

lx, y) -7 lx, x + y) . (3.5)

By removing the cyclic structure of the XOR gate we have produced a gate
which is no longer its own inverse. Thus, in addition to the Fourier transform
and this generalized XOR gate we include their inverses to our list of useful
gates. This generalized XOR operation performs translations over the entire
real line, which are related to the infinite additive group on JR.. The characters
X of this group satisfy the multiplicative property X(x + y) = X(x)x(y) for all
x, y E JR. and obey the sum rule

11
-
7f
00

-00
dx X(x) = J(x) , (3.6)

where X( x) = e2ix . Interestingly, this sum rule has the same form as that found
by Chau in higher-spin codes [10). Once we have recognized the parallel, it
is sufficient to take the code of a spin-~ system as a basis for our continuous-
variable code.
Based on these parallel group properties, we are tempted to speculate a
much more general and fundamental relation: We conjecture that n-qubit error
correction codes can be paralleled with n-wave-packet codes by replacing
the discrete-variable operations (Hadamard transform and XOR gate) by their
continuous-variable analogs (Fourier transform, generalized-XOR and their
inverses). As a last remark before embarking on the necessary substitutions (in
a specific example), we point out that the substituti?n conjecture is only valid
for qubit codes whose circuits involve only these (H and XOR) elements. We
shall therefore restrict our attention to this class of codes.
An example of a suitable 5-qubit code was given by Laflamme et al. [16). We
show an equivalent circuit in Fig. 3.1 [17]. As we perform the substitutions,
we must determine which qubit-XOR gates to replace with the generalized-
XOR and which with its inverse. To resolve this ambiguity, two conditions
are imposed. First, we demand that the code retain its properties under the
parity operation (on each wave packet). We conclude that either gate may be
chosen for the first operation on initially zero-position eigenstates. Ambiguity
remains for the last four XOR substitutions. As a second step, the necessary
and sufficient condition for quantum error correction [5,6],
22 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

must be met. Here IXencode) encodes a single wave-packet's position eigen-

state in a multi-wave-packet state, to. is a possible error that can be handled
by the code and Ao.{3 is a complex constant independent of the encoded states.
[Condition (3.7) says that correctable errors do not mask the orthogonality of
encoded states.]

I~)

10)
10)
10)
10)
Figure 3.1 Quantum error correction circuit from [17]. The qubit I'I/J) is rotated into a 5-particle
subspace by the unitary operations represented by the operations shown in this circuit. Note that
the 3-qubit gates are simply pairs of XORs.

In the case of a single wave-packet error, for our 5-wave-packet code, it turns
out that amongst the conditions ofEq. (3.7) only (x~n odeltlo. t5{3IXencode)'
having errors on wave packets 4 and 5, is affected by t~e ambiguity (see detail
below). An explicit calculation of all the conditions shows that the circuit of
Fig. 3.2 yields a satisfactory quantum error correction code (as do variations of
this circuit due to the extra freedom with respect to the choice of operator acting
on wave-packets 1-3). By analogy with the results for higher-spin codes, we
know that this code is optimal (though not perfect) and that no four-wave-packet
code would suffice [10]. The code thus constructed has the form

Ix encod)
e
= _1_
7[3/2
J dw dy dz e2i (wy+xz)

xlz,y+X,w+X,w-z,y-z). (3.8)
 Error correction/or continuous quantum variables 23

----------------~F ~.--

10)

10)

10)

10)
Figure 3.2 This "circuit" unitarily maps a one-dimensional single-wave-packet state 1'1/1) into
a 5-wave-packet error correction code. Here the auxiliary wave packets 10) are initially zero-
position eigenstates. For degrees-of-freedom larger than qubits the ideal XOR is not its own
inverse; here the daggers on the XOR gates represent the inverse operation.

Let us demonstrate the calculation of one of the conditions specified by

Eq. (3.7):
, At A
(xencode 1£4a £5{3lxencode) (3.9)

= Z wy +xz-w '"
- 1 / dw' dy' dz' dw dy dz e2"( y -x z ')
11"3

xc5(z' - z) c5(y' - y + x' - x) o(w' - w + x' - x)

x (w' - z'lt!lw - z)(y' - z'lt{3ly - z)
-2i(x' _x)2 /
e dw dy dz e2i (x' -x)(w+y-z)
11"3

x(w - x' + x - zlt!lw - z)(y - x' + x - zlt{3ly - z}.

Making the replacements w --+ w + z and y --+ y + z in this last expression we
obtain
-2i(x' _x)2 /
e dw dy dz e2i (x' -x)(w+y+z)
11"3

x(w - x' + xlt!lw}(y - x' + xlt{3ly} (3.10)

O(x'-x)/ At A
11"2 dw dy (wl£al w ) (YI£{3IY) == o(x' - x) Aa{3 .
24 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

For the other cases we find by explicit calculation, for wave-packets j =I- k, that
I At A _ I
(xencodel£ja £k,Blxencode) - <5(x - x) )..a,B • (3.11)

For j = k this constant is found to be

(3.12)

where C is formally infinite.

We shall argue that this infinity vanishes when the syndrome is read with only
finite precision, which is always going to be the real situation. However, this
requires us to demonstrate that our codes are robust: that for a sufficiently good
precision we may correct single-wave-packet errors to any specified accuracy.
In order to understand how the error syndromes are measured, let us consider a
simpler code, namely, the continuous version of Shor's original 9-qubit code,

Ix d ) = _1_! dwdydz e 2ix (w+y+z)

enco e 7r 3/ 2

x Iw, w, w, y, y, y, z, z, z) , (3.13)

where parity alone removes all ambiguity. (This code has been independently
obtained by Lloyd and Slotine [12].) Since this 9-wave-packet code corrects
position errors and momentum errors separately, it is sufficient to study the
subcode

IXencode) = lx, x, x) , (3.14)

designed to correct position errors on a single wave packet. The most general
position error (on a single wave packet) is given by some function of the
momentum of that system E(p) and need not be unitary on the code subspace
[Eq. (3.7)]. The action of such an error on a wave packet may be written in the
position basis as

E(p)lx) = ~ ! dy dp e 2ip (y-x) £(p)ly) = ! dy £(y)lx - y), (3.15)

where £(x) is the Fourier transform of £(p). Thus the most general position
error looks like a convolution of the wave packet's ket with some unknown
(though not completely arbitrary) function. Suppose this error occurs on wave
packet 1 in the repetition code (3.14). Further, let us use auxiliary wave-packets
(so-called ancillae) and compute the syndrome as shown in Fig. 3.3, then the
resulting state may be written as

! dy£(y)lx-y,x,x,-y,O,y). (3.16)
 Error correction/or continuous quantum variables 25

t
10)
t
10) '- '-1./

t
10) I'~ I'
'-

Figure 3.3 Syndrome calculation and measurement: A state with a sing1e-wave-packet position
error (here on wave packet 1) enters and the differences of each pair of positions is computed.
The syndrome {81, 82,83} may now be directly measured in the position basis.

Everything up till now has been unitary and assumed ideal. Now measure
the syndrome: Ideally it would be { -y, 0, y} collapsing the wave packet for
a specific y. Correcting the error is now easy, because we know the location,
value and sign of the error. Shifting the first wave packet by the amount y
retrieves the correctly encoded state lx, x, x) Note that this procedure uses only
very simple wave-packet-gates: The comparison stage is done classically, in
contrast to the scheme of Lloyd and Slotine, where the comparison is performed
at the amplitude level and involves significantly more complicated interactions
[12].
It is now easy to see what imprecise measurements of the syndromes will do.
Suppose each measured value of a syndrome sj is distributed randomly about
the true value sj according to the distribution Pmeas (sj - sj). We find two
conditions for error-correction to proceed smoothly. First, Pmeas (x) must be
narrow compared to any important length scales in t (x). This guarantees that
the chance for "correcting" the wrong wave packet is negligible and reduces the
position-error operator to an uninteresting prefactor. If the original unencoded
J
state had been dx ~ (x) Ix) then after error correction we would obtain the
mixed state

J dx ' dx dz ~(x) ~* (x') Pmeas(z) Ix - z, x, x) (x' - z, x', x'I· (3.17)

Thus, unless Pmeas (x) is also narrow compared to any important length scales
in ~ (x), decoherence will appear in the off-diagonal terms for wave packet 1
of the corrected state (3.17). This second condition is also seen in the quantum
teleportation of continuous variables due to inaccuracies caused by measure-
26 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

ment [13]. These conditions roughly match those described by Lloyd and
Slotine [12]. We note that any syndrome imprecision will degrade the encoded
states, though this precision may be improved by repeated measurements of the
syndromes. For our 5-wave-packet example (3.8), syndromes consist of sums
of two or more wave-packet positions or momenta and are measured similarly.
It should be noted that Chau's higher-spin code [10] could have been imme-
diately taken over into a quantum error correction code for continuous quantum
variables in accordance with our substitution procedure. However, we have pro-
duced an equivalent code with a more efficient circuit prescription: Whereas
Chau gives a procedure for constructing his higher-spin code using 9 general-
ized XOR operations, the circuit in Fig. 3.2 requires only 7 such gates or their
inverses. In fact, we could run this substitution backwards to obtain a cleaner
5-particle higher-spin code based on Eq. (3.8).
In order to consider potential implementations of the above code let us restrict
our attention to a situation where the wave packets are sitting in background
harmonic-oscillator potentials. By the virial theorem the form of a wave packet
in such a potential is preserved up to a trivial rotation in phase-space with
time. The two operations required may be implemented simply as follows:
The rotation in phase-space, Eq. (3.4), may be obtained by delaying the phase
of one wave packet relative to the others, and the XOR operation, Eq. (3.5),
should be implemented via a quantum non-demolition (QND) coupling. There
exists extensive experimental literature on these operations both for optical
fields and for trapped ions [13, 18, 19, 20, 21].
The conjecture put forth in this Letter leads to a simple, 2-step design of
error correction codes for continuous quantum variables. According to this
conjecture, any qubit code, whose circuit operations include only a specific
Hadamard transformation, its inverse and the ideal XOR, may be translated
to a continuous quantum-variable code, by substituting these operators with
their continuous analogs and then imposing two criteria - parity invariance
and the error-correction condition - which remove any ambiguities in the
choice of operators. We demonstrate the success of this coding procedure
in two examples (one based on Shor's 9-qubit code [2], and a second based
on a variation of the Laflamme et al. 5-qubit code [16, 17]). The 5-wave-
pa.:;ket code presented here is the optimal continuous encoding of a single
one-dimensional wave packet that protects against arbitrary single-wave-packet
errors. We show that this code (and in fact the entire class of codes derived
in this manner) are robust against imprecision in the error syndromes. The
potential implementation of the proposed class of circuits in optical-field and
ion-trap set-ups is an additional incentive for further investigation of the robust
manipulation of continuous quantum variables.
 Error correctionjor continuous quantum variables 27

This work was funded in part by EPSRC grant GR/L91344. The author appreci-
ated discussions with N. Cohen, H. J. Kimble, D. Gottesman and S. Schneider.

Appendix: Addendum
The above work demonstrates that an error correction code may be designed,
though for implementation it requires some kind of QND couplings. Soon after
this work was complete the author realized that much of the same work could
be done with straight linear optics, in particular coupling based upon beam
splitter operations [22].
First, consider two light fields Ix) and Iy) incident on an "ideal" (phase-free)
beam splitter. The output light fields are given by

812 (0) Ix, y) = Ix cos 0 - y sinO, y cos 0 + x sinO) , (3.A.l)

where the subscripts 12 refer to the wave packets acted upon. From these ideal
beam splitters we construct a 3-port device called a tritter [23]

~ ~ ~ -1 1
7123 = 8 23 (11"/4) 8 12 {cos y'3), (3.A.2)

from which we may construct the three wave-packet subcode of Shor's 9-wave-
packet code via

(3.A.3)

In fact, the only difference between this code and the "ideal" version of it
Ix, x, x) is a simple common scaling in each of the code's wave packets.
Finally, we combine these elements together with this freedom to scale, in
order to produce our encoding device out of linear optics. A suitable choice for
the 9-wave-packet code is a 9-port beam-splitter, which we call a nona-splitter

(3.A.4)

This device could be implemented as a series of eight ordinary beam splitters

or as a single (mass-produced) integrated-optics element [22].
More recently, a more robust set of quantum error-correcting codes over
continuous variables have been constructed which protect the state of a finite
dimensional quantum system from decoherence [24]. The advantage of this
variation over those described in this chapter are that they allow the effective
protection against small "diffusive" errors [24], which are closer to typical
realistic loss mechanisms. In the codes studied above, small errors comparable
or smaller than readout errors cannot be corrected and are additive with each
"protective" operation.
28 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

References
[1] P. W. Shor in Proc. 35th Annual Symposium on the Foundations of Com-
puter Science, edited by S. Goldwasser (IEEE Computer Society Press,
Los Alamitos, California, 1994), p.124.
[2] P. W. Shor, Phys. Rev. A 52, R2493 (1995).
[3] A. M. Steane, Proc. Roy. Soc. London 452, 2551 (1996).
[4] A. R. Calderbank and P. W. Shor, Phys. Rev. A 54, 1098 (1996).
[5] c. H. Bennett, D. P. DiVincenzo, J. A. Smolin and W. K. Wootters, Phys.
Rev. A 54,3824 (1996).
[6] E. Knill and R. Laflamme, Phys. Rev. A, 55,900 (1997).
[7] A. R. Calderbank, E. M. Rains, P. W. Shor and N. J. A. Sloane, Phys. Rev.
Lett. 78, 405 (1997).
[8] E. Knill, LANL report LAUR-96-2717, preprint quant-phl9608048.
[9] H. F. Chau, Phys. Rev. A 55, R839 (1997).
[10] H. F. Chau, Phys. Rev. A 56, Rl (1997).
[11] E. M. Rains, LANL preprint quant-phl9703048.
[12] S. Lloyd and J.-J. E. Slotine, LANL preprint quant-phl9711021.
[13] S. L. Braunstein and H. J. Kimble, "Teleportation of continuous quantum
variables," Phys. Rev. Lett., submitted; S. L. Braunstein, H. J. Kimble, Y.
Sorensen, A. Furusawa and N. Ph. Georiades, "Teleportation of continu-
ous quantum variables," IQEC 1998, abstract submitted.
[14] Although we consider multi-wave-packet states as one-dimensional we
would not want them to physically overlap so they could, for example, be
displaced one from another in an orthogonal direction.
[15] An example using as few as 7 qubits to correct an arbitrary single-qubit
error in a l-qubit encoded state can be found in [3].
[16] R. Laflamme, C. Miquel, J. P. Paz and W. H. Zurek, Phys. Rev. Lett. 77,
198 (1996).
[17] S. L. Braunstein and J. A. Smolin, Phys. Rev. A 55,945 (1997).
[18] S. F. Pereira, Z. Y. Ou and H. J. Kimble, Phys. Rev. Lett. 72, 214 (1994).
[19] K. Bencheikh, j. A. Levenson, P. Grangier and O. Lopez, Phys. Rev. Lett.
75,3422 (1995).
[20] R. L. de Matos Filho and W. Vogel, Phys. Rev. Lett. 76, 4520 (1996).
[21] R. Bruckmeier, H. Hansen and S. Schiller, Phys. Rev. Lett. 79, 1463
(1997).
[22] S. L. Braunstein, Nature (London) 394, 47 (1998).
 Error correctionjor continuous quantum variables 29

[23] M. Zukowski, Laser Phys. 4, 690 (1994).

[24] D. Gottesman, A. Kitaev and J. Preskill, "Encoding a qubit in an oscilla-
tor," quant-phl0008040.
Chapter 4

DEUTSCH-JOZSA ALGORITHM
FOR CONTINUOUS VARIABLES

Arun K. Pati
Institute of Physics, Bhubaneswar-751005, Orissa, INDIA
Theoretical Physics Division, BARe, Mumbai, INDIA

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 1 Ur, United Kingdom
[email protected]

Abstract We present an idealized quantum continuous variable analog of the Deutsch-

Jozsa algorithm which can be implemented on a perfect continuous variable
quantum computer. Using the Fourier transformation and XOR gate appropriate
for continuous spectra we show that under ideal operation to infinite precision
that there is an infinite reduction in number of query calls in this scheme.

In principle, quantum computers can have remarkable computational powers

which classical computers cannot [1, 2]. In the last few years it has been shown
that it is possible for quantum computers to perform certain computational tasks
faster than any classical computer [3, 4, 5, 6, 7, 8, 9]. Quantum computation
exploits quantum interference and entanglement to outperform its classical
counterparts. The first algorithm promising benefits from quantum parallelism
was discovered by Deutsch and Jozsa [6]. Soon after this Shor discovered [10]
his now famous algorithm for factoring large numbers [11]. Subsequently, a
fast quantum search algorithm was discovered by Grover [12, 13]; in addition,
the time-dependent generalization of Grover's algorithm and its stability under
unitary perturbation has been studied [14].
It may be mentioned that all these algorithms are usually designed for qubits.
However, in nature there are other classes of quantum systems whose observ-
abIes form, for example, continuous spectra. Usually a continuous variable can
be anything, e.g., position, momentum, energy (unbounded), the amplitudes of
the electromagnetic field, etc. It is important to know how these algorithms can
31
SL Braunstein andA.K. Pati (eds.), Quantum Information with Continuous Variables, 31-36.
© 2003 Kluwer Academic Publishers.
32 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

be generalized to continuous quantum variables. By learning how, we might

make progress towards discovering new algorithms which are perhaps more
naturally formulated using continuous variables.
Recently continuous quantum information has played an important role in
teleportation [15] and even error-correction codes [16, 17] with a possible
implementation using linear devices [18]. Moreover, quantum computation
over continuous variables has also been studied. It was found that the universal
continuous variable quantum computation can be effected using simple non-
linear operations with coupling provided solely by linear operations [19]. Just as
standard quantum computation can be thought of as the coherent manipulation
of two-level systems (qubits), continuous quantum computation can be thought
of as the manipulation of "qunats."
The first algorithm to have been studied for potential implementation us-
ing continuous quantum variables was the Grover virtual database search [20].
Here, we go on to generalize the Deutsch-Jozsa algorithm to continuous vari-
ables. This scheme naively gives an infinite speed-up over classical function
evaluation.
To start with, let us recall the standard Deutsch-Jozsa algorithm for qubits.
In this case we are given a number i E {O, ... ,2n - 1} = En and a "black

°
box" or "oracle query" that computes a binary function f (i) : En -t E.
Further, the function f which only takes values or 1 is promised to be either
constant or balanced (with an equal number of each type of outcome over all
input strings). The aim is to determine this property for f, i.e., whether it is
constant or balanced. On a classical computer in the worst case the oracle query
requires O(2n) function evaluations. However, if one calculates the function
using reversible quantum operations then only a single function evaluation is
required to achieve the goal [6, 21].
In the continuous variable setting we pose the problem in the following
way. Suppose there is a particle located somewhere along the x-axis. Since
x E lR. is a continuous variable it can take value from -00 to +00 (in practice
it may be from - L to L, where L is some length scale involved, but still the
number of possible values of x is infinite). Suppose there are two persons
Alice and Bob playing a game [22]. Alice tells Bob a value of x and Bob
calculates some function f(x) which takes values 0 or 1. Further, Bob has
promised Alice that he will use a function which is either constant or balanced.
A constant function is 0 or 1 for all values of x E (-L, +L). For a balanced
function, f (x) = 0 or 1 for exactly half of the cases. One can define the
balanced function more precisely in the following manner. Imagine that the
interval for the continuous variable x has been divided into n sub-intervals.
Let p, be the Lebesgue measure on R A function f(x) is balanced provided
the Lebesgue measure of the support for where the function is zero is identical
to the Lebesgue measure of the support for where the function is one, i.e.,
 Deutsch-Iozsa algorithm for continuous variables 33

JL{{x E IRlf{x) = O}) = JL{{x E IRlf{x) = 1}). Now, Alice wants to

know whether Bob has chosen a constant or balanced function. In the classical
scenario since there are an infinite number of possiblities for x Alice needs to
ask Bob (who has the oracle) an infinite number of times! However, we can
show if we use a perfect continuous variable quantum computer and unitary
operators that can be implemented on them, then a single function evaluation
is required to know this global property of the function.
Let us consider a continuous variable system whose Hilbert space is infinite
dimensional and spanned by a basis state Ix) satisfying the Dirac orthogonality
condition (xix') = 6{x-x'). In a continuous variable scheme a basic operation
is the Fourier transformation between position and momentum variables in
phase space (analog to the Walsh-Hadamard transformation for qubits). We
can define the Fourier transformation as an active operation on a qunat state
Ix) as

Fix) = ~/ dy e2iXY ly) , (4.1)

where both x and y are in the position basis. This has been used in developing
error correction codes [16, 18] and Grover's algorithm for continuous variables
[20]. This Fourier transformation can be easily applied in physical situations.
For example, when Ix) represents quadrature eigenstate of a mode of the
electromagnetic field, Fix) is simply an eigenstate of the conjugate quadrature
produced by a 'If' /2 phase delay.
Another useful gate on a continuous variable quantum computer is XOR
gate (analogous to the controlled NOT gate for qubits but without the cyclic
condition) defined as [18]

Ix) Jy) -+ Jx) Jx + y) . (4.2)

Further, we assume that given a classical circuit for computing f (x) there is a
quantum circuit which can compute a unitary transformation Uj on a continuous
variable quantum computer. If a quantum circuit exists that transforms

Jx)Jy) -+ Jx)Jy + f{x)) , (4.3)

then by linearity it can also act on any superposition of qunat states. For
example, if we evaluate the function on a state (4.1) along with another qunat
state Jz), we have

Uj{FJx)Jz)) = ~/ dy e2ixY Jy)Jz + f{y)) . (4.4)

This shows that using quantum parallelism for idealized qunat computers one
can evaluate all possible values of a function simultaneously with one applica-
tion ofUj.
34 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Now, we present the Deutsch-Jozsa algorithm for a continuous variable

quantum computer. The set of instructions for deciding the constant or balanced
nature of function f(x) are give below (and is illustrated as a quantum circuit
in Fig. 4.1):

17r/2) ---I 1----- FI7r /2)

Figure 4.1 Quantum cirquit for the continuous-variable Deutsch-Iozsa algorithm.

(i) Alice stores her query in a qunat register prepared in an ideal position
eigenstate Ixo) and attaches another qunat in a position eigenstate 17r/2). So
the two quants are in the state Ixo)I7r/2)
(ii) She creates superpositions of qunat states by applying the Fourier trans-
formation to the query qunat and the target qunat. The reSUlting state is given
by

Fl xo)FI7r/2) = ~! dxdye2ixox+i7TYlx)ly) . (4.5)

(iii) Bob evaluates the function using the unitary operator U!. The state
transforms as

~ ! dx e2ixox+i7T!(X) Ix)FI7r /2) . (4.6)

Here, the key role is played by the ancilla qunat state 17r /2). To see how the
function evaluation takes place consider the intermediate steps given by

If the function f (x) = 0 there is no sign change and if f (x) = 1 there is a sign
change. After the third step performed by Alice, she has a quant state in which
the result of Bob's function evaluation is encoded in the amplitude of the qunat
superposition state given in (4.6). To know the nature of the function she now
performs an inverse Fourier transformation on her qunat state.
(iv) The qunat states after the inverse Fourier tranform is given by

Iq) = ~ ! dx dx' e 2ix (xo-x') (_l)!(x) Ix')FI7r /2) . (4.8)

 Deutsch-lozsa algorithm/or continuous variables 35

(v) Alice measures her qunat by projecting onto the original position eigen-
state Ixo). In an ideal continuous variable scheme the correct projection oper-
ator is defined as [23]

P.6.xo = 1XO +.6.XO/2

xo-l:!.xo/2
dyly)(yl· (4.9)

As has been explained in [20, 23] if the observable has a continuous spectrum
then the measurement cannot be performed precisely but must involve some
spread L\xo. Therefore, the action of projection onto the qunat state after step
(iv) is given by

(4.10)

Now consider two possibilities. If the function is constant then the above
equation r~duces to ±lxo)FI.1T /2). [In simplifying we ~eed to use th~ Di~ac
delta function (1/71') f dx e2~x(xO-Y) = 8(xo - y).] ThiS means that if Allce
measures Ixo) she is sure that f(x) is definitely constant. In the other case, i.e.,
when the function is balanced she will not get the measurement outcome to be
Ixo). In fact, in the balanced case the outcome is orthogonal to the constant
case as the result gives zero. Therefore, a single function evaluation (follwed
by a measurement onto Ixo)) in a qunat quantum computer can decide whether
the promised function is constant or balanced. Unlike the qubit case, in the
idealized continuous variable case the reduction in the number of query calls
is from infinity to one.
In conclusion, we have generalised the primitive quantum algorithm (Deutsch-
Jozsa algorithm) from the discrete case to the idealized continuous case. It may
be worth mentioning that as in error correction codes for continuous-variablees
[16], if one replaces the Hadamard transform and XOR gate by their continuous-
variable analogs in original Deutsch-Jozsa algorithm for qubit case, then the
idealized algorithm works perfectly. This theoretically demonstrates the power
of quantum computers to exploit the superposition principle giving an infinite
speed up compared to classical scenario. This idealized analysis has not con-
sidered the affects of finite precision in measurement or state construction and
so whether it may be implemented remains an open question for further study.
Part of the difficulty in extending this work in this direction is that defining an
oracle for continuous variables appears to be a difficult task, one that we have
carefully avoided here. An alternate way forward might be to consider some
sort of "hybrid" approach involving both qunats and qubits. This is precisely
what Seth Lloyd considers in the following chapter.
36 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

AKP thanks P. van Loock and R. Simon for useful feedback. AKP also thanks
G. Giedke for discussions during Benasque Science Center-2000 in Spain on
defining balanced function for continuous variables.

References
[1] S. Lloyd, Science, 261, 1569 (1993).
[2] S. Lloyd, Science, 273, 1073 (1996).
[3] P. Benioff, Phys. Rev. Lett. 48, 1581 (1982).
[4] R. Feynman, Int. J. Theor. Phys. 21, 467 (1982).
[5] D. Deutsch, Proc. R. Soc. London A 400, 97 (1985).
[6] D. Deutsch and R. Jozsa, Proc. R. Soc. London, A 439,553 (1992).
[7] E. Bernstein and U. Vazirani, in Proc. of the 25th Annual Symposium on
the Theory of Computing (ACM Press, New York, 1993), p. 11-20.
[8] D. R. Simon, in Proc. of the 35th Annual Symposium on Foundations
of Computer Science, edited by S. Goldwasser (IEEE Computer Society,
Los Alamitos, CA, 1994), p. 116-123.
[9] D. DiVincenzo, Science, 270, 255 (1995).
[10] P. W. Shor, in Proceedings of the 37th Annual Symposium on Foundations
of Computer Science (IEEE Computer Society Press, Los Alamitos, CA,
1996), pp. 56-65.
[11] A. Ekert and R. Jozsa, Rev. Mod. Phys. 68, 733 (1996).
[12] L. K. Grover, Phys. Rev. Lett. 79, 325 (1997).
[13] L. K. Grover, Phys. Rev. Lett. 80, 4329 (1998).
[14] A. K. Pati, "Grover's algorithm, time-dependent search and unitary per-
turbation" (preprint), (1999).
[15] S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998).
[16] S. L. Braunstein, Phys. Rev. Lett. 80,4084 (1998).
[17] S. Lloyd and J.-J. E. Slotine, Phys. Rev. Lett. 80, 4088 (1998).
[18] S. L. Braunstein, Nature (London) 394, 47 (1998).
[19] S. Lloyd and S. L. Braunstein, Phys. Rev. Lett. 82,1784 (1999).
[20] A. K. Pati, S. L. Braunstein and S. Lloyd, quant-ph/0002082.
[21] R. Cleve, A. Ekert, C. Macciavello and M. Mosca, Proc. R. Soc. London
A 454, 339 (1998).
[22] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum
Information, (Cambridge University Press, 2000).
[23] C. Cohen-Tannoudji, B. Diu and F. Laloe, Quantum mechanics (John
Wiley & Sons, New York, 1977).
Chapter 5

HYBRID QUANTUM COMPUTING

Seth Lloyd
d'Arbeloff Laboratory for Information Systems and Technology
Department of Mechanical Engineering
Massachusetts Institute of Technology
Cambridge, Massachusetts 02139
[email protected]

Abstract Necessary and sufficient conditions are given for the construction of a hybrid
quantum computer that operates on both continuous and discrete quantum vari-
ables. Such hybrid computers are shown to be more efficient than conventional
quantum computers for performing a variety of quantum algorithms, such as
computing eigenvectors and eigenvalues.

Quantum computers are devices that process information in a way that

preserves quantum coherence [1,2,3,4,5,6,8,9, 10, 11]. The most common
model of quantum computation deals with coherent logical operations on two-
state quantum variables known as qubits. Quantum computation can also be
performed on variables with three or more states, and is well-defined even
when the underlying degrees of freedom are continuous [12, 13, 14, 15]. This
paper investigates hybrid quantum computers that operate on both discrete and
continuous quantum variables. It is shown that a simple set of operations (hybrid
quantum logic gates) can be used to approximate arbitrary tranformations of
the variables. Hybrid versions of quantum algorithms are discussed and a
hybrid version of an algorithm for finding eigenvalues and eigenvectors is
presented. Hybrid quantum algorithms can have a number of advantages over
conventional quantum algorithms, including lower computational complexity
and an enhanced resistance to noise and decoherence.
The primary reason for investigating hybrid quantum computers is that na-
ture contains both discrete quantum variables such as nuclear spins, photon
polarizations, and atomic energy levels, and continuous variables such as po-
sition, momentum, and the quadrature amplitudes of the electromagnetic field.
In conventional quantum computation, continuous variables are something of a
37
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 37-45.
© 2003 Kluwer Academic Publishers.
38 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

nuisance: either they figure as sources of noise and decoherence, as in the case
of environmental baths of harmonic oscillators, or they must be restricted to a
discrete set of states by cooling, as in the case of the oscillatory modes of ions
in ion-trap quantum computers. In hybrid quantum computation, by contrast,
the full range of continuous quantum variables can be put to use.
The basic model for performing quantum computation using a hybrid of
continuous and discrete variables follows the normal model for performing
quantum computation using discrete or continuous variables on their own [7,
15]. Assume that one has the ability to 'tum on' and 'tum off' the members of a
set Hamiltonian operators {±Hj}, corresponding to the ability to apply unitary
transformations of the form e±~Hjt. The set of transformations that can be
constructed in this fashion is the set of transformations of the form e- iHt where
His a member of the algebra generated from theHj via commutation: i.e., since
eiH2teiHlte-iH2te-iHlt = e-[Hl,H2]t 2 + O(t 3 ), the ability to tum on and tum
off ±H1 and ±H2 allows one effectively to tum on and off H = ±i[H1 , H 2],
etc. Transformations of the form e- iHt for non-inifinitesimal t can then be
built up from infinitesimal transformations to any desired degree of accuracy.
For the sake of ease of exposition, concentrate here on discrete variables
(qubits) that are spins, characterized by the usual Pauli operators (Yx, (Y y, (Y z and
to continuous variables (qunats) that are harmonic oscillators characterized by
the usual annihilation and creation operators a, at ([a, at] = 1), and by the
'position' and 'momentum' operators X = (a + at )j2, P = (a - a t )j2i,
([X, P] = i). It is convenient to think of the harmonic oscillators as modes
of the electromagnetic field with X and P proportional to the quadrature
amplitudes of the mode. The generalization to discrete variables with more
than two states and to other forms of continuous variable is straightforward and
will be discussed below.
To perform quantum computations one must be able to prepare one's vari-
ables in a desired state, perform quantum logic operations, and read out the
results. Assume that it is possible to prepare the discrete variables in the state
10) == I t)z, and the continuous variables in the vacuum state 10): alO) = O.
Assume that it is possible to measure (Y z for the discrete variables and X for
the continuous variables.
N ow look at performing transformations of the variables. Begin with just a
pair - one spin and one oscillator. Suppose that one can tum on and tum off
the Hamiltonians

(5.1)

As will now be seen, the ability to tum on and off Hamiltonians from this set
allows one to enact Hamiltonians that are arbitrary polynomials of the (Y's, X
and P. Note that these Hamiltonians all represent interactions between qubits
and oscillators: this is physically realistic in the sense that transformations on
 Hybrid quantum computing 39

physical spins or atoms are accomplished by making the spins interact with the
electromagnetic field, and vice versa. In physically realizable situations, such
as the ion traps and optical cavities discussed below, the interactions in 5.1 are
turned on and off by applying laser or microwave pulses to couple discrete to
continuous degrees of freedom.
Now investigate what can be accomplished by turning on and off these inter-
actions. If the spin is prepared in the state IO}, then turning on the Hamiltonian
a zP is equivalent to turning on the Hamiltonian P for the oscillator on its own.
The Hamiltonian X can be turned on in a similar fashion. In order to apply this
Hamiltonian for a finite amount of time, the spin must be constantly reprepared
in the state IO} or new spins in this state must be supplied. This operation allows
the construction of coherent states of the oscillator.
Now start constructing effective Hamiltonians by the method of commutation
above. Since i[P, axX] = ax, we can effectively tum on the Hamiltonian ax.
Similarly for the Hamiltonian ±i[P, azX] = ±az . And since i[az , ax] = 2a y ,
any single qubit transformation e-iO"t E SU(2) can be enacted by turning on
and off Hamiltonians in the set. Since i[azP, azX] = 2, an arbitrary overall
phase can also be turned on and off. That is, we can enact arbitrary single qubit
transformations.
Now systematically build up higher order transformations. Since i[azX,
axX] = 2ayX2, and i[ayX2, axX] = 2az X 3 , etc., we can effectively tum on
and off Hamiltonians of the form aX n , for arbitrary a, n. Similarly, we can
tum on and off Hamiltonians of the form a pn. By preparing the spin in the state
IO} and turning on and off the Hamiltonians azxm, azpn, we can enact sin-
gle oscillator transformations corresponding to Hamiltonians that are arbitrary
Hermitian polynomials in X and P. (Not all such Hamiltonians are bounded.
Nonetheless, one can build up infinitesimal versions of such Hamiltonians and
apply them for finite time to states for which they are bounded.)
So the simple set of Hamiltonians above allows the construction of arbitrary
single qubit transformations and arbitrary polynomial transformations of the
continuous variable, along with arbitrary interactions between the spin and the
oscillator. Let us now look at more than one spin and one oscillator.
Since i[a~P, a;X] = a~a;, we can tum on the interaction Hamiltonian
a~a; between two spins 1 and 2 by making them both interact with the same
oscillator. But the ability to tum on this Hamiltonian together with the ability to
tum on arbitrary single-spin Hamiltonian translates into the ability to perform
arbitrary transformations on sets of spins: that is, one can perform arbitrary
quantum logic operations on the qubits alone.
Similarly, since i[ayXl' a x X 2] = 2azX1X2, the ability to make two os-
cillators interact with the same spin, initially in the state IO}, allows one to
tum on the Hamiltonian X 1 X2 between the two oscillators 1 and 2. But this
ability, together with the ability to tum on single oscillator Hamiltonians that
40 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

are arbitrary Hermitian polynomials in X and P, translates into the ability to

turn on Hamiltonians that are arbitrary Hermitian polynomials of Xi, Pi for all
the oscillators together. So one can perform universal quantum computation
on the continuous variables on their own.
Continuing with constructing Hamiltonians via commutation, the ability
to prepare the 10) states for spins and oscillators, together with the ability
to turn on and off the simple set 5.1 of Hamiltonians given above, allows
one to effectively turn on and off Hamiltonians that are arbitrary Hermitian
polynomials in 1, o{, at, at:, Xi:' , P'k. That is, one can perform universal
quantum computation on the hybrid quantum computer.
How might such a hybrid quantum computer be realized? As it turns out,
many existing designs for quantum computers are easily modified to perform
hybrid quantum computation. For example, ion trap quantum computers [8, 10]
operate by coupling together the internal states of ions in an ion trap (qubits) via
their motional state (harmonic oscillators). Existing schemes for performing
quantum computation using ion traps only use the ground and first excited
state of the oscillator corresponding to the fundamental mode of the ions in
the trap, effectively treating the oscillator as a qubit. But the same methods
that are used to couple the ions to the oscillator can just as well be used to
apply the Hamiltonians in the set 5.1 above. An ion trap with many ions has
many modes, each of which can be used as a continuous variable in the hybrid
quantum computation. Similarly, the Pellizzari scheme for coupling together
trapped atoms (qubits) via a cavity mode of the electromagnetic field can readily
be altered to use the quadrature amplitudes of the modes of the cavity, rather
than simply using the lowest two energy eigenstates of a mode as a qubit [11].
Other potential continuous variables that might be used for hybrid quantum
computation are the translational states of atoms in a Bose condensate, the
continuum states of electrons in semiconductors, or the state of a Josephson
junction circuit. Essentially any hybrid system that affords precise control over
the interactions between discrete and continuous variables is a good candidate
for a hybrid quantum computer.
An important concern in the construction of hybrid quantum computers is
the problem of noise and decoherence. At first it might seem that continuous
variables are likely to be more susceptible to noise than discrete variables.
It is indeed true that more things can go wrong with a continuous variable
than with a discrete variable. However, quantum error correction routines
for continuous variables have been developed and require no greater overhead
than those for discrete variables [12, 13, 14, 16]. Although these routines
are not yet technologically practical on existing devices, it may well be that
improved versions of these routines combined with existing discrete quantum
error correction routines will allow efficient quantum error correction for hybrid
devices. In addition, as noted above, hybrid devices have the advantage that
 Hybrid quantum computing 41

they include in the computation states and degrees of freedom that would
normally be sources of noise, decoherence, and loss.
Now tum to applications of hybrid quantum computers. Where does the
ability to perform manipulations of continuous variables as well as qubits give
an advantage? The first point to note in constructing hybrid algorithms is that
we must be careful to assume physically reasonable uses of hybrid variables-
i.e., uses that do not require infinite or exponentially high precision. Even in
the classical case, the use of continuous variables can give remarkable compu-
tational speed ups (the ability to solve NP-complete problems in polynomial
time, the ability to find the the answer to uncomputable problems in finite
time, etc.) if one allows arbitrary precision in manipulating and measuring
continuous variables. By giving an explicit construction of the operations that
can be used to perform continuous variable and hybrid quantum computation,
however, we have implicitly avoided the use of infinite or excessive precision:
all such operations would require infinite or excessive computational resources
to construct, manipulate, and measure the desired over-precise states.
With this caveat in mind, tum to the operations that are relatively easy to
perform using continuous quantum variables. A particularly useful subroutine
in a variety of quantum algorithms is the quantum Fourier transform: Ix) -+
z=~= 1 e ixy Iy). In the case of discrete quantum variables the quantum Fourier
transform on N qubits takes on the order of N quantum logic operations to
perform. Although this is an efficient algorithm it is nonetheless difficult at
present to perform quantum Fourier transforms on more than a few qubits (the
current record is three) [17]. By contrast, in the case of the continuous quantum
variables X and P, the quantum Fourier transform is trivial. If the eigenstates
of X with eigenvalue x are written Ix), then theeigenstates of P with eigenvalue
p can be written Ip) = (1/V2n) f~oo eiPXlx)dx. That is, the eigenstates of
P are the quantum Fourier transform of the eigenstates of X. Coupling to P
instead of X then allows immediate access to the Fourier transformed variable.
The quantum Fourier transform on a continuous variable is accomplished by a
zero-step operation. The ease of performing the quantum Fourier transform on
continuous variables suggests that in devising algorithms for hybrid quantum
computers we look for problems in which the quantum Fourier transform plays
a central role.
Perhaps the best known quantum algorithm in which the quantum Fourier
transform plays a central role is Shor's algorithm for factoring large numbers
[4]. Setting aside the difficulty of performing the other operations in this
algorithm (such as modular exponentiation), it is immediately clear that using
a continuous variable as the register on which to perform the quantum Fourier
transform in Shor's algorithm would require an exponentially high precision in
the preparation and manipulation of the continuous variable. (Hybrid quantum
42 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

computation might still be used to speed up some aspects of Shor's algorithm;

this possibility will be investigated elsewhere.)
A second problem in which the quantum Fourier transform plays a key role
is that of simulating the dynamics of quantum systems [1, 18, 19, 20, 21].
Comparison with [18] shows that the ability of hybrid quantum computers
to turn on and off simple Hamiltonians involving a few discrete and a few
continuous variables at a time translates into the ability to perform efficient
quantum simulations of hybrid systems.
A particularly valuable type of quantum simulation is one that allows the
computation of spectra: using methods developed in [22, 23, 20] Abrams and
Lloyd have developed algorithms for computing eigenvalues and eigenvectors
of quantum systems and for obtaining improved estimates of the ground state
[21]. In its original discrete form, the algorithm is somewhat involved. How-
ever, the fact that quantum Fourier transforms are straightforward to perform on
continuous variables makes the Abrams-Lloyd algorithm particularly simple
in the case of hybrid quantum computation. Here we show how to perform
a quantum computation that computes the eigenvectors of a hybrid system
and that writes the eigenvalues of the system onto a register consisting of a
single continuous variable. The algorithm is a hybrid version of the discrete
algorithms proposed in [22, 23, 20, 21] and is closest in form to the discrete
algorithm proposed in [20] for simulating von Neumann measurements on a
quantum computer. Independently, Travaglione and Milburn [24] have shown
how methods of hybrid quantum computation can be used to compute the
eigenvectors of a continuous system and write the eigenvalues onto a discrete
register.
First, prepare a single continuous variable such as a mode of the electro-
magnetic field in the squeezed state Ix = 0) = (1/V27f) J~oo Ip)dp. In any
practical experiment, of course, such perfectly squeezed states are unavailable.
Imperfectly squeezed or un squeezed states will also work, however. As dis-
cussed below, the effect of imperfect squeezing is to decrease the resolution to
which the spectrum can be obtained. Prepare a second system in the state 1'IjJ)
whose decomposition into energy eigenstates IV;) = Li 'ljJiIEi) one wishes to
obtain. Here we assume that the system is discrete; in general, however, system
may be continuous, discrete, or a hybrid of continuous and discrete variables.
Next, using the methods of hybrid quantum computation described above,
couple the system to the continuous variable via the coupling Hamiltonian H P,
where H is the Hamiltonian whose eigenvalues and eigenvectors are to be ob-
tained. For H to be efficiently simulatable, it must be equal to Lk Hko where
each Hk acts on only a few variables at a time. Since H P = Lk HkP, if His
efficiently simulatable, so is H P, by the methods of hybrid quantum computa-
tion described above. Writing H = Lj Ej IEj) {Ej I, the time evolution of the
 Hybrid quantum computing 43

state of the system and the continuous variable is

1'I/J)lx = 0)
-+ e-iHPtl'I/J)lx = 0)
= e- iEj EjIEj){EjIPt L: j 'l/JjIEj) IX = 0)
L:j e-iEjtP'l/JjIEj)lx = 0)
= L:j 'l/JiIEj) Ix = Ejt) , (5.2)

since e -iPt Ix) = Ix + t). Clearly, at this point, a measurement of the variable
X on the continuous variable will yield the result x = tEi with probability
'l/Ji, leaving the system in the state lEi = x/t). That is, one can sample
the spectral decomposition of I'I/J), obtaining the eigenvalues Ei together with
their corresponding weights l'l/Jil 2 and eigenvectors lEi). The process is highly
efficient, requiring only the ability to prepare the initial squeezed state Ix = 0)
and to apply the Hamiltonian H P.
The hybrid eigenvalue and eigenvector finding algorithm using a continuous
variable to register the eigenvalue is more efficient than the corresponding
algorithm using qubits to register the eigenvalue. Since the quantum Fourier
transform is performed implicitly in the continuous register, fewer steps are
required in the hybrid algorithm. In addition, unlike the conventional version
of the algorithm, the hybrid version is insensitive to approximate decoherence
of the register in the course of the computation: measuring the value x of the
register in the course of the coupling does not affect the ability of the algorithm
to find eigenvectors and eigenvalues.
The requirement that the initial state of the continuous variable be perfectly
squeezed can also be relaxed. Suppose that the initial state is in a Gaussian state
J e-.B x2 / 2 Ix)dx. For example, {3 = 1 gives the un squeezed n = 0 vacuum
state, while {3 > 1 gives partial squeezing in X. With this initial state for the
continuous variable, after the algorithm has been run, the continuous variable
and the system are in the state

~
J
f e-.B x2 / 2 IE j) Ix + Ejt)dx. (5.3)

That is, the eigenvalues and eigenvectors are resolved to within an accuracy
1/t..J!3. By coupling the system to the continuous variable for a sufficiently
long time, the eigenvectors and eigenvalues of H may be determined to an
arbitrary degree of accuracy, even when the initial state is unsqueezed. Note
that resolving the eigenvalues of a system with an exponentially large number of
states requires exponential squeezing of the pointer state. But as noted in [21],
this algorithm still provides a potentially exponential speedup over classical
44 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

algorithms even when the eigenvalues are not determined to an exponential

degree of accuracy.
Hybrid quantum computers are devices that perform quantum computations
using both discrete variables such as spins and continuous variable such as
position and momentum, or the quadrature amplitudes of the electromagnetic
field. Hybrid quantum computation represents a natural extension of quantum
computation using quantum bits alone: as the example of finding eigenvalues
and eigenvectors presented here shows, hybrid quantum computations can be
more efficient and less sensitive to noise and decoherence than conventional
quantum computations. Nature supplies us with both discrete and continuous
quantum variables: it is advantageous to use them.

This work was supported by DARPAIARO under the QUIC initiative.

References
[1] R. P. Feynman, Optics News 11, 11 (1985); reprinted in Found. Phys. 16,
507 (1986).
[2] D. Deutsch, Proc. R. Soc. London A 400,97-117 (1985).
[3] S. Lloyd, Science 261, 1569-1571 (1993).
[4] P. Shor, in Proceedings 0/ the 35th Annual Symposium on Foundations
o/Computer Science, edited by S. Goldwasser, (IEEE Computer Society,
Los Alamitos, CA, 1994), pp. 124-134.
[5] S. Lloyd, Sci. Am. 273, 140-145 (1995).
[6] D. Divincenzo, Science 270,255-261 (1995).
[7] S. Lloyd, Phys. Rev. Lett. 75, 346 (1995).
[8] J. I. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091-4094 (1995).
[9] Q. A. Turchette, C. J. Hood, W. Lange, H. Mabuchi and H. J. Kimble,
Phys. Rev. Lett. 75, 4710-4713 (1995).
[10] C. Monroe, D. M. Meekhof, B. E. King, W. M. Itano and D. J. Wineland,
Phys. Rev. Lett. 75, 4714-4717 (1995).
[11] T. Pellizzari, S. A. Gardiner, J. I. Cirac and P. Zoller, Phys. Rev. Lett. 75,
3788-3791 (1995).
[12] S. Lloyd and J. J.-E. Slotine, Phys. Rev. Lett. 80,4088 (1998).
[13] S. L. Braunstein, Phys. Rev. Lett. 80,4084 (1998).
[14] S. L. Braunstein, Nature 394,47 (1998).
[15] S. Lloyd, S. L. Braunstein, Phys. Rev. Lett. 82, 1784-1787 (1999).
[16] J. Preskill and A. Kitaev, to be published.
[17] Y. Weinstein, S. Lloyd and D. Cory, quant-ph/9906059.
 Hybrid quantum computing 45

[18] S. Lloyd, Science 273, 1073 (1996).

[19] S. Wiesner, quant-phl9603028.
[20] C. Zalka, Proc. R. Soc. London A 454,313-322 (1998).
[21] D. Abrams and S. Lloyd, Phys. Rev. Lett. 83(24), 5162 (1999).
[22] A. Yu. Kitaev, quant-phl9511026.
[23] R. Cleve, A. Ekert, C. Macchiavello and M. Mosca, Proc. R. Soc. London
A 454 (1969),339 (1998).
[24] B. C. Travaglione and G. J. Milburn, 'Generation of eigenstates using the
phase estimation algorithm,' to be published.
Chapter 6

EFFICIENT CLASSICAL SIMULATION OF

CONTINUOUS VARIABLE QUANTUM
INFORMATION PROCESSES*

Stephen D. Bartlett and Barry C. Sanders

Department of Physics and Centre for Advanced Computing
- Algorithms and Cryptography
Macquarie University
Sydney, New South Wales 2109, Australia

Samuel L. Braunstein and Kae Nemoto

Informatics, Bangor University, Bangor LL57 1 UT, United Kingdom
[email protected]

Abstract We obtain sufficient conditions for the efficient simulation of a continuous vari-
able quantum algorithm or process on a classical computer. The resulting theorem
is an extension of the Gottesman-Knill theorem to continuous variable quantum
information. For a collection of harmonic oscillators, any quantum process that
begins with unentangled Gaussian states, performs only transformations gener-
ated by Hamiltonians that are quadratic in the canonical operators, and involves
only measurements of canonical operators (including finite losses) and suitable
operations conditioned on these measurements can be simulated efficiently on a
classical computer.

Quantum mechanics allows for information processing that could not be

performed classically. In particular, it may be possible to perform an algorithm
efficiently on a quantum computer that cannot be performed efficiently on a
classical one. Significant effort is now underway to construct quantum algo-
rithms and processes that yield such a speedup. The Gottesman-Knill (GK)

• S. D. Bartlett, B. C. Sanders, S. L. Braunstein and K. Nemoto, Physical Review Letters 88, 097904/1-4
(2002).
Copyright (2002) by the American Physical Society.

47
48 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

theorem [1] for discrete-variable (qubit) quantum information provides a valu-

able tool for assessing the classical complexity of a given process. Essentially,
it states that any quantum algorithm that initiates in the computational basis
and employs only a restricted class of gates (Hadamard, phase, CNOT, and
Pauli gates), along with projective measurements in the computational basis,
can be efficiently simulated on a classical computer. (For a precise formulation
and proof of this remarkable theorem, see [2], page 464.) The GK theorem
reveals that a large class of quantum algorithms do not provide a speedup over
classical processes. In fact, recent work has placed even stronger constraints
on the potential speedup offermionic quantum computers [3].
In addition to the successes of qubit-based algorithms, quantum information
over continuous variables (CV) has yielded many exciting advances, both
theoretically and experimentally, in fields such as quantum teleportation [4,5],
quantum cryptography [6, 7,8,9], and potentially quantum computation [10].
CV algorithms could also perform computational tasks more efficiently than is
possible classically. To assess the computational complexity of these tasks, it is
necessary to develop an extension of the GK theorem: What continuous variable
processes can be efficiently simulated on a classical computer? As a CV
quantum information process involves coupled canonical systems, this question
of efficient classical simulation is related to asking under what conditions a
quantum mechanical system can be modeled by a classical one. As noted by
Feynman [11], a key advantage of a quantum computer is its ability to simulate
quantum systems that cannot be efficiently simulated classically.
The issue of efficient classical simulation of a CV process is more involved
than for the discrete case. One notable difference is that the quantum states and
the unitary transformations involved are described by real-valued (as opposed
to integer-valued) parameters, and these parameters must be described on a
discrete classical computer with some assumption of error or limited precision.
Also, the states used in CV experiments are approximations to the idealized
computational basis. These basis states are infinitely squeezed states whereas
any experimental implementation will involve finitely squeezed states thus
deviating from their idealized form [10]. A good classical simulation must
be robust against such deviations. Measurements are part of the quantum
computation and, even in the computational basis, are subject to experimental
constraints (such as photodetection efficiency). Classical simulation must also
incorporate these measurements.
Despite these complications, we prove in the following an extension of
the GK theorem for continuous variables; i.e., we present a set of sufficient
conditions for a CV quantum information process which, if satisfied, ensure that
it can be efficiently simulated on a classical computer. To prove this theorem,
we employ the techniques of stabilizers [2] that are used for qubits. Using
the stabilizer formalism, it is often possible to simulate a quantum information
Efficient classical simulation ofcontinuous variable quantum information processes 49

process by following the evolution of a set of operators, the Pauli operators,

rather than the evolution of quantum states. For CV processes, we show that it is
more natural to analyze stabilizers in terms of the algebras (i.e., Hamiltonians)
that generate them, rather than the groups themselves. We define analogs
of the Pauli and Clifford algebras and groups for CV and construct sets of
gates (as unitary transformations) that can efficiently simulate any arbitrary
transformation in these groups. Any algorithm or process constructed out of
these Clifford group transformations can be efficiently modeled by following
the evolution of the Pauli operators rather than the states of the system.
The standard Pauli group On for CV quantum computation on n coupled
oscillator systems is the Heisenberg-Weyl group [HW(n)], which consists of
phase-space displacement operators for the n oscillators. Unlike the discrete
Pauli group for qubits, the group HW(n) is a continuous (Lie) group, and can
therefore only be generated by a set of continuously-parameterized operators.
The algebra hw(n) that generates this group is spanned by the 2n canonical
operators iii, Pi, i = 1, ... ,n, along with the identity operator t, satisfying
the commutation relations [iii,Pj] = iMijt. For a single oscillator, the n = 1
algebra is spanned by the canonical operators {ii,p,i} which generate the
single oscillator Pauli operators

(6.1)

with q,p E R The Pauli operator X(q) is a position-translation operator

(translating by an amount q), whereas Z(p) is a momentum boost operator
(kicking the momentum by an amount p). These operators are non-commutative
and obey the identity

X(q)Z(P) = e-iqpZ(p)X(q). (6.2)

On the computational basis of position eigenstates {Is); s E ~} [10,12,13],

the Pauli operators act as

X(q)ls) = Is + q) ,
~
Z(P)ls) = exp(/iPs) Is) . (6.3)

Note that it is conventional to use highly squeezed states to approximate position

eigenstates; these states satisfy the orthogonality relation (sis') = o(s - s') in
the limit of infinite squeezing.
The Pauli operators for one system can be used to construct a set of Pauli
operators {Xi (qi), Zi(Pi); i = 1, ... ,n} for n systems (where each operator
labeled by i acts as the identity on all other systems j t= i). This set generates
the Pauli group On. Note that the Pauli group is only a subgroup of all possible
unitary transformations. It is not possible to construct an arbitrary unitary
transformation using only the Pauli operators X (q) and Z (P); the Pauli group
50 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

only describes transformations generated by Hamiltonians that are linear in the

canonical variables.
For issues of classical simulation, we will be interested in transformations
that lie in the Clifford group. The Clifford group N(Qn) is the group of
transformations, acting by conjugation, that preserves the Pauli group gn; i.e.,
it is the normalizer of the Pauli group in the (infinite-dimensional) group of all
unitary transformations.
Theorem 1: The Clifford group N (Qn) for continuous variables is the semidi-
rect product group [HW(n)]Sp(2n', ~), consisting of all phase-space transla-
tions along with all one-mode and two-mode squeezing transformations. This
group is generated by inhomogeneous quadratic polynomials in the canonical
operators.
Proof: The most straightforward method to identify the Clifford group will
be to identify its algebra. The Clifford algebra consists of all Hamiltonian
operators He satisfying [Hhw, He] E hw(n) for all Hhw E hw(n). This alge-
bra must obviously include the algebra hw(n), and thus hw(n) is a subalgebra
of the Clifford algebra. In addition, this algebra includes all homogeneous
quadratic polynomials in the canonical operators {qi, Pi j i = 1, . .. ,n}. This
algebra of quadratics consists of Hamiltonians that generate one-mode squeez-
ing transformations [for example, the Hamiltonian Hs = ~(qp + pq)], and
also interaction Hamiltonians that generate two-mode squeezing transforma-
tions (for example, the interaction Hamiltonian Hint = ql ® P2). The algebra
of homogeneous quadratic polynomials in the canonical operators is known as
the linear symplectic algebra sp(2n, ~).
Together, the algebras hw(n) and sp(2n,~) form a larger algebra, con-
sisting of inhomogeneous quadratic Hamiltonians in the canonical operators
{qi, Pi j i = 1,... ,n}. This algebra is the semidirect sum algebra [hw(n)]
sp(2n, ~), with hw(n) as an ideal. The group generated by this algebra is the
semidirect product group [HW(n)]Sp(2n, ~). This group includes phase-space
displacements (the Pauli group), as well as the squeezing transformations (both
single- and two-mode) of quantum optics [14]. (QED)
In order to describe a quantum information process as a circuit, it is necessary
to find a set of transformations (gates) that generate the Clifford group; these
gates will serve as building blocks for arbitrary Clifford group transformations.
Following the derivation by Gottesman et al. [15], a set of gates will be defined
in terms of the elements of the Clifford algebra (i.e., the Hamiltonians) that
generate the transformations.
The SUM gate is the CV analog of the CNOT gate and provides the basic
interaction gate for two oscillator systems 1 and 2; it is defined as

(6.4)
Efficient classical simulation ofcontinuous variable quantum information processes 51

This gate is an interaction gate operation on the Pauli group (12 for two systems.
Referring to the definition (6.1) for the Pauli operators for a single system, the
action of this gate on the 92 Pauli operators is given by

SUM: X1(q) ® 12 -+ X1(q) ® X2(q) ,

Zl (P) ® 12 -+ Zl (p) ® 12 ,
h ® X 2 (q) -+ h ® X 2 (q) ,
h ® Z2(P) -+ Zl(p)-l ® Z2(p). (6.5)

This gate describes the unitary transformation used in a backaction evasion or

quantum nondemolition process [14].
The Fourier transform F is the CV analog of the Hadamard transformation.
It is defined as

F = exp ( *~ (q2 + p2)) , (6.6)

and the action on the Pauli operators is

F : X(q) -+ Z(q) ,
Z(p) -+ X(p)-l. (6.7)

The "phase gate" P(",) is a squeezing operation for CV, defined by

P(",) = exp C/i",q2) ,

i (6.8)

and the action on the Pauli operators is

i 2
P(",) : X(q) -+ e 2r. "1Q X (q ) Z (",q) ,
Z(P) -+ Z(p). (6.9)

[The operator P(",) is called the phase gate, in analogy to the discrete-variable
phase gate P [15], because of its similar action on the Pauli operators.]
For discrete variables, it is possible to generate the Clifford group using only
the SUM, F, and P gates [15]. However, for the CV definitions above, the
operators SUM, F, and P(",) are all elements of Sp(2n, lR); they are generated
by homogeneous quadratic Hamiltonians only. Thus, they are in a subgroup of
the Clifford group. In order to generate the entire Clifford group, one requires
a continuous HW(I) transformation [Le., a linear Hamiltonian, that generates a
one-parameter subgroup of HW(I)] such as the Pauli operator X(q). This set
{SUM, F, P(",), X(q); "', q E lR} generates the Clifford group.
We now have the necessary components to prove the main theorem of this
paper regarding efficient classical simulation of a CV process. We employ the
stabilizer formalism used for discrete variables and follow the evolution of the
52 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Pauli operators rather than the states. To start with, let us consider the ideal
case of a system with an initial state in the computational basis of the form
Iq1, q2, ... , qn}. This state may be fully characterized by the eigenvalues of
the generators of n Pauli operators {q1, q2, . .. , qn}. Any continuous variable
process or algorithm that is expressed in terms of Clifford group transformations
can then be modeled by following the evolution of the generators of these n
Pauli operators, rather than by following the evolution of the states in the Hilbert
space £2 (Rn ). The Clifford group maps linear combinations of Pauli operator
generators to linear combinations of Pauli operator generators (each qi and Pi
is mapped to sums of qj, Pj, j = 1, ... ,n in the Heisenberg picture). For
each of the n generators describing the initial state, one must keep track of 2n
real coefficients describing this linear combination. To simulate such a system,
then, requires following the evolution of 2n2 real numbers.
In the simplest case, measurements (in the computational basis) are per-
formed at the end of the computation. An efficient classical simulation involves
simulating the statistics of linear combinations of Pauli operator generators. In
terms of the Heisenberg evolution, the qj are described by their initial eigen-
values, and the pj in the sum by a uniform random number. This prescription
reproduces the statistics of all multi-mode correlations for measurements of
these operators.
Measurement in the computational basis plus feed-forward during the com-
putation may also be easily simulated for a sufficiently restricted class of feed-
forward operations; in particular, operations corresponding to feed-forward
displacement (not rotation or squeezing, though this restriction will be dropped
below) by an amount proportional to the measurement result. Such feed-
forward operations may be simulated by the Hamiltonian that generates the
SUM gate with measurement in the computational basis delayed until the end
of the computation. In other words, feed-forward from measurement can be
treated by employing conditional unitary operations with delayed measure-
ment [2], thus reducing feed-forward to the case already treated.
In practice, infinitely squeezed input states are not available. Instead, the
initial states will be of the form
(6.10)

where 10) is a vacuum state and S(r), r E R is the squeezing operation

which can be expressed directly in terms of elements of the Clifford group.
Now the vacuum states may also be described by stabilizers {q1 + iPl, q2 +
iP2, ... , qn + iPn} which are complex linear combinations of the generators.
Combining the initial squeezing operators into the computation, a classical
simulation requires following the evolution of 4n 2 numbers (twice that of
infinitely squeezed inputs due to the real and imaginary parts). Measurements
in the computational basis are again easily simulated in terms of this Heisenberg
Efficient classical simulation ofcontinuous variable quantum information processes 53

evolution, by treating each of the qi and Pi as random numbers independently

sampled from a Gaussian distribution with widths described by the vacuum
state. Simulation of measurement plus feed-forward follows exactly the same
prescription as before.
The condition for ideal measurements can be relaxed. Finite efficiency de-
tection can be modeled by a linear loss mechanism [16]. Such a mechanism
may be described by quadratic Hamiltonians and hence simulated by the Clif-
ford group. Note that the Clifford group transformations are precisely those
that preserve Gaussian states; i.e., they transform Gaussians to Gaussians; this
observation allows us to remove our earlier restriction on feed-forward gates
and allow for classical feed-forward of any Clifford group operation. Note that
non-Gaussian components to the states cannot be modeled in this manner.
Finally, it should be noted that modeling the evolution requires operations
on real-valued (continuous) variables, and thus must be discretized when the
simulation is done on a discrete (as opposed to analog) classical computer. The
discretization assumes a finite error, which will be bounded by the smaller of
the initial squeezing or the final detector "resolution" due to finite efficiency,
and this error must remain bounded throughout the simulation. As only the
operations of addition and multiplication are required, the discretization error
can be kept bounded with a polynomial cost to efficiency.
Thus, we have proved the extension of the GK theorem for continuous
variables:
Theorem 2 (efficient classical simulation): Any continuous variable quantum
information process that initiates with Gaussian states (products of squeezed
displaced vacuum states) and performs only (i) linear phase-space displace-
ments (given by the Pauli group), (ii) squeezing transformations on a single os-
cillator system, (iii) SUM gates, (iv) measurements in position- or momentum-
eigenstate basis (measurements of Pauli group operators) with finite losses,
and (v) Clifford group [HW(n)]Sp(2n,~) operations conditioned on classical
numbers or measurements of Pauli operators (classical feed-forward), can be
efficiently simulated using a classical computer.
We could summarize the conditions (i-iii) by simply stating (i-iii) transfor-
mations generated by Hamiltonians that are inhomogeneous quadratics in the
canonical operators {iii, Pi; i = 1,... ,n}, which is equivalent. Thus, any
circuit built up of components described by one- or two-mode quadratic Hamil-
tonians [such as the set of gates SUM, F, P(-,,), and X(q)], that initiates with
finitely squeezed states and involves only measurements of canonical variables
may be efficiently classically simulated.
As with the discrete-variable case, these conditions do not mean that entan-
glement between the n oscillator systems is not allowed; for example, starting
with (separable) position eigenstates, the Fourier transform gate combined with
the SUM gate can lead to entanglement. Thus, algorithms that produce entan-
54 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

glement between systems may still satisfy the conditions of the theorem and
thus may be simulated efficiently on a classical computer; included are those
used for CV quantum teleportation [4], quantum cryptography [6,7,8,9], and
error correction for CV quantum computing [12, 13]. Although these processes
are of a fundamentally quantum nature and involve entanglement between sys-
tems, this theorem demonstrates that they do not provide any speedup over a
classical process. Thus, our theorem provides a valuable tool in assessing the
classical complexity of simulating these quantum processes.
As shown in [10], in order to generate all unitary transformations given by
an arbitrary polynomial Hamiltonian (as is necessary to perform universal CV
quantum computation), one must include a gate described by a Hamiltonian
other than an inhomogeneous quadratic in the canonical operators, such as a
cubic or higher-order polynomial. Transformations generated by these Hamil-
tonians do not preserve the Pauli group, and thus cannot be described by the
stabilizer formalism. Moreover, any such Hamiltonian is sufficient [10]. One
example would be to include an optical Kerr nonlinearity [17], but there is a lack
of sufficiently strong nonlinear materials with low absorption. Alternatively,
it has recently been proposed that a measurement-induced nonlinearity (using
ideal photodetection) could be used in an optical scheme without the need for
nonlinear materials in the computation [15, 18]. The physical realization of
such nonlinearities is an important quest for quantum information theory over
continuous variables. These nonlinear transformations can be used in CV al-
gorithms that do not satisfy the criteria of this theorem, and which may provide
a significant speedup over any classical process.

This project has been supported by an Australian Research Council Large

Grant. S.L.B. and K.N. are funded in part under project QUICOV as part of
the IST-FET-QJPC programme.

References
[1] D. Gottesman, in Proceedings of the XXII International Colloquium on
Group Theoretical Methods in Physics, edited by S. P. Corney et ai.,
(International Press, Cambridge, MA, 1999), p. 32.
[2] M. A. Nielsen and 1. L. Chuang, Quantum Computation and Quantum In-
formation (Cambridge University Press, Cambridge, U.K., 2000), p. 464.
[3] E. Knill, quant-ph/0108033; B. M. Terhal and D. P. DiVincenzo, quant-
ph/O 1080 10.
[4] S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998).
[5] A. Furusawa et al., Science 282, 706 (1998).
[6] T. C. Ralph, Phys. Rev. A 61, 010303(R) (2000).
Efficient classical simulation ofcontinuous variable quantum information processes 55

[7] M. Hillery, Phys. Rev. A 61, 022309 (2000).

[8] M. D. Reid, Phys. Rev. A 62, 062308 (2000).
[9] D. Gottesman and J. Preskill, Phys. Rev. A 63,022309 (2001).
[10] S. Lloyd and S. L. Braunstein, Phys. Rev. Lett. 82, 1784 (1999).
[11] R. P. Feynman,Found. Phys. 16, 507 (1986).
[12] S. L. Braunstein, Phys. Rev. Lett. 80, 4084 (1998).
[13] S. L. Braunstein, Nature (London) 394, 47 (1998).
[14] D. F. Walls and G. J. Milburn, Quantum Optics (Springer, Berlin, 1994).
[15] D. Gottesman et at., Phys. Rev. A 64, 012310 (2001).
[16] H. P. Yuen and J. H. Shapiro, IEEE Trans. Inf. Theory 26, 78 (1980).
[17] G. J. Milburn and D. F. Walls, Phys. Rev. A 28, 2065 (1983).
[18] S. D. Bartlett and B. C. Sanders, Phys. Rev. A 65, 042310 (2002).
II

QUANTUM ENTANGLEMENT
Chapter 7

INTRODUCTION TO ENTANGLEMENT-BASED
PROTOCOLS

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 1 UT, United Kingdom
[email protected]

Arun K. Pati
Institute of Physics, Bhubaneswar-751005, Orissa, INDIA
Theoretical Physics Division, BARC, Mumbai, INDIA
[email protected]

Abstract We give a brief introduction to entanglement, teleportation, entanglement swap-

ping and purification protocols for finite-dimensional Hilbert space systems,
primarily for qubits.

1. INTRODUCTION
Quantum entanglement was first introduced by SchrOdinger and explored by
Einstein-Podolsky-Rosen (EPR) in their famous paper on the incompleteness of
quantum theory [1]. As SchrOdinger put it, the phenomenon of entanglement
is one of the quintessential features in quantum mechanics that has no ana-
logue in classical physics. Typically, when two quantum systems interact, the
wavefunction of one can get intertwined with the wavefunction of other. The
combined wavefunction of the composite system can show strong correlations
even though they may be widely separated. In an attempt to understand these
correlations Bell constructed an inequality that must be satisfied for all local
realistic models [2]. Surprisingly, he noted that entangled states violate such
inequalities. Thus demonstrating that indeed entanglement resists any classical
explanation.
Over the last decade it has been realised that quantum entanglement is an
important resource in quantum information and computation. In particular,
59
S.L Braunstein andA.K. Pati (eds.J, Quantum Information with Continuous Variables, 59-66.
© 2003 Kluwer Academic Publishers.
60 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

there are numerous quantum communication protocols that require shared en-
tanglement between a pair of parties, Alice and Bob. Utilizing this shared
resource together with local operations and classical communication (LOCC),
Alice and Bob can perform various feats: Alice can teleport an unknown state
to Bob using an entangled pair and two classical bits [3]; dense coding of
classical information is possible with shared entanglement between sender and
receiver [4]; and one may create entanglement between a pair of systems that
have never interacted in the past, called entanglement swapping [5].
For these protocols to work perfectly one requires ideal EPR pairs. This
is a problem since such states are highly prone to errors during transportation
or storage. The resulting imperfect EPR pairs behave like a noisy channel
in the above protocols and no longer work in an ideal manner. Fortunately,
there is yet another protocol that is capable of purifying such imperfect EPR
pairs (provided they are still entangled). In purification, two parties start with
a given number of imperfect pairs and by performing LOCC operations they
produce a smaller number of more highly entangled pairs. This process may
be repeated, provided the resource has not been exhausted, until the EPR pairs
are sufficiently ideal for carrying out the desired protocol.
In this article we briefly discuss entanglement, teleportation, entanglement
swapping and purification protocols for finite-dimensional systems (especially
for qubits). We hope this will motivate readers for the continuous variable
generalizations discussed in the following chapters.

2. QUANTUM ENTANGLEMENT
Entanglement is always a property of a composite system viewed as a system
consisting of subsystems. If the whole system is always viewed as a single
object then we would not bother about the role of entanglement. Entanglement
is a feature of states in Hilbert spaces which have been given a tensor product
structure. In the case of a composite system consisting of two subsystems A
and B (called a bipartite system) the combined state lives in llAB = llA 01lB.
If such a system can be described by a state which can be written as

(7.1)

where I~)A and I¢)B are states of the subsystems A and B, respectively, then
the state is not entangled (which is synonymous with it being separable). If
one cannot assign a definite pure state to each sub-system then it is entangled.
This defines entanglement for pure states.
 Introduction to entanglement-based protocols 61

If the dimensions of 1-l A is N A and of 1-l B is N B then an arbitrary entangled

state may be written as
NA,NB
IW)AB = Li,j Cijlxi)AIYj)B, (7.2)

where {IXi)} and {IYj)} are orthonormal basis states in the Hilbert space 1-lA
and 1-lB, respectively. Now the Schmidt decomposition theorem tells us that
by performing a singular-value decomposition of the complex matrix (Cij), we
may write any arbitrary bipartite state as [6]
min(NA,NB)
IW)AB = L JPili)Ali')B , (7.3)

where li)A and li')B are the singular-value transformed bases and the Pi'S are
non-zero eigenvalues of the reduced density matrices of the subsystem A or
B. We note the.../Pi are called Schmidt numbers and EiPi = 1. One can see
then that for pure states if there is more than one Schmidt number, then the
state is entangled. It should be remembered that the Schmidt decomposition
theorem holds only for a bipartite system and does not hold for tripartite
systems or higher. (However, there are special conditions under which Schmidt
decomposition theorem may be found for tripartite systems [7].)
More generally, states of a composite system need not be pure. If the
composite system is described by a mixed state, then we need a different
definition for entanglement. In general, we say that a state p AB is separable
(not entangled) on 1-lA ® 1-lB if it can be written as [8]

(7.4)

where p~) and p~) are states in 1-lA and 1-lB, respectively and the qi are non-
negative real weights. Such states display only classically correlations. This is
because two parties may always prepare such states from locally created states
plus classical communication.
One may ask how much entanglement such states contain. But to answer this
question we must define a measure of entanglement. It has been proposed that
any measure of entanglement E(p) for a state p (whether p is pure or mixed)
should satisfy three conditions [9]:
1. E(p) should be zero if and only if p is separable.
2. E (p) should be invariant under local unitary operations. This means that
if p -+ (UA ® UB)p(UA ® UB)t = p', then E(p) = E(p').
62 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

3. On average, entanglement cannot increase under general local measure-

ments, local unitary operations and classical communication (LOCC)
and post-selection. Mathematically, we may write this condition as

(7.5)

where Pi = AipA! and the Ai are local Kraus operators [11] satisfying
L::i A! Ai = 1.

One universally accepted measure of entanglement for pure bipartite systems

is the von Neumann entropy of either of the reduced density matrices PA or PB
[10] which is given by

and has units of entangled bits or "ebits." This can reach maximum value
when all the Pi are equal. Hence the maximum amount of entanglement in this
system is log N assuming N = min( N A, N B). One can define a maximally
entangled state I\[! max) AB to be

(7.7)

in 1{ A ~ 1{ B which is a state in a space of dimension N 2 .

For two qubits, EPR pairs are the most widely studied states in quantum
information theory. As an example, one canonical EPR state is given by

(7.S)

with 10), 11) being the computational basis states for qubits. One can easily
check that if we adopt the measure of entanglement as the von Neumann entropy
given in Eq. (7.6), then an EPR state contains exactly 1 ebit of entanglement
which is the maximum amount of entanglement that any 2 x 2-dimensional
Hilbert space can admit. Hence, this EPR pair is also a maximally entangled
state of two-qubits. It may be worth observing that, if we have a bipartite sate
in an N x N-dimensional Hilbert space with N = 2n , then the state I\[! max) is
equivalent to n (qubit-based) EPR pairs.
We now consider a number of communication protocols that may be achieved
with shared entanglement.
 Introduction to entanglement-based protocols 63

3. QUANTUM TELEPORTATION
Suppose that Alice and Bob are at widely separated locations. Alice wants
to send an unknown qubit to Bob without physically sending it! How is it
possible?
Let Alice and Bob share an EPR pair IEPR) 12 and have access to particles 1
and 2, respectively. If Victor gives a qubit to Alice (whose identity is unknown
to her) of the form

(7.9)

then the combined state of the input and EPR pair is just I'l/J) a ® IW-) 12. This
may be expressed in terms of a basis of "Bell-states" for particles a and 1 as

I'l/J)a ® IEPRh2 = ~ (1<P+)al ® R(I)I'l/J)2 + I<P-)al ® R(2)I'l/Jh

+ IW+)al ® R(3)I'l/Jh -IW-)al ® R(4)I'l/Jh) ,
(7.10)

where R(l) = iO"y, R(2) = O"x, R(3) = o"z are Pauli matrices and R(4) = IT.
Here the states I<p±) and IW±) are the canonical Bell-states first introduced in
Ref. [12] as

~(IO)IO) ± 11)11))
~(IO)ll) ± 11)10)) . (7.11)

We can see that if Alice now performs a joint measurement on particles a

and 1 in this Bell-basis she will obtain one of four possible outcomes {I<p±),
Iw±)}, corresponding to two classical bits of information. Then Alice sends
these two bits to Bob, who applies the appropriate unitary operation to his half
of the entangled state. This converts it into the original unknown state supplied
by Victor. For example, if the outcome is I<P +) then after receiving this classical
information Bob will apply R(l)t = -iO"y in order to "recreate" the original
state on particle 2 at his location. This completes the quantum teleportation
of a single unknown qubit. This protocol works for any of the four possible
outcomes of Alice's measurements and so it is successful every time. It should
be noted that in the process the original state that was handed to Alice is
destroyed and is "recreated" at Bob's location. Hence, it does not violate
the no-cloning theorem. Finally, we note that neither the quantum channel
(the shared EPR pair) nor the classical communication channel provide any
information about the state to be teleported. It is only when these two channels
are combined that the unknown state may be transferred to Bob.
64 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

One can easily generalize quantum teleportation to higher-dimensional quan-

tum systems (say for an N-dimensional Hilbert space). In this case the shared
entangled resource is a maximally entangled state in the N x N-dimensional
Hilbert space. A similar protocol to the above allows Alice to send her unknown
state to Bob with the use of a classical communication channel. In this case,
the communication cost is 210g 2 N classical bits or "cbits" of information.

4. DENSE CODING
Usually, from a transmitted qubit one can extract only one classical bit upon
measurement. However, if the sender and receiver share prior entanglement
then sending a suitably modulated version of the shared half (corresponding to
one qubit) to receiver, one can now extract two classical bits of information.
This doubling of classical capacity of quantum channel (with the assistance of
entanglement) is called dense coding or sometimes super dense coding.
Let us imagine that Alice and Bob share a canonical Bell-state

(7.12)

One important property of the Bell-states is that one can convert anyone of
them to any other by acting locally on one half of the system. In particular,
Alice can apply a set of local unitary operations giving:

1<I>+h2 I (8) II<I>+h2

1<I>-h2 a z (8) I 1<I>+h2
1\11+)12 ax (8) II<I>+h2
1\11-)12 iay (8) II<I>+h2 . (7.13)

After applying any of these four local operations, Alice sends her particle to
Bob. Then, by performing a Bell-measurement Bob can distinguish each case
thus, extracting two classical bits of information. It may be noted that dense
coding is closely related to quantum teleportation. In teleportation, one sends
a single qubit via two cbits plus shared entanglement, whereas in dense coding
one sends two cbits via one qubit plus shared entanglement.

5. ENTANGLEMENT SWAPPING
Entanglement swapping is a method to create entanglement between two
independent particles that have never interacted in the past. Though at first
glance this sounds strange, it is indeed possible in the quantum world [5].
Let us consider two independent sources each emitting an EPR pair 1,2 and
3,4, respectively. This means particles 1 and 2 are entangled and similarly for
particles 3 and 4. However, neither 1 and 4 nor 2 and 3 are entangled. Let Alice
 Introduction to entanglement-based protocols 65

take particle 1, Charlie take particle 4 and Bob, the man in the middle, will take
particles 2 and 3. The combined state of the system can then be written as

1'lt-h21'lt-h4 = ~ (1\!J+h4I\l1+)23 - 1'lt-h41'lt-h3

- 1.p+)14I.p+h3 + 1.p-h41.p-h3)' (7.14)

We can now see that if Bob performs a Bell-measurement on particles 2 and

3 that he will obtain one of four possible outcomes. For instance, if the outcome
of Bob's projection is the state 1.p±h3 then Alice and Charlie are found to share
the entangled state 1.p±h4. Similarly for all of Bob's measurement outcome
this leaves Alice and Charlie in one of the maximally entangled Bell-states. If
two of the particles are subjected to a Bell-measurement then the remaining
halves become entangled, despite their never having interacted.
Further, if Bob communicates to Alice and Charlie (or even to one is suf-
ficient) which of the four Bell-states he found, then they may convert their
now entangled state into any standard maximally entangled state. In a sense
the entanglement has been "swapped" between the partners to the pair that
Bob measured. Interestingly, this protocol can also be simply viewed as just
the teleportation of an entangled state. By performing a Bell-measurement
on the particles 2 and 3 Bob is able to teleport the entanglement to either (or
equivalently both) recipients.

6. ENTANGLEMENT PURIFICATION
As we have already noted, in practice ideal EPR pairs may undergo deco-
herence and become mixed entangled states. In the absence of a maximally
entangled resource one can no longer faithfully perform teleportation, dense
coding or entanglement swapping. This suggests that it would be useful if we
could improve the quality of imperfect EPR pairs through some other protocol.
Fortunately, this is possible through "entanglement purification" [13]. Since
we wish to restrict this protocol to LOCC operations we will not be able to win
on average. However, provided we have some entanglement there is still the
possibility of probabilistic ally purifying our imperfect EPR pairs.
Suppose Alice and Bob share multiple copies of an imperfectly entangled
EPR pair of the form

(7.15)

This state is a Werner mixture with the spin-singlet appearing with probability
f. This mixture is entangled provided f > ~ [13]. The simplest protocol
works on pairs of such imperfect states at a time. By performing suitable
LOCC operations [13] Alice and Bob will be able to post-select a single pair
66 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

having the form PAB(J/) with f' > f provided f > ~. This process only
occurs with a finite probability, however, every time it works a higher quality
entangled resource is purified out of several lesser quality states. Provided pairs
are available, this process may be repeated until a suitably high quality EPR
pair is created. In principle, the fidelity relative to an ideal EPR can approach
unity.

7. CONCLUSION
We have briefly touched on basic concepts of quantum entanglement and
its utility in quantum information processing. (More details may be found in
Refs. [11, 14].) Of these protocols, only entanglement purification does not
straightforwardly go over to continuous variable schemes, as we shall see in
the following chapters.

References
[1] A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935)
[2] J. S. Bell, Physics 1, 195 (1964)
[3] c. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres and W. K.
Wootters, Phys. Rev. Lett. 70, 1895 (1993).
[4] C. H. Bennett and S. J. Wiesner, Phys. Rev. Lett. 69, 2881 (1992).
[5] M. Zukowski, A. Zeilinger, M. A. Horne and E. Ekert, Phys. Rev. Lett.
71,4278 (1993).
[6] A. Peres, Quantum Theory: Concepts and Method (Kluwer Academic
Publisher, 1995).
[7] A. K. Pati, Phys. Lett. A 278, 118 (2000), and references therein.
[8] R. F. Werner, Phys. Rev. A 40, 4277 (1989).
[9] V. Vedral and M. Plenio, Phys. Rev. A 57, 1619 (1998).
[10] S. Popescu and D. Rohrlich, Phys. Rev. A 56, R3319 (1997).
[11] M. A. Nielsen and 1. L. Chuang, Quantum Computation and Quantum
Information, (Cambridge University Press, Cambridge, England, 2000).
[12] S. L. Braunstein, A. Mann and M. Revzen, Phys. Rev. Lett. 68, 3259
(1992).
[13] c. H. Bennett, H. J. Bernstein, S. Popescu, B. Schumacher, J. A. Smolin
and W. K. Wootters, Phys. Rev. Lett. 76, 722 (1996).
[14] J. Preskill, Lecture Notes, available at
http://www.theory.caltech.edu/people/preskill/ph229
Chapter 8

TELEPORTATION OF CONTINUOUS
QUANTUM VARIABLES*

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 I Ur, United Kingdom
[email protected]

H. J. Kimble
Norman Bridge Laboratory of Physics 12-33
California Institute of Technology
Pasadena, CA 91125

Abstract Quantum teleportation is analyzed for states of dynamical variables with con-
tinuous spectra, in contrast to previous work with discrete (spin) variables. The
entanglement fidelity of the scheme is computed, including the roles of finite
quantum correlation and nonideal detection efficiency. A protocol is presented
for teleporting the wave function of a single mode of the electromagnetic field
with high fidelity using squeezed-state entanglement and current experimental
capability.

Quantum mechanics offers certain unique capabilities for the processing of

information, whether for computation or communication [1]. A particluarly
startling discovery by Bennett et ai. is the possibility for teleportation of a
quantum state, whereby an unknown state of a spin-~ particle is transported by
"Alice" from a sending station to "Bob" at a receiving terminal by conveying
2 bits of classical information [2]. The enabling capability for this remark-
able process is what Bell termed the irreducible nonlocal content of quantum
mechanics, namely that Alice and Bob share an entangled quantum state and
exploit its nonlocal characteristics for the teleportation process. For spin-~

* S. L. Braunstein and H. J. Kimble, Physical Review Letters 80, 869-872 (1998).

Copyright (1998) by the American Physical Society.

67
68 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

particles, this entangled state is a pair of spins in a Bell state as in Bohm's

version of the Einstein, Podolsky, and Rosen (EPR) paradox [3] and for which
Bell formulated his famous inequalities [4].
Beyond the context of dichotomic variables, Vaidman has analyzed telepor-
tation of the wave function of a one-dimensional particle in a beautiful variation
of the original EPR paradox [5]. In this case, the nonlocal resource shared by
Alice and Bob is the EPR state with perfect correlations in both position and
momentum. The goal of this Letter is to extend Vaidman's analysis to incorpo-
rate finite (nonsingular) degrees of correlation among the relevant particles and
to include inefficiencies in the measurement process. The "quality" of the re-
sulting protocol for teleportation is quantified with the first explicit computation
of the fidelity of entanglement for a process acting on an infinite dimensional
Hilbert space. We further describe a realistic implementation for the quantum
teleportation of states of continuous variables, where now the entangled state
shared by Alice and Bob is a highly squeezed two-mode state of the electro-
magnetic field, with the quadrature amplitudes of the field playing the roles of
position and momentum. Indeed, an experimental demonstration of the origi-
nal EPR paradox for variables with a continuous spectrum has previously been
carried out [6, 7], which when combined with our analysis, forms the basis of a
realizable experiment to teleport the complete quantum state of a single mode
of the electromagnetic field.
Note that up until now, all experimental proposals for teleportation have
involved dichotomic variables in SU(2) [2,8,9, 10, 11] with optical schemes
accomplishing the Bell-operator measurement with low efficiency. Indeed, the
recent report of teleportation via parametric down conversion [12] succeeds
only a posteriori with rare post-selected detection events. By constrast, our
scheme employs linear elements corresponding to operations in SU(l, 1) [13]
for Bell-state detection and thus should operate at near unit absolute efficiency,
enabling a priori teleportation as originally envisionaged in Ref. [2].
As shown schematically in Fig. 8.1, an unknown input state described by
the Wigner function Win (a) is to be teleported to a remote station, with the
teleported (output) state denoted by Wout(a). In analogy with the previously
proposed scheme for teleportation of the state of a spin- ~ particle, Alice (at the
sending station) and Bob (at the receiving terminal) have previously arranged
to share an entangled state which is sent along paths 1 and 2. Within the context
of our scheme in SU(l, 1), the entangled state distributed to Alice and Bob is
described by the Wigner function WEPR(al, (2) [4]

WEPR(al; (2) 42 exp{ -e- 2r [(XI - X2)2 + (PI + P2)2]

7r

- e+ 2r [(xI + X2)2 + (PI - p2)2]}

-+ C 8(XI + X2) 8(PI - P2) , (8.1)
 Teleportation of continuous quantum variables 69

Ollt

2
W;,,(a) 1

Figure 8.1 Scheme for quantum teleportation of an (unknown) input state Win (a) from Alice's
sending station S to Bob's remote receiving terminal R, resulting in the teleported output state
Wout(a).

where aj = Xj + ipj. Here, the real quantities (Xj, pj) correspond to canoni-
cally conjugate variables for the relevant pathways and describe, for example,
position and momentum for a massive particle, and quadrature amplitudes for
the electromagnetic field. Note that for r -+ 00, the state described by Eq. (8.1)
becomes precisely the EPR state of Ref. [3] employed by Vaidman [5] and pro-
vides an ideal entangled "pair" shared between the teleportation sending and
receiving stations, albeit with divergent energy in this limit.
As for the protocol itself, the first step in teleporting the (unknown) state
Win(ain) is to form new variables {3a,b along paths (a, b) which are linear
superpositions of those of the initially independent pathways in and 1 at the
sending station S of Fig. 8.1, namely {3a,b = ~(al ± ain). The resulting
Wigner function in the variables ({3a; (3b; (2) exhibits "entanglement" between
the paths (a, b) and the remote path 2. Step 2 at S is then to measure the
observables corresponding to Re{3a = ~(Xl + Xin) == Xa and Im{3b =
~(Pl - Pin) == Pb at the detectors (Da, Db) shown in Fig. 8.1, with the
reSUlting classical outcomes denoted by (ixa' i pb ), respectively. We define
ideal measurement of (Xa,Pb) to be that for which the distribution Pab( iXa; i pb )
is identical to the associated Wigner function Wab(Xa;Pb). With the entangled
70 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

state of paths (1,2) given by Eq. (8.1), we find

2 Jd2a Win(a)G v[J2(i xa - i i pb ) - a]

2[Win 0 Gv][J2(i xa - ii pb )] , (8.2)

with 0 denoting convolution and Gv as a complex Gaussian distribution with

variance 1/ = cosh 2r /2. Note that such ideal detectors provide "perfect" in-
formation about (Xa,Pb) via (ixa,i pb ), while all information about (Pa,Xb) ==
(Im,6a = ~(pl + Pin), Re,6b = ~(Xl - Xin)) is lost. Furthermore, al-
though (i xa , i pb ) contains a small amount of information about the fiducial
state Win(a) = Win(Xin,Pin), this information goes to zero for r --* 00.
Nonetheless, the third and final step at the sending station is to transmit this
classical information to the receiving terminal.
As illustrated in Fig. 8.1, receipt of (i xa , i pb ) allows Bob to construct the
teleported state Wout (a2) from component 2 of the EPR state. That this resur-
rection is possible can be understood by examining the (unnormalized) Wigner
function for the system obtained by integrating out (Pa, Xb) in correspondence
to Alice's detection of (Xa,Pb), namely

where the variance T = sech2r /2. Note that as r --* 00, G T (a) quickly
approaches a delta-function, while Gv(a) describes a broad background state.
Thus, for large r, the reduced state of mode 2 is described by a broad pedestal
with negligible probability upon which sits a randomly located peak at 0:'2 ~
.J2(i xa - i i pb ) closely mimicing the incoming state Win(a). The location
of this random "displacement" is distributed according to Eq. (8.2), and is the
classical information that Alice sends to Bob.
By way of the actuator Ax,p shown in Fig. 8.1, Bob thus performs linear
displacements of the real and imaginery components of the complex amplitude
a2 to produce aout = a2 + .J2(i xa - i i pb ), where the quantities (ixa' i pb ) are
scaled to (Xa,Pb). Integrating out iXa and ipb yields the ensemble description
of states produced at the output of the teleportation device on an ensemble of
input states Win, namely

Wout = Win 0 Gu , (8.4)

where (J = e- 2r is the variance of the complex Gaussian G u , thus completing

the teleportation process.
Clearly, for r ---+ 00 the teleported state of Eq. (8.4) reproduces the original
unknown state Win [5]. However, note that as r --* 0, Wout also mimics Win,
now with two extra units of vacuum noise (Le., (J = !+ ~). One of these
 Teleportation of continuous quantum variables 71

noise contributions arises from Alice's attempt to measure both (Xin,Pin) [14],
while the second comes from Bob's use of this necessarily noisy information
to generate a coherent state at V2( iXa - i i pb )' In this way quantum mechanics
extracts two tariffs (one at each instance of the border crossing between quantum
and classical domains), each of which we term the quantum duty (or quduty).
Note that the limit r = 0 corresponds to what might be considered "classical"
teleportation for which the "best measurement" of the coherent amplitude of
the unknown state is made [14] and sent to the receiving station, where it is
used to produce a coherent state of that classical amplitude. For any r > 0, our
quantum teleportation protocol beats this classical scheme.
Before calculating an actual figure of merit for our protocol, we now spe-
cialize from general continuous variables to the case of a single mode of the
electromagnetic field and thereby to actual physical implementations of the
various transformations shown in Fig. 8.1. Beginning with the EPR state itself,
we note that such a state can be generated by nondegenerate parametric ampli-
fication with the quantities (x j, pj) as the quadrature-phase amplitudes of the
field [6], as has been experimentally confirmed via Type II down-conversion
[7]. The linear transformation i3a,b = ~(al ± ain) is accomplished by the
simple superposition of modes in and 1 at a 50/50 beam splitter. The detectors
(D a, Db) of Fig. 8.1 are now just balanced homodyne detectors with the phases
of their respective local oscillators set to record (Xa,Pb) in the observed pho-
tocurrents (ixa' i pb )' Note that for unit efficiency, homodyne detection provides
an ideal quantum measurement of the quadrature amplitudes required for our
protocol [15, 16, 17].
Non-ideal detectors, each having (amplitude) efficiency 'f/, may be modeled
by using a pair of auxiliary beam splitters at (Da, Db) to introduce noise from a
pair of vacuum modes described by annihilation operators (Ca,b, da,b) [15, 18].
It is then convenient to introduce annihilation operators corresponding to the
"modes" of the photocurrents described by

~ ~
Za,b = 'f/i3a,b + V~-2-(Ca,b + da,b) ,
A A A

(8.5)

where these fictitious objects allow us to apply an analog of the Wigner-function

formalism to the photocurrents and to incorporate the effects of nonidea1 pho-
todetection in a straightforward fashion. For example, loss in the response of
Alice's detectors [Eq. (8.2)] leads to the convolution

Pab(i xa , ipb ) = ~[Pab 0 Gd((ix a + i ipb)/'f/) , (8.6)

'f/

where G( has variance ( = (1 - 'f/2)/2'f/2, which goes to zero for 'f/ --+ 1 in
correspondence with the ideal character of homodyne detection. Substituting
72 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

for Pab from Eq. (8.2) then gives

Pab(ix a, i pb ) = 22 [Win 0 Gv](V2(i xa - i ipb )) , (8.7)

." ."
where D = ~ cosh2r + (1 _ .,,2)/.,,2.
Within the context of the electromagnetic field, Bob can efficiently perform
the required phase-space displacement of mode 2 based upon the classical
information (ixa' i pb ) received from Alice by combining the field of mode 2
with a (classical) coherent state of mean amplitude E It, where E = v'2 (ixa -
i i pb ) /.", at a highly reflecting mirror of transmissivity t --7 O. The mean state
after this shift is the final teleported state, namely

W out = Win 0 Gij , (8.8)

where Gij(a) = 1flij exp (-l a I2 /0-) with 0- = e- 2r + (1 _ .,,2);"'2.

The teleportation evolution described by Eq. (8.8) may be written in density
matrix form as

(8.9)

where Pin is the original state being teleported and b(a) is the displacement
operator. The dynamics associated with Eq. (8.9) were first studied by Glauber
[19] and Lachs [20] for an "incoming" vacuum state p = 10)(0 Iand for squeezed
vacuum by Vourdas and Weiner [21]. The detailed behavior of the photocount
statistics under this dynamics was investigated by Musslimani et ai. [22].
These references also relate the development of the convolutional formalism
used here (see also Refs. [23,24]).
To illustrate the protocol, consider teleportation of the coherent superposition
state

(8.10)

with corresponding Wigner function Win(a) illustrated in Fig. 8.2(a). The

teleported Wigner function Wout(a) as computed from Eq. (8.8) is shown in
Fig. 8.2(b) for parameters corresponding to -10 dB of squeezing (i.e., r = 1.15)
with efficiency .,,2 = 0.99, which should be compared to the parameters of
Ref. [25] [namely squeezing r = 0.69 (i.e., 6 dB of squeezing), and detectors
with absolute quantum efficiency .,,2 = 0.99 ± 0.02]. Note that the quantum
character of the state survives teleportation, including negative values for W out
associated with quantum interference for the off-diagonal components of Pin.
For comparison, note that for classical teleporation (i.e., r = 0), wg~t consists
of the (incoherent) superposition of two distributions centered at ±a, each of
which is broadened by the quduty.
 Teleportation of continuous quantum variables 73

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

a) W.In b) W
out
-0.6 -0.6

2 2
0
-2 -1 0 -1

Figure 8.2 (a) Wigner function Win(a) for the input state ofEq. (8.10) with a = 1.5i and
¢;= 7r. (b) Teleported output state Wout(a) for r = 1.15 and 1/ 2 = 0.99.

To provide a quantitative measure of the "quality" of the output state, we

note that the strongest measure of fidelity of a teleported state relative to the
input state is given by the entanglement fidelity [26]. For processes described
by Eq. (8.9), it is given by

Fe = Jd2~Gi't(~)lxVl1n (~)12 , (8.11)

where X . (~) = tr D(i~) Pin is the characteristic function for the incoming
Vl1n
state's Wigner function.
For the coherent superposition of Eq. (8.10) direct substitution yields a
fidelity of entanglement Fe of

(8.12)

For the state shown in Fig. 8.2(b) this fidelity is 0.6285 for r = 1.15 and rt 2 =
0.99 compared to 0.2487 for r = 0 and the same detector efficiency. This latter
fidelity precludes observation of any quantum features in the classically tele-
ported state, while the former case yields observable quantum characteristics
as seen in Fig. 8.2.
74 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Beyond anyone particular state, let us now concentrate on high fidelity

teleportation in general. In this case the Gaussian weighting described by G ii
is sufficiently narrow so that only the lowest terms in an expansion about ~ = 0
of Xur.
V~n
will contribute. That is, Ix ~n
. (~)12 may be approximated by

(8.13)

where IDoal 2 == (laI 2) - l(a)12 averaged over Win(a). Thus, the condition
for high fidelity teleportation (i.e., 1 - Fe « 1) becomes I/IDoal 2 » a. Now
IDoal 2 is just the number of photons (plus ~) in the incoming state after it
has been shifted so as to have no coherent amplitude. Roughly speaking it is
the maximal rms spread of the Wigner function of the unknown quantum state
being teleported and so its reciprocal bounds the size of "important" small scale
features in that state, though there can indeed be smaller features. Apparently
then the condition for high entanglement fidelity says that features in the Wigner
function smaller than I/IDoal do not give a significant contribution to the state's
identity.
In conclusion, our analysis suggests that existing experimental capabilities
should suffice to teleport manifestly quantum or nonclassical states of the
electromagnetic field with reasonable fidelity. For such experiments, extensions
of our analysis to the teleportation of broad bandwidth information must be
made and will be discussed elsewhere. In qualitative terms, our scheme should
allow efficient teleportation every inverse bandwidth, in sharp constrast to
relatively rare transfers for proposals involving weak down conversion for spin
degrees of freedom. Although our analysis is the first to obtain explicitly the
fidelity of entanglement on an infinite dimensional Hilbert space, an unresolved
issue is whether or not our protocol is "optimum," either with respect to this
measure or with regard to other criteria in the area of quantum communication
(e.g., the ability to teleport optimally an "alphabet" {j} of orthogonal states
Win)' More generally, the work presented here is part of a larger program
to extend classical communication with complex amplitutes into the quantum
domain.

S.L.B. was funded in part by EPSRC Grant No. GRlL91344 and a Humboldt
Fellowship. H.J.K. acknowledges support from DARPA via the QUIC Insti-
tute administered by ARO, from the Office of Naval Research, and from the
National Science Foundation. Both appreciate the hospitality of the Institute
for Theoretical Physics under National Science Foundation Grant No. PHY94-
07194.
 Teleportation of continuous quantum variables 75

References
[1] A. Steane, LANL Report No. quant-phl9708022; A. S. Holevo, LANL
Report No. quant-phl9708046.
[2] C. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres and W. K.
Wootters, Phys. Rev. Lett. 70, 1895 (1993).
[3] A. Einstein, B. Podolsky and N. Rosen, Phys. Rev. 47, 777 (1935).
[4] J. S. Bell, in Speakable and Unspeakable in Quantum Mechanics (Cam-
bridge Univ. Press, 1988), p. 196.
[5] L. Vaidman, Phys. Rev. A 49, 1473 (1994).
[6] M. D. Reid and P. D. Drummond, Phys. Rev. Lett. 60, 2731 (1988); M.
D. Reid, Phys. Rev. A 40,913 (1989).
[7] (a) z. Y. Ou, S. F. Pereira, H. J. Kimble and K. C. Peng, Phys. Rev. Lett.
68,3663 (1992); (b) Appl. Phys. B 55, 265 (1992).
[8] L. Davidovich, N. Zagury, M. Brune, J. M. Raimond and S. Haroche,
Phys. Rev. A 50, R895 (1984).
[9] J.1. Cirac and A. S. Parkins, Phys. Rev. A 50, R4441 (1994).
[10] T. Sleator and H. Weinfurter, Ann. N. Y. Acad. Sci. 755, 715 (1995).
[11] S. L. Braunstein and A. Mann, Phys. Rev. A 51, RI727 (1995); 53, 630(E)
(1996).
[12] D. Boumeester et al., Nature (London) 390, 575 (1997).
[13] B. Yurke, S. L. McCall and J. R. Klauder, Phys. Rev. A 33,4033 (1986).
[14] E. Arthurs and J. L. Kelly Jr., Bell. Syst. Tech. J. 44, 725 (1965).
[15] H. P. Yuen and J. H. Shapiro, IEEE Trans. Inf. Theory 26, 78 (1980).
[16] S. L. Braunstein, Phys. Rev. A 42,474 (1990).
[17] Z. Y. Ou and H. J. Kimble, Phys. Rev. A 52,3126 (1995).
[18] K. Banaszek and K. W6dkiewicz, Phys. Phys. ASS, 3117 (1997).
[19] R. J. Glauber, Phys. Rev. 131,2766 (1963).
[20] G. Lachs, Phys. Rev. 138, B1012 (1965).
[21] A. Vourdas and R. M. Weiner, Phys. Rev. A 36, 5866 (1987).
[22] Z. H. Musslimani, S. L. Braunstein, A. Mann, and M. Revzen, Phys. Rev.
A 51, 4967 (1995).
[23] M. S. Kim and N. Imoto, Phys. Rev. A 52,2401 (1995).
[24] K. Banaszek and K. W6dkiewicz, Phys. Rev. Lett. 76, 4344 (1996).
[25] E. S. Polzik, J. Carri and H. J. Kimble, Phys. Rev. Lett. 68, 3020 (1992);
(b) Appl. Phys. B 55, 279 (1992).
[26] B. Schumacher, Phys. Rev. A 54,2614 (1996).
Chapter 9

EXPERIMENTAL REALIZATION
OF CONTINUOUS VARIABLE
TELEPORTATION

Akira Furusawa
Department of Applied Physics, University of Tokyo
7-3-1 Hongo, Bunkyo-ku, Tokyo113-8656, Japan

H. J. Kimble
Norman Bridge Laboratory of Physics
California Institute of Technology, mcl2-33
Caltech, Pasadena, CA9II25, USA

1. INTRODUCTION
Quantum teleportation is a method of quantum state transportation with a
classical channel and a quantum channel [1]. In this technique, the "informa-
tion" contained in a quantum state is transferred from a sending station (Alice)
to a receiving station (Bob), with the original quantum state thereby recon-
structed at Bob's place with the received information and previously shared
entanglement. Note that it is impossible to perform the state transformation
represented by quantum teleportation only with a classical channel, which can
be qualitatively explained as follows. If one attempts to obtain complete in-
formation with some particular measurement on an unknown quantum state of
motion, for example, then both position and momentum (canonically conjugate
variables) must be determined simultaneously with negligible error, which is
of course impossible [2]. It is thus impossible for Alice to obtain complete
information on the unknown quantum state, so that she certainly cannot send
enough information for the reconstruction of the state to Bob. He then is unable
to reconstruct the complete state at his place. By contrast, in quantum tele-
portation, Alice and Bob neatly circumvent constraints that would otherwise
be imposed on Alice's state measurement and Bob's state generation, and are
thereby able to reconstruct the original state at Bob's place.
77
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 77-93.
© 2003 Kluwer Academic Publishers.
78 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

The essential character for this trick is quantum entanglement. For the case
of continuous quantum variables, the quantum entanglement is that discussed
by Einstein, Podolsky, and Rosen, in now the famous "EPR paradox" [3] .
The sharing of quantum entanglement between Alice and Bob is the critical
resource employed to enable the kind of quantum information transfer implicit
in quantum teleportation.

Unitary
Bell measurement transfonnation
Source of
Alice entanglement Bob

Victor

Classical infonnation

Figure 9.1 Quantum teleportation.

In general terms, teleportation protocols proceed as follows. Alice and Bob

must first share two components of an entangled state. Alice makes a joint
"measurement" (generalized Bell-state measurement) of an unknown state 17/1)
(which is to be teleported) together with her component of the EPR pair. For
the purpose of verifying the protocol, 17P) is created by Victor (the "verifier").
Conditioned upon Alice's measurement, the overall state "collapses" into a state
for which Bob's component of EPR state is related to 17P) by a simple unitary
transformation, which is however unknown to Bob. Fortunately, receipt of the
measurement result (classical information) from Alice enables Bob to perform
an appropriate (unitary) transformation of his component of the original EPR
state to obtain a "recreation" of the original state 17P) at his place, as can then be
verified by Victor for the output from Bob's station. In some sense, Alice and
Bob share "quantum uncertainty" and subtract it at Bob's place to reconstruct
17P)·
 Experimental realization of continuous variable teleportation 79

Quantum teleportation was originally proposed by Bennett et al. [1], with

4
two-dimensional systems (e.g., the states of spin particles) having received
the greatest attention. Initial experiments were directed toward the polarization
states of single photons [4, 5]. However, the nature of the down conver-
sion process used for the generation of the initial state and of the quantum
entanglement together with the overall low efficiencies precluded these experi-
ments from crossing the boundary between classical and quantum teleportation
[6, 7, 8, 9]. An essential difficulty was the lack of ability to perform a complete
4
set of Bell-state measurements, which for spin particles (or photon polariza-
tion) requires simple (but as yet unattainable) quantum logic between Alice's
component of the entangled Bell pair and the unknown state.
On a different front, quantum teleportation with continuous variables in
an infinite dimensional Hilbert space was first proposed by Vaidman [10].
His proposal was further investigated theoretically by Braunstein and Kim-
ble, who introduced a teleportation scheme with non-singular squeezed-state
entanglement [11]. This latter scheme was experimentally demonstrated by
the Quantum Optics group at Caltech in 1998 [12]. Somewhat remarkably, in
this scheme complete Bell-state measurements can be performed by way of
quadrature-phase measurements with homodyne techniques whose detection
efficiency can be close to unity. The high detection efficiency of this scheme
together with the EPR entanglement generated via summing of independent
squeezed beams enabled the boundary between the classical and quantum tele-
portation to be crossed for the first time. More specifically, a teleportation
fidelity of 0.58 ± 0.02 was obtained for the teleportation of coherent states,
where the relevant quantum-classical boundary is 0.50 for this experiment
[8, 9]. "Bona fide" quantum teleportation was thus realized for the first time in
this experiment with continuous variables, which we discuss in more detail in
this chapter.

2. EXPERIMENTAL REALIZATION
OF CONTINUOUS VARIABLE TELEPORTATION
2.1 EPR CORRELATION
The quantum entanglement relevant to our continuous variable teleportation
experiment is the EPR correlation as in the original 1935 paper [3], which
relates to the canonically conjugate position and momentum variables. As
was originally pointed out by Reid and Drummond [13], the EPR gedanken
experiment can be realized via the quadrature-phase amplitudes of a single-
mode of the electromagnetic field, since these variables are also canonically
conjugate in direct correspondence to position and momentum. Indeed, the first
realization of the EPR experiment was accomplished by way of the quadrature-
phase amplitudes using nondegenerate parametric down-conversion process
80 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

with type-II phase matching [14]. An equivalent EPR state can also be produced
with type-I phase matching and a half beam splitter [12]. This state is called a
two-mode squeezed vacuum. In the Heisenberg representation, the quadrature-
phase amplitude operators (Xj, pj) are transformed as follows [15, 16].

1 r ,,-(0) 1 -r ,,-(0)
Xl = yl2e Xl + .,fie X2 ,
1 -r ,,-(0) 1 r ,,-(0)
PI = yl2e PI + yl2e P2

(9.1)

where o'j = Xj + ipj (o'j : annihilation operator), a superscript (0) denotes

initial vacuum modes, r is the squeezing parameter, and path 1, 2 correspond
to the pathways to Alice and Bob, respectively. Here, for r -+ 00 the state
becomes the ideal EPR state: (Xl - X2) -+ 0, (PI + P2) -+ 0. Note that r is

°
certainly not infinite in any real experiment; however, entanglement exists for
any finite value of r > (even infinitesimal r and even in the presence of loss).
[17, 18]

EPR
1 " be:uns
"

Figure 9.2 Experimental setup for the test of the EPR correlation of continuous variables.
 Experimental realization of continuous variable teleportation 81

Figure 9.2 shows the experimental setup for the realization ofEPR correlation
of continuous variables [12, 14]. Two squeezed vacuum beams are created by
parametric down conversion in a subthreshold optical parametric oscillator
(OPO). The OPO consists of a ring cavity and a potassium niobate K Nb0 3
crystal and has two independent opposite circulations. Two pump beams
(430nm) drive the OPO in the opposite directions; these pump beams are
created by way of second harmonic generation with another K Nb0 3 crystal
in an external buildup cavity which is itself driven by the fundamental 860nm-
output of a titanium:sapphire Ti : Al 2 0 3 laser. The two squeezed vacuum
beams are combined with a half beam splitter, where the relative phases of
the input squeezed vacuum beams are locked to be ~ with respect to each
other. The beams emerging from the beam splitter are in a two-mode squeezed
vacuum, thereby realizing entangled EPR beams. One of the EPR beams goes
to Alice's station and the other beam goes to Bob's station. Alice and Bob's
stations consist in the first instance of homodyne detectors which enable us to
confirm the correlation (and hence entanglement [17, 18]) between the beams
by way of the difference of the photocurrent from the detectors. This difference
signal corresponds to (Xl - X2) or uh +'fJ2) depending upon the relative phase
chosen between Alice and Bob's local oscillators (LO Alice, LOBob) [14].
Operationally speaking, the electromagnetic fields employed in experiments
such as these are not single-mode but rather have finite bandwidth. The single-
mode treatment must be generalized to the case of multimode fields of finite
bandwidth, as discussed in detail in Ref. [14]. In this situation, the relevant
quantities are the spectral components (x(O),p(O)) of the quadrature-phase
amplitudes, where a general quadratue-phase amplitude at phase 8 is defined
by
11!1+Ll!1
2(0,8) == -2 dO' [a(O') exp( -i8) + at (-0') exp( +i8)] (9.2)
!1-Ll!1
with a(a t ) as the annihilation (creation) operator for the field at offset 0 from
the optical carrier, with (x(O),p(O)) = (2(0,0),2(0,11"/2)), and with the
integration extending over a small interval .6.0 about O. Then, the spectral
density of photocurrent fluctuations \]! (0) obeys the following relation,
(9.3)
In simple terms, the overall treatment for the multimode case remains essen-
tially unchanged from the single-mode case. However, now the relevant state
describes the electromagnetic field at frequency offset ±O within a bandwidth
.6.0 about the carrier WL (laser frequency); that is to say, AM and FM modula-
tion sidebands.
Figure 9.3 shows an experimental result for the setup shown in Figure 9.2.
In Figure 9.3, the horizontal axis corresponds to the relative phase between
82 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Alice locked, Bob scanned

Tim. (sec)

Figure 9.3 EPR correlation for Alice and Bob beyond the vacuum-state limit.

Alice and Bob's local oscillators. Actually the phase of Alice's local oscillator
(LO) is locked to the phase of the EPR beam, and the phase of Bob's LO is
scanned. The vertical axis represents the photocurrent fluctuation W(O)EPR
for the difference output between Alice and Bob's photocurrents. Note that
W(D) E P R = OdE == W0 corresponds to the classical limit of the difference
output without quantum correlation. It is determined experimentally without
inputs to Alice and Bob's homodyne detectors (i.e., with vacuum inputs to their
detectors), and is the level arising from uncorrelated vacuum fluctuations at each
of the balanced detectors (i.e., two units of vacuum noise). Most importantly
is to examine the minima for wmin(D)EPR, which corresponds to a quantum
correlation between Alice and Bob's output [(:h - 1:2) or (lh +P2)] if and only
if Wmin < wo, which is indeed the case in Figure 9.3. In fact, the work of Refs.
[17, 18] ensures that the observation W min < W 0 is sufficient as to guarantee
that the beams are entangled. Of course, for the ideal EPR state, the level Wmin
would become arbitrarily small as compared to Wo (tending to -00 when
expressed on a logarithmic scale as in Figure 9.3). The maximum value Wmax
of the periodical curve indicates the level of anticorrelation implicit in the EPR
state. The shape of the curve rising and falling between W min and W max shows
the appearance of correlation and an anticorrelation between the quadrature
amplitudes at Alice and Bob's station as the relative phase between Alice and
 Experimental realization of continuous variable teleportation 83

Bob is scanned. The existence of quantum correlation and hence entanglement

in both quadrature-phase amplitudes (x and p) is thereby confirmed.

2.2 TELEPORTATION EXPERIMENT

Figure 9.4 Experimental setup for the continuous variable teleportation [12].

Having discussed briefly the operational verification of entanglement for the

EPR beams, we next turn to the experimental setup for the continuous-variable
teleportation as shown in Figure 9.4. First Alice and Bob share the EPR
beams that are created with the technique described in the previous subsection.
The state of the electromagnetic field to be teleported (Xin ,Pin) is created by
Victor, which for our experiment is more precisely a particular set of modulation
sidebands (coherent state). The beam to be teleported is combined with Alice's
EPR beam (Xl , PI) by using a half beam splitter. This process creates states
84 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

described by the quadrature amplitudes (xu,Pu) and (xv,Pv), where

Pu

Pv (9.4)

Alice makes a measurement of both quadrature-phase amplitudes Xu and

Pv by using two homodyne detectors and gets the classical values Xu and Pv'
This measurement corresponds to the Bell-state measurement of continuous
variables. In the ideal case (r -+ 00), Alice cannot get any information
on the state itself because the amount of quantum "noise" (e 2T ) in (xl,pd
is big enough to hide the information on the state to be teleported, which
is a key feature of the measurement protocol. If, by contrast, she were to
attempt to measure directly both quadrature-phase amplitudes of the state to
be teleported simultaneously without the entangled input, she would get only
partial information [2], and in the process destroy the state.
Figure 9.5 shows the outputs of one of Alice's homodyne detectors. The
horizontal axis corresponds to the phase of Alice's local oscillator ()Ax, which
is being swept in time. The vertical axis w(o)~lice corresponds to spectral
density of photocurrent fluctuations associated with the quadrature amplitude
xu(O, (}Ax) , where the maxima in w(o)~lice give the power (relative to the
vacuum state) for the amplitude xu(O, (}Ax = 0) == xu(O). In this experiment,
Victor generates a coherent state which consists of a classical phase-space
displacement and one unit of vacuum noise. The peak value in the periodic
modulation of \II(o):lice in the figure corresponds to the power associated
with Jz of the coherent displacement (-3dB), which is 22dB higher than the
vacuum noise level in this particular case. The reduction by 3dB arises because
the intensity of the unknown state is reduced by half by the beam splitter for
mixing the unknown state and Alice's EPR beam. The minima in the periodic
variation of w(o)~lice are equivalent to the level of the corresponding flat
trace A~lice, which is the quantum noise level with Alice's EPR beam present.
The associated level without Alice's EPR beam is ip~~ce (with a vacuum-
state input). The figure shows the quantum noise level with her EPR beam
is higher than the level without her EPR beam, namely A~lice > ip~~ce , in
correspondence to a loss of information by Alice for quantum telep~rtation.
Note that the quantum noise level with her EPR beam would diverge in the
ideal case (r -+ 00).
 Experimental realization of continuous variable teleportation 85

(a) 30

25

o 20 40 60 80

Time [msec)

Figure 9.5 Output of one of Alice's homodyne detectors (Dx). =

n/27r 2.9M H z and
!J.n/27r = 30kHz. The part (b) is the expanded view with a ten-trace average for the input
state which has no displacements, namely vacuum state. [12]

By way of Equation (9.4), Bob's EPR beam (X2,P2) is represented as follows

[15],

X'2 '
Xin - v 1n2 x 2 - V1n2'
L:e -r"(O) L:X u ,

P'2 '+ V1n2L:e -r PI

= Pin ,,(0) - V1n2 '
L:pv· (9.5)
86 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Alice's generalized Bell-state measurement results effectively in the quantum

variables Xu and Pv being transformed into the classical variables Xu and Pv in
the Equation (9.5). When the quantum efficiency of the homodyne detectors
('fJ) is less than unity, the Xu and Pv fluctuate under the influence of the invasion
of vacuum noise. In this case, Xu and Pv in the Equation (9.5) are replaced by
'fJXu + ~x~O) and 'fJPv + J1 - 'fJ2p~0), respectively, where x~O) and p~O)
are the quadrature amplitudes of the respective invading vacua.
Alice sends the measurement results Xu and Pv to Bob. He uses this infor-
mation to modulate a (coherent) light beam in both amplitude and phase, with
some overall gain 9 [12]. This modulated beam is then combined coherently
at the highly reflecting mirror mBob shown in Figure 9.4 to interfere with his
component of the entangled EPR beam (X2,P2), thereby creating the teleported
output state (Xtel,Pted. This procedure corresponds to a simple phase-space
displacement of Bob's EPR beam as follows:

Xtel X2 + g..,fixu,
Ptel P2 + g..,fipv. (9.6)

In the absence of losses ('fJ = 1) and for unity gain (g = 1), the quadrature
operators associated with the teleported state become

1n2 -r ,,(0)
Xtel Xin - Y X2 ,
A A

~e

Ptel
A

= Pin
A + Y1n2~e -r"-(O)
PI . (9.7)

For r -+ 00, Xtel -+ Xin, Ptel -+ Pin, in correspondence to perfect telportation.

Of course in an actual experiment, the gain 9 must be determined opera-
tionally. For the particular case of Figure 9.5, the displacement of the input
coherent state determined by Alice's homodyne detectors (22dB above the
vacuum-state limit) corresponds to half of the input signal power. If Bob's out-
put (as verified by Victor) carries twice the power specified by Alice's output
(namely, 25dB in the case at hand), the gain 9 is then determined to be unity,
namely OdB. Precisely speaking, the 9 should be corrected by the detection
efficiency ( associated with Alice's homodyne detection (propagation, homo-
dyne efficiency, and detector quantum efficiency). But since ( ~ 0.97 is almost
unity in our experiment, the aforementioned procedure for fixing 9 = 1 (OdB)
can be used with small error.
In somewhat more global terms, the actual procedure for determining 9 =
1 (OdB) is illustrated by Figure 9.6. This figure gives the variation of the
coherent amplitude and of the variance with gain g2 without EPR beams.
Since these two dependences are different and both agree with theory without
adjustable parameters, we can conclude that our setup functions in agreement
with our simple model. When Aout equals to Ain (here, Ain = Aout = 21dB),
 Experimental realization of continuous variable teleportation 87

10
. .. . •• ... ~

.....•.-.
5 ... . . .-•
--............. -

-2 o 2 4 6 A

Gain g? [dB)

Figure 9.6 The variation of the coherent amplitude Aout and of the variance ow with gain l
without EPR beams. The input amplitude Ain is +21dB above the vacuum-state limit in this
particular case. The solid lines are the theoretical curves for ( = 1.

we can detennine 9 = 1. From the Figure 9.6, we can see aw = 4.8dB for
9 = 1, whose meaning will be presented later.
Moving then to the case of teleportation in the presence of entangled EPR
beams, Bob combines his modulated beam with his EPR beam and reconstructs
the state to be teleported. In this process, the "noise" arising from the EPR
beam is effectively "subtracted" from Bob's modulated beam by destructive
interference at mBob.
Experimental results from this protocol are shown in Figure 9.7. The hori-
zontal axis corresponds to the phase of Victor's local oscillator, which is being
swept in time. The vertical axis 1lTVictor corresponds to the spectral density
of photocurrent fluctuations associated with the quadrature amplitudes Xtel (n)
and Pte I (n) measured by Victor for a fixed (but arbitrary) phase for the input
state. The maximum value of the periodic curve corresponds to a coherent
amplitude for the output state approximately 25dB above the vacuum-state
level <p~ictor; here, the gain has been set to be 9 ~ 1 as in the previous discus-
sion. This result shows the classical phase-space displacement is successfully
reconstructed.
88 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

(a)

-5
6
. y Victor
(b) 5
o
4 --' --- . --~- .-. ~ .~ ---~-~ _._.,- AVictor
3

,
2

0 -- -'~~--~-------- __
._- - <l>Viclor
-1 o
0 20 40 60 80

Time [msec]

Figure 9.7 Bob's output verified by Victor. The part (b) is the expanded view wi th a ten-trace
average for the vacuum input for Alice. [12]

The minima of the trace for WVi ctor correspond to the variance of the output
state for the quadrature orthogonal to that of the coherent amplitude, and are
equivalent to the level AVictor shown by the labeled flat trace. The various
phase-independent traces in the figure correspond to the quantum noise levels
with the EPR beams present for Alice and Bob (A Victor), without these EPR
beams at both locations (Y6i ctor) , and with a vacuum-state input to Victor's
 Experimental realization of continuous variable teleportation 89

homodyne detector (<p ~ ictor). Of course, "without the EPR beams" means
that vacuum noise (r = 0) invades Alice and Bob's stations, leading to a
degradation of the "quality" of teleportation.
Indeed, for teleportation of coherent states in the absence of shared entan-
glement between Alice and Bob (no EPR beams), Equation (9.7) shows that the
quantum noise for Bob's output becomes three units of vacuum noise (in either
quadrature, (f:l!i;~el)' (f:lP~el)). One unit comes from the original quantum noise
of the input coherent state, and the other two units correspond to successive
"quantum duties", the first being to cross the boundary from the quantum to
classical world (Alice's attempt to detect both quadrature amplitudes) and the
second from the classical to quantum (Bob's generation of a coherent displace-
ment) [11]. The experimental result T~ictor ~ 4.8dB in correspondence to a
factor of 3 above the vacuum-state limit in Figures 9.6 and 9.7 indicates almost
perfect performance of the "classical" teleportation with near unity detection
efficiency (recall ( = 0.97). As discussed in more detail in Ref. [8, 9], T~ictor
is the limit of "classical" teleportation, where explicitly we mean teleportation
without shared entanglement.
From Figure 9.7 and similar measurements, we determine that AVictor lies
1.ldB-lower than T~ictor. This means that quantum te1eportation is suc-
cessfully performed beyond the classical limit, as clarified by the following
discussion. To quantify the "quality" of the teleportation for a pure state l.,pin) ,
we calculate the teleportation fidelity F == (.,pin IPout l.,pin) [8,9]. For the case
of teleportation of coherent states, the boundary between classical and quantum
teleportation has been shown to be fidelity F = 0.50 [8, 9]. We stress that this
limit applies to the specific case of coherent states and only to the distinction
between what Alice and Bob can accomplish with and without shared entangle-
ment. Teleportation to accomplish other tasks in quantum information science
requires yet higher values for the fidelity.
Nonetheless, when the input state is a coherent state, the fidelity F of the
teleported output can be represented as follows [15]:

where aQ and a~ are the variances of the Q function of the teleported field
for the corresponding quadratures. The relevant variances aQ and ~ can be
determined from the measured efficiency factors in the experiment and are
90 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

given by the following equation [12]:

where rx,p are the squeezing parameters for the respective quadrature compo-
nents, 6,2 characterize the (amplitude) efficiency with which the EPR beams
are propagated and detected along paths(1,2), and'fJ gives the (amplitude) effi-
ciency for detection of the unknown input state by Alice. We stress that all of
these quantities can be directly measured, so that the comparison of theory as
in the above equation and the experimentally recorded variances can be made
with no adjustable parameters.

(Jw X.P 6

(dB] 7
Without EPR beams
6

With EPR beams

I~-L~~~~~~~~~~~

-.. -3 -2 -I 0 234

Gain g' (dB]

Figure 9.8 Variances a~l of the teleported field measured by Victor [12]. Open and filled
symbols in the figure are experimental results. The open squares represent the results for the case
with the slight imbalance of amount of squeezing in the two-mode squeezed vacuum. The filled
squares and triangles represent the results for the case of the balanced amount of squeezing. The
solid lines represent the theoretical predictions of Equation (9.9).

Following such a procedure, we show in Figure 9.8 the experimental results

for the variances a«f, as well as the theoretical prediction of Equation (9.9),
again with no adjustable parameters. By using these measured values of aQ
 Experimental realization of continuous variable teleportation 91

and a~ together with the independently measured values for the gain g, we can
use Equation (9.8) to arrive at an experimental estimate of the fidelity Fexp,
with the results shown by the points in Figure 9.8 for the cases with and without
the EPR beams present. We can also calculate Ftheory by way of Equations
(9.8, 9.9), with this theoretical prediction shown by the curves in Figure 9.9.
The agreement between theory and experiment is evidently quite good.

0.7

O.l' ~/With EPR beams

},
' \
0 .5 ; .,~ Without EPR beams
u. /t"t
II ~\
>- 0 .1
.~

\
Qi
'U 0.3
u:
1"
/
0.2

0 .1 ,I

.!ol.
~.
0.0
~ ·3 ·2
-
.,
,IV'"

0 2 3

Gain 92 [dB]

Figure 9.9 Fidelity F inferred from the measurement of Victor [12]. Open and filled squares
in the figure are experimental estimates of the fidelity F e xp. The open squares represent the
Fe x p for the case with the slight imbalance of amount of squeezing in the two-mode squeezed
vacuum. The filled squares represent the F ex p for the case of the balanced amount of squeezing.
The solid lines represent the theoretical predictions Ftheory.

From the Figure 9.9, we see that the fidelity Fexp for the case with EPR
beams exceeds the classical limit Fo = 0.50 for 9 = 1 (OdE), with the
maximum value F exp = 0.58 ± 0.02 obtained. Fexp > Fo is an unambiguous
demonstration of the quantum character of the teleportation protocol.

3. SUMMARY
The fidelity F exp = 0.58±0.02 has been obtained in an experiment with con-
tinuous variable quantum teleportation. This value exceeds the classical limit
for the fidelity for the teleportation of coherent states of the electromagnetic
field. As discussed in more detail in Ref. [8, 9], this is the first demonstration
92 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

of "bona fide" quantum teleportation for which every state entering Alice's
sending station is actually teleported to emerge from Bob's receiving station
with a fidelity exceeding that which would be possible if Alice and Bob shared
only a classical communication channel.

Acknowledgments
The authors express their gratitude to S. L. Braunstein, J. L. Sorensen, C. A
Fuchs, E. S. Polzik, and S. J. van Enk. This work was supported by the NSF,
by the ONR, and by DARPA via the QUIC Institute administered by the ARO.
AF acknowledges T. Ide for preparing the figures.

References
[1] Bennett, C. H., et al.,(1993) Teleporting an unknown quantum state via
dual classical and Einstein-Podolsky-Rosen Channels, Phys. Rev. Lett. 70,
1895.
[2] Arthurs, E. and Kelly Jr., J. L., (1965) On the simultaneous measurement
of a pair of conjugate variables, Bell. Syst. Tech. J. 44, 725.
[3] Einstein, A, Podolsky, B., and Rosen, N.,(1935) Can quantum-
mechanical description of physical reality be considered complete?, Phys.
Rev. 47, 777.
[4] Boschi, D., et al.,(1998) Experimental realization of teleporting an un-
known pure quantum state via dual classical and Einstein-Podolsky-Rosen
channels, Phys. Rev. Lett. 80, 1121.
[5] Bouwmeester, D., et al.,(1997) Experimental quantum teleportation, Na-
ture 390, 575.
[6] Braunstein, S. L., and Kimble, H. J., (1998) A posteriori teleportation,
Nature 394, 840.
[7] Kok, P., and Braunstein, S. L., (2000) Postselected versus nonpostselected
quantum teleportation using parametric down-conversion, Phys. Rev. A 61,
042304.
[8] Braunstein, S. L., Fuchs, C. A, and Kimble, H. J., (2000) Criteria for
continuous-variable quantum teleportation, J. Mod. Opt. 47, 267.
[9] Braunstein, S. L., Fuchs, C. A, Kimble, H. J., and van Loock, P., (2001)
Quantum versus classical domains for teleportation with continuous vari-
ables, Phys. Rev. A 64, 022321.
[10] Vaidman, L., (1994) Teleportation of quantum states, Phys. Rev. A 49,
1473.
[11] Braunstein, S. L., and Kimble, H. J., (1998) Teleportation of continuous
quantum variables, Phys. Rev. Lett. 80, 869.
 Experimental realization of continuous variable teleportation 93

[12] Furusawa, A, Sorensen, J. L., Braunstein, S. L., Fuchs, C. A, Kimble, H.

J., and Polzik, E. S., (1998) Unconditional quantum teleportation, Science
282,706.
[13] Reid, M. D. and Drummond P. D., (1988) Quantum correlations of phase
in nondegenerate parametric oscillator, Phys. Rev. Lett. 60, 2731; Reid M.
D., (1989) Demonstration of the Einstein-Podolsky-Rosen paradox using
nondegenerate parametric amplification, Phys. Rev. A 40, 913.
[14] Ou, Z. Y., Pereira, S. E, Kimble, H. J., and Peng, K. c., (1992) Realization
of the Einstein-Podolsky-Rosen paradox for continuous variables, Phys.
Rev. Lett. 68, 3663.
[15] van Loock, P., Braunstein, S. L., and Kimble, H. J., (2000) Broadband
teleportation, Phys. Rev. A 62,022309.
[16] Furusawa, A, (1999) Quantum teleportation, bit 31, no. 10, 2, in Japanese.
[17] Duan, L.-M., Giedke, G., Cirac, J. I., and Zoller, P., (2000) Inseparability
criterion for continuous variable systems, Phys. Rev. Lett. 84, 2722.
[18] Simon, R., (2000) Peres-Horodecki separability criterion for continuous
variable systems, Phys. Rev. Lett. 84, 2726.
Chapter 10

DENSE CODING FOR

CONTINUOUS VARIABLES*

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 1 UT, United Kingdom
[email protected]

H. J. Kimble
Norman Bridge Laboratory of Physics 12-33
California Institute of Technology, Pasadena, CA 91125

Abstract A scheme to achieve dense quantum coding for the quadrature amplitudes of the
electromagnetic field is presented. The protocol utilizes shared entanglement
provided by nondegenerate parametric down-conversion in the limit of large
gain to attain high efficiency. For a constraint in the mean number of photons
fj, associated with modulation in the signal channel, the channel capacity for
dense coding is found to be In(1 + fj, + fj,2), which always beats coherent-state
communication and surpasses squeezed-state communication for fi > 1. For
fi ~ 1, the dense coding capacity approaches twice that of either scheme.

An important component of contemporary quantum information theory is

the investigation of the classical information capacities of noisy quantum com-
munication channels. Here, classical information is encoded by the choice of
one particular quantum state from among a predefined ensemble of quantum
states by the sender Alice for transmission over a quantum channel to the re-
ceiver Bob. If Alice and Bob are allowed to communicate only via a one-way
exchange along such a noisy quantum channel, then the optimal amount of clas-
sical information that can be reliably transmitted over the channel has recently
been established [1,2].

• S. L. Braunstein and H. J. Kimble, Physical Review A 61, 042302/1-4 (2002).

Copyright (2000) by the American Physical Society.
96 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Stated more explicitly, if a classical signal a taken from the ensemble Po.
is to be transmitted as a quantum state Po., then Holevo's bound for a bosonic
quantum channel says that the mutual information H (A: B) between the sender
A (Alice) and receiver B (Bob) is bounded by [1]

(10.1)

where S(p) is the von Neumann entropy associated with the density operator
P = f d2 a Po. Po. for the mean channel state.
By contrast, if Alice and Bob share a quantum resource in the form if
an ensemble of entangled states, then quantum mechanics enables protocols
for communication that can circumvent the aforementioned bound on channel
capacity. For example, as shown originally by Bennett and Wiesner [3], Alice
and Bob can beat the Holevo limit by exploiting their shared entanglement to
achieve dense quantum coding. Here, the signal is encoded at Alice's sending
station and transmitted via one component of a pair of entangled quantum states,
with then the second component of the entangled pair exploited for decoding
the signal at Bob's receiving station. In this scheme, the cost of distributing
the entangled states to Alice and Bob is not figured into the accounting of
constraints on the quantum channel (e.g., the mean energy). Such neglect
of the distribution cost of entanglement is sensible in some situations, as for
example, if the entanglement were to be sent during off-peak times when the
communication channel is otherwise under utilized, or if it had been conveyed
by other means to Alice and Bob in advance (e.g., via a pair of quantum
CDs with stored, entangled quantum states). Note that in general, no signal
modulation is applied to the second (i.e., Bob's) component of the entangled
state, so that it carries no information by itself.
Although quantum dense coding has most often been discussed within the
setting of discrete quantum variables (e.g., qubits) [3,4], in this paper we show
that highly efficient dense coding is possible for continuous quantum variables.
As in our prior work on quantum teleportation [5, 6, 7], our scheme for achieving
quantum dense coding exploits squeezed-state entanglement, and therefore
should allow unconditional signal transmission with high efficiency, in contrast
to the conditional transmission with extremely low efficiency achieved in Ref.
[4]. More specifically, for signal states a associated with the complex amplitude
of the electromagnetic field, the channel capacity for dense coding is found to
be In(1 + n + n 2 ), where n is the mean photon number for modulation in
the signal channel. The channel capacity for dense coding in our scheme
thus always beats coherent-state communication and surpasses squeezed-state
communication for n > 1. For n » 1, the dense coding capacity approaches
twice that of either scheme.
 Dense coding for continuous variables 97

Alice

Quantum Channel

Entanglement Distribution

Figure 10.1 Illustration of the scheme for achieving super-dense quantum coding for signal
states over the complex amplitude ex = x + ip of the electromagnetic field. The quantum
resource that enables dense coding is the EPR source that generates entangled beams (1,2)
shared by Alice and Bob.

As illustrated in Fig. 10.1, the relevant continuous variables for our protocol
are the quadrature amplitudes (x,]3) of the electromagnetic field, with the
classical signal a = (x) +i(p) then associated with the quantum state Pex drawn
from the phase space for a single mode of the field. The entangled resource
shared by Alice and Bob is a pair of EPR beams with quantum correlations
between canonically conjugate variables (X,]3)(1,2) as were first described by
Einstein, Podolsky, and Rosen (EPR [8]), and which can be efficiently generated
via the nonlinear optical process of parametric down conversion, resulting in
a highly squeezed two-mode state of the electromagnetic field [9, 10]. In the
ideal case, the correlations between quadrature-phase amplitudes for the two
beams (1, 2) are such that

(10.2)

albeit it with an concomitant divergence in the mean photon number fi in each

channel.
Component 1 of this entangled pair of beams is input to Alice's sending
station, where the message M~ corresponding to the classical signal ain is
encoded as the quantum state Pain by a simple phase-space offset by way of the
displacement operator D(ain) applied to 1 [11]. The displacement D(ain) can
98 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

be implemented in a straightforward fashion by amplitude and phase offsets

generated by the (suitably normalized) classical currents (i xa , iPa) as in Ref. [7].
The state corresponding to Alice's displacement of the EPR beam constitutes
the quantum signal and is transmitted along the quantum channel shown in
Fig. 10.1 to Bob's receiving station, (Fig. 10.2) where it is decoded with the aid
of the second component 2 of the original EPR pair of beams and the homodyne
detectors (d x , dp). The resulting photocurrents (iXb' i pb ) suitably normalized
to produce aout = iXb + i ipb constitute the message Mf: received by Bob. In
the limit n -7 00, Eq. (10.2) ensures aout = ain, so that the classical message
would be perfectly recovered. However, even for finite n as is relevant to a
channel constrained in mean energy, the finite correlations implicit in the EPR
beams enable quantum dense coding with enhanced channel capacity relative
to either coherent state or squeezed state communication, as we now show.

Beam /12
(a) Ppz (b) a

Signal Beam /11

Channel
PIJI

EPRBeam2

Figure 10.2 Depiction of signal decoding at Bob's receiving station. (a) At Bob's 50 - 50
beam splitter mb, the displaced EPR beam 1 is combined with the component 2 to yield
two independent squeezed beams, with the /31,2 beams having fluctuations reduced below the
vacuum-state limit along (Xf3llPf3J. Homodyne detection at (d""d p ) (Fig. 10.1) with LO
phases set to measure (Xf31' Pf3J, respectively, then yields the complex signal amplitude Gout
with variance set by the associated squeezed states. (b) The net effect of the dense coding
protocol is the transmission and detection of states of complex amplitUde G with an effective
uncertainty below the vacuum-state limit (indicated by the dashed circle).
 Dense coding for continuous variables 99

Consider the specific case of EPR beams (1,2) approximated by the two-
mode squeezed state with Wigner function

WEPR(a1,a2) = 42 exp[-e- 2T (a1 -

7r
a2)~ - e2T (a1 - a2)J (10.3)

where the subscripts R and I refer to real and imaginary parts of the field
amplitude a, respectively (i.e., aR,I = x,p). Note that for r -t 00, the field
state becomes the ideal EPR state as described in Eq. (10.2), namely,

(10.4)

As shown in Fig. 10.1, signal modulation is performed only on mode 1, with

mode 2 treated as an overall shared resource by Alice and Bob (and which
could have been generated by Alice herself). The modulation scheme that we
choose is simply to displace mode 1 by an amount ain. This leads to a displaced
Wigner function given by WEPR (a1 - ain, a2), corresponding to the field state
that is sent via the quantum channel from Alice to Bob.
Upon receiving this transmitted state (consisting of the modulated mode 1),
the final step in the dense-coding protocol is for Bob to combine it with the
shared resource (mode 2) and retrieve the original classical signal ain with as
high a fidelity as possible. As indicated in Fig. 10.1, this demodulation can
be performed with a simple 50 - 50 beam splitter that superposes the modes
(1,2) to yield output fields that are the sum and difference of the input fields
and which we label as (31 and (32, respectively. The resulting state emerging
from Bob's beam splitter has Wigner function

The classical signal that we seek is retrieved by homodyne detection at detectors

(d x , dp ), which measure the analogs of position and momentum for the sum and
difference fields ((31, (32). For ideal homodyne detection the resulting outcomes
are distributed according to

where (3 = (31R + i(32J and represents a highly peaked distribution about the
complex displacement a/ V2. For large squeezing parameter r this allows us
to extract the original signal a which we choose to be distributed as

1
Pa = -2
7ra-
exp( -Ial
2
/a- 2 ) . (10.6)
100 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Note that mode 1 of this displaced state has a mean number of photons given
by

(10.7)

In order to compute the quantity of information that may be sent through this
dense coding channel we note the unconditioned probability for the homodyne
statistics is given by

p _ 2 ex ( -21,61 2 ) (10.8)
(,6) - 7r((]"2 + e-2r) p (]"2 + e-2r .

The mutual information describing the achievable information throughput of

this dense coding channel is then given by

(10.9)

For a fixed n in Eq. (10.7) this information is optimized when n = e r sinh r,

i.e., when (]"2 = sinh r cosh r so yielding a dense coding capacity of

(10.10)

which for large squeezing r becomes

cdense rv 4r . (10.11)

How efficient is this dense coding in comparison to single channel coding?

Let us place a "common" constraint of having a fixed mean number of photons
n which can be modulated. For a single bosonic channel Drummond and
Caves [12] and Yuen and Ozawa [13] have used Holevo's result to show that the
optimal channel capacity is just that given by photon counting from a maximum
entropy ensemble of number states. In this case the channel capacity (the
maximal mutual information) achieves the ensemble entropy, see Eq. (10.1),
so

c= S(p) = (1 + n) In(l + n) - n In n . (10.12)

Substituting n = e r sinh r into this we find

C rv 2r , (10.13)

for large squeezing r. This is just one-half of the asymptotic dense coding
mutual information, see Eq. (10.11). Thus asymptotically, at least, the dense
coding scheme allows twice as much information to be encoded within a given
 Dense coding for continuous variables 101

state, although it has an extra expense (not included within the simple constraint
n) of requiring shared entanglement.
It is worth noting that this dense coding scheme does not always beat the
optimal single channel capacity. Indeed, for small squeezing it is worse. The
break-even squeezing required for dense coding to equal the capacity of the
optimal single channel communication is

rbreak-even ::= 0.7809 , (10.14)

which corresponds to roughly 6.78 dB of two-mode squeezing or to n ::=

1.884. This break-even point takes into account the difficulty of making highly
squeezed two-mode inxsqueezed states. No similar difficulty has been factored
into making ideal number states used in the benchmark scheme with which our
dense coding scheme is compared.
A fairer comparison is against single-mode coherent state communication
with heterodyne detection. Here the channel capacity is well known [14,15,16]
for the mean photon number constraint to be

ccoh = In(1 + n) , (10.15)

which is always beaten by the optimal dense coding scheme described by

Eq. (10.10).
An improvement on coherent state communication is squeezed state com-
munication with a single mode. The channel capacity of this channel has been
calculated [16] to be

C sq = In(1 + 2n) , (10.16)

which is beaten by the dense coding scheme ofEq. (10.10) for n > 1, i.e., the
break-even squeezing required is

r~':eak-even ::= 0.5493 , (10.17)

which corresponds to 4.77 dB.

In summary, we have shown how to perform dense quantum coding for
continuous quantum variables by utilizing squeezed state entanglement. For a
constraint in the mean number of photons that may be modulated n, the dense
coding capacity is found to be In(1 +n+n2). This scheme always beats single-
mode coherent-state communication and surpasses single-mode squeezed-state
communication for n > 1. Note that in terms of actual implementation, our
protocol should allow for high efficiency, unconditional transmission with en-
coded information sent every inverse bandwidth time. This situation is in
contrast to implementations that employ weak parametric down conversion,
where transmission is achieved conditionally and relatively rarely. In fact Mat-
tIe et al. [4] obtained rates of only 1 in 107 per inverse bandwidth time [17].
102 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

By going to strong down conversion and using a characteristically different

type of entanglement, our scheme should allow information to be sent with
much higher efficiency and should simultaneously improve the ability to detect
orthogonal Bell states. Indeed, these advantages enabled the first experimen-
tal realization of unconditional quantum teleportation within the past year [7].
Beyond the particular setting of quantum communication discussed here, this
research is part of a larger program to explore the potential for quantum in-
formation processing with continuous quantum variables. Such investigations
are quite timely in light of important recent progress concerning the prospects
for diverse quantum algorithms with continuous variables, including universal
quantum computation [18] and quantum error correction [19, 20, 21], with
quantum teleportation being a prime example [5, 22, 23]. Although still in
its earliest stages, theoretical protocols have been developed for realistic phys-
ical systems that should allow a variety of elementary processing operations
for continuous quantum variables, including significantly quantum storage for
EPR states [24, 25].

S .L.B. was supported in part by the UK Engineering and Physical Sciences

Research Council and the Royal Academy of Engineering. The work ofH.J.K.
is supported by DARPA via the QUIC Institute which is administered by ARO,
by the Office of Naval Research, and by the National Science Foundation.

References
[1] A. S. Holevo, IEEE Trans. Info. Theory 44,269 (1998).
[2] B. Schumacher and M. D. Westmoreland, Phys. Rev. A 56, 131 (1997).
[3] c. H. Bennett and S. J. Wiesner, Phys. Rev. Lett. 69, 2881 (1992).
[4] K. MattIe, H. Weinfurter, P. G. Kwiat and A. Zeilinger, Phys. Rev. Lett.
76,4656 (1996).
[5] S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998).
[6] P. van Loock, S. L. Braunstein, and H. J. Kimble, "Broadband teleporta-
tion," LANL preprint, quant-phl9902030.
[7] A. Furusawa, J. Sl<1rensen, S. L. Braunstein, C. Fuchs, H. J. Kimble, and
E. S. Polzik, Science 282, 706 (1998).
[8] A. Einstein, B. Podolsky, N. Rosen, Phys. Rev. 47, 777 (1935).
[9] M. D. Reid and P. D. Drummond, Phys. Rev. Lett. 60,2731 (1988); M.
D. Reid, Phys. Rev. A 40,913 (1989).
[10] Z. Y. Ou, S. F. Pereira, H. J. Kimble, and K. C. Peng, Phys. Rev. Lett. 68,
3663 (1992); Appl. Phys. B: Photophys. Laser Chern. 55, 265 (1992).
[11] L. Mandel and E. Wolf, Optical Coherence and Quantum Optics, (Cam-
bridge University Press, Cambridge, England, 1995).
 Dense coding for continuous variables 103

[12] C. M. Caves and P. D. Drummond, Rev. Mod. Phys. 66, 481 (1994).
[13] H. P. Yuen and M. Ozawa, Phys. Rev. Lett. 70, 363 (1993).
[14] J. P. Gordon, Proc. IRE 50, 1898 (1962).
[15] C. Y. She, IEEE Trans. Inf. Theory IT·14, 32 (1968).
[16] Y. Yamamoto and H. A. Haus, Rev. Mod. Phys. 58, 1001 (1986).
[17] H. Weinfurter (private communication).
[18] S. Lloyd and S. L. Braunstein, Phys. Rev. Lett. 82, 1784 (1999).
[19] S. Lloyd and J. J.-E. Slotine, Phys. Rev. Lett. 80, 4088 (1998).
[20] S. L. Braunstein, Phys. Rev. Lett. 80, 4084 (1998).
[21] S. L. Braunstein, Nature (London) 394, 47 (1998).
[22] C. H. Bennett et al., Phys. Rev. Lett. 70, 1895 (1993).
[23] L. Vaidman, Phys. Rev. A49, 1473 (1994).
[24] A. S. Parkins and H. J. Kimble, e-print quant-phl9904062.
[25] A. S. Parkins and H. J. Kimble, e-print quant-phl9907049.
Chapter 11

MULTIPARTITE
GREENBERGER-HORNE-ZEILINGER
PARADOXES FOR CONTINUOUS VARIABLES

Serge Massar and Stefano Pironio

Service de Physique TMorique, CP 225,
Universite Libre de Bruxelles
1050 Brussels, Belgium

Abstract We show how to construct Greenberger-Home-Zeilinger type paradoxes for con-

tinuous variable systems. We give two examples corresponding to 3-party and
5-party paradoxes. The paradoxes are revealed by carrying out position and
momentum measurements. The structure of the quantum states which lead to
these paradoxes is discussed.

When studying continuous variables systems, described by conjugate vari-

ables with commutation relation [x,p] = i, it is natural to inquire how non-
locality can be revealed in those systems. Experimentally the operations that
are easy to carry out on such systems involve linear optics, squeezing and
homodyne detection. Using these operations, the states that can be prepared
are Gaussian states and the measurements that can be performed are measure-
ments of quadratures. But Gaussian states possess a Wigner function which
is positive everywhere and so provide a trivial local-hidden variable model for
measurement of x or p.
To exhibit non-locality in these systems, it is thus necessary to drop some
of the requirements imposed by current day experimental techniques. For in-
stance one can invoke more challenging measurements such as photon counting
measurements or consider more general states that will necessitate higher order
non-linear couplings to be produced. Using these two approaches it has recently
been possible to extend from discrete variables to continuous variables systems
the usual non-locality tests: Bell inequalities [1, 2, 3], Hardy's non-locality
proof [4] and the Greenberger-Horne-Zeilinger paradox [5, 6, 7].
105
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, lO5-lO9.
© 2003 Kluwer Academic Publishers.
106 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Greenberger-Horne-Zeilinger (GHZ) paradoxes [8] as formulated by Mer-

min [9] are particularly elegant and simple ways of demonstrating the non-
locality of quantum systems since the argument can be carried out at the level
of operators only. The existence of a generalization of the original GHZ para-
dox for qubits to continuous variables was first pointed out by Clifton [5] and
was studied in more details in [6] and [7]. The paradox presented in [7] in-
volve measurements of the parity of the number of photons, while in [5] and
[6], it is associated with position and momentum variables. It is this last case
that we will consider here. We shall summarize the results of [6] and show
that the multipartite multidimensional GHZ paradoxes introduced in [10] can
easily be generalized to the case of continuous variables by exploiting the non-
commutative geometry of the phase space. This idea is closely related to the
technique used to embed finite-dimensional quantum error correcting code in
the infinite-dimensional Hilbert space of continuous variables systems [11].
Let us introduce the dimensionless variables

- x d - pL
x = y'7fL an p = y'7f , (11.1)

where L is an arbitrary length scale. Consider the translation operators in phase

space

x a = exp(iax) and y,B = exp(i{3jJ) . (11.2)

These unitary operators obey the commutation relation

Xay,B = eia ,B/7rY,B X a , (11.3)

which follows from [x,jJ] = i/7r and the identity eAe B = e[A,BleBe A (valid
if A and B commute with their commutator). The continuous variable GHZ
paradoxes will be built out of these operators.
Let us first consider the case of three spatially separated parties, A, B,
C, each of which possess one part of an entangled system described by the
canonical variables XA,PA, XB,PB, xc and pc. Consider the operators X[7r
and Yj±7r acting on the space of party j (j = A, B, C). Since a{3 = ±7r 2 ,
it follows from (11.3), that these operators obey the commutations relations
X[7rYj±7r = - Yj±7r X[7r. Using these operators let us construct the following
four GHZ operators:

VI X A7r X B7r X C7r

V2 X- 7r y-7r y7r
A B C
V3 y7r X- 7r y-7r (11.4)
A B c
y-7r y7r X- 7r
V4 A B C
 Multipartite GHZ paradoxes 107

These four operators give rise to a GHZ paradox as we now show. First note
that the following two properties hold:

1. VI, 112, V3, V4 all commute. Thus they can be simultaneously diagonal-
ized (in fact there exists a complete set of common eigenvectors).

2. The product VI V2 V3 V4 = -IABC equals minus the identity operator.

These properties are easily proven using the commutations relations Xf7r 1j±7r =
_1j±7r Xf7r. Any common eigenstate of VI, V2, V3, V4 will give rise to a GHZ
paradox. Indeed suppose that the parties measure the hermitian operators x j
or Pj, j = A, B, C on this common eigenstate. The result of the measurement
associates a complex number of unit norm to either the Xj or Yj unitary op-
erators. If one of the combinations of operators that occurs in eq. (11.4) is
measured, a value can be assigned to one of the operators Vi, V2, V3, V4. Quan-
tum mechanics imposes that this value is equal to the corresponding eigenvalue.
Moreover - due to property 2 - the product of the eigenvalues is -1.
But this is in contradiction with local hidden variables theories. Indeed in
a local hidden theory one must assign, prior to the measurement, a complex
number of unit norm to all the operators Xj and Yj. Then taking the product of
the four c-numbers assigned simultaneously to VI, V2, V3, V4 yields +1 instead
of -l.
Remark that all other tests of non-locality for continuous variable systems [1,
2, 3, 4, 7] use measurements with a discrete spectrum (such as the parity photon
number) or involving only a discrete set of outcome (such as the probability that
x > 0 or x < 0). In our version of the GHZ paradox for continuous variables
this discrete character doesn't seem to appear at first sight. However it turn out
that it is also the case thought in a subtle way because eq. (11.4) can be viewed
as an infinite set of 2 dimensional paradoxes (see [6] for more details).
In [10], GHZ paradoxes for many parties and multidimensional systems
where constructed. These paradoxes where build using d-dimensional unitary
operators with commutation relations:

(11.5)

which is a generalization of the anticommutation relation of spin operators for

two-dimensional systems. Using xa and y.B and choosing the coefficients a
and {J such that a{J = 27[2/ d with d and integer, this commutation relation
can be realized in a continuous variable systems and so all the paradoxes
presented in [10] can be rephrased with minor modifications in the context of
infinite-dimensional Hilbert space.
Let us for instance generalise to continuous variables the paradox for 5 par-
ties each having a 4 dimensional systems described in [10]. We now consider
108 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

the operators x±q, yq and y-3q where q = 7r / J2. They obey the commu-
tation relation x±qyq = e±i7r /2yq x±q and X±qy-3q = e±i7r /2y-3q x±q.
Consider now the six unitary operators
WI X1 Xq
W2 X A YB~q
q
W3 Yl Xi/ (11.6)
W4 Yl y~
W5 Yl y~
W,6 y-3q yq
A . B
One easily shows that these six unitary operators commute and that their product
is minus the identity operator. Furthermore if one assigns a classical value to
x j and to Pj for j = A, B, C, D, E, then the product of the operators takes the
value + 1. Hence, using the same argument as in the three party case, we have
a contradiction.
There is a slight difference between the paradox (11.6) and the 4-dimensional
paradox described in [10]. The origin of this difference is that in a d-
dimensional Hilbert space, if unitary operators X, Y obey XY = e i7r / dy X,
then X d = yd = I (up to a phase which can be set to 1), or equivalently,
X d- I = xt and yd-I = yt. In the continuous case these relations no longer
hold and the GHZ operators Wi'S must be slightly modified, i.e. the operator
X- q = x qt and y-3q = y3qt have to be explicitly introduced in order for
the product of the Wi'S to give minus the identity. Note that the same remark
applies for the previous paradox (11.4) where in the descrete 2-dimensional
version xt = X and yt = y.
As we mentioned earlier the GHZ states are not Gaussian states. A detailed
analysis of the common eigenstates of VI, V2 , V3 , V4 is given in [6]. Let us give
an example of such an eigenstate. Define the following coherent superpositions
of infinitely squeezed states:
1
L
00 _

It) = J2 (Ii = 2k) + iii = 2k + 1))

k=-oo

L
00
1
I-!-) J2 (Ii = 2k) - iii = 2k + 1)) , (11.7)
k=-oo

where Ii) = Ix = v"ifLi). Then a common eigenstate of the four GHZ

operators VI, V2 , V3 , V4 is
(I t)AI t)BI t)c -I -!-)AI -!-)BI -!-)o) /v2.
However as shown in [6], for any choice of the eigenvalues of the operators
Vb there is an infinite family of eigenvectors, ie. the eigenspace is infinitely
degenerate.
 Multipartite GHZ paradoxes 109

In summary we have shown the existence of multipartite GHZ paradoxes

for continuous variable systems. These paradoxes involve measurements of
position and momentum variables only, but the states which are measured are
complex and difficult to construct experimentaly.

Acknowledgments
We would like to thank N. Cerf for helpful discussions. We acknowledges
funding by the European Union under project EQUIP (lST-FET program).
S.M. is a research associate of the Belgian National Research Foundation.

References
[1] K. Banaszek and K. W6dkiewicz, Phys. Rev. A 58,4345 (1998).
[2] A. Kuzmich, I. A. Walmsley and L. Mandel, Phys. Rev. Lett. 85, 1349
(2000).
[3] Z. Chen, 1. Pan, G. Hou and Y. Zhang, Phys. Rev. Lett. 88040406 (2002).
[4] B. Yurke, M. Hillery and D. Stoler, Phys. Rev. A 60,3444 (1999).
[5] R. Clifton, Phys. Lett. A 271, 1 (2000).
[6] S. Massar and S. Pironio, Phys. Rev. A 64 062108 (2001).
[7] Z. Chen and Y. Zhang, Phys. Rev. A 65 044102 (2001).
[8] D. M. Greenberger, M. Home, A. Zeilinger, in Bell's Theorem, Quan-
tum Theory, and Conceptions of the Universe, M. Kafatos, ed., Kluwer,
Dordrecht, The Netherlands (1989), p. 69.
[9] N. D. Mermin, Phys. Rev. Lett. 65, 3373 (1990) and Phys. Today, 43(6),
9 (1990).
[10] N. Cerf, S. Massar, S. Pironio, Greenberger-Home-Zeilinger paradoxes
for many qudits, quant-phlOl07031, to be published in Phys. Rev. Lett.
[11] D. Gottesman, A. Kitaev and J. Preskill, Phys. Rev. A 64 012310 (2001).
Chapter 12

MULTIPARTITE ENTANGLEMENT
FOR CONTINUOUS VARIABLES

Peter van Loock

ZentrumJUr Moderne Optik, Universitiit Erlangen-NiJ.rnberg, 91058 Erlangen, Germany
[email protected]

Samuel L. Braunstein
lnfonnatics, Bangor University, Bangor LL57 1Ur, United Kingdom
[email protected]

Abstract First, we show how the quantum circuits for generating and measuring mUlti-party
entanglement of qubits can be translated to continuous quantum variables. We
derive sufficient inseparability criteria for N -party continuous-variable states and
discuss their applicability. Then, we consider a family of multipartite entangled
states (multi-party multi-mode states with one mode per party) described by
continuous quantum variables and analyze their properties. These states can be
efficiently generated using squeezed light and linear optics.

Keywords: Multipartite entanglement, squeezed light

1. INTRODUCTION
What is the main motivation to deal with continuous variables for quantum
communication purposes? Quantum communication schemes rely on state
preparation, local unitary transformations, measurements, and classical com-
munication. In addition, sometimes shared entanglement is part of the protocol.
Within the framework of quantum optics, these ingredients can be efficiently
implemented when they are applied to the continuous quadrature amplitudes
of electromagnetic modes. For example, the tools for measuring a quadrature
with near-unit efficiency or for displacing an optical mode in phase space are
provided by homodyne detection and feed-forward techniques, respectively.
Continuous-variable entanglement can be efficiently produced using squeezed
111
S.L. Braunstein and A.K. Pati (eds.). Quantum Information with Continuous Variables, 111-143.
© 2003 Kluwer Academic Publishers.
112 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

light and linear optics. In this chapter, we consider a rather general manifes-
tation of continuous-variable entanglement, namely that between an arbitrary
number of parties (modes). We will see that even those N-party entangled
states where none of the N parties can be separated from the others in the
total state vector are comparatively "cheap" in tenns of the resources needed:
their generation only requires one single-mode squeezed state and N - 1 beam
splitters.

2. MULTIPARTITE ENTANGLEMENT
The main subject of this section is mUlti-party entanglement of infinite-
dimensional states described by continuous variables. After a few general
remarks on entanglement between two and more parties in arbitrary dimen-
sions, we will show how the quantum circuits for creating and measuring qubit
entanglement may be translated to continuous variables. Then we derive in-
equalities that may serve as sufficient multi-party inseparability criteria for
continuous-variable states. These are applicable both for a theoretical test and
for an indirect experimental verification of multi-party entanglement. Finally,
we focus on a family of genuinely multi-party entangled continuous-variable
states whose members are fully inseparable with respect to all their parties.

2.1 TWO PARTIES VERSUS MANY PARTIES

Bipartite entanglement, the entanglement of a pair of systems shared by
two parties, is easy to handle for pure states. For any pure two-party state,
orthonormal bases of each subsystem exist, {Iu n )} and {Iv n )}, so that the total
state vector can be written in the "Schmidt decomposition" [1] as

(12.1)
n

where the summation goes over the smaller of the dimensionalities of the two
subsystems. The Schmidt coefficients Cn are real and non-negative, and satisfy
I:n c~ = 1. The Schmidt decomposition may be obtained by transforming the
expansion of an arbitrary pure bipartite state as

mk nmk n

with Cnn == Cn. In the first step, the matrix a with complex elements amk
is diagonalised, a = ucv T , where u and v are unitary matrices and c is a
diagonal matrix with non-negative elements. In the second step, we defined
lun ) == I: mumnlm) and Ivn) == I:k vknl k ) which form orthonormal sets due
to the unitarity of u and v and the orthonormality of 1m) and Ik). A pure state of
two d-Ievel systems ("qudits") is now maximally entangled when the Schmidt
 Multipartite entanglement/or continuous variables 113

coefficients of its total state vector are all equal. Since the eigenvalues of the
reduced density operator upon tracing out one half of a bipartite state are the
Schmidt coefficients squared,

PI = Tr2,o12 = Tr2!1f;h2(1f;! = L c~!unh (un! ,

n
(12.3)

tracing out either qudit of a maximally entangled state leaves the other half
in the maximally mixed state n/ d. A pure two-party state is factorizable (not
entangled) if and only if the number of nonzero Schmidt coefficients is one. Any
Schmidt number greater than one indicates entanglement. Thus, the "majority"
of pure state vectors in the Hilbert space of two parties are nonmaximally
entangled. Furthermore, any pure two-party state is entangled if and only if
for suitably chosen observables, it yields a violation of inequalities imposed by
local realistic theories [2]. A unique measure of bipartite entanglement for pure
states is given by the partial von Neumann entropy, the von Neumann entropy
(-Tr,olog,o) of the remaining system after tracing out either subsystem [3]:
Ev.N. = -Tr,ollogd,ol = - L:n c~ logd c~, ranging between zero and one (in
units of "edits").
Mixed states are more subtle, even for only two parties. As for the quan-
tification of bipartite mixed-state entanglement, there are various measures
available such as the entanglement of formation and distillation [4]. Only for
pure states, these measures coincide and equal the partial von Neumann en-
tropy. The definition of pure-state entanglement via the non-factorizability of
the total state vector is generalized to mixed states through non-separability
(or inseparability) of the total density operator. A general quantum state of a
two-party system is separable if its total density operator is a mixture (a convex
sum) of product states [5],

(12.4)

Otherwise, it is inseparable 1. In general, it is a non-trivial question whether

a given density operator is separable or inseparable. Nonetheless, a very
convenient method to test for inseparability is Peres' partial transpose criterion
[6]. For a separable state as in. Eq. (12.4), transposition of either density matrix
yields again a legitimate non-negative density operator with unit trace,
~, (~)T ~ (12.5)
P12 = L.....; Fi Pil
""' D
® Pi2 ,

since (,oil) T = (,oil) * corresponds to a legitimate density matrix. This is a nec-

essary condition for a separable state, and hence a single negative eigenvalue
of the partially transposed density matrix is a sufficient condition for insepa-
rability. In the (2 x 2)- and (2 x 3)-dimensional cases (and, for example, for
114 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

two-mode Gaussian states, see below), this condition is both necessary and suf-
ficient. For any other dimension, negative partial transpose is only sufficient for
inseparability [7] 2. Other sufficient inseparability criteria include violations
of inequalities imposed by local realistic theories (though mixed inseparable
states do not necessarily lead to such violations), an entropic inequality [namely
Ev.N. (PI) > Ev.N. (h2), again with PI = Tr 2P12] [11], and a condition based
on the theory of majorization [12]. Concluding the discussion of two-party
entanglement, we emphasize that both the pure-state Schmidt decomposition
and the partial transpose criterion for mixed states are also applicable to infinite
dimensions. An example for the infinite-dimensional Schmidt decomposition
is the two-mode squeezed vacuum state in the Fock (photon number) basis
[13]. The unphysical operation (a positive, but not completely positive map)
that corresponds to the transposition is time reversal [14]: in terms of contin-
uous variables, any separable two-party state remains a legitimate state after
the transformation (XI,PI, X2,P2) -+ (Xl, -PI, X2,P2), where (Xi,Pi) are the
phase-space variables (positions and momenta) for example in the Wigner
representation. However, arbitrary inseparable states may be turned into un-
physical states, and furthermore, inseparable two-party two-mode Gaussian
states always become unphysical via this transformation [14].
Multipartite entanglement, the entanglement shared by more than two
parties, is a more complex issue. For pure multi-party states, a Schmidt
decomposition does not exist in general. The total state vector then cannot be
written as a single sum over orthonormal basis states. There is, however, one
very important representative of multipartite entanglement which does have the
form of a mUlti-party Schmidt decomposition, namely the Greenberger-Horne-
Zeilinger (GHZ) state [15]

IGHZ) = ~ (1000) + 1111)) , (12.6)

here given as a three-qubit state. Although there is no rigorous definition of

maximally entangled mUlti-party states due to the lack of a general Schmidt
decomposition, the form of the GHZ state with all "Schmidt coefficients" equal
suggests that it exhibits maximum multipartite entanglement. In fact, there
are various reasons for assigning the attribute "maximally entangled" to the
N-party GHZ states [(1000··· 000) + 1111···111) )/V2]. For example, they
yield the maximum violations of mUlti-party inequalities imposed by local
realistic theories [16]. Further, their entanglement heavily relies on all parties,
and if examined pairwise they do not contain simple bipartite entanglement
(see below).
 Multipartite entanglement/or continuous variables 115

For the case of three qubits, any pure and fully entangled state can be
transformed to either the GHZ state or the so-called W state [17],

IW) = ~ (1100) + 1010) + 1001)) , (12.7)

via stochastic local operations and classical communication ("SLOCC", where

stochastic means that the state is transformed with non-zero probability). Thus,
with respect to SLOCC, there are two inequivalent classes of genuine tripartite
entanglement, represented by the GHZ and the W state. Genuinely or fully
tripartite entangled here means that the entanglement of the three-qubit state is
not just present between two parties while the remaining party can be separated
by a tensor product. Though genuinely tripartite, the entanglement of the W
state is also "readily bipartite". This means that the remaining two-party state
after tracing out one party,

'IhIW)(WI = ~ (100)(001 + 110)(101 + 101)(011 + 101)(101 + 110)(011) ,

(12.8)

is inseparable which can be verified by taking the partial transpose [the eigen-
values are 1/3, 1/3, (1 ± v'5)/6]. This is in contrast to the GHZ state where
tracing out one party yields the separable two-qubit state

Tr1IGHZ)(GHZI ~ (100)(001 + 111)(111) (12.9)

= ~ (10)(01 ® 10)(01 + 11)(11 ® 11)(11)

n
Note that this is not the maximally mixed state of two qubits, 18>2/4. The
maximally mixed state of one qubit, however, is obtained after tracing out two
parties of the GHZ state. Maximum bipartite entanglement is available from
the GHZ state through a local measurement of one party in the conjugate basis
{(10) ± 11) )/J2} (plus classical communication about the result),

~ (I0)r ± 11)r)(1 (01 ± 1(11) IGHZ) 1 ±

H(I0)r ± 11)r)(1 (01 ± 1(11) IGHZ)II = J2(10)r ± 11)r) ® I<p ).
(12.10)

Here, I<p±) are two of the four Bell states, I<p±) = (100) ± 111) )/J2, IW±) =
(101) ± 11O))/J2·
What can be said about arbitrary mixed entangled states of more than two
parties? There is of course an immense variety of inequivalent classes of multi-
party mixed states [e.g., five classes of three-qubit states of which the extreme
cases are the fully separable (p = :Ei Pi Pil ®Pi2®Pi3) and the fully (genuinely)
116 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

inseparable states [18]]. In general, multi-party inseparability criteria cannot be

formulated in such a compact form as the two-party partial transpose criterion.
Similarly, the quantification of multipartite entanglement, even for pure states,
is still subject of current research. Existing multi-party entanglement measures
do not appear to be unique as is the partial von Neumann entropy for pure
two-party states. Furthermore, violations of mUlti-party inequalities imposed
by local realism do not necessarily imply genuine mUlti-party inseparability. In
the case of continuous variables, we may now focus on the following questions:
How can we generate, measure, and (theoretically and experimentally) verify
genuine multipartite entangled states? How do the continuous-variable states
compare to their qubit counterparts with respect to various properties?

2.2 CREATING MULTIPARTITE ENTANGLEMENT

A compact way to describe how entanglement may be created is in terms
of a quantum circuit. Quantum circuits consist of a sequence of unitary trans-
formations (quantum gates), sometimes supplemented by measurements. A
quantum circuit is independent of a particular physical realization.
Let us consider the generation of entanglement between arbitrarily many
qubits. The quantum circuit shall turn N independent qubits into an N -partite
entangled state. Initially, the N qubits shall be in the eigenstate 10). All we
need is a circuit with the following two elementary gates: the Hadamard gate,
acting on a single qubit as

1 1
10) ---+ v'2 (10) + 11)) , 11) ---+ v'2(IO) -11)) , (12.11)

and the controlled-NOT (C-NOT) gate, a two-qubit operation acting as

100) ---+ 100) , 101) ---+ 101) , 110) ---+ 111) , 111) ---+ 110) .(12.12)

The first qubit (control qubit) remains unchanged under the C-NOT. The second
qubit (target qubit) is flipped if the control qubit is set to 1, and is left unchanged
otherwise. Equivalently, we can describe the action of the C-NOT gate by
IYl, Y2) -+ IYl, Yl EEl Y2) with Yl, Y2 = 0,1 and the addition modulo two EEl.
The N -partite entangled output state of the circuit (see Fig. 12.1) is the N -qubit
GHZ state.
Let us translate the qubit quantum circuit to continuous variables [19].
The position and momentum variables x and p (units-free with n = 1,
[Xl,Pk] = idlk/2) may correspond to the quadrature amplitudes of a single
electromagnetic mode, i.e., the real and imaginary part of the single mode's
annihilation operator: a = x + ip. At this stage, it is convenient to consider
position and momentum eigenstates. We may now replace the Hadamard by a
 Multipartite entanglementfor continuous variables 117

10)
C-NOT
10)
10)
I
I

10) ED
Figure 12.1 Quantum circuit for generating the N-qubit GHZ state. The gates (unitary trans-
formations) are a Hadamard gate ("H") and pairwise acting C-Naf gates.

Fourier transform,

Flx)position = ~
y 7r
1 00

-00
dy e2ixYIY)position = Ip = X)momentum , (12.13)

and the C-NOT gates by appropriate beam splitter operations 3. The input
states are taken to be zero-position eigenstates Ix = 0). The sequence of beam
splitter operations Bjk (()) is provided by a network of ideal phase-free beam
splitters (with typically asymmetric transmittance and reflectivity) acting on
the position eigenstates as

BI2(())lxl,X2) = IXlsin()+x2coS(),XICOS()-x2sin()) = Ix~,x~).

(12.14)

Now we apply this sequence of beam splitters (making an "N -splitter"),

BN-IN(7r/4)BN-2N-l (Sin- l 1/J3") x ... X B12 (Sin- l 1/m) ,

(12.15)

I
to a zero-momentum eigenstate Ip = 0) oc dx Ix) of mode 1 (the Fourier
transformed zero-position eigenstate) and N - 1 zero-position eigenstates
Ix = 0) in modes 2 through N. We obtain the entangled N-mode state
I dx lx, x, . .. ,x). This state is an eigenstate with total momentum zero and
all relative positions Xi - Xj = 0 (i, j = 1,2, ... ,N). It is clearly an analogue
to the qubit GHZ state with perfect correlations among the quadratures. How-
ever, it is an unphysical and unnormalizable state (e.g., for two modes, it cor-
responds to the maximally entangled, infinitely squeezed two-mode squeezed
vacuum state with infinite energy). Rather than sending infinitely squeezed
position eigenstates through the entanglement-generating circuit, we will now
use finitely squeezed states.
118 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

In the Heisenberg representation, an ideal phase-free beam splitter operation

acting on two modes with annihilation operators Ck and Cl is described by

sin ()
( C'k ) -_ ( cos cos () ) ( Ck ) (12.16)
c
1 (} - sin (} Cl .

Let us now define a matrix Bkl((}) which is an N-dimensional identity matrix

with the entries Ikk, hi, Ilk, and III replaced by the corresponding entries of
the above beam splitter matrix. Thus, the matrix for the N -splitter becomes

U(N) == BN-IN (sin- 1 ~) BN-2N-l (sin- 1 ~)

x··· X B12 (sin- 1 .J-&) . (12.17)

The entanglement-generating circuit is now applied to N position-squeezed

vacuum modes. In other words, one momentum-squeezed and N - 1 position-
squeezed vacuum modes are coupled by an N -splitter,

(12.18)

where the input modes are squeezed [13] according to

0,1 cosh rl O,~O) + sinh rl O,~o)t ,

o'i = cosh r2o'~O) - sinh r2O,~o)t , (12.19)

with i = 2,3, ... ,N and vacuum modes labeled by the superscript '(0)'. In
terms of the input quadratures, we have

(12.20)

for o'j = Xj + ipj (j = 1,2, ... , N). The squeezing parameters rl and r2
determine the degree of squeezing of the momentum-squeezed and the N - 1
position-squeezed modes, respectively. The correlations between the output
quadratures are revealed by the arbitrarily small noise in the relative positions
and the total momentum for sufficiently large squeezing rl and r2,

e- 2T2 /2 ,
Ne- 2'fl /4, (12.21)

for k i- l (k, l = 1,2, ... , N) and O,~ = x~ + ip~. Note that all modes involved
have zero mean values, thus the variances and the second moments are identical.
 Multipartite entanglement for continuous variables 119

2.3 MEASURING MULTIPARTITE ENTANGLEMENT

Rather than constructing a circuit for generating entangled states, now our
task shall be the measurement of multi-party entanglement, i.e., the projection
onto the basis of maximally entangled multi-party states. For qubits, it is well-
known that this can be achieved simply by inverting the above entanglement-
generating circuit (a similar strategy also works for d-Ievel systems [20]). The
GHZ basis states for N qubits read
1
IW n ,m1,m2, ... ,mN_1) = y'2 (10) ®lml) ®l m 2) ® ... ®lmN-I)
+ (-l)nll) ®Il EEl ml) ®Il EEl m2) ® ... ®Il EEl mN-I)), (12.22)

where n, ml, m2, ... , mN-I = 0,1. The projection onto the basis states
{I wn,m1 ,m2 , ... ,mN -I)} is accomplished when the output states of the inverted
circuit (see Fig. 12.1),

(CNOTN- I NCNOTN-2N-I··· CNOT I 2 H d- 1

= H I CNOT 12 CNOT 23··· CNOTN-1N, (12.23)

are measured in the computational basis. Eventually, {IW n ,m1,m2, ... ,mN_J}
are distinguished via the measured output states

Reentering the domain of continuous variables, let us now introduce the

maximally entangled states

..;:;r 1
00
-1 dx e2ivX lx) ®Ix - U1)
-00

®Ix - U1 - U2) ® ®Ix - UI - U2 - ... - UN-I) . (12.25)

Since J~oo Ix) (xl = n and (xix') = J(x - x'), they form a complete,

= n®N, (12.26)
and orthogonal,

(w( v, U1, U2, ... , UN -1) Iw( v', U~, U;, ... , U~ -1))
= J(v - V')J(U1 - U~)J(U2 - U;)··· J(UN-1 - U~_l)' (12.27)

set of basis states for N modes. For creating continuous-variable entan-

glement, we simply replaced the C-NOT gates by appropriate beam split-
ter operations. Let us employ the same strategy here in order to measure
120 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

continuous-variable entanglement. In other words, a projection onto the

continuous-variable GHZ basis {Iw(v, Ul, U2, ... , UN-I)}} shall be performed
by applying an inverse N -splitter followed by a Fourier transform of mode
1 and by subsequently }De!lsurin~ the positions of all modes. For an N-
mode state with modes bl, b2 , ..• , bN, this means that we effectively measure
Iill bAI1 = AI R b AI R b R bAI - AI
PI' e 2 = X2' e 3 = X3, ... , e N = XN , WI
- AI - AI - AI ·th

(12.28)

For instance, in the three-mode case, the measured observables are

1 (A A A)
AI
PI - V3 PI + P2 + P3 ,

AI
X2 V(L 1 (A A)
3"Xl - v'6 X2 + X3 ,

1 (A A)
AI
X3 V2 X2 - X3 , (12.29)

where here bj = Xj + ipj. In fact, in a single shot, the quantities v/V3,

fi73(Ul + U2/ 2), and U2/V'i are determined through these measurements,
and so are all the parameters v == PI +P2+P3, Ul == Xl -X2, andu2 == X2 -X3
required to detect the basis state Iw(v, Ul, U2)) from Eq. (12.25) with N = 3.
In general, for arbitrary N, the measurements yield p~ = v / v'N and

= Is (UN-4 + ~ (UN-3 + ~ (UN-2 + ~UN-l))) ,

= ~ (UN-3 + ~ (UN-2 + ~UN-l)) ,

= Ii(UN-2+~UN-l)'
1
V'i UN - l , (12.30)

where v == PI + P2 + ... + PN, Ul == Xl - X2, U2 == X2 - X3,···, and UN-l ==

X N -1 - X N. This confirms that the inverse N -splitter combined with the appro-
priate homodyne detections (that is tools solely from linear optics) enables in
principle a complete distinction of the basis states {I w(v, Ul, U2, ... , UN -1)) } in
 Multipartite entanglementfor continuous variables 121

Eq. (12.25). More precisely, the fidelity ofthe state discrimination can be arbi-
trarily high for sufficiently good accuracy of the homodyne detectors. We may
conclude that the requirements of such a "GHZ state analyzer" for continuous
variables are easily met by current experimental capabilities. This is in contrast
to the GHZ state analyzer for photonic qubits [capable of discriminating or
measuring states like those in Eq. (12.22)]. Although arbitrarily high fidelity
can be approached in principle using linear optics and photon number detec-
tors, one would need sufficiently many, highly entangled auxiliary photons and
detectors resolving correspondingly large photon numbers [21, 20]. Neither of
these requirements is met by current technology. Of course, the C-NOT gates
of a qubit GHZ state measurement device can in principle be implemented
via the so-called cross Kerr effect using nonlinear optics. However, on the
single-photon level, this would require optical nonlinearities of exotic strength.
In this section, we have shown how measurements onto the maximally
entangled continuous-variable GHZ basis can be realized using linear optics and
quadrature detections. These schemes are an extension of the well-known two-
party case, where the continuous-variable Bell basis [Eq. (12.25) with N = 2] is
the analogue to the qubit Bell states [Eq. (12.22) with N = 2]. The continuous-
variable and the qubit Bell states form those measurement bases that were used
in the quantum teleportation experiments [22] and [23, 24], respectively. The
extension of measurements onto the maximally entangled basis to more than
two parties and their potential optical realization in the continuous-variable
realm might be relevant to mUlti-party quantum communication protocols such
as the multi-party generalization of entanglement swapping [25]. However, the
entanglement resources in a continuous-variable protocol, namely the entangled
continuous-variable states that are producible with squeezed light and beam
splitters, exhibit only imperfect entanglement due to the finite degree of the
squeezing. When can we actually be sure that they are multi-party entangled at
all? In the next section, we will address this question and discuss criteria for the
theoretical and the experimental verification of mUltipartite continuous-variable
entanglement.

2.4 SUFFICIENT INSEPARABILITY CRITERIA

For continuous-variable two-party states, an inseparability criterion can be
derived that does not rely on the partial transpose. It is based on the variances
of quadrature combinations such as :h - X2 and PI + P2, motivated by the fact
that the maximally entangled bipartite state I dx Ix, x) is a (zero-)eigenstate
of these two combinations [26]. Similarly, in a continuous-variable Bell mea-
surement, the quadrature combinations Xl - X2 and PI + P2 are the relevant
observables to be detected. Hence, for two modes, applying the variance-
based inseparability criterion and measuring in the maximally entangled basis
122 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

can both be accomplished by equal means, namely a single beam splitter and
two homodyne detectors. In other words, the effectively inverse circuit for
the generation of bipartite entanglement provides the recipe for both measur-
ing maximum entanglement and verifying nonmaximum entanglement. When
looking for multi-party inseparability criteria for arbitrarily many modes, it
seems to be natural to pursue a similar strategy. We are therefore aiming at a
criterion which is based on the variances of those quadrature combinations that
are the measured observables in a continuous-variable GHZ measurement.
Let us consider three modes. According to Eq. (12.29), we define the
operators

(12.31)

where we added a factor of /2 in w compared to the first line of Eq. (12.29).

Let us further assume that the three-party state p is fully separable and can be
written as a mixture of tripartite product states,

(12.32)

Using this state, we can calculate the total variance of the operators in Eq. (12.31),

((~U)2)p + ((~v)2)p + ((~w)2)p

= L Pi ((U 2)i + (V 2)i + (W 2)i) - (U)~ - (V)~ - (w)~

L Pi ~ ((xi)i + (X~)i + (X~)i + (Pi)i + (P~)i + (P~)i)

i

- LPi ~ ((X1)i(X2)i + (Xl)i(X3)i + (X2)i(X3)i

~
 Multipartite entanglementfor continuous variables 123

L Pi ~ (((~;h)2)i + ((~X2)2)i + ((~X3)2)i

+ ((~PI)2)i + ((~p2)2)i + ((~P3)2)i)

+ ~P;(u)i- (~P;(U)i)' + ~P;(i»);- (~Pi (V).)'

+ ~ Pi (w)1- (~Pi (W)i)' , (12.33)

where (... )i means the average in the product state Pil 0 Pi2 0 Pi3. Similar
to the derivation in Ref. [26], we can apply the Cauchy-Schwarz inequality
2:i Pi(u); 2: (2:i Pi l(u)iI)2, and see that the last two lines in Eq. (12.33) are
bounded below by zero. Also taking into account the sum uncertainty relation
((b.Xj)2)i + ((b.pj)2)i 2: I[xj,pjll = 1/2 (j = 1,2,3), we find that the total
variance itself is bounded below by 1 (using 2:i Pi = 1). Any total variance
smaller than this boundary of 1 would imply that the quantum state concerned
is not fully separable as in Eq. (12.32). But would this also imply that the
quantum state is genuinely tripartite entangled in the sense that none of the
parties can be separated from the others (as, for example, in the pure qubit
states IGHZ) and IW)? This is obviously not the case and a total variance
below 1 does not rule out the possibility of partial separability. The quantum
state might still not be a genuine tripartite entangled state, since it might be
written in one or more of the following forms 4 [18]:

P= L Pi Pil2 0 Pi3, P = L PI PiI3 0 Pi2, P = L PI' Pi23 0 Pil.(12.34)

i

Thus, in general, a violation of ((b.u)2) P + ((b.v)2) P + ((~tV )2) p 2: 1 does not

necessarily witness genuine tripartite entanglement (a counterexample will be
given below). However, it does witness genuine tripartite entanglement when
the quantum state in question is pure and totally symmetric with respect to all
three subsystems [28]. In that case, a possible separation of any individual
subsystem,

(12.35)

implies full separability, P = PI 0 P2 0 P3. Hence a total variance below 1

negates the possibility of any form of separability in this case.
By extending the quadrature combinations in Eq. (12.31) from 3 to N parties
(corresponding to the output modes of an inverse N-splitter) and performing
a similar calculation as for N = 3 with an additional factor of VN - 1 in
the total momentum operator tV, we find that any N -mode state with modes
124 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

hI, h2' ... , hN which isfully separable, P = I:i Pi Pil ® Pi2 ® ... ® PiN, obeys
the inequality

(12.36)

Here, PIAl == I m b 1, X2 == Re b2, ... , XN==

A I AI AI AI Re bAIN are the correspond'mg out-
put quadratures of the inverse N -splitter applied to the modes hI, h2' ... , hN
[Eq. (12.28)]. Alternatively, one can also derive the following necessary con-
dition for full separability [28],

(12.37)

In this inequality, Xij = Xi - Xj and P = I:~I Pi are the relative positions

and the total momentum of the relevant state with modes hj = Xj + ipj.
The choice of the operators in the inequalities Eq. (12.36) and Eq. (12.37)
relies upon the fact that the quantum fluctuations of these observables simulta-
neously vanish for maximum GHZ entanglement. This must be in agreement
with their commutation relations, and indeed, we have

(12.38)

for any N and i, j. Similarly, the output quadratures of the inverse N -splitter
yield, for instance, for N = 3,

[PI + P2 + P3, X2 - X3] = 0, [PI + P2 + P3, 2XI - (X2 + X3)] = °.

(12.39)

Both criteria in Eq. (12.36) and Eq. (12.37) represent necessary conditions
for full separability, though they are not entirely equivalent [i.e., there are
partially inseparable states that violate Eq. (12.37), but satisfy Eq. (12.36),
see below]. Moreover, the criterion in Eq. (12.37) contains in some sense
redundant observables. As we know from the previous section, N observables
suffice to measure an N -party GHZ entangled state. These N observables are
suitably chosen quadratures of the N output modes of an inverse N -splitter.
Their detection simultaneously determines the total momentum and the N - 1
relative positions Xl -X2, X2 -X3, ... , and XN-I -XN [Eq. (12.30)] 5. However,
in Eq. (12.37), there are 1 + [N(N -1)]/2 different operators. Nevertheless, for
two parties and modes, the conditions in Eq. (12.36) and Eq. (12.37) coincide
and correspond to the necessary separability condition for arbitrary bipartite
states given in Ref. [26].
In summary, we have shown in this section that the circuit for measur-
ing GHZ entanglement also provides a sufficient inseparability criterion for
 Multipartite entanglementfor continuous variables 125

arbitrary multi-party continuous-variable states (pure or mixed, Gaussian or

non-Gaussian) of arbitrarily many parties. This criterion is experimentally ac-
cessible via linear optics and homodyne detections. The disadvantage of not
being a necessary inseparability condition (not even for Gaussian states, see
below) might be unsatisfactory from a theoretical point of view, but would not
be an obstacle to experimental inseparability proofs. A more serious drawback,
in particular when considering an experimental verification of mUlti-party en-
tanglement, is the fact that without additional assumptions, for arbitrary states,
the criteria presented in this section are in general not sufficient for genuine
mUlti-party inseparability. They verify only partial inseparability. In the next
section, we will give a simple example for this.

2.5 MULTI-PARTY ENTANGLED STATES

2.5.1 Partial multipartite entanglement. Let us investigate how the
following pure three-mode state described by the Heisenberg operators

behaves with respect to the mUlti-party inseparability criteria. Modes 1 and

2 are in a two-mode squeezed vacuum state [Eq. (12.18) and Eq. (12.20) for
N = 2 and r = rl = r2], and mode 3 is in the vacuum state. This three-party
state is obviously only partially 6 entangled (it is an example for the second class
of the five classes of three-mode Gaussian states in Ref. [27]). Applying an
inverse "tritter" (three-splitter) to these modes, calculating the relevant output
variances, and inserting them into Eq. (12.36) yields

_1 (1_e+ 2r + e- 2r ) + _
1 > _
1 (12.41)
4 3 6-2'

as a necessary condition for full separability. We find that equality holds for
r = 0 which doesn't tell us anything, though we know, of course, that the
state is fully separable in this case and must obey Eq. (12.36). For some finite
squeezing r, the inequality Eq. (12.41) is illustrated in Fig. 12.2. Similarly,
application of the criterion in Eq. (12.37) to the above state leads to

1 3
"4 (3e- 2r + cosh2r + 2) ~ "2 . (12.42)

Again, equality holds for r = O. In Fig. 12.2, also this condition is depicted
for some finite squeezing r.
126 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

1.6~--------------~--~

1.5~---------------,~
0.55
1.4
0.5~---------T--------~

1.3
0.45
1.2

o 0.2 0.4 0.6 0.8 1

o 0.2 0.4 0.6 0.8 1
squeezing r
squeezing r

Figure 12.2 Application of the necessary conditions for full three-party separability. On the
left: the inequality Eq. (12.42) with the boundary 3/2 as a function of the squeezing r; for
o < r < 1, the inequality is violated nearly everywhere. On the right: the inequality Eq. (12.41)
with the boundary 1/2 as a function of the squeezing r; for 0 < r < 1, the inequality is satisfied
nearly as much as it is violated.

The comparison between the two conditions in this example demonstrates

that they are not equivalent. For some squeezing, the partially entangled three-
mode state violates Eq. (12.37) while satisfying Eq. (12.36). Moreover, both
conditions can apparently be violated by an only partially entangled state.
Hence, both their violation, though ruling out full separability, does not imply
the presence of genuine multi-party entanglement. Another observation is that
both conditions are satisfied for sufficiently large squeezing when the partial
entanglement is sufficiently good. This confirms that the two conditions are
necessary for full separability, but not sufficient, not even for a Gaussian state
like that in our example. In fact, also the bipartite separability condition
of Ref. [26] is both necessary and sufficient only for Gaussian states in a
certain standard form (where any Gaussian state can be transformed into this
standard form via local operations). The partially entangled three-mode state
here lacks the symmetry that is required for a state to violate the separability
conditions whenever it contains some entanglement. We will now tum to a
family of multipartite entangled states which are totally symmetric with respect
to all their parties, which do always violate both conditions for full multi-party
separability, and which are indeed genuinely multi-party entangled.

2.5.2 Genuine multipartite entanglement. Let us consider the output

states of the entanglement-generating circuit in section 2.2. There, we applied
an N-splitter to one momentum-squeezed (rl) and N - 1 position-squeezed
(r2) vacuum modes to obtain the modes a~, a~, ... , a'tr [Eq. (12.18)]. Now one
can easily see that the first multi-party separability condition is violated for
any nonzero squeezing rl > 0 or r2 > 0, because application of an inverse
 MUltipartite entanglement/or continuous variables 127

N -splitter means

Ut(N) (a~ a2 ~I
aN
)T (12.43)
U t (N) U(N) (al a2 ~
aN )T ,
with aI, a2, ... , aN fromEq. (12.19). SinceUt(N)U(N) = I (identity matrix),
the squeezed quadratures ofEq. (12.20) can be directly inserted into Eq. (12.36)
yielding a violation of (e- 2r1 + e- 2r2 )/4 ~ 1/2 for any Tl > 0 or T2 > o.
Alternatively, using Eq. (12.21), the criterion in Eq. (12.37) becomes

( N2) 2(N N e- 2r1 N N

e- 2r2
- 1) + 4 = 4" (e- 2TI + e- 2r2 ) ~ 2" . (12.44)

This condition is also violated for any Tl > 0 or T2 > O. Due to their purity
and total symmetry we conclude that the members of the family of states which
emerge from the N -splitter circuit are genuinely mUlti-party entangled for any
Tl > 0 or T2 > O. This applies in particular to the case where Tl > 0 and
T2 = 0, i.e., when only one squeezed light mode is required for the creation of
genuine multipartite entanglement.
Independent of the inequalities Eq. (12.36) and Eq. (12.37), there are also
other ways to see that these particular states are genuinely multi-party entangled.
One simply has to find some form of entanglement in these states. For example,
by tracing out modes 2 through N of the pure N -mode state given in Eq. (12.18),
one finds that the remaining one-mode state is mixed, provided Tl > 0 or T2 > 0
[28]. Thus, the pure N -mode state is somehow entangled and hence genuinely
mUlti-party entangled due to its complete symmetry. Note that in order to infer
even only partial entanglement via tracing out parties, the N -mode state here
has to be pure. In contrast, the mUlti-party inseparability criteria of section 2.4
may verify partial inseparability for any N -mode state.
From a conceptual point of view, it is very illuminating to analyze which
states of the above family of N -mode states can be transformed into each other
via local squeezing operations [29]. For example, by applying local squeezers
with squeezing 81 and 82 to the two modes of the bipartite state generated with
only one squeezer [Eq. (12.20) for N = 2 with T2 = 0], we obtain

(12.45)
With the choice of 81 = 82 = TI/2 == T, the state in Eq. (12.45) is identical to
a two-mode squeezed state built from two equally squeezed states [Eq. (12.20)
128 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

for N = 2 with rl == r, r2 == r]. The latter and the state produced with only
one squeezer [Eq. (12.20) with rl = 2r and r2 = 0] are equivalent under local
squeezing operations. This means that Alice and Bob sharing the state produced
with one squeezer rl = 2r have access to the same amount of entanglement as
in the "canonical" two-mode squeezed state with squeezing r = rI/2, Ev.N. =
[cosh(rI/2)j2 log[cosh(rI/2)j2 - [sinh(rI/2)j2 log[sinh(rI/2)j2 [30]. For a
given amount of entanglement, however, the canonical two-mode squeezed
vacuum state has the least mean photon number. Conversely, for a given mean
energy, the canonical two-mode squeezed vacuum state contains the maximum
amount of entanglement possible.
Similar arguments apply to the states of more than two modes. From the
family of N -mode states, the state with the least mean photon number is
determined by the relation

c±'" = (N - 1) sinh2r2 [ (12.46)

This relation is obtained by requiring each mode of the N -mode states to be

symmetric or "unbiased" in the x and p variances [29]. Only for N = 2, we
obtain rl = r2. Otherwise, the first squeezer with rl and the N - 1 remaining
squeezers with r2 have different squeezing. In the limit of large squeezing, we
may use sinh 2r2 ~ e+ 2r2 /2 and approximate e+ 2r1 of Eq. (12.46) by

(12.47)

We see that in order to produce the minimum-energy N-mode state, the single
rl -squeezer is, in terms of the squeezing factor, N - 1 times as much squeezed
as each r2-squeezer. However, also in this general N -mode case, the other
N -mode states of the family can be converted into the minimum-energy states
via local squeezing operations. This applies in particular to the N-mode states
produced with just a single squeezer and to those built from N equally squeezed
states. As a result, due to the equivalence under local entanglement-preserving
operations, with a single sufficiently squeezed state and beam splitters, arbi-
trarily many genuinely multi-party entangled modes can be created just as well
as with N squeezers and beam splitters.
In contrast to the three-mode state given by Eq. (12.40), the output states of
the N -splitter are totally symmetric under interchange of parties. This becomes
more transparent when we look at the states in the Wigner representation. For
simplicity, let us assume r = rl = r2. The position-squeezed input states of
the N -splitter circuit, for instance, have the Wigner function

(12.48)
 Multipartite entanglement/or continuous variables 129

Through the linear N -splitter operation, the total input Wigner function to the
N -splitter (one momentum-squeezed and N - 1 position-squeezed vacuum
modes),

Win(X, p) = (~) N exp( _2e- 2r Xl 2 _ 2e+2rpI2)

x exp( _2e+ 2r X2 2 - 2e- 2r P2 2)

x exp(-2e+ 2r x3 2 - 2e- 2r p3 2)
x··· x exp(-2e+ 2r xN 2 - 2e- 2r pN 2) , (12.49)

is transformed into the output Wigner function

Here we have used x = (Xl, X2, ... , XN) and p = (PI,P2, .. ·,PN). The pure-
state Wigner function Wout (x, p) is always positive, symmetric among the N
modes, and becomes peaked at Xi - Xj = 0 (i,j = 1,2, ... , N) and PI +
P2 + ... + PN = 0 for large squeezing r. For N = 2, it exactly equals the
well-known two-mode squeezed vacuum state Wigner function [13], which is
proportional to 8(XI - x2)8(PI + P2) in the limit of infinite squeezing. As
discussed previously, the state Wout (x, p) is genuinely N -partite entangled for
any squeezing r > O. The quantum nature of the cross correlations XiXj and
PiPj appearing in Wout (x, p) for any r > 0 is also confirmed by the purity of
this state. This purity is guaranteed, since beam splitters tum pure states into
pure states (it can also be checked via the correlation matrix of the Gaussian
state Wout (x, p) [28]).
A nice example for a mUlti-party entangled state which is not a member
of the above family of states and not totally symmetric with respect to all its
modes is the (M + I)-mode state described by the Wigner function
130 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

(:;2)M+l exp { - 2e- 2T1

(sin()o Xl
() M+l
+ c~ ~ Xi )2

(sm()OPI +
. COS ()o M+I )2
_2e+ 2T1 ..JM ~ Pi

1 . () {Jf;
VM + 1 <- sm 0 <_ . M + 1 '
- r=:=::=== (12.52)

e- 2T! == ..JM sin ()o - cos ()o , -2T2 _ ..JM cos ()o - sin ()o
..JM sin ()o + cos ()o e = ..JM cos ()o + sin ()o .
The significance of this (M + I)-mode state is that it represents a kind of
multiuser quantum channel ("MQC") enabling optimal 1 -+ M "telec1oning"
of arbitrary coherent states from one sender to M receivers [31]. Though
not completely symmetric with respect to all M + 1 modes (but to modes 2
through M + 1), it is a pure Gaussian state which is indeed genuinely multi-
party entangled. This can be seen, because none of the modes can be factored
out of the total Wigner function. Despite its "asymmetry", this state is not
only partially multi-party entangled as is the asymmetric pure three-mode state
given by Eq. (12.40). Of course, the bipartite entanglement between mode 1
on one side and modes 2 through M + 1 on the other side is the most important
property of WMQc(X, p) in order to be useful for 1 -+ M telec10ning [31].
The generation of the state WMQc(X, p) is very similar to that of the above
family of multi-party entangled states produced with an N -splitter: first make
a bipartite entangled state by combining two squeezed vacua, one squeezed in
P with Tl and the other one squeezed in X with T2, at a phase-free beam splitter
with reflectivity/transmittance parameter () = eo.
Then keep one half (the
mode 1) and send the other half together with M - 1 vacuum modes through an
 Multipartite entanglement for continuous variables 131

M -splitter. The annihilation operators of the initial modes aj before the beam
splitters, j = 1,2, ... , M + 1, are then given by

0,1 = cosh rl ala) + sinh rl alO)t ,

0,2 cosh r2a~O) - sinh r2a~O)t ,
ai a~O) , (12.53)

where i = 3,4, ... ,M + 1.

By using the ideal phase-free beam splitter operation from Eq. (12.16), with
Bkl(B) this time representing an (M + I)-dimensional identity matrix with the
entries hk, hi, Ilk> and III replaced by the corresponding entries of the beam
splitter matrix in Eq. (12.16), the MQC-generating circuit can be written as

(b l b2 ... bM+l f = UMQc(M + 1) (0,1 0,2 ... aM+! f,

(12.54)

with

UMQc(M + 1) = BM M+l (sin- l ~) BM-IM (Sin- l ~)

x ... X B34 (sin- l 1
v'M-I
) B 23 (sin- l ~)
yM
x B12 (B o) . (12.55)

The first beam splitter, acting on modes 0,1 and 0,2, has reflectivity/transmittance
parameter B == Bo. The remaining beam splitters represent an M -splitter. In
Eq. (12.54), the output modes bj correspond to the M + 1 modes of the
MQC state described by WMQC in Eq. (12.51). Let us now return to the
totally symmetric multipartite entangled states given by Eq. (12.18) and explore
some of their properties. For simplicity, we will thereby focus on those states
emerging from the N -splitter circuit which are created with input states equally
squeezed in momentum and position, r = rl = r2.

2.5.3 Nonlocality and other properties. In this paragraph, we will dis-

cuss some of the properties of the state Wout (x, p) in Eq. (12.50). This will
further illustrate the character of W out (x, p) as a nonmaximally entangled
mUlti-party state. One such property is that this state, despite having an al-
ways positive Wigner function, violates N -party Bell-type [2] (or Mermin-type
[32]) inequalities imposed by local realism for any squeezing r > 0 [33]. The
observables producing these violations are displaced photon-number parities
rather than continuous variables such as x and p [34]. Like for the qubit states
[32], the violations increase as the number of parties N grows. However, this
132 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

increase becomes steadily smaller for larger N, as opposed to the exponential

increase for the maximally entangled qubit GHZ states [32]. This discrepancy
may be explained by the fact that the violations are exposed only for finite
squeezing where the state W out (x, p) is a nonmaximally entangled multi-party
state [33]. Note that, in general, the violations of N -party inequalities imposed
by local realism do not necessarily imply the presence of genuine multipartite
entanglement. However, for the pure and symmetric states Wout(x, p), once
again, proving some kind of entanglement means proving genuine multipartite
entanglement.
In order to prove the nonlocality exhibited by the state W (x, p) == W out (x, p),
let us now use the fact that the Wigner function is proportional to the quantum
expectation value of a displaced parity operator [35, 34]:

(12.56)

where 0 = x + ip = (aI, a2, ... , aN) and 11(0) is the quantum expectation
value of the operator
N N
fr(o) = @fri(ai) = Q9Di(ai)(-l)TtiD!(ad . (12.57)
i=l i=l

The operator Di (ad is the displacement operator,

D(a) = exp(aa t - a*a) , (12.58)

acting on mode i. Thus, fr( 0) is a product of displaced parity operators given

by

(12.59)

with the projection operators

L 12k)(2k ID! (ai),

00

Di(ai) (12.60)
k=O

L 12k + 1)(2k + II D !(ad ,

00

fr~-)(ad = Di(ai) (12.61)

k=O

corresponding to the measurement of an even (parity + 1) or an odd (parity -1)

number of photons in mode i. This means that each mode is now characterized
by a dichotomic variable similar to the spin of a spin-1I2 particle or the single-
photon polarization. Different spin or polarizer orientations from the original
 Multipartite entanglement/or continuous variables 133

qubit based Bell inequality are replaced by different displacements in phase

space. This set of two-valued measurements for each setting is just what we
need for the nonlocality test.
In the case of N -particle systems, such a nonlocality test is possible using the
N -particle generalization of the two-particle Bell-CHSH inequality [16]. This
inequality is based on the following recursively defined linear combination of
joint measurement results (in this paragraph, the symbol B does not refer to a
beam splitter operation),

BN == ~[o-(aN) + o-(a',y )]BN-1

+~[o-(aN) - o-(a',y)]B~_l = ±2, (12.62)

where o-(aN) = ±1 and o-(a',y) = ±1 describe two possible outcomes for two
possible measurement settings (denoted by aN and a',y) of measurements on
the Nth particle. Note, the expressions B~ are equivalent to BN but with all
the ai and a~ swapped. Provided that B N -1 = ±2 and B~ -1 = ±2, Equation
(12.62) is trivially true for a single run of measurements where 0-( aN) is either
+1 or -1 and similarly for o-(a',y). Induction proves Eq. (12.62) for any N
when we take

B2 == [0-(a1) + 0-(aD]0-(a2)
+[o-(ad - o-(a~)]o-(a~) = ±2 . (12.63)

Within the framework of local realistic theories with hidden variables A =

(AI, A2, ... , AN) and the nonnalized probability distribution P(A), we obtain
an inequality for the average value of BN == BN(A),

II dA1dA2 ... dANP(A)BN(A) I: ; 2. (12.64)

By the linearity of averaging, this is a sum of means of products of the o-(ad

and o-(aD. For example, if N = 2, we obtain the CHSH inequality

(12.65)

with the correlation functions

C(a1' a2) = I dA1dA2P(A1, A2)0-(a1' Ada(a2' A2) . (12.66)

Following Bell [2], an always positive Wigner function can serve as the hidden-
variable probability distribution with respect to measurements corresponding
to any linear combination of x and p. In this sense, the finitely squeezed two-
mode squeezed state Wigner function could prevent the CHSH inequality from
134 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

being violated when restricted to such measurements: W(XI,PI, X2,P2) _

P()I1, A2). The same applies to the Wignerfunction inEq. (12.50): W(x, p) =
P("\) could be used to construct correlation functions

C(a) = ! dAldA2 ... dANP(..\)

x a(al' AI)a(a2' A2)··· a(aN, AN), (12.67)

where a = (aI, a2, ... , aN). However, for parity measurements on each mode
with possible results ±1 for each differing displacement, this would require
unbounded 8-functions for the local objective quantities a(ai' Ai) [34], as in
this case we have

C(a) =II(a) = (7r/2)NW(a) . (12.68)

This relation directly relates the correlation function to the Wigner function
and is indeed crucial for the nonlocality proof of the continuous-variable states
in Eq. (12.50).
Let us begin by analyzing the nonlocal correlations exhibited by the entan-
gled two-party state. For this state, the two-mode squeezed state in Eq. (12.50)
with N = 2, we may investigate the combination [34]

82 = II(O, 0) + II(O, (3) + II(a, 0) - II(a, (3) , (12.69)

which according to Eq. (12.65) satisfies 182 1:::; 2 for local realistic theories.
Here, we have chosen the displacement settings al = a2 = 0 and a~ = a,
a2'-(3
- .

t
Writing the states in Eq. (12.50) as

II(a) = exp { - 2 cosh 2r 1",1' (12.70)

+ sinh2r [~ t(aiaj + aiaj) - f)a; + a i2 )] } ,

ZJ z=l

for N = 2 and a = (3 = iVY with the real displacement parameter :1 ~ 0 7,

we obtain 8 2 = 1 +2 exp( -2:1 cosh 2r) -exp( -4:1 e+ 2r ). In thelimitoflarge
r (so cosh 2r ~ e+ 2r /2) and small :1, 8 2 is maximized for :1 e+ 2r = (In 2) /3,
yielding B~ax ~ 2.19 [34], which is a c1ear violation of the inequality 182 1:::; 2.
Smaller violations also occur for smaller squeezing and larger:1. Indeed, for
any nonzero squeezing, some violation takes place [33].
We will now consider more than two parties. Let us first examine the three-
mode state by setting N = 3 in Eq. (12.50). According to the inequality of the
 Multipartite entanglement/or continuous variables 135

correlation functions derived from Eq. (12.62)-(12.64), we have

IC(al,a2,a~) + C(al,a~,a3) + C(a~,a2,a3) - C(a~,a~,a~)1 ~ 2.
(12.71)
Thus, for the combination
B3 = II(O, 0, ')') + II(O,,6, 0) + II(a, 0, 0) - II( a,,6, ')'), (12.72)
a contradiction to local realism is demonstrated by 1B31 > 2. The corresponding
°
settings here are al = a2 = a3 = and a~ = a, a~ = ,6, a~ = ')'. With the
choice a = v:Jei¢l, ,6 = v:Jei ¢2, and ')' = v:Jei ¢3, we obtain

B3 = L exp (- 2.1 cosh 2r -

3 2
3.1 sinh 2r cos 2<Pi ) (12.73)
i=l

exp { -6.1 cosh 2r - ~.1 sinh 2r ~[cos 2¢i - 4 cost ¢i + ¢j)] } .

Apparently, because of the symmetry of the entangled three-mode state,

equal phases <Pi should also be chosen in order to maximize B3. The best choice
is <PI = <P2 = <P3 = 1[/2, which ensures that the positive terms in Eq. (12.73)
become maximal and the contribution of the negative term minimal. Therefore,
we again use equal settings a = ,6 = ')' = iv:J and obtain
B3 = 3exp(-2.Jcosh2r+2.Jsinh2r/3) -exp(-6.Je+ 2T ). (12.74)
The violations of 1B31 ~ 2 that occur with this result are similar to the violations
of IB21 ~ 2 obtained for the two-mode state, but the N = 3 violations are even
more significant than the N = 2 violations [33]. In the limit of large r (and
small .1), we may use cosh 2r ~ sinh 2r ~ e+ 2T /2 in Eq. (12.74). Then B3
is maximized for .Je+2T = 3(ln3)/16: B!fax ~ 2.32. This optimal choice
requires smaller displacements .1 than those of the N = 2 case for the same
squeezing.
Let us now investigate the cases N = 4 and N = 5. From Eq. (12.62)-
(12.64) with N = 4, the following inequality for the correlation functions can
be derived:

~IC(al' a2, a3, a~) + C(al' a2, a~, a4) + C(al' a~, a3, a4)
+C(a~,a2,a3,a4) + C(al,a2,a~,a~) + C(al,a~,a3,a~)
+C(a~,a2,a3,a~) + C(al,a~,a~,a4) + C(a~,a2,a~,a4)
+C(a~,a~,a3,a4) - C(a~,a~,a~,a4) - C(a~,a;,a3,a~)
-C(a~,a2,a~,a~) - C(al,a;,a~,a~) - C(al,a2,a3,a4)
-C(a~,a;,a~,a~)1 ~ 2. (12.75)
136 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

It is symmetric among allfour parties as any inequality derived from Eq. (12.62)-
(12.64) is symmetric among all parties. For the settings al = a2 = a3 = a4 =
° and a~ = a, a~ = {3, a~ = "(, a~ = 8, complying with local realism means
184 1~ 2 where
1
84 = 2" [II(O, 0, 0, 8) + II(O, 0, ,,(, 0) + II(O, {3, 0, 0)
+II(a, 0, 0, 0) + II(O, 0, ,,(, 8) + II(O, {3, 0, 8)
+II(a, 0, 0, 8) + II(O, {3, ,,(, 0) + II(a, 0, ,,(, 0)
+II(a, {3, 0, 0) - II(a, {3, ,,(, 0) - II(a, {3, 0, 8)
-II(a, 0,,,(, 8) - II(0,{3,,,(,8) - II(O,O,O,O)
-II(a, {3, ,,(, 8)] . (12.76)

Similarly, for N = 5 one finds

1
2" [II(O, 0, 0, 8, E) + II(O, 0, ,,(, 0, E) + II(O, {3, 0, 0, E)
+II(a, 0, 0, 0, E) + II(O, 0, "(, 8, 0) + II(O, {3, 0, 8, 0)
+II(a, 0, 0, 8, 0) + II(O, {3, "(, 0, 0) + II(a, 0, ,,(, 0, 0)
+II(a, {3, 0, 0, 0) - II(a, {3, ,,(,8,0) - II(a, {3, ,,(, 0, E)
-II (a, {3, 0, 8,(0) -II(a,0,,,(,8,E) -II(0,{3,"(,8,E)
-II(O, 0, 0, 0, 0)] , (12.77)

which has to satisfy 185 1~ 2 and contains the same settings as for N = 4, but
°
in addition we have chosen a5 = and a~ = E.
We can now use the entangled states of Eq. (12.70) with N = 4 and N =
5 and apply the inequalities to them. For the same reason as for N = 3
(symmetry among all modes in the states and in the inequalities), the choice
a = {3 = "( = 8 = 10 = iVY appears to be optimal (maximizes positive terms
and minimizes negative contributions).
With this choice, we obtain

8 4 = 2 exp( - 2:1 cosh 2r + :1 sinh 2r)

- 2 exp( -6:1 cosh 2r - 3:1 sinh 2r)

+3 exp( -4:1 cosh 2r) - 2"1 exp ( -8:1e+2r ) - 2"1 '

85 5 exp(-4:1 cosh 2r + 4:1 sinh2rj5)
-~exp(-8:1cosh2r - 24:1sinh2rj5) - ~. (12.78)
2 2
Apparently, the maximum violation of IBNI ~ 2 (for our particular choice
of settings) grows with increasing number of parties N [33]. The asymptotic
 Multipartite entanglement/or continuous variables 137

analysis (large r and small .1) yields, for instance, for N = 5: Br ax ~

2.48 with .1e+ 2r = 5(ln 2) /24. For a given amount of squeezing, smaller
displacements .1 than those for N ~ 4 (at the same squeezing) are needed to
approach this maximum violation. Another interesting observation is that in
all four cases (N = 2,3,4,5), violations occur for any nonzero squeezing [33].
This implies the presence of N -partite entanglement for any nonzero squeezing.
Moreover, also for modest finite squeezing, the size of the violations (at optimal
displacement .1) grows with increasing N [33].
Larger numbers of parties N > 5 were also considered in Ref. [33]. The
degree of nonlocality of the continuous-variable states, if represented by the
maximum violation of the corresponding Bell-type inequalities, seems to grow
with an increasing number of parties. This growth, however, decelerates for
larger numbers of parties. Thus, the 'evolution' of the continuous-variable
states' nonlocality with an increasing number of parties and the corresponding
'evolution' of nonlocality for the qubit GHZ states are qualitatively similar but
quantitatively different with an exponential increase for the qubits. The reason
for this may be that the qubit GHZ states are maximally entangled, whereas the
continuous-variable states are nonmaximally entangled for any finite squeezing.
Similarly, the N -party version of the nonmaximally entangled qubit state IW)
yields a non-exponential increase of the maximum violations (by employing,
for example, an analysis analogous to that here [36]). Note that the observation
of the nonlocality of the continuous-variable states here requires small but
nonzero displacements .1 ex e- 2r , which is not achievable when the singular
maximally entangled states for infinite squeezing are considered.
Finally, the "unbiased" minimum-energy states of the family of entangled
N -mode states might yield larger violations. These states are not produced with
N equal squeezers (as those states whose nonlocality we have analyzed here),
but with one rl-squeezer and N -1 r2-squeezers related as in Eq. (12.46). With
growing N, the unbiased states increasingly differ from the states that we have
used for the nonlocality test [see Eq. (12.47) for large squeezing]. On the other
hand, the biased and the unbiased states are equivalent under local squeezing
operations and thus cannot differ in their potential nonlocality. In addition,
this equivalence shows that also the unbiased states are only nonmaximally
entangled for finite squeezing, which suggests that they also do not lead to an
exponential increase of the violations as for the qubit GHZ states.
In section 2.1, we discussed some properties of pure, fully entangled states
of three qubits. An important feature of these states is that a distinction can
be made between two inequivalent classes: states from the first class can be
converted into the state IGHZ) via SLOCC, but not into the state IW) (not even
with arbitrarily small probability). For the second class, exactly the opposite
holds. In several senses, the representative IGHZ) of the former class would be
best described as a maximally entangled state, whereas the representative IW)
138 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

of the latter class is nonmaximally entangled. A distinct feature of the maximum

entanglement of IGHZ) is that after tracing out one qubit, the remaining qubit
pair is in a separable mixed state 8. Apparently, the entanglement of IGHZ)
heavily relies on all three parties. By contrast, the entanglement of the state
IW) is robust to some extent against disposal of one qubit. When tracing
out one qubit of IW), the remaining pair shares a mixed entangled state. In
the continuous-variable setting, we can make analogous observations. By
J
interpreting the state dx lx, x, x) as the analogue of IGHZ), we see that

'Ih ! dxdx' Ix,x,x)(x',x',x'i = ! dx IX)22(xl ®lxh3(xl, (12.79)

which is clearly a separable mixed state (and indeed not the maximally mixed
J
state ex: dx dx' lx, x')(x, x'I). More interesting is the behaviour of a regular-
J
ized version of dx lx, x, x). In order to apply bipartite inseparability criteria,
let us trace out (integrate out) one mode of the Wigner function W out (x, p) in
Eq. (12.50) for N = 3,

Th W out (x, p) = ! dXl dPI W out (x, p)

+2r e+ 2r + 2e- 2r (2 2) -2r e- 2r + 2e+ 2r (2 2)

ex: exp [ - 2e e- 2r + 2e+ 2r x2 + x3 - 2e e+2r + 2e-2r P2 + P3
e+ 2r - e- 2r
+4e+2r e- 2r + 2e 4 -2r
+2 x2 x 3 + e
e- 2r - e+ 2r 1
r e-+--'2:---2---:
r + e- 2=-P2P3.
r

(12.80)
From the resulting Gaussian two-mode Wigner function, we can extract the
inverse correlation matrix. For Gaussian N -mode states with zero mean values,
the Wigner function is given by

W{e) - 1 exp {_~ eV - 1 eT } (12.81)

." - {27r)NJdetV 2"'''''
with the 2N -dimensional vector ehaving the quadrature pairs of all N modes
as its components,
e (Xl,Pl,X2,P2, ... ,XN,PN),
e = (Xl,Pl,X2,P2, ... ,XN,PN) , (12.82)
and with the 2N x 2N correlation matrix V having as its elements the second
moments (symmetrized according to the Weyl correspondence),
Tr[p (t::.eit::.ej + t::.ej t::.ei)/2] ((eie'} + tid/ 2)
= ! e
W{e) eiej d 2N = Vij, (12.83)
 Multipartite entanglement/or continuous variables 139

where b.~i = ~i - (~i) = ~i for zero mean values. The last equality defines
the correlation matrix for any quantum state, but for Gaussian states of the
form Eq. (12.81), the Wigner function is completely determined by the second-
moment correlation matrix. Now we can calculate the bipartite correlation
matrix of the state in Eq. (12.80),

+ 2C"
)
( e+2>- 0 2 sinh2r 0
V=~ 0
e- 2r + 2e+ 2r 0 -2 sinh2r
12 2 sinh2r 0 e+ 2r + 2e- 2r 0
0 -2sinh2r 0 e- 2r + 2e+ 2r
(12.84)

We could have also obtained this two-mode correlation matrix by extracting the
three-mode correlation matrix V of the state Wout(x, p) in Eq. (12.50) with
N = 3 and ignoring all entries involving mode 1 [or equivalently by explicitly
calculating the correlations between modes 2 and 3 with the Heisenberg opera-
tors in Eq. (12.18) for r = rl = r2 and N = 3]. The resulting two-mode state
is a (mixed) inseparable state for any nonzero squeezing r > O. Note that, for
instance, the total variance in Eq. (12.37) with N = 2 becomes for this state
(5e- 2r + e+ 2r )/6, which drops below the boundary of 1 only for sufficiently
small nonzero squeezing, but approaches infinity as the squeezing increases.
However, we can easily verify the state's inseparability for any r > 0 by look-
ing at the necessary two-party separability condition in product form given in
Ref. [37]. We find that

(12.85)

which drops below the separability boundary of 1/4 for any r > O. Of course,
also the necessary and sufficient partial transpose criterion from Ref. [14]
indicates entanglement for any r > 0 [28]. Recall that by first taking the
"infinite-squeezing limit" and then tracing out one mode, we had obtained a
separable state [Eq. (12.79)]. That was what we expected according to the
result for the maximally entangled qubit state IGHZ).
SO after all, we confirm what we had intuitively expected: the tripartite
state W out (x, p) for finite squeezing is a nonmaximally entangled state like the
qubit state IW). Only for infinite squeezing does it approach the maximally
entangled state f dx lx, x, x), the analogue of IGHZ). This result reflects
what is known for two parties. The two-mode squeezed state W out (x, p) with
N = 2 becomes a maximally entangled state f dx lx, x), such as the Bell state
(100) + 111))/ y'2, only for infinite squeezing. For finite squeezing, it is known
to be nonmaximally entangled.
140 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

3. CONCLUSIONS
Let us conclude by asking whether we were able to find answers to the
questions posed at the beginning of this chapter: how can we generate, measure,
and (theoretically and experimentally) verify genuine multipartite entangled
states for continuous variables? How do the continuous-variable states compare
to their qubit counterparts with respect to various properties?
As for the generation, we demonstrated that genuinely N -party entangled
states are producible with squeezed light resources and beam splitters. In
particular, one sufficiently squeezed light mode is in principle the only resource
needed to create any degree of genuine mUlti-party entanglement by means of
linear optics. The resulting states, though genuinely multi-party entangled, are
always nonmaximally entangled mUlti-party states due to the finite amount of
the squeezing. They behave like the N -party versions of the qubit state IW).
First, they also contain bipartite entanglement readily available between any
pair of modes (just as IW) and as opposed to the qubit state IGHZ). Secondly,
they yield a non-exponential increase of violations of mUlti-party Bell-type
inequalities for growing number of parties (as for IW) and different from the
qubit state IGHZ) for which the increase is exponential).
Furthermore, we have seen that by inverting the circuits for generating
genuine but nonmaximum mUlti-party entanglement, one can perform projec-
tion measurements onto the maximally entangled mUlti-party (GHZ) basis for
continuous variables. In contrast to the difficulties in performing such measure-
ments for photonic qubits within the framework of linear optics, continuous-
variable GHZ measurements only require beam splitters and homo dyne detec-
tors. In addition, we showed that the circuits for measuring maximum GHZ
entanglement are also applicable to the theoretical and experimental verifica-
tion of the nonmaximum entanglement of the multi-party states (which are
those producible in the laboratory). The circuits provide a necessary condition
for full separability of any N-partite N-mode state (pure or mixed, Gaussian
or non-Gaussian) with any number of modes N. However, this condition is
not sufficient for full separability and, more importantly, its violation does not
verify genuine but only partial multipartite entanglement. For the theoretical
verification of genuine multipartite entanglement, additional assumptions have
to be taken into account such as the total symmetry of the relevant states. There-
fore, an unambiguous experimental proof of genuine multipartite entanglement
of continuous-variable states was not proposed in this chapter. A possible ap-
proach to this would be to consider the violation of stricter N-party Bell-type
inequalities which cannot be violated by only partially entangled states. How-
ever, the experimental nonlocality test would then rely on observables such as
the photon number parity, and hence become unfeasible with current technol-
 Multipartite entanglementfor continuous variables 141

ogy. More desirable would be a test for genuine multipartite entanglement that
is solely based on linear optics and efficient homodyne detections.

Notes
1. Separable states also exhibit correlations, but those are purely classical. For instance, compare the
separable state P == ~(10)(010 10)(01 + 11)(11 0 11)(11) to the pure maximally entangled "Bell state"
I<J>+) == 0(10) 010) + 11) 011)) = 0(1+) 01+) + H 01-)) with the conjugate basis states
I±) = 0(10) ± 11)). The separable state p is classically correlated only with respect to the predetermined
basis {I 0), II)}. However, the Bell state I<J> +) is a priori quantum correlated in both bases {I 0), II)} and
{I +), 1-)}, and may become a posteriori classically correlated depending on the particular basis choice in a
local measurement. Similarly, we will see later that the inseparability criteria for continuous variables need
to be expressed in terms of the positions and their conjugate momenta.
2. Inseparable states with positive partial transpose cannot be distilled to a maximally entangled state
via local operations and classical communication. They are so-called "bound entangled" [8]. The converse,
however, does not hold. An explicit example of a bound entangled state with negative partial transpose was
given in Ref. [9]. In other words, not all entangled states that reveal their inseparability through negative
partial transpose are distillable or "free entangled". On the other hand, any state P12 that violates the
so-called reduction criterion, PI 0 !l - P12 2: 0 or !l 0 ih - Ih2 2: 0, is both inseparable and distillable
[10]. This reduction criterion is in general weaker than the partial transpose criterion and the two criteria
are equivalent in the (2 X 2)- and (2 x 3)-dimensional cases.
3. A possible continuous-variable generalization of the C-NOT gate is lXI, X2) -+ lXI, Xl + X2),
where the addition modulo two of the qubit C-NOT, IYl, YZ) -+ IYl, Yl ED Y2) with Yl, Y2 = 0,1, has
been replaced by the normal addition. However, for the quantum circuit here, a beam splitter operation as
described by Eq. (12.14) is a suitable substitute for the generalized C-NOT gate.
4. A full classification of tripartite Gaussian states is given in Ref. [27] in analogy to that for qubits
from Ref. [18]. In addition, necessary and sufficient three-mode inseparability criteria for Gaussian states
are proposed in Ref. [27].
5. The variances of the N - 1 relative positions :h - X2, X2 - X3, ... , and XN -1 - XN are also
available via the variances of the output quadratures of the inverse N -splitter. First, the variance of
x~ == 0 (x N -1 - XN ) corresponding to the last line in Eq. (12.30) is directly measurable. In addition,
by converting the measured photocurrent into a light amplitude and "displacing" (feed-forward) x~_l
according to x~_l -+ x'fv_l = x~_l - Jsx~ == !f(XN-2 - XN-l), one can directly measure
the variance of XN -2 - XN -1 etc. Similarly, one would also employ this feed-forward technique in a
mUlti-party quantum communication protocol that relies on the N classical results of an N -mode GHZ state
measurement.
6. note that we use the term "partially entangled" here for states which are not genuinely multi-
party inseparable. In the literature, sometimes "partial entanglement" is also referred to as nonmaximum
entanglement of two or more parties (in the sense that for two parties the Schmidt coefficients are not all
equal). As discussed later, also the genuinely multi-party entangled continuous-variable states are only
nonmaximally entangled due to the finite degree of the squeezing.
7. This choice of two equal settings leads to the same result as that of Banaszek and Wodkiewicz [34]
who used opposite signs: Q = v:J and f3 = -v:J.
8. However, remember that this is not the maximally mixed state for two qubits. Only when tracing
out two parties do we end up having the maximally mixed one-qubit state.

References

[1] E. Schmidt, Math. Annalen 63, 433 (1906).

[2] J. S. Bell, Physics (N.Y.) 1, 195 (1964).
[3] c. H. Bennett et ai., Phys. Rev. A 53,2046 (1996).
142 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[4] c. H. Bennett et ai., Phys. Rev. A 54,3824 (1996).

[5] R. F. Werner, Phys. Rev. A 40,4277 (1989).
[6] A. Peres, Phys. Rev. Lett. 77, 1413 (1996).
[7] M. Horodecki, P. Horodecki, andR. Horodecki, Phys. Lett. A 223, 1 (1996).
[8] M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. Lett. 80, 5239
(1998).
[9] D. P. DiVincenzo et aI., Phys. Rev. A 61,062312 (2000).
[10] M. Horodecki and P. Horodecki, Phys. Rev. A 59, 4206 (1999).
[11] R. Horodecki, P. Horodecki, and M. Horodecki, Phys. Lett. A 210, 377
(1996).
[12] M. A. Nielsen and J. Kempe, Phys. Rev. Lett. 86, 5184 (2001).
[13] D. F. Walls and G. J. Milburn, Quantum Optics, Springer Verlag Berlin
Heidelberg New York (1994).
[14] R. Simon, Phys. Rev. Lett. 84, 2726 (2000).
[15] D. M. Greenberger, M. A. Horne, A. Shimony, and A. Zeilinger, Am. J.
Phys.58, 1131 (1990).
[16] D. N. Klyshko, Phys. Lett. A 172, 399 (1993); N. Gisin andH. Bechmann-
Pasquinucci, Phys. Lett. A 246, 1 (1998).
[17] W. Dur, G. Vidal, and J. I. Cirac, Phys. Rev. A 62, 062314 (2000).
[18] W. Dur, J. I. Cirac, and R. Tarrach, Phys. Rev. Lett. 83, 3562 (1999).
[19] P. van Loock and S. L. Braunstein, Phys. Rev. Lett. 84, 3482 (2000).
[20] M. Dusek, Los Alamos arXive quant-ph/0107119 (2001).
[21] E. Knill, R. Laflamme, and G. J. Milburn, Nature 409, 46 (2001).
[22] A. Furusawa et at., Science 282, 706 (1998).
[23] D. Bouwmeester et at., Phys. Rev. Lett. 82, 1345 (1999).
[24] D. Boschi et at., Phys. Rev. Lett. 80, 1121 (1998).
[25] S. Bose, V. Vedral, and P. L. Knight, Phys. Rev. A 57,822 (1998).
[26] L.-M. Duan et at., Phys. Rev. Lett. 84, 2722 (2000).
[27] G. Giedke et at., Phys. Rev. A 64,052303 (2001).
[28] P. van Loock, Fortschr. d. Phys., to appear.
[29] W. P. Bowen, P. K. Lam, and T. C. Ralph, Los Alamos arXive quant-
ph/0104108 (2001).
[30] S. J. van Enk, Phys. Rev. A 60,5059 (1999).
[31] P. van Loock and S. L. Braunstein, Phys. Rev. Lett. 87, 247901 (2001).
[32] N. D. Mermin, Phys. Rev. Lett. 65, 1838 (1990).
[33] P. van Loock and S. L. Braunstein, Phys. Rev. A 63, 022106 (2001).
 Multipartite entanglement for continuous variables 143

[34] K. Banaszek and K. Wodkiewicz, Phys. Rev. A 58,4345 (1998).

[35] A. Royer, Phys. Rev. A 15, 449 (1977); H. Moya-Cessa and P. L. Knight,
Phys. Rev. A 48,2479 (1993).
[36] G. Li et ai., J. of Mod. Opt. 49, 237 (2002).
[37] S. M. Tan, Phys. Rev. A 60, 2752 (1999).
Chapter 13

INSEPARABILITY CRITERION FOR

CONTINUOUS VARIABLE SYSTEMS

Lu-Ming Duan 1 ,2, G. Giedke 1 , J. 1. Cirac l , and P. Zoller l

1 Institut fur Theoretische Physik, Universitiit Innsbruck, A -6020 Innsbruck, Austria
2 Laboratory of Quantum Communication and Quantum Computation, University of Science
and Technology of China, Hefei 230026, China
Email: [email protected]

As with discrete systems, quantum entanglement also plays the basic role in
quantum information protocols with continuous variables. A problem of great
importance is then to check whether a continuous variable state, generally
mixed, is entangled (inseparable). For discrete systems, there is the Peres-
Horodecki inseparability criterion [1, 2], based on the negativity of the partial
transpose of the composite density operator. This negativity provides a neces-
sary and sufficient condition for inseparability of 2 x 2 or 2 x 3-dimensional
systems. In this section, we will describe an entirely different inseparability
criterion for continuous variable states, which was first proposed in Ref. [3].
The Peres-Horodecki criterion was also successfully extended to the continu-
ous variable systems shortly afterwards, which will be described in the next
section by Simon.
The inseparability criterion described here is based on the total variance
of a pair of Einstein-Podolsky-Rosen (EPR) type operators. For any separable
continuous variable states, this total variance is bounded from below by a certain
value resulting from the uncertainty relation, whereas for entangled states
this bound can be exceeded. So violation of this bound provides a sufficient
condition for inseparability of any continuous variable state. Furthermore, for
the set of Gaussian states, which are of great practical importance, this criterion
turns out to be a necessary and sufficient condition for inseparability. In fact,
for any Gaussian state the compliance with the low bound by a certain pair
of EPR type operators guarantees that the state has a P-representation with
positive distribution, so the state must be separable.
145
S.L. Braunstein and A.K. Pari (eds.), Quantum Information with Continuous Variables, 145-153.
© 2003 Kluwer Academic Publishers.
146 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

We say a quantum state p of two modes 1 and 2 is separable if and only if it

can be expressed in the following form

p =L PiPil ® Pi2, (13.1)

where we assume Pil and Pi2 to be normalized states of the modes 1 and 2,
respectively, and Pi ~ 0 to satisfy L:
Pi = l.
i
A maximally entangled continuous variable state can be expressed as a co-
eigenstate of a pair of EPR-type operators [4], such as Xl + X2 and Pi - P2.
SO the total variance of these two operators reduces to zero for maximally
entangled continuous variable states. Of course, the maximally entangled
continuous variable states are not physical, but for the physical entangled
continuous variable states-the two-mode squeezed states [5], this variance will
rapidly tend to zero by increasing the degree of squeezing. Interestingly, we
find that for any separable state, there exists a lower bound to the total variance.
To be more general, we consider the following type of EPR-like operators:

u
~

= Ia I~Xl + -X2,
1~ (13.2)
a
v
~

= Iapi
I~ 1~
- -P2, (13.3)
a

where we assume a is an arbitrary (nonzero) real number. For any separable

state, the total variance of any pair of EPR-like operators in the form of Eq.
(13.2,13.3) should satisfy a lower bound indicated by the following theorem:
Theorem 1 (sufficient criterion for inseparability): For any separable
quantum state P, the total variance of a pair of EPR-like operators defined by
Eq. (13.2,13.3) with the commutators [Xj,Pj'] = ~8jj' (j,/ = 1,2) satisfies
the inequality

((.6.U)2) p + ((.6.v)2) p ~ ~ (a 2 + :2). (13.4)

u v
Proof. We can directly calculate the total variance of the and operators
using the decomposition (13.1) of the density operator p, and finally get the
following expression:
 Inseparability criterion for continuous variable systems 147

((LlU)2) P + ((Llv)2) p = ~ Pi ((U2)i + (V2)J - (U)~ - (V)~

~

~ Pi (a 2 (XD i + :2 (X~)i + a2 (fJI)i + :2 (~)i)

~

+2 1: 1
(~Pi (Xl)i (X2)i - ~ Pi (Pi)i (P2)i) - (U)~ - (v)~ (13.5)
~ ~

= ~Pi [a 2 ((( LlXl)2)i + ((LlPi)2)J

~

+ :2 (( (LlX2)2) + ((Llp2)2) J]
i

+ ~P' M; - (~P;(U)')' + ~p;(V); - (~> (V),)'

In Eq. (13.5), the symbol (···)i denotes average over the product density
operator Pil Q9 Pi2. It follows from the uncertainty relation that ((LlXj)2) i +
((Llpj)2) i ~ I[xj,pjll = 1/2 for j = 1,2, and by applying the Cauchy-

Schwarz inequality (~Pi) (~Pi (u) ~) ~ (~Pi 1(u) I) 2 , we know that

i

the last line of Eq. (13.5) is bounded from below by zero. Hence, the total
u v
variance of the two EPR-like operators and is bounded from below by
a 2 + -%- for any separable state. This completes the proof of the theorem.
a
Note that this theorem in fact gives a set of inequalities for separable states.
The operators Xj ,Pj (j = 1,2) in the definition (13.1) can be any local operators
satisfying the commutators [Xj,Pj'] = ~8jj'. In particular, if we apply an
u v,
arbitrary local unitary operation Ul Q9 U2 to the operators and the ineqUality
(13.4) remains unchanged. Note also that without loss of generality we have
taken the operators Xj and Pj dimensionless.
For inseparable states, the total variance of the u and v operators is required
by the uncertainty relation to be larger than or equal to ~ Ia 2 - ~ I ' which
reduces to zero for a = 1. For separable states the much stronger bound given
by Eq. (13.4) must be satisfied. A natural question is then how strong the
bound is. Is it strong enough to ensure that if some inequality in the form of Eq.
(13.4) is satisfied, the state necessarily becomes separable? Of course, it will be
very difficult to consider this problem for arbitrary continuous variable states.
However, in recent experiments and protocols for quantum communication
[6, 7, 8, 9, 10, 11, 12, 13], continuous variable entanglement is generated by
two-mode squeezing or by beam splitters, and the communication noise results
148 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

from photon absorption and thermal photon emission. All these processes
lead to Gaussian states. So, we will limit ourselves to consider Gaussian
states, which are of great practical importance. We find that the inequality
(13.4) indeed gives a necessary and sufficient inseparability criterion for all the
Gaussian states. To present and prove our main theorem, we need first mention
some notations and results for Gaussian states.
It is convenient to represent a Gaussian state by its Wigner characteristic
function. A two-mode state with the density operator p has the following
Wigner characteristic function [5]

tr [pexp !
(AlaI - Ai a + A2 a2 - A;an]
tr {pexp [2i (A{X1 + Afp1 + A~X2 + A~P2)J XP.6)
where the parameters Aj = Af + iA§, and the annihilation operators aj =
Xj + ipj, with the quadrature amplitudes Xj,Pj satisfying the commutators
[Xj ,Pj'] = ~ 15j j' (j, j' = 1, 2). For a Gaussian state, the Wigner character-
istic function x(w) (AI, A2) is a Gaussian function of Af and A§ [5]. Without
loss of generality, we can write x(w) (AI, A2) in the form

x(W) (AI, A2) = exp [-~ (Ai, Af, A~, A~) M (Ai, Af, A~, A~)T] (13.7)

In Eq. (13.7), linear terms in the exponent are not included since they can
be easily removed by some local displacements of Xj, pj and thus have no
influence on separability or inseparability of the state. The correlation property
of the Gaussian state is completely determined by the 4 x 4 real symmetric
correlation matrix M, which can be expressed as

M = (~~ g2)' (13.8)

where G 1 , G 2 , and Care 2 x 2 real matrices. To study the separability property,

it is convenient to first transform the Gaussian state to some standard forms
through local linear unitary Bogoliubov operations (LLUBOs) Ul = U 1 (>9 U2 .
In the Heisenberg picture, the general form of the LLUBO Ul is expressed as
Ul (xj,pjf U/ = Hj (Xj,pj)T for j = 1,2, where Hj is some 2 x 2 real
matrix with det H j = 1. Any LLUBO is obtainable by combining the squeezing
transformation together with some rotations [14]. We have the following two
lemmas concerning the standard forms of the Gaussian state:
Lemma 1 (standard form I): Any Gaussian state Pc can be transformed
through LLUBOs to the standard form I with the correlation matrix given by
 Inseparability criterion for continuous variable systems 149

n c
MI = (
scm :
n
) ,(n,m ~ 1) (13.9)
c'
Proof. A LLUBO on the state PG transforms the correlation matrix M in the
Wigner characteristic function in the following way

(13.10)

where VI and V2 are real matrices with det VI = det V2 = 1. Since the
matrices GI and G 2 in Eq. (13.8) are real symmetric, we can choose first
a LLUBO with orthogonal VI and V2 which diagonalize G I and G2, and
then a local squeezing operation which transforms the diagonalized G I and
G 2 into the matrices G~ = nI2 and G~ = mI2, respectively, where 12 is
the 2 x 2 unit matrix. After these two steps of operations, we assume the
matrix 0 in Eq. (13.8) is changed into 0', which always has a singular value
decomposition, thus it can be diagonalized by another LLUBO with suitable
orthogonal VI and V2. The last orthogonal LLUBO does not influence G~
and G~ any more since they are proportional to the unit matrix. Hence, any
Gaussian state can be transformed by three-step LLUBOs to the standard form
I. The four parameters n, m, c, and c' in the standard form I are related to
the four invariants det G I, det G 2 , det C, and det M of the correlation matrix
under LLUBOs by the equations det GI = n 2 , det G2 = m 2 , det C = cc',
and det M = (nm - c2) (nm - c'2) .
Lemma 2 (standard form ll): Any Gaussian state PG can be transformed
through LLUBOs into the standard form II with the correlation matrix given by

(13.11)

where the ni, mi and Ci satisfy

nI-1 n2 -1
, (13.12)
mI-1 m2- 1
hl-l c21 J(ni -1) (mi -1) - J(n2 - 1) (m2 -1).(13.13)
Proof. First, any Gaussian state can be tranformed through LLUBOs to the
standard form I. We then apply two additional local squeezing operations on
the standard form I, and get the state with the following correlation matrix
150 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

nrl J r lr2 C ,
, B..
Tl
c
.jTIT2
M= (13.14)
J r lr2 e , mr2
c m
.jTIT2 T2

where rl and r2 are arbitrary squeezing parameters. M' in Eq. (13.14) has the
standard form M! I (13 .11) if rl and r2 satisfy the following two equations
B..-1 m_1
Tl = _T'-"2'---_, (13.15)
nrl -1 mr2-1

y'rir2lcl- _Ie'l = !(nr, -1)(mr2 -I) - V(~ -1) (m - 1).

J r lr2 r2 rl
(13.16)

Our task remains to prove that Eqs. (13.15) and (13.16) are indeed satisfied by
some positive rl and r2 for arbitrary Gaussian states. Without loss of generality,
we assume lei ~ Ie' I and n ~ m. From Eq. (13.15), r2 can be expressed as a
continuous function of rl with r2 (rl = 1) = 1 and r2 (rl) Tl--+f m. Substitut-
ing this expression r2 (rl) into Eq. (13.16), we construct a function f (rl) by
subtracting the right hand side ofEq. (13.16) from the left hand side. Obviously,
f (rl = 1) = lei-Ie' I ~ 0, and f (rl) T~ Jrlm (lcl- In (m - !)) s
0, where the inequality J
lei s n (m - rk) results from the physical condition
((~11:0)2) + ((~VO)2) ~ 1[11:0,0'0]1 with 11:0 = Jm-
!Xl - rcrvnX2 and
0'0 = '(;: P2. It follows from continuity that there must exist a ri E [1, (0)
which makes f (rl = ri) = 0. So Eqs. (13.15) and (13.16) have at least one
solution. This proves lemma 2.
We remark that corresponding to a given standard form I or II, there are
a class of Gaussian states, which are equivalent under LLUBOs. Note that
separability or inseparability is a property not influenced by LLUBOs, so all
the Gaussian states with the same standard forms have the same separability
or inseparability property. With the above preparations, now we present the
following main theorem:
Theorem 2 (necessary and sufficient inseparability criterion for Gaus-
sian states): A Gaussian state Pa is separable if and only if when expressed
in its standard form II, the inequality (13.4) is satisfied by the following two
EPR-type operators
 Inseparability criterion for continuous variable systems 151

CI
u = aOxI- --X2,
~

Icd ao
1~
(13.17)

C2
v = aOPI- --P2,
~ 1~
(13.18)
IC21 ao
2 =
where a o Jm1-1 = Jm -1.
nl-l n2-1
2

Proof. The 'only if' part follows directly from theorem 1. We only need
to prove the 'if' part. From lemma 2, we can first transform the Gaussian
state through LLUBOs to the standard form n. The state after transformation
is denoted by pM.Then, substituting the expression (13.17,13.18) of and u
vinto the inequality (13.4), and calculating ((~u)2) + ((~v)2) using the
correlation matrix MIl, we get the following inequality

2nl + n2 ml + m2 1 1 (2 1)
ao 4 + 4a5 - 2' (I cII + Ic21) 2: 2' ao + a5 ' (13.19)

which, combined with Eq. (13.12,13.13), yields

ICII < v(nl - 1) (ml - 1). (13.20)

IC21 < v(n2 - 1) (m2 - 1) (13.21)

The inequality (13.20,13.21) ensures that the matrix M:I -

I is positive semi-
definite. So there exists a Fourier transformation to the following normal
characteristic function of the state pM

This means that pM can be expressed as

(13.23)

where P (a, (3) is the Fourier transformation of X(~) (AI, A2) and thus is a
positive Gaussian function. Eq. (13.23) shows pg
is separable. Since the
original Gaussian state PG differs from pM
by only some LLUBOs, it must
also be separable. This completes the proof of theorem 2.
152 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Now we have a necessary and sufficient inseparability criterion for all the
Gaussian states. We conclude the paper by applying this criterion to a simple
example. Consider a two-mode squeezed vacuum state e 1 2
r(atat -a1( 2 )
Ivae)
with the squeezing parameter r. This state has been used in recent experiment
for continuous variable quantum teleportation [13]. Suppose that the two
optical modes are subject to independent thermal noise during transmission
with the same damping coefficient denoted by 'fJ and the same mean thermal
photon number denoted by n. It is easy to show that after time t, the standard
correlation matrix for this Gaussian state has the form of Eq. (13.9) with n =
m = cosh (2r) e- 21/t + (2n + 1) (1 - e- 21/t ) and e = -e' = sinh (2r) e- 21/t
[15]. So the inseparability criterion means that if the transmission time t
satisfies

t< 2~ In (1 + 1 -2~-2r) , (13.24)

the state is entangled; otherwise it becomes separable. Interestingly. Eq.

(13.24) shows that if there is only vacuum fluctuation noise, i.e., n = 0 (this
seems to be a good approximation for optical frequency), the initial squeezed
state is always entangled. This result does not remain true if thermal noise
is present. In the limit n » 1, the state is not entangled any more when the
transmission time t ~ 14~~2r.

References
[1] A. Peres, Phys. Rev. Lett. 77, 1413 (1996).
[2] M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Lett. A 223, 1
(1996).
[3] L. M. Duan, G. Giedke, J. I. Cirac, and P. Zoller, Phys. Rev. Lett. 84, 2722
(2000).
[4] A. Einstein, B. Podolsky, and R. Rosen, Phys. Rev. 47, 777 (1935).
[5] C. W. Gardiner and P. Zoller, Quantum Noise (2nd. Ed.), Springer-Verlag
(1999).
[6] L. Vaidman, Phys. Rev. A 49, 1473 (1994).
[7] S. L. Braunstein, H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998).
[8] S. L. Braunstein, Nature 394, 47 (1998).
[9] S. L. Braunstein and S. Lloyd, Phys. Rev. Lett. 82,1789 (1999).
[10] G. J. Milburn and S. L. Braunstein, quant-phl9812018.
[11] P. Loack, A. L. Braunstein, and H. J. Kimble, quant-phl9902030.
[12] A. S. Parkins, and H. J. Kimble, quant-phl9904062.
 Inseparability criterion/or continuous variable systems 153

[13] A. Furusawa et ai., Science 282, 706 (1998).

[14] S. L. Braunstein, quant-phl9904002.
[15] L. M. Duan and G. C. Guo, Quantum Semic1ass. Opt. 9, 953 (1997).
Chapter 14

SEPARABILITY CRITERION
FOR GAUSSIAN STATES

R. Simon
The Institute of Mathematical Sciences, Tharamani, Chennai 600 113, India

Abstract The PPT (positivity under partial transpose) criterion is studied in the context of
separability of continuous variable bipartite states. The partial transpose oper-
ation admits, in the Wigner representation of quantum mechanics, a geometric
interpretation as momentum reversl or mirror reflection in phase space. This
recognition leads to uncertainty principles, stronger than the traditional ones, to
be obeyed by all PPT (separable as well as bound entangled) states. In the special
case of bipartite two-mode systems, the PPT crrterion turns out to be necessary
and sufficient condition for separability, for all Gaussian states: a 1 + 1 syatem
has no bound entangled Gaussian state. The symplectic group oflinear canonical
transformations and the representation of these transformations through (meta-
plectic) unitary Hilbert space operators play an important role in our ananysis.

A major part of the effort in quantum information science has traditionally

been in respect of systems with finite number of Hilbert space dimensions,
more specifically in respect of qubits corresponding to Hilbert space dimension
two. But recently there has been much interest in the canonical case of con-
tinuous variable systems [1,2,3,4,5,6]. We may mention, in particular, the
experimental realization of quantum teleportation of coherent states [7]. This
achievement has acted as a significant impetus for this interest.
Entanglement or inseparability plays a central role as the primary resource in
almost all branches of the emerging field of quantum information and quantum
computation [8]. Issues related to enranglement are considerably richer in the
context of mixed states compared to the situation with pure states. For instance,
while it is nearly trivial to test if a given bipartite state is separable, we do not
yet have an effective algorithm to test if a given mixed state of a bipartite system
is separable or entangled.
A particularly elegant criterion for checking if a (bipartite mixed) state is
separable or not was formulated by Peres [9]. This criterion based on partial
155
S.L Braunstein andA.K. Pati (eds.), Quantum Information with Continuous Variables, 155-172.
© 2003 Kluwer Academic Publishers.
156 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

transposition turns out to be necessary and sufficient for separability for all
states of 2 x 2 and 2 x 3 dimensional bipartite systems, but ceases to be so in
higher dimensions as shown by Horodecki [10]. Entanglements which are not
witnessed by the partial transpose operation cannot be distilled, and for this
reason such entanglements have come to be known as bound entanglements.
With increasing Hilbert space dimension, tests for separability will be ex-
pected to become more and more difficult to implement in practice, and less
and less definitive in their outcome. On the other hand, in the infinite dimen-
sional case corresponding to continuous variable systems Gaussian states are
of particular interest from the point of view of experiments, and for this reason
a test for separability which is decisive for all Gaussian states could be of
considerable value in the continuous case, even if it fails to be decisive for non-
Gaussian states. Hence is the importance of the result establised in Ref. [11]
that the partial transpose condition proves both necessary and sufficient (NS)
condition for separability, for all Guassian states of a bipartite system of two
harmonic oscillators. We should hasten to add that the partial transpose crite-
rion of separability ceases to be NS, even for Gaussian states, when both sides
of the system (Alice and Bob) have two or more oscillators each, as shown by
Werner and Wolf [12].
An interesting approach to separability of Gaussian states, based on the total
variance of a pair of Einstein-Podolsky-Rosen type operators, was indepen-
dently formulated by Duan et al. [13]. This approach too leads to a criterion
which proves NS in the 1 + 1 case. The same authors have proposed also
an entanglement purification protocol to generate maximally entangled states
from two-mode squeezed states or from mixed Gaussian entangled states [14].
The origin of the fact that the issue of separability is easily tractable in
the case of Gaussin states can be traced as follows. First, a Guassian state is
fully determined by its first and second moments. The Weyl group of unitary
operators which effect translations in phase space are manifestly separable,
and hence their action does not affect the separability or otherwise of a state.
We may thus assume, wihout loss of generality, that the first moments of the
state of our interest all vanish. In other words, separability of a Gaussian state
is determined entirely by the variance (or covariance) matrix of the state. The
problem thus gets reduced from study of the infnite dimensional density matrix
to analysis of the finite dimensional variance matrix.
Secondly, under the unitary action of the symplectic group of real linear
canonical transformation the moments of a quantum state in general, and the
variance matrix in particular, undergo simple geometric changes. One may
thus use local symplectic transformtions (local transformations have no effect
on separability) to convert the variance matrix into a canonical form in which
the separability issue becomes particularly easy to settle.
 Separability criterionjor Gaussian states 157

Chracterization of a Guassian state in term of its variance matrix (and first

moments) and the action of symplectic transformations on the state take a
particularly revealing and convenient form in the Wigner representation of
quantum mechanics. Further, the partial transpose operation acquires a sug-
gestive geometric interpretation in this representation. For these reasons the
Wigner representation proves convenient for analysis of the issue of separability
for Gaussian states.

1. LINEAR CANONICAL TRANSFORMATIONS

Consider a system of n oscillators described by phase space variables
ql, PI; q2, P2; ... qn, Pn or by the corresponding canonical operators iiI, ih ;
ih,'P2; ... iin,Pn. It is convenient to arrange these phase space variables and
hermitian operators into column vectors

ql iiI
PI PI
q2 ii2
~= P2 t= P2 (14.1)

qn qn
Pn Pn
The complete set of canonical commutation relations

[iij, iik] 0, [Pj,17k] = 0;

[iij,17k] 8jk, j,k = 1,2 1 ••• n (14.2)
can be written in the useful compact form [15]
[ta" e,B] = i na,B, O!, f3 = 1,2, ... 2n; (14.3)

where
J 0 0 o
o J 0 o
o 0 J o (14.4)

000 J
Linear homogeneous canonical transformations in the 2n-dimensional phase
space of our n-oscillator system act as Hilbert space unitary operators in the
quantum description. With a set of canonical coordinates chosen for the phase
space, such a transformation is identified by its matrix S:
S: ~ -+ e = S~. (14.5)
158 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Clearly, this transformation is canonical if and only if S respects

snsT = n, (14.6)

which is recognized as the defining condition for Sp(2n, R), the group of
symplectic matrices. Thus, linear canonical transformations are in one-to-one
correspondence with elements of the symplectic group Sp(2n, R). It is useful
to note that det S = 1, \:IS E Sp(2n, R), and that S E Sp(2n, R) implies
sr E Sp(2n, R).
Let U (S) be a Hilbert space unitary operator corresponding to the canonical
transformation S. This means we have the following evolution for the operators
t
2n
U(S): ta -+ t~ =U(S)t taU(S) = "L Sa/3t- (14.7)
/3=1

We could have written this in the abreviated form t' = U(S)t tU(S) = S t-
In order that the unitaty operators U(S) give a representation of Sp(2n, R),
we should be able to choose the former in such a way that U(S)U(S') =
U(SS'), \:IS, s' E Sp(2n.R). Clearly, U(S)t tU(S) = S t allows, in the cor-
respondence S -+ U(S), freedom of a S-dependent phase factor exp( i¢(S) )
mUltiplying U(S). Can we make a one-to-one choice S -+ ¢(S) such that
U(S)U(S') = U(SS'), \:IS, s' E Sp(2n.R)? The answer turns out to be
in the negative. The maximum simplification one can obtain in this regard is
a one-to-two correspondence S -+ ±U(S). In other words this correspon-
dence yields a two-valued unitary 'representation' of the symplectic group.
Conversely, these unitary operators close to form the defining representation
of a new group called the metaplectic group Mp(2n, R). This group is a
double cover of the symplectic group. A detailed discussion of the symplectic-
metaplectic connection, with further references, can be found in Ref. [16]. The
situation is somewhat analogous to the well known connection between SO(3)
and SU(2). The major diference is that SO(3) is doubly connected and so its
double cover SU(2) is simply connected (and hence is the universal cover of
SO(3) ); but Sp(2n, R) is infinitely connected, and so also is its double cover.
The point being made is that what is natural for the quantum description of a
system of oscillators at the state vector level is the metaplectic group rather than
the symplectic group. The symlectic group is recoverd in the description at
the density operator level (or, equivalently, in the Wigner description), though.
This subtle aspect can be appreciated by considering just one oscillator. After
evolution through one period, which corresponds to the identity element of
Sp(2, R), the position and momentum - and indeed the entire phase space-
return to their original configuration. But the unitary evolution of a statevector
17J;) of the same oscillator through one period returns - 17J;) rather than 17J;)
 Separability criterion for Gaussian states 159

itself. The state vector is recovered in full only after two periods. Originating
in this subtlety is not only the zero-point energy, but also the Gouy phase picked
up by a light beam as it crosses a focus [17], the 7r/4 phase picked up by the
WKB wavefunction at the turning points, and the Maslov index.

2. THE WIGNER DESCRIPTION

The Wigner description of quantum mechanics is at the density operator
level, rather than at the state vector level, and it treats pure states and mixed
states on the same footing. In this description density operators of a qantum
system are put into one-to-one correspondence with real-valued functions over
the (classical) phase space through the rule [18, 15]

W(q,p) = 7r- nJd nq, (q - q'l p Iq + q') exp(2i q' . p). (14.8)

where q = (ql, q2, ... qn) andp = (Pl,P2, ... Pn). We will often write W(e)
in place of W (q, p). It is clear that this map is invertible, and thus the function
W (q, p) captures the dendity matrix in its entirety. In most aspects of structure
and dynamics, the Wigner distribution function W(q,p) behaves exactly like
the phase space density in classical statistical mechanics, but W (q, p) is not
pointwise nonnegative for most quantum states. Indeed, the Wigner function
of a pure state is pointwise nonnegative if and only if the state is Gaussian [19]!
The defining properties of Wigner distributions are transcriptions, through
the map (14.8), of those of density operators. Thus, hermiticity of density
operator p is equivalent to the corresponding phase space distribution being
real, while the condition trp = 1 transcribes into Jd2n e W(e) = 1. The
nonnegativity of p, which can be stated as the requirement that tr(pp') ~ 0 for
every den sty operator p', gets translated into the condition

tr(pp') = (27rt J d2n e W(e)W'(e) ~ 0, (14.9)

for every Wigner distribution W'(e). In particular, a phase space distribution

W(e) which is known to be a Wigner distribution corresponds to a pure state
if and only if

(14.10)

In the nonnegativity condition tr(pp') ~ 0, it is sufficient to restrict p' to pure

states; so also in the condition (14.9) it proves sufficient to restrict W'(e) to
pure state Wigner functions.
The Wigner description offers several advantages in respect of our primary
concern, namely the issue of separability of Gaussian states. Foremost among
160 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

these is the action of canonical transformations. We have noted that the sym-
plectic group Sp(2n, R) acts unitarily and irreducibly on the n-mode Hilbert
space [20]. The (infinite dimensional) unitary operator U(S) corresponding
to S E Sp(2n, R) transforms the state vector i"p) to i"p') = U(S)i"p), and
hence the density operator p to p' = U (S) PU (S) t. This transformation takes
a strikingly simple form in the Wigner description, and this is one reason for
the effectiveness of the Wigner picture in handling canonical transformations
[20]:
S: p ---7 U(S) pU(S)t ¢:::::? W(e) ---7 W'(e) = W(S-le). (14.11)
That is, W'(Se) = W(e) for every canonical transformation S E Sp(2n, R),
and the Wigner function transforms as a Sp(2n, R) scalar field.
Secondly, the transpose map T and the partial transpose map PT take tran-
parent geometric form in the Wigner description. Indeed, it follows from the
definition of Wigner distribution that transpose operation on the density opera-
tor, which is equivalent to complex conjugation of the elements of the density
matrix in the position representation, transcribes faithfully into momentum
reversal operation in the Wigner description:
T: W(q,p) -+ W'(q,p) = W(q, - p) = W(Ae),
A = diag(I,-I; 1. -1; ... ; 1,-1)
= a3 EEl a3 EEl ••• EEl a3· (14.12)

This mirror reflection, which inverts the p coordinates leaving q coordinates

unchanged, is consistent with our expectation that transpose operation on the
density operator is the same thing as time reversal.
The transposition map takes density operators to density operators and,
equivalently, Wigner distributions to Wigner distributions. This statement
is, in a sense, worded stronger than what it really is: transposition cannot be
implemented as a physical process. Transposition with respect to one subsystem
of a composite system (partial transpose) can take Wigner distributions (density
operators) into phase space distributions (operators) which are not Wigner
distributions (density operators). Thus, partial transpose acts as a potential
entanglement witness, for it does not necessarily preserve the 'Wigner quality'
of the phase space distribution of an entangled state.
Finally, moments of the canonical operators are related to the Wigner distri-
bution in exactly the same manner they are related to the phase space density in
classial statistical mechanics. The first moments (€a) == tr(€aP) are given by

(14.13)

Let us define difference or fluctuation operators b.~ through b.~ = ~ - (~).

Clearly, the 2n components of b.~ obey the same commutation relations as
 Separability criterion for Gaussian states 161

those of { Similarly, we may define Llea = ea -

(ea)' The uncertainties
are defined as the expectations of the hermitian operators {Llta, Llt(3 }
(Llta Ll t(3 + Llt(3Llta)/2:
({Llta, Llt(3}) = tr ({Llta, Llt(3},8) = Jd 2ne Llea Lle(3 W(O. (14.14)

It proves useful to arrange the uncertainties ({ Llta, Llt(3}) into a real 2n x 2n

matrix V defined through Va(3 == ({ Llta, Llt(3} ). This matrix is variously
known as the variance, covariance, or noise matrix.

3. GAUSSIAN STATES AND THEIR NOISE MATRIX

Given an infinite-dimensional hermitian operator of unit trace, it is not always
easy to test its nonnegativity and establish that it is a valid density operator.
Testing the Wigner quality of a real phase space distribution is equally difficult.
This issue can, however, be elegantly resolved for Gaussian states of a system
of n oscillators, for an arbitrary value of n.
The dstinguishing feature of Guassian states is that they are fully charac-
terised by their first and second moments. Indeed, the Wigner distribution of a
Gaussian state has the form

W(e) = ((27ft v'det V) -1 exp ( -~(e - (0) TV- 1 (e - (0)) .(14.15)

The first moments given by the 2n components of (e) can assume arbitrary
real values. They can be changed at will by the unitary action of the Weyl
displacement operators. Further, the value of the first moments affect in no
way the Wigner quality of the phase space distribution. We can therefore
assume, without loss of gnerality, that our Gaussian state has vanishing first
moments; it is then completely specified by the positive definite variance matrix
V!
What are the additional conditions that the noise matrix V > 0 should satisfy
in order that the Gaussian distribution in (14.15) is a Wigner distribution? To
obtain these conditions, which arise basically from the uncertainty principle,
note first of all that the symplectic transformation law (14.11) implies that the
first moments change as
(14.16)
and that the variance matrix undergoes the congruence
s: V --+ V' = S V ST. (14.17)
In view of (14.6), this congruence is equivalent to the conjugation
S: vn --+ v'n = svns- 1 . (14.18)
162 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Thus, f31. = tr(Vn)21., i = 1,2, ... , n are symplectic invariants. And these
are the only inepenent invariants of V, (Vn)k being traceless for odd values
of k.
It is elear that not every rotation in the 2n-dimensional phase space is a
canonical transformation. Indeed, the set of all phase space rotations forming
a subgroup of Sp(2n, R) is isimorphic to the n 2 -dmensional unitary group
U(n). And n 2 is smaller than 2n2 - n, the dimension of the full roration
group SO(2n), for n > 1. We do not, therefore, expect in general to be
able to diagonalize a real symmetric matrix V using symplectic congruence
V -+ V' = sv ffI', S E Sp(2n, R). Williamson theorem [21, 15, 22]
guaranttes that if V possesses the additional quality of being positive definite,
it can be diagonalized through symplectic congruence. That is, V > 0 implies
that there exists an Sv E Sp(2n, R) such that

Sv: V -+ Vwc = Sv V SJ = diag( ~l, ~l; ~2, ~2; ... ; ~n, ~n X.l4.19)
Comparing with the Gaussian Wigner distribution in (14.15), it is clear that
the Williamson canonical form Vwc corresponds to product of single-mode
thermal state Wigner distributions, the temperature parameter of the j -th mode
being determined by ~j. The invariant traces are asily computed to be
n
f3<
n -
.-
- 2 "'(~
~ J.)21. , i -
-I" 2 ... ,n. (14.20)
j=l

We may as well take these ~j 's as a complete set of algebraically independent

symplectic invariants associated with the variance matrix V. These preparatory
remarks enable us to give a complete characterization of variance matrices, and
hence of Gaussian states.

Proposition 1: The following three conditions are equivalent to one another,

and form necessary and sufficient condition for a real symmetric positive defi-
nite 2n x 2n matrix V to be a physical variance matrix:
1. V + ~n ~ o.
2. V ~ inTv-ln = inv-1nT > O.
3. V ~ ~S ST, for some S E Sp(2n, R).
Proof: Note that the above conditions, each of which subsumes V > 0, are
Sp(2n, R) -invariant, as they should be, and this is the key to their proof: the
uncertainty principle must be satisfied not only by the components of t, but
t, t,
also by those of = S for every S E Sp(2n, R).
For a single mode state with diagonal variance matrix V = diag (~, ~),
the acceptability condition or uncertainty principle reads ~ ~ 1/2. Since the
 Separability criterion for Gaussian states 163

Williamson normal form corresponds to product of single mode states, the

acceptability condition for the normal form follows from that for the single
mode case. That is Vwc = diag(~I' ~lj ~2, ~2j ... j ~n, ~n) is acceptable
as a variance matrix if and only if ~j 2: 1/2, for j = 1,2,··· n. We can now
employ the symplectic metric n to rewrite this condition as Vwc + ~n 2: o.
Covariance of this condition under symplectic congruence leads to V + ~n 2: 0
in the general case.
The variable canonically conjugate to e~ is given by (n e')a. Product ofthe
variances of this pair of conjugate variables should be bounded below by 1/4,
and this may be written as var(e~) 2: :H
var( (n e')a )]-1. Stating this in a
manifestly covariant form we have V 2: inT V-I n = in V-I nT.
Finally, the acceptability condition on Vwc can be written as the requirement
Vwc 2: ~. The statement that V is acceptable if and only if V 2: ~S ST, for
some S E Sp(2n, R) is simply a covariant rendering of this requirement.
This completes proof of the proposition, and gives a complete characteri-
zation of Gaussian states: the Gaussian phase space distribution (14.15) is a
Wigner distribution if and only if V satisfies the above acceptability condition.
We may note in passing that the formulation of the necessary and sufficient
condition above is invariant under the transpose or momentum reversal map
which takes V into AV AT = AV A or, eqivalently, n into A-I V (AT)-1 =
- n. While the first two versions are manivestly invariant, invariance of the
third one follows from the fact that A SA E Sp(2n, R), 'liS E Sp(2n, R),
even though A itself is not an element of Sp(2n, R). We conclude this Section
with the following observation.

Proposition 2: The variance matrix of a state has the special form V = ~ SST,
with S E Sp(2n, R), if and only if the state under consideration is a Gaussian
pure state.

4. SEPARABILITY CRITERION FOR THE

VARIANCE MATRIX
We are now equiped to consider a bipartite system consisting of n modes,
with m modes in the possession of Alice and the remaining n - m modes in
Bob's possession. Let us introduce the following notations:

qA = (ql, q2,··· , qm), PA = (Pr,P2,··· ,Pm),

eA = (ql,Plj q2,P2j··· j qm,Pm)j
qB = (qm+l, qm+2,··· , qn), PB = (Pm+l,Pm+2,··· ,Pn),
eA = (qm+l,Pm+1j qm+2,Pm+2j··· j qn,Pn)j
e (14.21)
164 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Consistent with this notation we may write the symplectic metric n as a direct
sum of the symplectic metrics of the two subsystems:
o
Q= ( (14.22)
o
Let P be a bipartite density matrix, and W(e) = W(eA, eB) the correspond-
ing Wigner distribution. The partial transpose map with respect to the Bob
subsystem corresponds to
W(e) -+ W'(e) = W(A'e) = W(eA,ABeB),

A' = ( I:A :B ) = IdA B- ffiA (14.23)

Though W(Ae) is a Wigner distribution for every Wigner distribution W(e),

the disribution W(A'e) need not be Wigner. Stated differently, the momentum
reversal map is positive but not completely positive. Since it is not completely
positive, it does not correspond to any physically realizable process or channel.
It does, however, prove to be of value in exposing or witnessing entanglement.
By definition, a quantum state Pof the bipartite system is separable if and
only if Pcan be expressed in the form

P= LPkPAk0PBk, (14.24)
k

with nonnegative Pk'S, where PAk'S and PBk'S are density operators of the
subsystems of Alice and Bob respectively. Clearly, product states correspond
to product Wigner distributions and separable states separable Wigner distribu-
tions:
W(e) = LPk WAk(eA)WBk(eB). (14.25)
k

o
V= ( (14.26)
o VB
 Separability criterion for Gaussian states 165

where VA and VB are the variance matrices of the subsystems. Let VA k EI1 VB k
be the variance matrix of the product state Wk(e) = WAk(eA)WB k(eB),
and let e(k) be its first moments. Then the variance matrix of the separable
state (14.25) is given by [recall that (e) has been assumed to vanish for W(e) ]

V = L Pk VAk EI1 VBk +L Pk e(k)e(k)T. (14.27)

k k

Since convex sum of variance matrices is a variance matrix, and since

L Pk VAk EI1 VBk = (2: Pk VAk) EI1 (2: Pk VBk), (14.28)

k k k
the first sum on the right hand side of (14.27) is a variance matrix. Since it has
the block-diagonal form VA EI1 VB where VA and VB are acceptable variance
matrices, it follows from the third form of the condition in Proposition 1 that
there exist symplectic matrices SA E sp(2m, R) and SB E Sp(2n - 2m, R)
such that

(L Pk VAk) EEl (L Pk VB k)
k k
1 T T
> "2 SA SA EElSBSB· (14.29)

Since the second sum on the right hand side of (14.27) is a manifestly nonneg-
ative matrix, we have estalished a necessary condition on the variance matrix
of a separable state.

Proposition 3: If V is the variance matrix of a separable state, then there are

matrices SA E Sp(2m, R) and SB E Sp(2n - 2m, R) such that.

1 T T
V ~ "2 SASA EI1 SBSB· (14.30)

The validity of this condition is quite general: the number of modes in the
possession of Alice and Bob can be arbitrary, and the state under reference
need not be Gaussian.
If the state under consideration is Gaussian, the above result can be con-
siderbly strengthened. To this end, we begin by noting the following.

Proposition 4: The variance matrix of a state has the special form V

~ sAsI EI1 SBS,£, with SA E Sp(2m,R) and SB E Sp(2n - 2m,R), if and
only if the state under consideration is a Gaussian state which is both a pure
state and a product state.
166 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Let us denote the block-diagonal matrix! SAsI EB SBS,£ by VGPP to

indicate the fact that it represents a Gaussian, pure, and product state. If the
condition in Proposition 3 is met, we have V = VGPP + Vclass, for some Vclass 2:
0, and the subscript of Vclass anticipates a classical probability distribution we
are going to associate with Vclass' Let r be the rank (dimension of the range)
of Vclass, and let the r x r matrix "V;;~:~ be the restriction of Vclass to its range.
Then V = VGPP + Vclass implies

d et v(res)]
class J dTt." '

X exp [ -! eT("V;;~:~)-le] WVGPp(e - e'), (14.31)

where dTe indicates integraton over the range of Vclass, and WVGPp(e - e) is
the Wigner distribution of the GPP state. Thus we have

Proposition 5: A Gaussian state with variance matrix V is separable if and

only if
1 T T
V 2: 2" SASA EB SBSB (14.32)

for some SA E Sp(2m, R), SB E Sp(2n - 2m, R).

5. SEPARABILITY AND UNCERTAINTY PRINCIPLE

Positivity under partial transpose (PPT) has important implications on the
uncertainty principle, and it is useful to explore this aspect in some detail. The
uncertainty principle V + ~ n 2: 0 is a direct consequence of the canonical
commutation relations (14.3) and the nonnegativity of the density matrix p.
It is equivalent to the statement that for every pair of 2n-dimensional phase
space vectors d = (dA, dB), d' = (d~, d~), the hermitian operators X(d) =
d
T
e= 2: a daea and X(d) = d e= 2: a daea obey
A A A, ,T A , A

((~X(d))2) + ((~X(d'))2) > I [X(d), X(d')ll

Id,Tndl
IdInAd~ + dInBd~l· (14.33)

Partial transpose acts on the Wigner distribution as momentum reversal on

Bob's side and, as a result, the variance matrix undergoes the change V -+
if = A'V A'. Since W(A'e) has to be a Wigner distribution if the state under
consideration is separable, we have
-
V
z
+ 2" n 2: 0, if = A'V A'
,
n= (nA
0
0)
nB '
(14.34)
 Separability criterion for Gaussian states 167

as a necessary condition for separability. Equivalently,

V + in>
2 -,
0 n= A'nA' = (nA0 0)
-nB '
(14.35)

so that separability of p implies the additional restriction

((~X(d))2) + ((~X(d'))2) > Id'Tn dl

IdInAd~ - d];nBd~l· (14.36)
Combining this partial transposed uncertainty principle with the original (14.33)
we have
(14.37)
\;j d, d' E n2n.
This restriction, to be obeyed by all PPT states, is generically stronger than
the usual uncertainty principle (14.33). For instance, let X(d) commute with
X(d'), i.e., let ~nd' = O. What our result shows is that if the state is PPT,
then X (d) and X (d ') cannot both have arbitrarily small uncertainties unless
dTn d' = 0 as well, i.e., unless dInAdA and d];nBdB vanish individually.
An example in the 1 + 1 case may help. The two operators X = Xl + 'lh +
X2 + lh Y = Xl - PI - X2 + P2 commute, but the sum of their uncertainties
in any PPT state has to be ~ 4.

6. TWO-MODE SYSTEMS
In this case wherein Alice and Bob hold a single mode each, the set of all
homogeneous real linear canonical transformations constitute the ten-parameter
real symplectic group Sp( 4, R), and nand becomen
n= ( JO)
0J
- (J0 -J0) .
' n= (14.38)

The variance matrix can be written in the block form

V=(;tT ~). (14.39)

Separability of a state is not affected by local transformations which act inde-

pendently on the Hilbert spaces of Alice's and Bob's subsystems. Of particular
interest in the present case is the unitary metaplectic action of the six-parameter
Sp(2, R) ® Sp(2, R) subgroup of Sp( 4, R) corresponding to independent local
linear canonical transformations on the subsystems of Alice and Bob:
Slocal E Sp(2, R) ® Sp(2, R) c Sp( 4, R) :

Slocal (~l ~2)' SlJSf = J = S2JS'f (14.40)

168 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

It is possible and desirable to cast the stronger uncertainy principle (14.37) for
separable states in an Sp(2, R) ® Sp(2, R) invariant form.
The congruence V -+ Slocal V ~cal by the local group changes the blocks
of V in the following manner:

A -+ SlASi, B -+ S2BS'f, C -+ SlCS'f,

Thus, the Sp(2, R)®Sp(2, R) invariants associated with V are II = det A, 12 =
detB, 19 = detC, and 14 = tr(AJCJBJCT J) [detV is an obvious invari-
ant, but it is a function of the h's, namely det V = hh + 19 2 - 14].
It turns out that the (matrix) uncertainity inequality V + ~n ;::: 0 is equivalent,
in the two-mode case, to the Sp(2, R) ® Sp(2, R) invariant (scalar) inequality

detA detB + (~- detc) 2 - tr(AJCJBJCT J)

1
;::: 4(detA + detB). (14.41)

To prove this result, first note that the two inequalities are equivalent for variance
matrices of the special form

Vo = (14.42)

But any variance matrix can be brought to this special form by effecting
a suitable local canonical transformation corresponding to some element of
Sp(2, R) x Sp(2, R). In veiw of the manifest Sp(2, R) ® Sp(2, R) invariant
structure of (14.41), it follows that the two inequalities are indeed equivalent
for all two-mode variance matrices.
Under partial transpose or momentum reversal on Bob's side represented by
the phase space mirror reflectin A' = Id EEl C73, we have V -+ V = A'V A'.
That is, C -+ CC73 and B -+ C73BC73, while A remains unchanged [C73 is the
diagonal Pauli matrix: C73 = diag (1, -1)]. As a consequence, 13 = det C flips
signature while h, hand 14 remain unchanged. Thus, condition (14.34) for V
takes a form identical to (14.41) with only the signature in front of det C in the
second term on the left hand side reversed. Thus the PPT requirement that the
variance matrix of a separable state has to obey A'V A' + ~n ;::: 0 inaddition to
the fundamental uncertainty principle V + ~n ;::: 0, takes the form

detA detB + (~-ldetCI) 2 - tr(AJCJBJCT J)

1
~ 4(detA + detB). (14.43)
 Separability criterion for Gaussian states 169

This is the final form of the implication of PPT on the variance matrix of a
two-mode bipartite state. This condition is invariant not only under phase space
mirror reflection (partial transpose) A', but also under Sp(2, R) ® Sp(2, R), as
it should be! It constitutes a complete description of the implication PPT has
for the second moments of any state.
For the standard form Va, our condition (14.43) reads

(14.44)

But the point is that the separability (PPT) check (14.43) can be applied directly
on V, with no need to transform it to the special form Vo.
To summarise, conditions (14.33), (14.41), and V + ~n 2: 0 are equivalent
statements of the fundamental uncertainty principle, and hence will be satisfied
by every physical state. The mutually equivalent statements (14.37), (14.43),
and (14.34) constitute the PPT criterion at the level ofthe second moments, and
should necessarily be satisfied by every separable state, Gaussian or otherwise.
As the reader may anticipate, we can make stronger statements for Gaussan
states: separability and PPT become equivalent in the two-mode Gaussian case.
To this end, note that states with det C 2: 0 definitely satisfy the PPT condi-
tion (14.43), which in this case is subsumed by the physical condition (14.41).
We will begin by establishing that such Gaussian states are indeed separable.

Lemma: Gaussian states with det C 2: 0 are separable.

Proof: First consider the case det C > O. We can arrange a 2: b, Cl 2:

C2 > 0 in the special form Vo in (14.42). Let us do a local canonical trans-
formation Slocal = diag (x, x-I, x-I, x), corresponding to reciprocal local
scalings (squeezings) at the Alice and Bob ends, and follow it by Sl~cal =
diag (y, y-I, y, y-l), corresponding to common local scalings at these ends.
We have, as a result,

(14.45)

Choose x such that cr/(x 2a - x- 2b) = C2/(x- 2a - x 2b). That is, x =

[(era + C2b)/(C2a + clb)P/4. With this choice, Vo' acquires such a structure
that it can be diagonalized by rotation through equal amounts in the ql, q2 and
170 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

PI, P2 planes:

Vto' -+ Vt"
0 = d'lag (~+, " );
~+, ~-, ~_ .
(1446)

~± ~y2 { x 2a + x- 2b ± [(x 2a - x- 2b)2 + 4clp/2} , (14.47)

~~ ~y-2 { x- 2a + x 2b ± [(x- 2a - x 2b)2 + 4c?P/2} . (14.48)

Such an equal rotation is a canonical transformation; and therefore VJ' is a valid

variance matrix. For our diagonal VO", the uncertainty principle Vo" + ~n ~
o simply reads that the product ~_~_' ~ 1/4 (and subsumes ~+~+' ~
1/4). It follows that we can choose y such that ~_, ~~ ~ 1/2 (for instance,
choose y so as to render ~_ = ~~). Then Vo" ~ 1/2Id4, where Id4 is
the four-dimensional unit matrix. Since Vo' and Vo" are rotationally related,
this implies Vo' ~ 1/2Id4. Hence we conclude, by Proposition 4, that the
Gaussian state corresponding to VO' is separable. This in turn implies that the
original V corresponds to a separable state, since V and VO' are related by local
transformation. This completes proof for the case det C > O.
Now suppose det C = 0, so that in Va we have CI ~ 0 = C2. Carry out
a local scaling corresponding to Slocal = diag (J2li", 1/J2li", V2b, 1/V2b),
taking Va -+ Vo'; the diagonal entries of VO' are (2a 2 , 1/2, 2b 2 , 1/2), and
the two nonzero off diagonal entries equal 2abcl. With this form for VO',
the uncertainty principle Vo' + ~n ~ 0 implies Vo' ~ 1/2Id4, establishing
separability of the Gaussian state. This completes proof of our Lemma.
The main separability theorem for Gaussian states is an immediate conse-
quence of the Lemma, as we shall demonstrate.

Theorem: Positivity under partial transpose is necessary and sufficient condi-

tion for separability, for all 1 + 1 bipartite Gaussian states.

Proof: We consider in turn the two distinct cases det C < 0 and det C ~ O.
Suppose det C < O. Then there are two possibilities. If (14.33) [or (14.34)]
is violated, then the Gaussian state is definitely entangled since (14.33) is a
necessary condition for separability. If (14.33) is respected, then the mirror
reflected state is a physical Gaussian state with det C > 0 (recall that mirror
reflection flips the signature of det C), and is separable by the above lemma.
This implies separability of the original state, since a mirror reflected separable
state is separable. Finally, suppose det C ~ O. Condition (14.33) is definitely
satisfied since it is subsumed by the uncertainty principle V + ~n ~ 0 in the
det C ~ 0 case. By our lemma, the state is separable. This completes proof of
the theorem.
 Separability criterion for Gaussian states 171

7. BOUND ENTANGLED GAUSSIAN STATES

The considerations of Ref. [11] detailed above were restricted to the bipartite
situation wherein the subsysems of Alice and Bob consist of a single mode each,
the principal result being that there is no bound entangled Gaussian state in the
1 + 1 case. A rather remarkable work of Werner and Wolf [12] goes beyond
the 1 + 1 case to establish two important resuls. Introducing the notion of
minimal PPT, they show that the principal result of Ref. [11] presented as the
above theorem implies that there exists no bound entangled Gaussian state in
he 1 + N case either.
More importanly, they present example of a bound entanled Gaussian state
in the 2 + 2 case. The noise matrix of their example has a simple structure:

2 0 0 0 1 0 0 0
0 1 0 0 0 0 0 -1
0 0 2 0 0 0 -1 0
0 0 0 1 0 -1 0 0
V=~ (14.49)
2 1 0 0 0 2 0 0 0
0 0 0 -1 0 4 0 0
0 0 -1 0 0 0 2 0
0 -1 0 0 0 0 0 4

Their proof that this Gaussian state is PPT, and that it is not separable, is
distinguished by effective use of symmetry arguments.
To conclude, we note that a recent work of Giedke et. al. [23] formulates a
necessary and sufficient condition for separability of Gaussian states of bipartite
systems of arbitrary number of modes. These authors show that all bipartite
Gaussian states with nonpositive partial transpose are distillable.

References
[1] L. Vaidman, Phys. Rev. A 49, 1473 (1994); L. Vaidman and N. Yoran,
Phys. Rev. A 59, 116 (1999).
[2] A. S. Parkins and H. J. Kimble, quant-phl9904062; quant-phl9907049;
quant-phl9909021.
[3] S. L. Braunstein, Nature 394,47 (1998); quant-phl9904002; S. Lloyd and
S. L. Braunstein, Phys. Rev. Lett. 82, 1784 (1999) ..
[4] G. J. Milburn and S. L. Braunstein, quant-phl9812018.
[5] P. van Loock, S. L. Braunstein, and H. J. Kimble, quant-phl9902030; P.
van Loock and S. L. Braunstein, quant-phl9906021; quant-phl9906075.
[6] S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 88, 86991998).
172 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[7] A. Furusawa et al., Science 282, 706 (1998).

[8] C. H. Bennett, Phys. Today 48, 24 (1995);
D. P. DiVincenzo, Science 270, 255 (1995).
[9] A. Peres, Phys. Rev. Lett. 77, 1413 (1996).
[10] P. Horodecki, Phys. Lett. A 232,333 (1997).
[11] R Simon, Phys. Rev. Lett. 84, 2726 (2000).
[12] RE Werner and M.M. Wolf, Phys. Rev. Lett. 86, 3658 (2001).
[13] L. M. Duan, G. Giedke, J.I. Cirac, and P. Zoller, Phys. Rev. Lett. 84, 2722
(2000).
[14] L. M. Duan, G. Giedke, J.I. Cirac, andP. Zoller, Phys. Rev. Lett. 84, 4002
(2000).
[15] R Simon, E. C. G. Sudarshan, and N. Mukunda, Phys. Rev. A 36,3868
(1987).
[16] R Simon and N. Mukunda, in Symmetries in Science V, Ed. B. Gruber
(Plenum, NY, 1993), p. 659-689.
[17] R Simon and N. Mukunda, Phys. Rev. Lett. 70, 880 (1993).
[18] E. P. Wigner, Phys. Rev. 40, 749 (1932); R G. Littlejohn, Phys. Rep. 138,
193 (1986).
[19] R Hudson, Rep. Math. Phys. 6, 249 (1975).
[20] R Simon, E. C. G. Sudarshan, and N. Mukunda, Phys. Rev. A 37,3028
(1988).
[21] J. Williamson, Am. J. Math. 58, 141 (1936). See also v.1. Arnold, Math-
ematical Methods of Classical Mechanics (Springer-Verlag, NY, 1978),
Appendix 6.
[22] R Simon, N. Mukunda, and B. Dutta, Phys. Rev. A 49, 1567 (1994).
[23] G. Giedke, B. Kraus, M. Lewenstein, and J. I. Cirac, Phys. Rev. Lett. 84,
176904 (2001).
Chapter 15

DISTILLABILITY AND ENTANGLEMENT

PURIFICATION FOR GAUSSIAN STATES

Geza Giedke,I,2 Lu-Ming Duan,I,3 J. Ignacio Cirac 1 ,2 and Peter Zoller1

lInstitutefor Theoretical Physics
University of Innsbruck, Innsbruck, Austria
2 Max-Planck Institute for Quantum Optics
Garching, Germany
3 Institute for Quantum Information
Caltech, CA 9 II 25, USA
[email protected]

Abstract We review the concepts of distillability and entanglement purification. We prove

that all inseparable Gaussian two-mode continuous variable states are distillable.
We present a entanglement purification protocol for Gaussian states and discuss
a quantum optical implementation of this protocol.

1. ENTANGLEMENT PURIFICATION
Entanglement (cf. Ch. II.7) is one of the most striking and characteristic
features of quantum mechanics. In quantum communication it is also a valuable
resource: it is necessary for many applications such as quantum teleportation
[1] or quantum cryptography[2], and is typically used up in those processes.
Ideally, the communicating parties A(lice) and B(ob) share maximally entan-
gled states, which enable them to perfectly implement the desired protocols.
In reality, however, due to loss and decoherence in the channel connecting A
and B, it is only possible to generate mixed, partially entangled states between
distant locations, which are often not directly useful for the task at hand. A way
to overcome this problem is offered by entanglement purification or distillation
[3,4,5], which describes the process of obtaining at least one maximally en-
tangled state out of many partially entangled ones. In general, any sequence of
local quantum operations (possibly correlated by classical communication) that
allows A and B to transform a (sufficiently large) number of entangled states
173
S.L. Braunstein andA.K. Pati (eds.), Quantum 11!formation with Continuous Variables, 173-192.
© 2003 Kluwer Academic Publishers.
174 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

p into at least one pure maximally entangled state comprises an entanglement

purification protocol (EPP). EPPs establish the feasibility of quantum commu-
nication in a realistic setting; in particular, combining quantum teleportation
and entanglement purification in the quantum repeater protocol [6] allows for
efficient quantum communication through noisy channels over arbitrary dis-
tances. Thus EPPs playas important a role for quantum communication as do
quantum error correcting codes (QECC) for quantum computing. Indeed, there
is a close connection between QECCs and EPPs [4].
Unfortunately, it is currently not known, in general, which mixed states
p can be distilled in this manner. At the moment we only have conditions
that are either necessary or sufficient for distillability, but not both. First,
obviously the state p must be inseparable (entangled) for it to be distillable.
Moreover, there exists a stronger necessary condition, namely that p must have
non-positive partial transpose [7]. Surprisingly enough, there are entangled
states which cannot be transformed into maximally entangled states at all (i.e.
which are undistillable) since their density matrices remain positive under
partial transposition [8]; these states are also called bound entangled. Bound
entanglement has been found in bipartite systems of dimension N > 6 [9] and
also in continuous variable (CV) systems [10, 19]. There is evidence that this
condition is in general not sufficient, since there seem to exist states that have
non-positive partial transpose but that nevertheless are not distillable [11]. On
the other hand, a useful sufficient condition for N x N systems, the so-called
reduction criterion, has been established [12]. It states that if there exists some
vector I~) such that

(~I (trBP ® :0. - p) I~) < o. (15.1)

then the state p is distillable. Here, trB stands for the partial trace with respect
to the second subsystem. An important aspect of this criterion is that if one can
find a state I~) satisfying (15.1), then one can explicitly construct a protocol to
distill p [12].
The question of distillability of general CV states is a formidable task that
we do not attempt to tackle here. Rather we consider in the remainder of this
chapter a simple but important class of CV states, namely Gaussian states of two
modes. There are several reasons to focus on these states. First, most of the CV
states that can currently be prepared in the lab are Gaussian states [13]. Second,
important CV quantum communication protocols are based on Gaussian states
(teleportation [14, 15], cryptography [16]), and finally, Gaussian states are
mathematically well understood, e.g., there exists an inseparability criterion.
Thus the largest set of potentially distillable states is easily characterized. In
the next section we will show that indeed all inseparable two-mode Gaussian
states can be distilled.
 Distillability and entanglement purification for Gaussian states 175

2. GAUSSIAN STATES
We consider states that are defined on the Hilbert space L2(JR) (or N copies
»,
thereof: L2(JR)®N = rJ(JR N that is, e.g., states of one or more modes of
the electromagnetic field. (We will in the following often use quantum optical
terms, like "modes" or "beam splitters", because quantum optics currently
offers the most promising setting for the realization of CV systems; this does
not limit the presented results to quantum optical systems.)

General definitions For such systems, it is convenient to describe the state

p by its characteristic function [17, 18]
(15.2)

Here e= (qI, . .. ,qN ,PI,· .. ,PN) E JR2N is a real vector and

W(e) = e-i~r:=l(qkRk+PkRk+N), (15.3)
where Rk are the canonical operators of the kth system, respectively, Xk =
Rk, P k = Rk+N, satisfying commutation relations [Rk, Rl] = -iJkl:D. (Ii = 1,
where Jkl are the entries of the 2N x 2N matrix J defined by J(qI, .. . ,qN,
PI, ... ,PN)T = (-PI, ... , -PN,qI,··· ,qN)T. W(e) is called the Weyl op-
erator [and is identical to the quantum optical displacement operator D(a) for
a = (PI - iqI,··. ,PN - iqN)J.
For a general density operator p we define (following [19]) the mean or
displacement as the vector d
dk = tr(pRk)
and the covariance (or correlation) matrix M by
Mkl = 2tr [p(Rk - dk:D.)(Rl - dl:D.)] + iJkl . (15.4)
The characteristic function contains all the information about the state (see
e.g., [17, 18]), that is, one can find p starting from x.
The Gaussian states are exactly those for which X is a Gaussian function of
e [20]
(15.5)
i.e., a Gaussian state is fully characterized by the correlation matrix M and the
displacement vector d.
Not all2N x 2N matrices are allowed, since the characteristic function must
correspond to the density operator of a physical state, i.e., to a positive operator.
It has been shown [20, 19] that a given correlation matrix M corresponds to a
physical state if and only if (iff) M > 0 and
M - iJ 2: 0 (15.6)
176 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Quasifree transformations There is an important set of unitary transforma-

tions intimately related to Gaussian states: the so called quasifree or linear
Bogoliubov transformations. They map a state with characteristic function X
to one with Xdefined by X(e) = X(S~ + d) where S is a symplectic! 2N x 2N
matrix and d a real vector in 1R2N. Clearly, Gaussian states are mapped to
Gaussian states by quasifree transformations.
Quasifree transformations can be implemented with Hamiltonians that are at
most quadratic in the quadrature operators X k, Pk. For optical fields this means
that they can be performed with beam splitters, phase shifters and squeezers,
i.e., currently available technology.

Gaussian states of two modes Any Gaussian state of two modes can be
transformed into what we called the standard form, using local quasifree trans-
formations (LQT) only [21, 22]. For a state in standard form the corresponding
characteristic function has displacement d = 0 and the correlation matrix M
has the simple form

(15.7)

where the block matrices are diagonal; and M a , Mb proportional to the identity:

MA = 0 na0) ' M B = ( nb0 nb0) ' M AB = (kx. 0 kp0) .

( na (15.8)

The four real parameters (na, nb, kx, kp) fully characterize a Gaussian state
up to LQTs. They can be easily calculated from the LQT-invariants IMAI,
IMBI, IMABI and IMI via:
(IS.9a)

(IS.9b)

where IMI denotes the determinant of M; without loss of generality we can

choose kx ~ Ikpl·
Recall that not all values of these parameters are allowed, since the correla-
tion matrix must be positive and satisfy the condition (15.6). In terms of the
parameters (15.9) we can reexpress these conditions as

(nanb - k;)(nanb - k;) +1 > n~ + n~ + 2kxkp, (15. lOa)

nanb - k; > 1. (IS. lOb)
 Distillability and entanglement purification/or Gaussian states 177

Inseparability On the other hand, in [22] it was shown that a Gaussian state is
entangled iff it does not transform into a proper state under partial transposition.
Starting from this result it is easy to show that a Gaussian state is entangled iff
the corresponding parameters satisfy [28]

(15.11)

3. DISTILLABILITY OF GAUSSIAN STATES

In the following we present the proof given in [23] that a Gaussian state
p of two modes is distillable if and only if its parameters fulfill (15.11), that
is, iff it is entangled. We will proceed as follows: first, we recall the proof
of the reduction criterion in [12] and extend the result to infinite dimensions,
providing a sufficient condition for distillability of CV states. Secondly, we
show for symmetric Gaussian states, i.e. states for which na = nb = n, that the
inseparability criterion (15.11) is equivalent to the reduction criterion. Thus all
symmetric inseparable states are distillable. Finally, we show that all the states
which are not symmetric (na =I nb) can be brought into a symmetric form by
using local operations maintaining inseparability and thus distillability.
We start by reviewing the proof of the reduction criterion for distillability
and generalizing it to infinite dimensions.

3.1 THE REDUCTION CRITERION

For N -level systems Let a density matrix p and the pure state 11/-1) =
2:n m a nm In) ® 1m) fulfill the condition (15.1), where the vectors In) form an
orth~normal basis. The coefficients a nm define a matrix A = (a nm ) satisfying
AAt = trB(11/-I) (1/-11). Distillation of p is divided into three steps.
(i) The first is a filtering operation: The operator AAt ® :n. can be viewed
as an element of a positive-operator-valued measure (POVM), which defines
a generalized measurement [24]. Conditional on the measurement outcome
corresponding to AA t ® :n. we obtain the state

p = At ® :n.pA ® :n./tr(pAAt ® :n.), (15.12)

which still satisfies (15.1) but now with 11/-1) = I~~) := IN 2:f=llk, k), the
symmetric maximally entangled state of two N-Ievel systems. In this case,
(15.1) implies tr(p Iq,~) (q,~j) > liN.
A state satisfying this inequality can be distilled by a generalization of the
protocol of Ref. [3], which consists of two steps: depolarization and joint
measurements.
(ii) Applying an operation of the form U ® U* (U a randomly chosen unitary)
depolarizes p, i.e. transforms it into a mixture of the maximally entangled state
Iq,~) (which is invariant under transformations of the form U ® U*) and the
178 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

completely mixed state ~:n.; the overlap of p with IcI>~) remains unchanged.
(iii) Taking two entangled pairs in this depolarized form, both A and B perform
the generalized XOR gate XORN : Ik) Il) t--+ Ik) I(l + k)modN) on their
respective systems. Then both measure the state of their second system in
the basis Ik). The first pair is kept, if they get the same result otherwise it
is discarded (as the second pair always is). The resulting state has a density
matrix p', which has a larger overlap with the maximally entangled state 1cI>~)
than the original p. Iterating the last two steps sufficiently often, the overlap
between the resulting state and 1cI>~) approaches 1, that is, the distilled stated
converges to the maximally entangled state 1cI>~).

For infinite dimensional systems That distillability is implied by Ineq. (15.1)

was proved in [12] for finite dimensional systems; to apply this condition to
CV states we have to extend the proof to dim1i = 00.
Let {Ik) : k = 0,1, ... } be an orthonormal basis of 1i, let 1i n = span{IO) ,
11) , ... , In)}, and let p be a density matrix on 1i 01i. Assume that :31'1/;) E
1i, E > 0 such that ('1/;1 (trBP 0 :n. - p) 1'1/;) :::; -E < O. Since Pn = Pll n PPll n
converges to p (e.g. in the weak operator topology), there is N ~ 0 such that
('1/;1 (trBPn 0 :n. - Pn) 1'1/;) :::; -E/2 for all n ~ N. Thus P can be projected by
local operations to a distillable state PN and is therefore itself distillable. _

3.2 DISTILLABILITY OF GAUSSIAN STATES

If both the states P and 'I/; occuring in (15.1) are Gaussian with displacements
dp = d1jJ = 0 and correlation matrices Mp, M1jJ respectively, then (15.1) takes
the form [25] (IMI denotes the determinant of M):

(15.13)
Let P be a symmetric Gaussian state in standard form and 'I/; the pure two-
mode squeezed state 1'1/;) = co;h T l:n tanh n r Inn). This is itself a Gaussian
state and the four parameters (15.9) are na = nb = cosh r, kx = -kp = sinh r.
In the limit of large r (keeping only the leading terms in eT ) Ineq. (15.13)
becomes after some simple algebra

(15.14)
But Ineq. (15.14) is - for symmetric states - implied by the inseparability
criterion: in that case, (15.11) simplifies to In 2 - kxkp - 11 < n(k x - kp),
which implies Ineq. (15.14), proving that all symmetric inseparable Gaussian
states are distillable.
If the state is not symmetric, it means that the reduced state at one of the
two sides has larger entropy than the other. This suggests to let a pure state
interact with the "hotter" side to cool it down. To do this without destroying the
 Distillability and entanglement purification for Gaussian states 179

entanglement of p, we proceed as follows: p is transformed such that the corre-

lation matrix of its Wigner function takes on its standard form with parameters
(Na, N b, Kx, Kp). The Wigner function is the symplectic Fourier transform
of the characteristic function (15.5) and therefore also Gaussian for Gaussian
states. The Wigner correlation matrix Mw is related to the (characteristic) cor-
relation matrix by Mw = -JM-IJ [26]. One can formulate the conditions
(15.10,15.11) similarly in terms of the parameters (Na , N b, Kx, Kp), which
are related to the parameters of X by

(Na,Nb,Kx,Kp) = (nb,na,-kp,-kx)/JiMT·

A state in the standard form (15.8) can be brought into the Wigner standard
form by local squeezing operations.
Now assume that Nb < N a, i.e., B is the hotter side. B takes an ancilla
mode in the vacuum state and couples it to its mode using a beam splitter with
transmittivity cos 2 (}. After a homodyne measurement of the ancilla results a
state p with Wigner correlation matrix M with2

M _.!. ( c2 Na + 8 2 Dx 0 )
A - v 0 c2 Na + 8 2 NaNb '

M _ .!. ( cKx 0 )
AB - v 0 cKpv '

where the abbreviations c = cos (}, 8 = sin (}, v = 8 2 Nb + c2 , and D x,p

NaNb - K;,p were used. Symmetry, i.e.IMAI = IMBI, requires
N2_N 2
tan2 (} = a b. (15.15)
Nb-DxNa

Checking (15.11) for M one easily sees that the sign of the left-hand side does
not change; therefore the transformed state is inseparable iff the original one
was inseparable. It remains to be shown that there always exists a (} to satisfy
(15.15), i.e., that the right hand side of Eq. (15.15) is positive. The numerator
is positive since Nb < N a, the denominator is positive for all states since the
second part of condition (15.10) implies that (Na - DxNb) > 0 and the first
part assures that (Na - DxNb)(Nb - DpNa ) ;::: (NaKx + NbKp) 2 ;::: 0, hence
all Gaussian states can be symmetrized this way. But since every Gaussian state
can be brought into Wigner standard form by local unitaries, this completes the
proof. _
180 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

This implies that the protocol of Ref. [12] can in principle be used to obtain
maximally entangled states in a finite dimensional Hilbert space from any given
inseparable Gaussian two-mode state.
Before discussing EPPs in more detail in the following, we end this section
on distillability by mentioning an interesting recent result. The equivalence of
inseparability and distillability of Gaussian states holds only for the case of two
modes. Werner and Wolf showed in [19] that if both A and B possess more that
one mode, there exist Gaussian states that are bound entangled, namely states
that are entangled but whose density matrices remain positive under partial
transposition. In later work [23] it was shown that for any number of modes
bipartite Gaussian states are distillable iff their partial transpose is not positive.

4. ENTANGLEMENT PURIFICATION FOR

GAUSSIAN STATES
In this section we discuss some processes that may be used to accomplish
entanglement purification.
For qubit systems, efficient entanglement purification protocols have been
found [3], but none has so far been realized experimentally due to the great
difficulty to perform repeated collective operations in realistic quantum com-
munication systems. Direct extensions of these schemes have been considered
for entanglement purification of Gaussian states, but until now none of these
extensions has provided an EPP for Gaussian states [27, 28]. Moreover, ex-
tensive numerical investigations of local quasifree transformations acting on
several pairs of entangled Gaussian states and ancillas have failed to tum up
any improvement in entanglement or related properties [28]. In fact, it was
shown very recently [35] that it is impossible to distill Gaussian states with
LQT and homodyne measurements.
Thus, the discussion should be extended to a larger class of operations to
distill entanglement of CV states. In [29] a protocol to increase the entangle-
ment for the special case of pure two-mode squeezed states has been proposed,
which is based on conditional photon number subtraction; the efficiency, how-
ever, seems to be an obstacle for its practical realization. In the following two
subsections we will discuss briefly the EPP for all Gaussian states based on
the proof in Sec. 3. and then in detail a more practical protocol [30, 31] that
circumvents the major difficulties of the general scheme, and allows to distill
entanglement of a subset of (pure and mixed) Gaussian states.

4.1 AN EPP FOR ALL GAUSSIAN STATES

Let us briefly consider how the steps for distillation of an inseparable Gaus-
sian state p using (the generalization of) the protocol of Ref. [12] might be
accomplished quantum optically.
 Distillability and entanglement purificationfor Gaussian states 181

As usual for EPPs A and B share a large number of identically prepared

entangled systems in the known state p.
0.) Symmetrization: Remove displacements d i- 0; symmetrize p if nec-
essary; bring the symmetric state into standard form. All these steps can
be performed by the local use of beam splitters, one-mode squeezers, ancilla
systems in coherent states, and a homodyne measurement.
For a state in symmetric standard form the filtering operation (15.12) required
in the EPP is unnecessary, since then p already satisfies Ineq. (15.1) with the
state 1'ljJ) ex lim>'--+l L:k A,k Ik) Ik) (in the photon number basis). This gives
anm = lim>'--+l A,n+mt5nm , hence A = (a nm ) = 1.
1.) Depolarization: Transform the state into a mixture of 1q,~+1) and the
maximally mixed state ex 1 by applying U ® U* with U randomly chosen.
However, the class of currently realizable unitaries is in fact very limited and
we do not know how to depolarize an arbitrary state quantum optically.
2.) Joint measurement: This is the central step of the distillation protocol.
A bilocal XOR is used to mutually entangle two entangled pairs. A subsequent
measurement selects a distilled subensemble.
This operation may be implemented by a measurement of the total photon
number Na = Nal + N a2 , a = A, B on both sides: The state conditional
on both A and B obtaining the same result N differs only by a local unitary
transformation (namely In, N - n)a f-7 In, N)a) from the one obtained by
directly following the steps described in Sec. 3., i.e., first projecting bi-Iocally
to the N + 1 dimensional subspace 1-lN+1 (p f-7 PN+l), then performing the
bi-Iocal XORN+1, and finally measuring the target system with result N3. As
shown before, for a sufficiently large value of N, the truncated state PN+1
is distillable and then step 2.) produces a state with larger overlap with the
N + I-level maximally entangled state 1q,~+1).

Each iteration of these two steps brings the state closer to a maximally entan-
gled state in the Hilbert space of dimension N, where (N + 1)2 is the last result
of the total photon number measurement. Hence with finite probability one can
get arbitrarily close to a maximally entangled state in any finite dimensional
space provided the initial supply of states P is sufficiently large.

4.2 A PRACTICAL EPP

In this subsection, we describe the entanglement purification scheme pre-
sented in [30, 31], having the following properties: (i) It produces maximally
entangled states in a finite dimensional subspace. For a relevant class of states
this is accomplished in a single step. (ii) For pure states it reaches the maximal
allowed efficiency in the asymptotic limit (when the number of pairs of modes
goes to infinity). (iii) It can be readily extended to distill maximally entangled
182 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

states from a relevant class of mixed Gaussian states which result from losses
in the light transmission. Furthermore, we propose and analyze how to imple-
ment this protocol experimentally using high finesse cavities and cross-Kerr
nonlinearities. We begin by describing the entanglement purification protocol
for pure two-mode squeezed states, and then extend the protocol to include
mixed Gaussian CV states, and describe a physical implementation in the next
subsection.

For pure states First assume that we have generated m entangled pairs Ai, Bi
(i = 1,2,··· m) between two distant sides A and B. Each pair of modes Ai, Bi
are prepared in the pure two-mode squeezed state I'll) AiBi' which in the number
basis has the form
00

I'll) AiBi = VI _.x2 L .xn In) Ai In) Bi ' (15.16)

n:=:O

where.x = tanh (r) ,and r is the squeezing parameter [18]. The entanglement
BJ
E (I'll) Ai of the two-component state (15.16) is uniquely quantified by the
von Neumann entropy of the reduced density operator of one subsystem. The
joint state I'll) (AiBd of the m entangled pairs is simply the product of all the
I\lI) Ai Bi' which can be rewritten as

1'lI)(AiBd = (1 - .x2) T f J
j:=:O
.xj fjm) 1J)(AiBd ' (15.17)

where (AiBi} is abbreviation ofthe symbol AI, B I , A 2 , B 2 , ... and Am, B m ,

and the normalized state Ij) (Ai B ,} is defined as

IiI, i 2 ,··· ,im)(Ai} ® IiI, i 2 ,··· ,im)(Bd· (15.18)

. f(m).
The fu nctIOn j m E q. (15 . 17) an d (15 . 17)·IS gIven
. b y f(m)
j = (j+m-l)!
j!(m-l)! .
Note that the state Ij) (A;Bi} represents a maximally entangled state in the
subspace corresponding to a local photon number of j at both sides. To con-
centrate the entanglement of these m entangled pairs, we perform a QND
measurement of the total photon number N Al + N A2 + ... + N Am on the A
side (we will describe later how to implement this measurement experimen-
tally). The QND measurement projects the state 1'lI)(AiBd onto the two-party
maximally entangled state Ij)(Ai B ,} with probability

Pj = (1- .x2)m .x 2j fjm).

 Distillability and entanglement purification/or Gaussian states 183

0.'1 r---~-"""----'----"------' 0.'1 r-----.-~----.----,----......--_,

0.3 0.3

O.ZI 0.2.\

0.2 0.2
c:

.
O.IS 0.1.

0.1 0.\ ,
,
0.05 0.01

0.'

Figure 15.1 The purification success probability versus entanglement increase ratio for two
(left) and four (right) pairs. Dotted line for the squeezing parameter r = 0.5, dashed line for
r = 1.0, and solid line for r = 1.5.

The entanglement of the resulting state Jj)(Ai B;} is given by E(Jj)(AiB;}) =

log Jjm). The quantity

rj = E (lj)(AiBd) IE (IW)AiBJ

defines the entanglement increase ratio, and ifrj > 1, we get a more entangled
state. Even with a small number m, the probability of obtaining a more
entangled state is quite high. Figs. 15.1a,b show the probability of achieving an
entanglement increase ratio r j for various values of initial entanglement and
initial number of pairs.
To measure how efficient the scheme is, we define the entanglement transfer
efficiency Y with the expression

Y= f ______
p;m) E (Jj) (AiBd)
~J=_o ______
~

(15.19)
mE (IW)AiBJ
It is the ratio of the average entanglement after concentration measurement
to the total initial entanglement contained in the m pairs. Obviously, Y :S 1
should always hold. With the squeezing parameter r = 0.5,1.0 or 1.5, the
entanglement transfer efficiency versus the number of pairs m is shown in Fig.
15.2.
From this figure, we see that the entanglement transfer efficiency is near to
1 for a large number of pairs. In fact, it can be proven that as m -+ 00, we
would get with unit probability a maximally entangled state with entanglement
mE (I w) Ai Bi ). To show this, we calculate the mean value and the variance of
184 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

0.9

0.8 ...
0.7

0.6

0.5

0.4

0.3

0.2

0.1

10 12 14 16 18 20
m

Figure 15.2 The entanglement transfer efficiency versus the number of pairs m in simultaneous
concentration. Dotted line for r = 0.5, dashed line for r = 1.0, and solid line for r = 1.5.

the distribution p;m), and find

(IS.20)
(1 - ).2)2'

The results show that if m tends to infinity, V(b.j) /J -+

2 0 and the distribution
p;m) tends to a 8-like function. Furthermore, around the mean value J, the
entanglement of the resulting state I)) (AiBd is
(1S.21)

so the entanglement transfer efficiency tends to unity. This proves that the
purification method described above is optimal in the asymptotic limit (m -+
(0), analogous to the purification protocol presented in [S] for the qubit case.
For any finite number of entangled pairs, this purification protocol is more
efficient than that in [S], since it takes advantage of the special relations between
the coefficients in the two-mode squeezed state.
An interesting feature of this entanglement purification protocol is that for
any measurement outcome j of. 0, we always get a useful maximally entangled
state in some finite Hilbert space, though the entanglement of the outcome state
I)) (AiBd does not necessarily exceed that of the original state I\II) A,B, if j is
small.

For mixed entangled states In reality, due to unavoidable losses during

light transmission, we will not start from an ideal two-mode squeezed state, but
 Distillability and entanglement purification/or Gaussian states 185

rather from a mixed state described by the following master equation

m

p= -i (HeffP - pH!ff) + L (17AaAiPa~i + 17BaBiPa1i) (15.22)

i=l

where p is the density operator of the m entangled pairs with p (0) = 1\l1)(Ai Bd
(\l1I, the pure two-mode squeezed state (15.16), and the effective Hamiltonian

(15.23)

In Eqs. (15.22) and (15.23), aai denotes the annihilation operator of the mode
ai, (a = A or B), and we have assumed that the damping rates 17 A and 17Bare
the same for all the m entangled pairs, but 17A and 17B may be different from
each other. Eq. (15.22) describes a situation in which single photon absorption
is the only relevant source of noise as is typically the case at optical frequencies.
Small Noise In many practical cases, it is reasonable to assume that the
light transmission noise is small. Let T denote the transmission time, then
17AT and 17BT are small factors. To the first order in 17AT and 17BT the fi-
nal state of the m entangled pairs is in the language of quantum trajectories
[18] either 1\l1(O))(Ai B;} DC e-iHeffT 1\l1)(Ai B;} (no quantum jumps occurred)
or 1\l1(a;))(Ai Bd DC y'17aTaai 1\l1)(Ai Bd (a jump occurred in the ai channel
(a = A,B andi = 1,2,··· ,m). The final density operator is a mixture of all
these possible states. To distill entanglement from this mixed state, we perform
QND measurements of the total photon number on both sides A and B, with
results jA and jB, respectively. We then compare jA and jB through classical
communication (eC), and keep the outcome state if and only if jA = jB.
Let pJ) and piP
denote the projections onto the eigenspaces of the corre-
m m
sponding total number operators L a~aAi and L a1.aBi with eigenvalue
i=l' i=l'
j, respectively. It is easy to show that

p(j) p(j) I\l1(O))

A B (AiBi}
p(j) p(j)
A B
1\l1(a;))
(AiBd
o. (15.24)

So if j A = jB = j, the outcome state is the maximally entangled state Ij) (Ai B;}
with entanglement log (tjm)). The probability to get the state jj) (AiBd is
now given by pj = (1- ,\2)m ,\2j fjm)e-('TJA+'TJB)Tj. It should be noted that
the projection operators pi!) piP cannot eliminate the states obtained from the
186 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

initial state IW)(AiBd by a quantum jump on each side A and B. The total
probability of this kind of quantum jumps is proportional to m 2n 2'T1A'TIBT2. So
the condition for small transmission noise requires m 2n 2'T1A'TIBT2 « 1, where
n = sinh2 (r) is the mean photon number for a single mode.
Asymmetric Noise In the purification of mixed entanglement, we need CC to
confirm that the measurement results on both sides are the same, and during this
CC, we implicitly assume that the storage noise for the modes is negligible.
In fact, that the storage noise is much smaller than the transmission noise
is a common assumption made in all the entanglement purification schemes
which need the help of repeated CC [3]. If we make this assumption for
continuous variable systems, there exists another simple configuration in which
the purification protocol works. Let the two-mode squeezed states be generated
at side A. After state generation, we keep the modes Ai on side A with a very
small storage loss rate 'TIA, and at the same time the modes Bi are transmitted
to the distant side B with a loss rate 'TIB » 'TIA. We call this a configuration
with an asymmetric transmission noise. In this configuration, the purification
protocol is exactly the same as that described in the above paragraph. We note
that the component in the final mixed density operator which is kept by the
projection P)1) pijl should be subject to the same times of quantum jumps
on each side A and B. We want this component to be a maximally entangled
state. This requires that the total probability of the same nonzero number of
quantum jumps on both sides to be very small. Clearly, this total probability is
always smaller than n'TIAT, no matter how large the damping rate 'TIB is. So the
working condition of the purification protocol in the asymmetric transmission
noise configuration is given by n'TIAT « 1. The loss rate 'TIB can be large. The
probability to obtain the maximally entangled state Ij) (Ai B ;} is still given by
pj = (1 - ,\2)m ,\2j fjm)e-(TJA+TJB)Tj.
A-posteriori Purification For continuous variable systems the assumption
of storage with a very small loss rate is typically unrealistic. In that case, we
can use the following simple method to circumvent the storage problem. Note
that the purpose of distilling maximally entangled states often is to directly use
them in some quantum communication protocol, such as quantum cryptography
or quantum teleportation. So we can modify the above purification protocol
by the following procedure: immediately after the state generation, we make a
QND measurement of the total photon number on side A (measurement result
jA). Then we do not store the resulting state on side A, but immediately
use it (e.g., perform the corresponding measurement as required by a quantum
cryptography protocol [16]). During this process, the modes Bi are being sent to
the distant side B, and when they arrive, we make another QND measurement
of the modes Bi and get a outcome j B. The resulting state on side Bean
be directly used (for quantum cryptography, for instance) if j A = j B, and
 Distillability and entanglement purification for Gaussian states 187

discarded otherwise. By this method, we formally get maximally entangled

states through posterior confirmation, and at the same time we need not store
the modes on both sides.

4.3 IMPLEMENTATION OF THE EPP

To experimentally implement the above purification scheme, we first have
to generate Gaussian entangled states between two distant sides, and then
perform a local QND measurement of the total excitation number of several
entangled pairs. Here we describe the experimental scheme proposed in [30,
31], which uses high finesse optical cavities to store CV entangled states and
cavity enhanced cross-Kerr interactions to realize the local QND measurement.
Gaussian entangled states between two distant cavities can be generated
as follows (cf. Fig. 15.3): we transmit and then couple the two outputs of
a nondegenerate optical parametric amplifier to distant high finesse cavities.
The steady state of the cavities is nothing but a Gaussian continuous entangled
state described by the solution of Eq. (15.22) after taking into account the
propagation loss [32, 31].

A B
Figure 15.3 Schematic setup for generating Gaussian entangled states between two distant
cavities.

The difficult part is to perform a QND measurement of the total photon

number contained in several local cavities. We use the setup depicted in Fig.
15.4 to achieve this goal. (For convenience, we use the two-cavity measurement
as an example to illustrate the method. Extension of the measurement method
to multi-cavity cases is straightforward.)
The measurement model depicted in Fig. 15.4 is an example of a cas-
caded quantum system [18]. The incident light bil can be expressed as
bil = b~l + g.,;r, where g.,;r (g is a large dimensionless factor) is a con-
stant driving field, and b~l is the standard vacuum white noise, satisfying
(b~l (t) b~l (tf)) = 0 and (b~l (t) b~l (tf)) = 8 (t - tf) . The Hamiltonian for
the Kerr medium is assumed to be Hi = nxNiblbi, (i = lor 2), where bi is
the annihilation operator for the ring cavity mode, and X is the cross-phase
modulation coefficient. The self-phase modulation can be made much smaller
than the cross-phase modulation with some resonance conditions for the Kerr
188 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

KH------H

KI+-......... II

Figure 15.4 Schematic experimental setup to measure the total photon number NI + N2
contained in the cavities I and II. The cavities I and II, each with a smaIl damping rate K, and
with a cross Kerr medium inside, are put respectively in a bigger ring cavity. The ring cavities
with the damping rate 'Yare used to enhance the cross Kerr interactions. A strong continuous
coherent driving light bi! (t) is incident on the first ring cavity, whose output bOI is directed to
the second ring cavity. The output bo 2 (t) of the second ring cavity is continuously observed
through a homodyne detection.

medium, and thus is negligible [33, 34]. In the frame rotating at the optical
frequency, the Langevin equations describing the dynamics in the two ring
cavities have the form

-iXNlbl - ~bl - "'/yb~l - g,,(,

(15.25)
-iXN2b2 - ~b2 - .../Ybi2 ,

with the boundary conditions (see Fig. 1) bi2 = b01 = b~l + g..,fY + ..,fYb 1
and bo2 = bi2 +..,fYbz. In the realistic case "( » X (Ni ) , (i = 1, 2), we
can adiabatically eliminate the cavity modes bi, and express the final output
bo2 of the second ring cavity as an operator function of the observable Nl +
N 2 . The experimentally measured quantity is the integration of the homodyne
photocurrent over the measurement time T. Choosing the phase of the driving
 Distillability and entanglement purification for Gaussian states 189

field so that 9 = i 191, the measured observable corresponds to the operator

XT
(15.26)

where X~) = ~ (bT+bt), and bT, satisfying [bT,bt] = 1, is defined

by bT = Jr Ji{ b~l (t) dt. Eq. (15.26) assumes, » X(Ni ) and e--yT « 1.
There are two different contributions in Eq. (15.26). The first term represents
the signal, which is proportional to Nl + N2, and the second term is the vacuum
noise. The distinguishability of this measurement is given by r5n = 8Ig~n'
If r5n < 1, i.e., if the measuring time T > 64Ig}x 2 ' we effectively perform a
measurement of Nl + N 2; and if T is also smaller than ,,-(1
i ) , the photon loss
in the cavities I and II during the measurement is negligible. So the setup gives
an effective QND measurement of the total photon number operator Nl + N2
under the condition

(15.27)

This condition seems to be feasible with the present technology. For ex-
ample, if we assume the cross-Kerr interaction is provided by the resonantly
enhanced Kerr nonlinearity as considered and demonstrated in [33, 34], the
Kerr coefficient X/27r '"" O.lMHz would be obtainable4 . We can choose the
decay rates K,/27r '"" 4MHz and, /27r '"" 100MHz, and let the dimensionless
factor 9 '"" 100 (for a cavity with cross area S '"" 0.5 x 1O-4 cm2 , 9 '"" 100
corresponds a coherent driving light with intensity about 40mWcm- 2 ). The
mean photon number (N1 ) = (N2 ) = sinh2 (r) '"" 1.4 for a practical squeez-
ing parameter r '"" 1.0. With the above parameters, Eq. (15.27) can be easily
satisfied if we choose the measuring time T '"" 8ns. More favorable values for
the parameters are certainly possible.
To bring the above proposal into a real experiment, there are several imper-
fections which should be considered. These include phase instability of the
driving field, imbalance between the two ring cavities, light absorption in the
Kerr medium and the mirrors, self phase modulation effects, light transmission
loss between the ring cavities, and inefficiency of the detectors. To realize
a QND measurement, the imperfections should be small enough. We have
deduced quantitative requirements for all the imperfections listed above [31].
190 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

With the parameters given in the above paragraph, all these requirements can
be met experimentally.

Acknowledgments
G.G. acknowledges financial support by the German Friedrich-Naumann-
Stiftung. This work was supported by the Austrian Science Foundation under
the SFB "Control and Measurement of Coherent Quantum Systems" (Project
11), the European Union under the TMR network ERB-FMRX-CT96-0087
and the project EQUIP (contract IST-1999-11 053), the European Science Foun-
dation, and the Institute for Quantum Information GmbH, Innsbruck.

Notes
1. S is called symplectic if ST JS = J.
2. See [35] for the general formalism to describe measurements on Gaussian states.
3. To be more precise: this equivalence holds on the infinite dimensional space, when XOR:ln, m) >-+
In, m + n). For states in a N dimensional subspace (as obtained after the first step) this equivalence is only
true for measurement outcomes NOt. :os: N
4. In fact, Ref. [33] considered a configuration, yielding a Kerr coefficient X ~ lOOMHz, to realize
a single-photon turnstile device. But the estimation there puts a stringent limit on the required cavity
parameters [K. M. Gheri et at., Phys. Rev. A 60, R2673, 1999]. We take a much more moderate estimation
of the relevant parameters and find X/27r rv O.lMHz is obtainable. This value of the Kerr coefficient
is large enough for performing the QND measurement, though it is certainly not enough for realizing a
single-photon turnstile device.

References

[1] Bouwmeester, J.-W. Pan, K. MattIe, M. Eibl, H. Weinfurter, A. Zeilinger,

Nature 390,575, (1997); D. Boschi, S. Branca, F. De Martini, L. Hardy,
S. Popescu, Phys. Rev. Lett. 80, 1121 (1998).
[2] A. Ekert, Phys. Rev. Lett. 67, 661 (1991); C.H. Bennett, G. Brassard, N.
Mermin, Phys. Rev. Lett. 68, 557 (1992).

[3] N. Gisin, Phys. Lett. A 210, 151 (1996); C.H. Bennett, G. Brassard, S.
Popescu, B. Schumacher, J.A. Smolin, W.K. Wootters, Phys. Rev. Lett.
76, 722 (1996).
[4] C.H. Bennett, D.P. DiVincenzo, J.A. Smolin, W.K. Wootters, Phys. Rev.
A 543824 (1996).
[5] C.H. Bennett, H.J. Bernstein, S. Popescu, B. Schumacher, Phys. Rev. A
53,2046 (1996).
[6] H.-J. Briegel, W. Dur, J.I. Cirac, P. Zoller, Phys. Rev. Lett. 81, 5932
(1998).
 Distil/ability and entanglement purification/or Gaussian states 191

[7] M. Horodecki, P. Horodecki, R Horodecki, Phys. Rev. Lett. 80, 5239

(1998).
[8] P. Horodecki, Phys. Lett. A 232,233 (1997).
[9] M. Lewenstein, D. Bruss, J.1. Cirac, B. Kraus, M. Kus, J. Samsonowicz,
A. Sanpera, R Tarrach, J. Mod. Opt. 47, 2481 (2000), quant-ph/006064.
[10] P. Horodecki, M. Lewenstein Phys. Rev. Lett. 85, 2657 (2000).
[11] W. Dur, J.1. Cirac, M. Lewenstein D. BruB, Phys. Rev. A 61, 062313
(2000), quant-ph/9910022; D. DiVincenzo, P. Shor, J. Smolin, B. Terhal,
A. Thapliyal, Phys. Rev. A 61, 062312 (2000), quant-ph/9910026.
[12] M. Horodecki, P. Horodecki, Phys. Rev. A 59,4206 (1999).
[13] H.-A. Bachor, A guide to experiments in quantum optics, Wiley-VCH,
Weinheim (1998).
[14] L. Vaidman, Phys. Rev. A 49, 1473 (1994); S.L. Braunstein, H.J. Kimble,
Phys. Rev. Lett. 80, 869 (1998).
[15] A. Furusawa, J. SSZlrensen, SL Braunstein, C.A. Fuchs, R.J. Kimble, E.S.
Polzik, Science 282, 706 (1998).
[16] T.e. Ralph, Phys. Rev. A 61, 010303R (2000); M. Hillery, Phys. Rev.
A 61, 022309 (2000); M. Reid, Phys. Rev. A 62, 062308 (2000),
quant-ph/9909030; T.e. Ralph, Phys. Rev. A 62,062307 (2000), quant-
phl0007024.
[17] D. Petz, An Invitation to the Algebra of Canonical Commutation Relations,
Leuven University Press, Leuven (1990).
[18] e. Gardiner, P. Zoller, Quantum Noise 2nd ed., Springer-Verlag, Berlin
(1999).
[19] RF. Werner, M.M. Wolf, Phys. Rev. Lett. 86, 3658 (2001), quant-
phl0009118 (2000).
[20] J. Manuceau, A. Verbeure, Comm. Math. Phys. 9, 293 (1968).
[21] L. Duan, G. Giedke, J.1. Cirac, P. Zoller, Phys. Rev. Lett. 84,2722 (2000).
[22] R Simon, Phys. Rev. Lett. 84, 2726 (2000).
[23] G. Giedke, L.-M. Duan, J.1. Cirac, P. Zoller, Quant. Inf. Compo 1(3), 79
(2001); quant-ph/0104072 and quant-phl0007061.
[24] C. Helstrom, Quantum detection and estimation theory, Associate Press,
London (1976).
[25] H. Scutaru, J. Math. Phys. 39, 6403 (1998).
[26] H. Scutaru, Phys. Lett. A 141, 223 (1989).
[27] S. Parker, S. Bose, M.B. Plenio, Phys. Rev A, 61, 032305 (1999).
[28] G. Giedke, PhD Thesis, Universitat Innsbruck (2001).
192 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[29] T. Opatrny, G. Kurizki, D.-G. Welsch, Phys. Rev. A 61, 032302 (1999).
[30] L.-M. Duan, G. Giedke, J.I. Cirac, P. Zoller, Phys. Rev. Lett. 84, 4002
(2000).
[31] L.-M. Duan, G. Giedke, J.I. Cirac, P. Zoller, Phys. Rev. A 62, 032304
(2000).
[32] A. S. Parkins, H. J. Kimble, Phys. Rev. A 61, 052104 (2000).
[33] A. Imamoglu, H. Schmid, G. Woods, M. Deutsch, Phys. Rev. Lett. 79,
1467 (1997); 81,2836 (1998).
[34] L.v. Hau, S.E. Harris, Z. Dutton, C.H. Behroozi, Nature 397,594 (1999).
[35] G. Giedke and J. I. Cirac, quant-phl0204085.
Chapter 16

ENTANGLEMENT PURIFICATION
VIA ENTANGLEMENT SWAPPING

S. Parker
Optics Section, The Blackett Laboratory, Imperial College
London, England, SW72BW
[email protected]

S. Bose
Centre for Quantum Computing, Clarendon Laboratory,
University of Oxford, England, OX] 3DU
[email protected]

M. B. Plenio
Optics Section, The Blackett Laboratory, Imperial College
London, England, SW72BW
[email protected]

Abstract In this section we aim to generalize the entanglement swapping procedure to

continuous variable systems and show that for certain types of pure continuous
variable states this process can increase the amount of entanglement between
separated parties i.e. achieve entanglement purification. To show this we will
need a generalization of the von Neumann entropy - the finite state entanglement
measure - to continuous variables together with a numerical procedure for its
calculation.

1. INTRODUCTION
The experimental [1, 2, 3, 4] and theoretical [5] realization of teleportation
in continuous variable systems has demonstrated that entanglement between
separated systems is essential in order to be able to perform teleportation and
193
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 193-209.
© 2003 Kluwer Academic Publishers.
194 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

that the efficiency of the procedure is strongly dependent on the amount of

entanglement therein.
It is therefore important that we are able to quantify the amount of entangle-
ment in these processes and provide procedures to concentrate this entangle-
ment. This problem of concentrating entanglement is more precisely phrased as
follows: given an entangled bipartite state shared between two spatially sepa-
rated parties (Alice and Bob) the aim is to increase the amount of entanglement
using only local operations and classical communication [6,7,8]. This process
is known as entanglement purification, and for continuous variable systems was
first considered in [9] and investigated further in a number of directions. Many
nice results on the distillability of Gaussian states have been obtained [10], in-
cluding distillability criteria and actual physical implementations of distillation
schemes. Also the structure of the entanglement of continuous variable states
has been investigated with nice results such as bound entangled states [11] .
A significant amount of progress has been made in the purification of discrete
systems [6, 7, 12]. An example to be used in this chapter for pure finite level
states is entanglement swapping [13], a process very similar to teleportation,
which can achieve purification of entanglement from two bipartite systems
into just one. In the swapping process one particle from each of two pairs of
entangled particles can become entangled without ever having interacted when
a joint measurement is performed on the other two particles. This is a proba-
bilistic process dependent on the measurement result, and the entanglement of
the first two particles is also probabilistic. On average it cannot increase but
for some of the measurement results the entanglement can be more than that
of either initial pair. This is one method, therefore, of achieving entanglement
purification.
We will present two classes of continuous entangled states and show that a
continuous generalization of the entanglement swapping procedure using the
continuous controlled-NOT and Hadamard gates introduced by Braunstein [14]
is able to produce purification in one of our two classes.
For the quantification of entanglement and of the efficiency of the concen-
tration procedures a measure of entanglement is required. For pure finite states
it is generally accepted that in the asymptotic limit of many copies the only
measure of entanglement that is sensible is the entropy of entanglement [6].
Here we use a direct generalization of this to verify whether or not the attempted
purification procedure has been successful.

2. ENTANGLING GATES AND CLASSES OF

ENTANGLED STATES
For finite level states two basic operations used in quantum information
procedures are the Hadamard transform and the controlled-NOT gate. The
 Entanglement purification via entanglement swapping 195

continuous analogue of the former single "particle" gate is the Fourier trans-
form:

Fix) = _1_! e2ixy/u2 Iy) dy . (16.1)

..foa
The scale length, a, included here is normally used to make the expression
dimensionless but we will also use it later as a convenient scale with which to
compare various lengths in the states we will be dealing with. If we set a = 1
and work in units Ii = 1/2 this is also, of course, the transform used to go from
the position to the momentum basis. The inverse, Ft is obtained by replacing
i by -i in Eq. (16.1) giving the result that FFt Ix) = FtF Ix) = Ix).
There are a number of generalizations we could use for the continuous
controlled-NOT gate: the phase-free beam splitter transformation B (i) lx, y) =
Ix:;!, x:}f) is one good candidate but we will use a slightly simpler version:
(16.2)

whose inverse is obtained by replacing the + with a - sign on the right hand
side.
We can now define the 'entangling' operation and its inverse:

"t _ ,.dCt (16.3)

"12 -.r1 12'
which is just the individual inverse operations performed in reverse order. Note
however that it is only the CNOT gate that is able to entangle states but in the
case of our gate C12 it can only do so if particle 1 is in a superposition state
in the basis we are dealing with. We therefore also require an operation that
creates a superposition state for the first particle. Thus we will use the Fourier
transform.
The operation £12 can be used to form entangled states from different initial
unentangled states of two particles. Applying it to two Gaussian wavepackets

(16.4)

and IG,6(x2))2 produces the state

i:
C12F1I Ga(x1))1IG,6(X2))2

exp [:2 ( _x 2a 2 -
IX)l Ix + y + X2)
~: + 2iX1X) ]
dxdy
IB a,6(x1' X2))12' (16.5)
196 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Such states can be used to demonstrate teleportation of an unknown state, the

fidelity of the teleportation approaching unity as a and f3 -+ 0, where the state
becomes like an infinitely squeezed two mode squeezed state or an EPR state
[15]. We will call the states IBa,B(Xl, X2))12 partially correlated entangled
states.
Our second class of states cannot easily be formed using this operation as
they are a different kind of entangled state. They are a direct generalization of
the so called Schrodinger cat states, which are a superposition of two coherent
states Ia). The Wigner function for one of these states is shown in figure 16.1.

. "~'"

0.1 " ", .:':'"

0.08
. '"

-":"'" ...
... ,....
0.06

W 0.02

-0.02

-0.04

-0.06
10

10

x
Figure 16.1 Wigner function for a Schrbdinger cat state, 1/V2 (I+a) + I-a» with coherent
states la) = 13).

Our generalizations we will call two-mode cat states [16, 17], as they are
Schrodinger cat states of two modes whose locations are correlated with each
 Entanglement purification via entanglement swapping 197

jC(d))12 = i:
other quantum mechanically:

[Aoe-CX-d)2_CY+d)2

jxh jY)2 dxdy

+ Ale-CX+d)2-CY-d)2]

1 f;00
-00
1 2
[A j e-(X-C-l)i d) -(y+C-l)id) ]
2

jxh jY)2 dxdy. (16.6)

The complex coefficients, A j , are such that jAoj2 + jAlj2 = 1 (so the state
is not normalized correctly). This state is a superposition of the first particle
being located around d and the second around -d and vice versa. The scale
length does not appear here (it is set to unity) as an increase in scale length is
equivalent to a decrease in the value of d.

3. QUANTIFICATION OF ENTANGLEMENT
The measure of entanglement for pure bipartite states, the entropy of entan-
glement, E(PI2), is just the Von Neumann entropy, S, of either partial density
operator, PI or P2, of the system [6]

(16.7)

where the Ai'S are the eigenvalues of PI or P2. We can very simply generalize
this to continuous variable systems but we need to know a little about the
eigenvalues of such systems. Let us express a continuous variable state as
follows

j'l/Jh2 = J
'l/J(x, y) jxh jY)2 dxdy, (16.8)

and find the partial density operator of one of the particles (the first, say) by
tracing out the other

PI J 2 (xj (j'l/J) ('l/Jjh2jx)2 dx

J PI (x, y) jx) 11 (yj dxdy. (16.9)

We now wish to find the eigenvalues of PI and so we form a continuous

eigenvalue equation

(16.10)
198 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

where Ai is the eigenvalue corresponding to <Pi (x) and PI is known as the

kernel of the equation. The mathematics we require is covered in the area
of integral eigenvalue equations [18]. In particular we will be interested in
Hermitian kernels (for which pi (y, x) = PI (x, y)) which are integrable over
the two continuous degrees of freedom, i.e. they are quadratically integrable.
Such kernels have many interesting properties [9, 18] of which we will
list the important ones for the work that follows. Firstly, the eigenvalues of
all Hermitian kernels are real and the set of eigenfunctions linearly indepen-
dent, complete and orthogonal. Secondly, those Hermitian kernels which are
quadratically integrable in general have infinitely many eigenvalues (although
certain kernels may have a finite number), and these eigenvalues have no accu-
mulation points (except at zero), i.e. the eigenvalues do not form a continuous
set.
It is now obvious that the generalization of the entropy of entanglement
to continuous systems is trivial and is again the Von Neumann entropy of
either partial density operator, except that the summation over eigenvalues in
Eq. (16.7) may have infinitely many terms. It should be noted however, that
such an entanglement measure is not continuous anymore. In fact, in any
neighbourhood of a product state lies an arbitrarily strongly entangled state.
Such surprising bevaviour is removed if one introduces the physical constraint
of states of bounded energy as was shown in Ref. [19].
One important property of the entropy of entanglement for discrete systems
is its invariance under local unitary transformation. This has been shown to be
true for its continuous generalization [9] and other important properties such
as concavity, subadditivity, strong subadditivity and the triangle inequality
also follow for this measure of entanglement as they do its finite partner [20]
provided the relevant quantities converge when we deal with infinite systems.
The measure could in principle be generalized to a mixed state measure
analogous to the entanglement of formation [7, 21] or the relative entropy of
entanglement [22], however, in this section we focus on the pure state case.

4. THE CALCULATION OF ENTANGLEMENT

We will now move on to calculating the entanglement in the two types
of entangled states we have formed. We will find that in general we cannot
calculate the entanglement easily and will need to employ a number of different
methods, both analytic and numeric.

4.1 SOME MATHEMATICAL PRELIMINARIES

Let us first look at our first class of states, the partially correlated entangled
states, IB a j3(XI, X2))12. We have two partial traces we could take but for pure
states both partial traces will have the same set of eigenvalues. We will consider
 Entanglement purification via entanglement swapping 199

only one of these, tracing out the second particle. The integral eigenvalue
equation corresponding to this partial trace (aside from normalization) is

! [1(
exp a 2 -
(2
a 1)
+ 2(32 (x 2 + X ,2 )

+2ixt(x - x') + ;') ] ¢i(X') dx' = )..i¢i(X). (16.11)

It can then be shown [9] that this eigenvalue equation is equivalent to (in that it
has the same set of eigenvalues, aside from normalization) the following one

! ~Xp [- (1 + P) (x 2 + x,2) + 2XX'l

v
¢(x') dx'

K(x,x')
)..¢(x) (16.12)

where K(x, x') is the kernel of our integral eigenvalue equation representing
the elements of the density operator and

(16.13)

Notice that the eigenvalues are independent of the scale length, a, and the
value of Xl or X2. It is only dependent on the product of widths of the original
Gaussian distributions with respect to a.
Similarly we calculate the partial trace of the two-mode cat state and find
that it's corresponding integral eigenvalue equation is

¢(X') dx'

v
K(x,x')
)..¢(x).(16.14)

This, however, is not easily transformed into a simpler form.

4.2 NUMERICAL PROCEDURE FOR THE

PARTIALLY CORRELATED STATES
We cannot solve many of the integral eigenvalue equations we encounter so
we must use some numerical approximation [23]. The most direct approach
is to solve a discrete eigenvalue equation by approximating the integral by the
rectangle rule. Our eigenvalue equation has infinite limits so there must also be
a cut-off point in the limits beyond which we do not approximate the integral.
200 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

First, therefore, we discretize the eigenvalue equation (16.11) into 2n + 1

parts (i = -n ... ,0, ... n) each of width 8 covering the range -w S x S w
where w = n8. Our eigenvalue equation then becomes
n
8 L Ppq<pp = A<Pq, (16.15)
p=-n

where for the partially correlated states

Ppq = exp [(-(1 + 2a.2(32)(p2 + q2) + 2pq) 82 ] . (16.16)
The entropy of entanglement is then approximated by

E(a.,(3) = ~ (E~TAS l092 (E~TAJ) , (16.17)

where the outer sum is over the set of eigenvalues and the sums over s and t
are to normalize the set of eigenvalues as the Ppq is not normalized.
What we are doing here is sampling the spectra of eigenvalues over discrete
ranges. If we were to take 8 -+ 0 and n -+ 00 we should converge to the exact
value of the entropy of entanglement as in Eq. (16.7). In fact, for these states
there is an analytical result for the entropy of entanglement [3] which we will
discuss in the next section. In practice for most values of a. and (3, 2n + 1 = 201
and w around 10 standard deviations from the mean were sufficient to produce
results in agreement with the analytic result accurate to 6 significant figures.
The numerical results for the entanglement are shown in figure 16.2 for
varying values of a. and (3. They were generated using numerical procedures
for eigenvalue problems from the NAG library.
We expect the entanglement to increase when the parameters a. and (3 are
reduced as this corresponds to a reduction in the spread or uncertainty in
the wavefunction of the two particles before the entangling operations were
performed. As these parameters approach zero the states become like the
entangled states of Vaidman [15]. These are simultaneous eigenstates of the
operators Xl + X2 and ih - 'P2 and are therefore maximally correlated. The
entanglement is then infinite and convergence in the numerical procedure is
difficult to achieve for these small values of a. and (3.

4.3 ANALYTICAL RESULTS FOR THE PARTIALLY

CORRELATED STATES
Here we use a result for a two mode squeezed state [3] with squeezing
parameter r [24]:

V;(x, y) = exp ( -l (e 2T (x + y)2 + e- 2T (x _ y)2)) . (16.18)

 Entanglement purification via entanglement swapping 201

5,------,-------,------,------,-------,------,

4.5

3.5

3
E
Q)
E
Q)
c;, 2.5

~ 2

1.5

Figure 16.2 Entanglement of a partially correlated state in terms of a and (3, the widths of the
Gaussians from which they are formed.

After taking the partial trace of this state the integral eigenvalue equation can
also be transfonned into the fonn of Eq. (16.11) with a parameter [9]

P = 2cosech 2 (2r). (16.19)

This tells us a number of things, firstly that our partially correlated states are
just a generalization of two-mode squeezed states and secondly, as the state can
be written analytically in the number basis [3]

1 00

l1fJh2 = cos h()

r I)tanh(r)t In)1In)2
(16.20)
n=O

we can calculate the entanglement exactly:

(16.21)

This in tum gives us an analytical result for our partially correlated states via
the substitution 2r = arcsinh(1/ a,B) into Eq. (16.21). This follows directly
from Eqs. (16.19) and (16.13).
202 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

4.4 NUMERICAL PROCEDURE FOR THE

TWO-MODE CAT STATE
Now we move on to the entanglement of the cat states. We must proceed
directly by discretizing the density operator of equation (16.13) and making it
the kernel of equation (16.15). Results for the entanglement of the cat states
are shown in figure 16.3 with varying parameters d and IAoI2 .

. ~ ..
'

1.5

o 0
d

Figure 16.3 Entanglement of the cat states against (half) the distance between Gaussians, d,
and the coefficient Ao. The entanglement is greatest for high values of d where the Gaussians
become orthogonal, and for IAol2 = 0.5 as with discrete entanglement.

Note that the entanglement for any given value of d is maximum when
lAo 12= 1/2 and that the entanglement increases with d for given values of Ao,
approaching the limit E = -lAo 1210g2 lAo 12 - IAll210g21All2 as d ~ 00,
where the separated Gaussians become orthogonal. Note also that only two
eigenvalues dominated the contribution to the entropy. These observations,
therefore, tell us that these states behave very much like discrete 2-level entan-
gled systems.
 Entanglement purification via entanglement swapping 203

5. ENTANGLEMENT PURIFICATION VIA

ENTANGLEMENT SWAPPING
Let us first briefly describe the process of entanglement swapping for pure
states of two levels [25]. Given two pairs of entangled states a Bell state
measurement on one of the particles from each of the pairs can leave the
remaining two particles entangled with each other. How entangled they are
depends on the measurement result. Let us assume that the initial two pairs are
in the same pure state:

(16.22)

and 11/1) 34. Then a Bell state measurement on particles 2 and 3 will, for certain
measurement results, leave particles 1 and 4 in states

(16.23)

with probability (a 4 + (34)/2, which is less entangled (provided the two pairs
are not already maximally entangled (a = (3 = 1/v'2», and states

(16.24)

with probability a 2(32, which is maximally entangled. Therefore the proba-

bility of obtaining a maximally entangled state increases with the amount of
entanglement in the initial states. Note also that the average entanglement
obtained at the end cannot be more than the initial amount of entanglement in
one of the pairs.
We can then see that this procedure can be used (by Alice and Bob) for
purification. Alice and Bob share an entangled state and Bob holds another
copy of the same entangled state as in the upper part of figure 16.4. Bob
performs entanglement swapping by making a Bell state projection on one
particle from each pair of entangled particles. With a certain probability, this
will produce a maximally entangled state.
Generalization of this procedure to continuous variable states is also straight
forward (see also [26]). However, we will replace the Bell state measurement
with the reverse entangling operation t;t followed by separate measurements on
each of the two particles. This is entirely equivalent to a Bell state measurement.
204 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

• ~
....
Bell state projection •
_------------I initial
•
Alice Bob

• • final

• •
Figure 16.4 The entanglement swapping procedure. Bob, holding a copy of the entangled
state shared by himself and Alice, performs a Bell state measurement on a particle from each
pair and, for certain measurement outcomes, the entanglement of the final shared pair is higher
than that of the initial shared pair.

5.1 PARTIALLY CORRELATED STATES

So we attempt the entanglement swapping procedure with the partially cor-
related states, starting with infinite resolution projective measurements:

2 (a14 (bl (EJ4 (IBaa(O, c)h 2 IB /1/1(O, C))34))

J exp [CT 2(a/+ ,82) (-x 2g (a,,8) - y2g(,8, a) + 2xy

+2b(y - x) - 2ia(ya 2 + x,82)) ] Ixh IY)3 (16.25)

where
(16.26)
Has the entanglement increased? Again we can take the partial trace of this
state and transform the kernel into the form of Eq. (16.11). This gives us a
parameter Pswap = 2[(a 4 + a 2,82 + 1)(,84 + ,82a2 + 1) - 1] where before
the swapping process the parameter was Po = 2a 4 or 2,84. Note firstly that
P swap does not depend on the measurement results a and b and is therefore not
probabilistic so any increase in entanglement would be deterministic thereby
breaking laws of the conservation of entanglement. However Pswap ~ Po and
the entanglement is strictly decreasing with increase in P so the swapped pair
has less entanglement and we have not achieved purification.
 Entanglement purification via entanglement swapping 205

Of course, our final projections in this method were onto the unphysical states
la) and Ib) but further calculations indicate that with finite width projections
(performed by projecting onto the Gaussian states of Eq. (16.4» the parameter
P still increases. Such calculations involve 6th degree polynomials in the
width parameters (O! and (3 etc.) so proving that P increases for all values of
these parameters is difficult and we have not been able to do so analytically.
However, numerical results indicate that this is true.

5.2 TWO-MODE CAT STATES

We now attempt a similar method with the two-mode cat states, but setting
the scale length (J' = 1 and making finite resolution measurements of width 1-':

11/J)14 = 2 (G/L(a)13 (G/L(b)1 (£J3 (IC(d)h2I C (d))34))

= f ,2: 1
A j A k e-(x-(-1)jd)2_(Y+(-1)k d)2
J,k=O
x e( dbh(tt) ( (-l)j +(1+1.12)( _l)k )+2d20jk)
x e(iadh(tt)((1+tt 2)(-1)L(-1)k)) IX)1IY)4 dxdy (16.27)
where

(16.28)

Writing this state out in full in the high precision measurement limit, I-' = 0

e-2d2+2db e-(x-d)2_(y+d)2
11/J}14 = / (AoAo
e-(x-d)2_(y-d)2
+AoA1 e 2iad
e- 2iad e-(x+d)2_(y+d)2
+AIAo
e-2d2-2db e -(X+d)2_(y-d)2)
+A1A1
IX)l IY)4 dxdy (16.29)
and looking at the particular case where the probabilistic measured values are
a = b = 0 we can see purification for high values of d as the middle two
terms now dominate and have coefficients of equal magnitude. As d -+ 00
they become maximally entangled. This is again very much like the action
of discrete entanglement under purification procedures: the coefficients of the
states have changed, not the states themselves.
For the results of figure 16.5 and 16.6 we have chosen the values Ao = y'Q.3
and d = 1.0. They show the entanglement of the resulting state (16.26) for a
range of values of a and b with I-' = 0 and 0.5 respectively.
206 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

o.g

0.7
.,c
E
~0.6
c
:l!
oll 05

0.4

0.3

0.2
-0.5

-0.5
0.5 -1

b
a/x

Figure 16.5 Entanglement of swapped cat states with J-t = O. Above the level of the plane
purification has been achieved.

The horizontal planes are at the level of entanglement of either cat state before
the purification procedure is performed, that is, above this plane purification
has been achieved.

6. SUMMARY
We should now address why it is that we have observed purification via
entanglement swapping in only one of the sets of states that we have considered.
As mentioned above the two-mode cat states have many characteristics like
those of discrete two-level entanglement and so it is not surprising that the
method of entanglement swapping generalized from these systems is successful.
The failure of the procedure for the partially correlated states is more difficult
to explain. These states are the kind of states used in continuous variable
teleportation experiments [1, 2, 3, 4, 5] and a simple method of purification
would be ideal. However, we were unable to find any simple continuous
procedure that would produce purification, indeed no procedure was found
where the final entanglement was in anyway probabilistic in the measurement
outcomes, an essential ingredient if a successful procedure is not to violate
conservation laws of entanglement. In fact it has been brought to our attention
 Entanglement purification via entanglement swapping 207

0'9h~~~
o,e

0.7

c
~ 0.6
0.
§c: 0.5
w

;,. :: . ......... .

0.5
-0.5
o
0,5 _1

b
a/x

Figure 16.6 Entanglement of swapped cat states with p, = 0.5. Again purification has been
achieved above the level of the plane, but with the inaccuracy in the measurement part of the
entanglement swapping process the amount of purification is reduced.

[27] that such methods could never increase the amount of entanglement for
Gaussian states. When completing this chapter it was found rigorously, that
Gaussian operations never allow entanglement distillation of Gaussian states,
see Ref. [28].
Fortunately, more recent work has now achieved entanglement purification
in Gaussian states [10] using non-demolition measurements with a number
of entangled pairs. However, what exactly the key difference is between the
two types of states presented here which allows purification by our methods in
one class but not the other is still unclear. There are obvious correspondences
between the form of entanglement in the two mode cat states and discrete
systems and it would be interesting to find a condition for continuous variable
purification, as it has been attempted here, which a state undergoing purification
must obey. The fact that purification has been demonstrated here and elsewhere
in continuous systems, however, are interesting results.
208 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Acknowledgments
This work is supported by the United Kingdom Engineering and Physical
Sciences Research Council (EPSRC), the Inlaks Foundation, The Leverhulme
Trust, the EU TMR-networks ERB 4061PL95-1412 and ERB FMRXCT96-
0066, the EU project EQUIP and the European Science Foundation programme
on quantum information theory.

References
[1] S. L. Braunstein, H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998).
[2] G. J. Milburn, S. L. Braunstein, Phys. Rev. A 60, 937 (1999).
[3] SJ. van Enk, Phys. Rev. A 60,5095 (1999).
[4] T. C. Ralph, P. K. Lam, Phys. Rev. Lett. 81, 5668 (1998).
[5] A. Furusawa, J. L. Sj2Irensen, S. L. Braunstein, C. A. Fuchs, H. J. Kimble,
E. S. Polzik, Science, 282, 706 (1998).
[6] C. H. Bennett, H. J. Herbert, S. Popescu, B. Schumacher, Phys. Rev. A 53,
2046 (1996); S. Popescu, D. Rohrlich, Phys. Rev. A 56, R3319 (1997).
[7] C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, W. K. Wootters, Phys.
Rev. A 54,3824 (1996).
[8] M. B. Plenio and V. Vedral, Contemp.Phys. 39,431 (1998)
[9] S. Parker, S. Bose, M. B. Plenio, Phys. Rev. A 61,032305 (2000).
[10] Lu-Ming Duan, G. Giedke, J.I. Cirac, P. Zoller, Phy~. Rev. A 62, 032304
(2000); Lu-Ming Duan, G. Giedke, J.I. Cirac, P. Zoller, Physical imple-
mentation for entanglement purification of Gaussian continuous variable
quantum states, LANL eprint: quant-ph 0003116.; Geza Giedke, Lu-Ming
Duan, J. Ignacio Cirac, Peter Zoller, All inseparable two-mode Gaussian
continuous variable states are distillable Lanl e-print quant-ph/0007061
[11] M. Lewenstein and P. Horodecki, lanl e-print quant-ph/OOO1035
[12] H. -K. Lo, S. Popescu, Concentrating entanglement by local actions -
beyond mean values, LANL eprint: quant-ph 9707038; M. A. Nielsen,
Phys. Rev. Lett. 83, 436 (1999); G. Vidal, Phys. Rev. Lett. 83, 1046
(1999);D. Jonathan, M. B. Plenio, Phys. Rev. Lett. 83, 3566 (1999); L.
Hardy, Phys. Rev. A 60, 1912 (1999).
[13] M. Zukowski, A. Zeilinger, M. A. Home, A. K. Ekert, Phys. Rev. Lett.
71,4287 (1993); J -W. Pan, D. Bouwmeester, H. Weinfurter, A. Zeilinger,
Phys. Rev. Lett. 80, 3891 (1998); S. Bose, V. Vedral, P. L. Knight, Phys.
Rev. A 57,822 (1998).
[14] S. Braunstein, Phys. Rev. Lett. 80,4084 (1998).
[15] L. Vaidman, Phys, Rev. A 49, 1473 (1994).
 Entanglement purification via entanglement swapping 209

[16] B. C. Sanders, Phys. Rev. A 45,6811 (1992).

[17] C. C. Gerry, Phys. Rev. ASS, 2478 (1997).
[18] F. G. Tricomi, Integral Equations, Wiley, New York (1967).
[19] J. Eisert, C. Simon and M. B. Plenio, J. Phys. A 35,3911 (2002).
[20] A. Wehr!, Rev. Mod. Phys. 50, 221 (1978).
[21] S. Hill, W. K. Wootters, Phys. Rev. Lett. 78, 5022 (1997); W. K. Wootters,
Phys. Rev. Lett. 80, 2245 (1998).
[22] V. Vedral, M. B. Plenio, Phys. Rev. A 57, 1619 (1998); V. Vedra!, M. B.
Plenio, K. Jacobs, P. L. Knight, Phys. Rev. A 56,4452 (1997); V. Vedra!,
M. B. Plenio, M. A. Rippin, P. L. Knight, Phys. Rev. Lett. 78, 2275 (1997);
M. Murao, M. B. Plenio, S. Popescu, V. Vedral, P. L. Knight, Phys. Rev.
A 57,4075 (1998); V. Vedral, M. B. Plenio, Phys.Rev. A 57, 1619 (1998).
[23] D.A. Koike, A. J. Post, J. Chern. Phys. 98, 4853 (1993).
[24] U. Leonhardt, Measuring the Quantum State of Light, Cambridge Uni-
versity Press
[25] S. Bose, V. Vedral, P. L. Knight, Phys. Rev. A 60, 194 (1999).
[26] R. E. S. Polkinghome, T. C. Ralph, Phys. Rev. Lett. 83,2095 (1999); P.
van Loock, S. Braunstein, Phys. Rev. A 61, 010302(R) (2000)
[27] T. Opatmy, G. Kurizki, D.-G. Welsch, Phys. Rev. A 61, 032302 (2000).
[28] J. Eisert, S. Scheel and M. B. Plenio, quant-phl0204052; and G. Giedke
and J.1. Cirac, quant-phl0204085.
Chapter 17

BOUND ENTANGLEMENT FOR CONTINUOUS

VARIABLES IS A RARE PHENOMENON

Pawel Horodeckil, J. Ignacio Cirac 2 and Maciej Lewenstein 3

1 Faculty of Applied Physics and Mathematics
Technical University of Gdmisk, 80-952 Gdmlsk. Poland
2 Institutjiir Theoretische Physik. Universitiit Innsbruck, A-6020 Innsbruck, Austria
3 Institutjiir Theoretische Physik, Universitiit Hannover
30167 Hannover, Germany

Abstract We discuss the notion of bound entanglement (BE) for continuous variables (CV).
We show that the set of non-distillable states (NDS) for CV is nowhere dense
in the set of all states, i.e., the states of infinite-dimensional bipartite systems
are generically distillable. This automatically implies that the sets of separable
states, entangled states with positive partial transpose, and bound entangled states
are also nowhere dense in the set of all states. All these properties significantly
distinguish quantum CV systems from the spin like ones. The aspects of the
definition of BE for CV is also analysed, especially in context of Schmidt numbers
theory. In particular the main result is generalised by means of arbitrary Schmidt
number and single copy regime.

1. INTRODUCTION
Bound entanglement [1] is the entanglement which cannot be distilled (pu-
rified), i.e. no pure state entanglement can be obtained from it by means of
local operations and classical communication (LOCC)[2]. So far, it has been
studied mainly for spin like systems. These studies has allowed to discover
many interesting properties of bound entanglement, for both bipartite[3], and
multiparticle systems[4]. Recently, much attention has been devoted contin-
uous variable (CV) systems (c.f. [5]). Bound entanglement has also been
considered for continuous variables (CV), and the first nontrivial examples of
BES for CV have been constructed[6] (see also [7]). Once we have some ex-
amples of BES for CV, it is interesting to ask how frequent is the phenomenon
211
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 211-228.
© 2003 Kluwer Academic Publishers.
212 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

of bound entanglement, i. e. how many states of that kind are in set of all CV
states?
The question of "how many quantum states having some interesting property
are there?" is very natural. In the context of entanglement it was first considered
in Ref. [8], where the problem of the volume of the subset of separable (non-
entangled) states in the set of all bipartite states of spin systems was considered.
Numerical evidence has shown that the volume of the set of separable states
approaches zero when the size of the spin goes to infinity. It was also shown
that for any finite spin system the volume of separable states is nonzero due to
the existence of a separable neighborhood, i.e. an open ball of separable states
in the vicinity of the maximally mixed state in arbitrary dimension. Further first
analystical bounds on the size of neighbourhood have been provided [9]. All
this raised a series of questions concerning the interpretation of experiments
of quantum computing based on high temperature NMR; many interesting
analyses have been performed in this context [to, 11].
The question of the "size" of the set representing separable states has been
recently answered [12] for CV; it has been show that for bipartite states this
subset is nowhere dense (relative to the trace-norm topology). This implies that
this set does not contain any open ball and also that CV states are generically
non-separable. On the other hand there exists another subset that is of interest
in the context of entanglement. This is the subset of of non-distillable states
(NDS), i.e. states that cannot be distilled. This subset contains the separable
states, and therefore it might well be that an appreciable fraction of all states
are in such a subset. In this paper we show that this is not the case; that is, the
subset of NDS is nowhere dense in the set of bipartite states of Cv. We present
two different proofs of this fact. One uses the uniform topology, and the other
one the trace-norm topology.
We also perform analsysis how one can relax conditions of NDS in context
of CV in comparison with the standard definition, and prove stronger version
of the main result with help of Schmidt numbers theory [13] and single copy
regime (see [15, 14]).
There are several results which follow from our proofs. In particular, since
the subset of NDS contains the subset of BES, we have that that subset is also
nowhere dense. The same thing occurs with the subset of states with positive
partial transpose (PPT) [1] and therefore with with those PPT states that are
entangled. Moreover, since the subset of separable states is also contained in
the one of NDS, our results include the ones given in Ref. [12].
 Bound entanglement/or continuous variables is a rare phenomenon 213

2. NON-DISTILLABLE STATES FOR

CONTINUOUS VARIABLES
The main subject of this paper is the question about whether generic CV
states are non-distillable. We present below the answer to this question: the
subset of NDS is nowhere dense. In this section we first discuss the definition
of NDS. Then, we present two proofs of our result. First, it will be proven using
the uniform topology and exploiting the fact that any density operator can be
considered as the limit of a sequence of density operators defined on a finite
support. Then, following the approach of Ref. [12] we shall prove the general
statement that any proper closed subset of bipartite states which is invariant
under local transformations is nowhere dense (in the trace norm topology).
This generalizes the results of Ref. [12] and, as we shall see, together with the
fact that the set of NDS is closed, proves our claims.

2.1 FREE AND BOUND ENTANGLED STATES

Let us denote by M set of all density operators acting on H A ® H B, where
HA ~ HB ~ L2(R). Let us consider a density operator p E M. The notion
of distillability of entanglement introduced in Ref. [2] has been operationally
characterised. Namely, according to Ref. [1], p is distillable (free entangled)
iff there exists some finite n E N, two rank two projectors PA, P B acting on
H A, H B, respectively, such that

(17.1)

otherwise, p is non-distillable. Entangled but non-distillable states are called

bound entangled states [1]. Here we call D and N the set of all distillable and
non-distillable density operators, respectively. Physically, this definition tells
us that a state is distillable iff out of a sufficiently large number of copies we
can obtain by local operations two qubits which are entangled. The reason for
that is clear. First, given the fact that one can distill maximally entangled states
out of all entangled states of qubits, this means that if the above condition is
true, we can always distill maximally entangled qubit states out of the original
state p. Second, if the above condition is not fulfilled for any n, then we will
not be able to produce (asymptotically) any qubit maximally entangled state by
using local operations alone.
In the following we will reexpress Eq. (17.1) as follows:

(17.2)

for some I\]!") E HA ® H B . Without loosing generality we can take I'll) =

P A ® p~B I'll), i.e. I'll) belongs to the 2 ® 2 subspace determined by PA, p~B .
214 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

It is easy to see then that

(lw)(wl)TB = L Akl¢k)(¢kl, (17.3)

k=l

where -1/2 :S Ak :S 1, i.e. IAk I :S 1 and the I¢k) are also in the same 2 x 2
subspace.

2.2 NON-DISTILLABLE STATES FOR CV

ARE NOWHERE DENSE: PROOF I
In this subsection we show that the set N of NDS states is nowhere dense in
the set of all states M. We will first recall some definitions. A subset A of a
topological space X is nowhere dense if its closure contains no open set. Note
that if A is already closed then it is nowhere dense iff it contains no open set.
For example, the subset of integer numbers Z in R (with the topology induced
by the absolute value metric) is nowhere dense since it is already close and no
neighborhood of any integer contains only integers.
In this subsection we will use the uniform topology for operators, which is
the one derived from the operator norm. First we will show that N is closed
with that topology by proving that its complement, D, is open. Then, we will
show that D is dense in the set of all density operators. From this last point it
follows that N (equivalently, its closure) contains no open set and therefore it
is nowhere dense.
In order to show that D is open, let us consider some p ED. According
to the definition of D, there exists some finite integer n, and a Schmidt rank
two state Iw) E HA ® HE such that Eq. (17.2) is fulfilled with (17.3). Let
us consider an open ball B,., (p) = {p', liP' - pil < 1]}. We will show that for
1] < IEI/4n, B17(P) cD.
To this aim we argue that

I (wi ((PA ® PB) (p@n - (p')@n)(PA ® pB)fB IW) I

4
= L Ak(¢klp@n - (p')@nl¢k)
k=l
:S 411p@n - (p')@nll :S 41]n. (17.4)

The latter inequality can be proven by induction, using the identity

(17.5)
 Bound entanglement/or continuous variables is a rare phenomenon 215

and the fact that both Ilpll and IIp'll are smaller than one. Thus, if", < lEI!4n,
we see that for any p' E B'T/ (p),

(WI((PA ® PB)(p,)®n(PA ® PB))TB I'll) < 0, (17.6)

ergo p' is distillable.

Now we show that D is dense in M. To this aim we observe that for any
p E M we can always find a sequence {Pn} ~=o' with Pn E D such that Pn ~ p.
We consider the spectral decomposition of p as
00

p = LPnlwn)(wnl, (17.7)
n=1

where we have chosen PI 2: P2, .... Note that since p is a trace class operator,
the sequence Pn converges monotonically to zero. On the other hand, we can
write the Schmidt decomposition of each Iw n ) as
00

Iwn) = L VAn,klun,k, Vn,k), (17.8)

k=1

where again we have chosen An,k 2: An,k+l 2: 0 and An,k converges monoton-
ically to zero as k -+ 00. Now, we define
N
PN == LPnIWN,n)(WN,nl, (17.9)
n=1

where
N
IWN,n) = L VAn,klun,k,Vn,k). (17.10)
k=1

It is clear that PN is supported on Hf

® Hff, where both H!1 have fi-
nite dimension. Thus, we can always find two pairs of orthogonal vec-
tors laf,2) E HA e H!1 and Ibf,2) E HB e Hff. Let us define Iq,N) =
(lal,b1) + la2,b2))/v'2 and
(17.11)

where KN is a normalization constant. It is clear that PN N::4oo p. On the

other hand, defining

lal)(all + la2)(a21, (17.12a)

Ibl)(bll + Ib2)(b21 (17.12b)
216 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

and taking n = 1 it is clear that PNED. This completes the proof. 0

Obviously the fact that the set of NDS states for CV is nowhere dense,
implies that the contained in it sets of BES and PPT states are also nowhere
dense. One can, however, prove the latter directly using the method used above.
The only difference would be that the state 1\lI) which in above proof belongs
to a 2 ® 2 subspace has to be substituted by a general vector 1\lI) of arbitrary
Schmidt rank, but at the same time there is no need to consider n-fold tensor
products, since the PPT property of p is maintained for arbitrary number of
its copies. One has to use, however, the property of the partially transposed
projector 11(I\lI)(\lIDTB II ~ l.

2.3 NON-DISTILLABLE STATES FOR CV

ARE NOWHERE DENSE: PROOF II
In order to demonstrate the statement of the previous section in another way,
we first need to recall notions of several necessary tools, which have been used
in Ref. [12]. Let B(1lI) ® B(1l2) stand for bounded operators on Hilbert space
1l = 111 ® 112 describing our bipartite system. We assume that at least one of
the subsystems is described by CV, and hence it has the infinite dimension.
Now, consider a third auxiliary system described by 1l3. It is convenient
to describe all states in 1l = 111 ® 112 as reduced states of some pure states
in the extended space 1l ® 1l3. If we have a pure state IVI23)(vI231 in the
extended space, then the reduced state Tr3(lvI23)(vI231) is denoted by l?12. Let
us denote by 7 the set of all states on B(1ld ® B(1l2). This is a set of unit
trace operators with nonnegative spectrum. We shall also endow this set with
the norm topology II . liT, IIAIIT == Tr(v' AtA). Now, one defines [12] the
map <I> from the unit sphere S representing all wavefunctions form 1l ® 113 to
the set of states 7 in the following way :

(17.13)

The map <I> : S --+ 7 is continuous (in the norm II . liT) and onto. In particular
it maps dense subsets onto dense subsets (see [12] for explanation).
Consider the set X of all vectors ul23 = A®I ®I(VI23) for all A E B(1lI).
The vector Vl23 is called i-cyclic (see [12]) if the closure of X in the norm II·IIT
turns out to be the whole space 111 ® 112 ® 1l3. The physical interpretation of
I-cyclic vectors in both finite dimensional, as well as in the CV case, is that
those are the vectors which have maximal possible Schmidt rank. Note, that
according to Lemma 2 of Ref. [25] they form a dense set in 111 ® 112 ® 1l3.
Now we consider the following simple
Observation 1.- Let the set NV be (i) a proper closed (in II . liT norm)
subset of the set of states 7 which is (ii) invariant under the operations A ® I.
 Bound entanglement/or continuous variables is a rare phenomenon 217

Then any vector V123 satisfying ~(V123) E Nv cannot be i-cyclic and Nv is

nowhere dense in T.
The above observation is a natural generalization of the Lemma 1 of Ref.
[12]. To show this, consider such vector v that its "reduction" ~(v) belongs to
NV, and take any vector

v' = A ® I ® l( v), (17.14)

defined for arbitrary A, such that Ilv'll = 1. We shall show first that ~(v') also
belongs to NV. Indeed (see [12]) we have ~(v') = A ® 1~(v)At ® I and
(because the norm of v' is one) the trace ~(v') is one. But, because the set NV
is closed under the operation A ® 1(·)At ® I, we see that ~(v') still belongs
to the set.
Now suppose that v were I-cyclic. Then, that the set M of all vectors v'
would be dense in the unit sphere S of all normalized vectors belonging to
1i ® 1i3. As the map ~ is continuous and onto, it certainly would map M onto
some new set denoted by ~(M), which would be dense in set of all bipartite
states T. Thus closure of ~(M) must have give all T. But, on the other hand
any element of ~ (M) (which is defined as ~ (v') for some vector v' of the form
(17.14» belongs to NV. As the latter is closed, the closure of ~(M) would
have to be a subset of NV. But Nv was supposed to be closed and strictly
smaller than the set T, so the closure of ~(M) cannot be equal to T. This
gives the required contradiction. The above reasoning follows the lines of the
proof of Ref. [12]. The only difference is that instead of the specific set of
separable states considered there, here we have considered an abstract set NV,
which has some special properties. Note, that the assumptions of Observation
1 and the fact that the I-cyclic vectors form a dense set in S imply that the
closed set NV is nowhere dense. If it had contained a open set, then, following
continuity of ~ this open set would have had to be an image of open subset
of 1i 1 ® 1i2 ® 1i3, which would have had to contain a ball, an thus a I-cyclic
vector. Now, to show that the set of NDS states is nowhere dense we have to
show that it is (i) invariant under local operations of the type A ® I, (ii) closed
in the trace norm II . liT. The first property (i) is immediate, since a NDS
cannot be converted into a free entangled state by means of local operations.
The second one is not so obvious for continuous variables, but it follows from
the results of the previous subsection. We thus have:
Observation 2.- The property of non-distillability is invariant under the one
side local action A ® IOAt ® I.
The proof is simple - the arguments of Ref. [1] can be applied (see also
[6]) to show that any local separable superoperator cannot cause that the state
looses the non-distillability property.
Observation 3. - The set of all NDS is closed in the norm II . liT.
218 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

To prove the closeness of the set of NDS, we prove that its complement, i.e.
the set of distillable states D is open in the trace norm. To this aim we repeat
the arguments of subsection A and consider some p E D, for which there exists
some finite integer number n, PA, PB, rank two projectors acting on H A, H B,
and a rank two vector Iw) E HA ® HB such that Eq. (17.2) is fulfilled. We
consider now an open ball in the trace norm, i.e. i3.,., (p) = {p', liP' - piiT < 'f/}.
Note that if p' E i3.,., (p) then the operatornorm fulfills lip' - pil :::; lip' - piiT < 'f/
[16]. Using the same argument as before we show that for 'f/ < IEI/4n,
i3.,., (p) CD, which completes the proof. 0
Combining the Observations 1.-3. we see that the set of NDS states is
nowhere dense in the trace norm, which implies the same property for the BES,
PPT states, and separable states.

3. BOUND ENTANGLEMENT: AN ANALYSIS

As mentioned in the introduction, non-trivial BES for continuous variables
(CV) have been discovered. In this subsection we discuss some of the details
regarding these states, as well as whether entangled states in CV with infinite
Schmidt number represent are generic in the set of entangled states.

3.1 CONTINUOUS VARIABLE

BOUND ENTANGLED STATES
The construction of non-trivial BES for CV systems [6] was based on an
idea similar to the one used for spin systems, for which it has been proven that
any entangled state with positive partial transpose[17] cannot be distilled[1].
The crucial element of the construction was to create the state in such a way
that it cannot be obtained simply by embedding a bound entangled state in a
finite dimensional Hilbert space into the Hilbert space of CV.
The particular example (! of CV BES proposed by us was first of all assumed
to satisfy the condition that its partial transpose (!TB , defined as
TB
(!m/L,nv -= ( m, /L I(!TBI n, v ) -- (!mv,n/L' (17.15)
has a nonnegative spectrum. Such requirement was, however, not sufficient, as
one could invent the following "trivial" example of a PPT entangled states for
CV[6]

(17.16)

The above state is build from infinitely many "copies" of the same 3 ® 3 1 BES
CJ labeled by CJ n . Each of CJ n has the matrix elements of the original CJ, but in
the basis Sn = {li,j)}7,j~~n' Here {Pdi=l is an infinite sequence of nonzero
probabilities, L~l Pi = 1. The bound entanglement of the CV state a is in a
 Bound entanglement/or continuous variables is a rare phenomenon 219

certain sense spurious, as it can in principle be reversibly converted by means

of local operations and classical communication into the 3 ® 3 entanglement.
One could easily construct another example, similar to the one above, with
(5n acting in k n ® k n Hilbert space with k n -+ 00, and (j- being block diagonal as

in (17.16). This example is much more interesting as far as CV are concerned,

because it can not be reversibly converted into any state of fixed spin. Thus to
some extend it might be regarded as generic BE. However, such a state would
still be a mixture of "locally orthogonal" spin states, which does not exploit
fully the CV Hilbert space structure, i.e. infinite dimension fully. In fact,
if such CV BES were produced by a random mixture, they could be easily
"decoupled " by local projective measurements and classical communication.
Thus we propose to define generic BES for CV in a stronger way, namely as
the states from which no pure entanglement can be distilled, and they cannot
be represented by the states of the above sort.
This is a somewhat phenomenological definition, but it implies to single out
some required properties of the generic CV BES. The first nontrivial examples
of the generic CV BES, presented in Ref. [6], fulfill those requirements. These
states have the form:
00 00

(! ex Iw)(wl + L L Iwmn)(wmnl, (17.17)

n=lm>n
with the following definitions of the symbols: Iw) = E~=l anln, n), Iw) E
11. = l2(C) ® l2(C) with the finite norm IIwl12 = E~=1IanI2 = q < 00, and
vectors

(17.18)

for n < m with (in general) complex an and en, such that (i) 0 < ICn+l1 <
Icnl < 1, (ii) E~=l E:>n IIwmnl1 2 is finite. The latter condition can be
achieved for example by setting an = an, en = cn, for some 0 < a < c < 1,
see [6]. Physically, the vector Iw), when normalized, may describe a state
of two modes of the quantized electromagnetic field, or more generally two
harmonic oscillators. The state (17.17) has the following properties: (i) it is
bound entangled, as it has the PPT property (i. e. it has the positive partial
transpose); (ii) it is not a simple "direct sum" of finite spin BES in a sense of
the "spurious" examples discussed above (Eq. (17.16».
Recently considerable attention has been devoted to the so called Gaussian
states. In systems of two harmonic oscillator modes (one of Alice, one of Bob),
i.e. in the, so called, Gaussian 1 x 1 case, it has been shown that no bound
entanglement exists - such Gaussian states are either separable [18, 19], or
distillable [20]. In another words, in this case PPT property is a necessary and
sufficient condition for separability, and non-distillability. This result can be
220 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

extended to the case 1 x N. Soon after realizing this facts Werner and Wolf
have found an example of a Gaussian BES with PPT property [7]. This result
has been achieved by considering first covariance matrices of Gaussian states
and their null subspaces. It was noted that the Gaussian state is separable,
iff its covariance matrix can be minorized by some block diagonal covariance
matrix. Second, the characterization of PPT states in terms of covariance
matrix has been found. The BES has been constructed using an elegant explicit
construction, performed using the analysis of the range and the "subtraction
method" first developed for spin systems in Refs. [21, 22, 23, 24]. In the
terminology of Refs. [21,22,23,24] the states found in Ref. [7] are examples
of the so called "edge states". The approach of Ref. [7] can be used further
to analyze multiparticle entanglement. In particular, one can try to "split" the
covariance matrix of n x n state in a way to get m x m x m state with some
bound entanglement properties. Indeed, we have recently managed to solve the
separability problem for the case of tripartite system with one mode per each
party [25]. The result of Werner and Wolf appeared first a little surprising in
the view of Refs. [18, 19, 20]. Recently, some of us have been able to clarify
this and solve ultimately the separability [26] and distillability [27] problems
for Gaussian states of two parties sharing arbitrary number of modes. While
the PPT property remains a valid necessary and sufficient condition for non-
distillability, the separability criterion has a complex form of a nonlinear map
for covariance matrices.
Finally, it is worth mentioning that it is not known yet whether there exist
BES which do not have PPT, even though there is a strong indication of this
fact [28]. If this were finally true, this would have important implications in
the context of distillation [29], since it may well happen that by mixing two
NDS one obtains a distillable one.

3.2 QUESTION OF GENERICITY:

THE STRUCTURAL POINT OF VIEW
There is one open question whether the given CV entanglement represents a
generic entanglement in the sense that it has infinite Schmidt number (see [13]),
i. e. whether it is the limit of matrices whose Schmidt number goes to infinity.
This means that, in principle, in order to generate the state, one would have to be
able to generate the states of arbitrary Schmidt rank. However, some of the CV
BES states similar to "spurious" ones could have also this property - if a finite
dimensional n (8) n PPT states with Schmidt rank of order O(nQ) with some
o < a ~ 1 existed, then we could put in the expression (17.16) the k n (8) k n
states O"n with the rank O(k~) where say kn+l = 2(L:~=1 k n ). Thus, we see
that in order to describe the generic CV entangled states it seems reasonable to
require the stronger version the notion of infinite Schmidt number. Intuitively,
 Bound entanglement/or continuous variables is a rare phenomenon 221

it should mean that the pure states with infinite Schmidt rank are necessarily
involved in the mixed state representation. One possible definition would be
that a generic CV state with infinite Schmidt rank should be necessarily of the
form (! = EiPilwi)(Wil, with IWi) not necessarily orthogonal, but with at
least one IWi) of infinite rank. Such states obviously exist - take for instance
one pure state of infinte Schmidt rank, or a convex combination of two such
states. However, in the above definition the precise notion of the decomposition
in the CV case in the sense of Ref. [30] has to be specified. Another possible
definition (which seems to be significantly weaker) would be to require the
generic CV state to be the limit of n ® n states of Schmidt rank nOt for some
O<a~1.
Concerning BES - we do not know whether there exists any BES for CV
with PPT property, having at the same time the feature of being a generic CV
state, whatever it would mean. It is worth stressing at this point that according
to the results of Ref. [31], PPT entangled state in n ® n space are expected
to have Schmidt number smaller than n. In fact, in the Appendix A, we
present the arguments analogous to those used in Ref. [31] that the typical PPT
bound entangled states in n ® n space either have the Schmidt number of order
0(1), or their partial transpose have this property. It is possible, however,
that the recently introduced Gaussian bound entangled states [7, 26] satisfy
all requirements as far as the CV genericity is concerned. It would thus be
interesting to analyze the Schmidt number of those states.

3.3 QUESTION OF GENERICITY:

DISTILLATION POINT OF VIEW
In former section we have dealed with question of genericity of CV state as far
as the structure of the state is concerned. Bound (nondistillable) entanglement
is directly related to distillation procedures. It is important to address the
question from different point of view i. e. analysing the output of distillation
procedure from the point of view of genericity.
The present result of section II clearly shows that NDS in the sense of
standard definition (that no entanglment can be distilled from given state) is
nowhere dense in set of CV states. This is an important theorem generalising
previous results. However in the classical definition of NDS treats both finite
and infinite dimensional entanglement nondistillable. For sake of many appli-
cations the next step of study which would operationally distinguish those two
quantities would be desirable.
In such approach "fully CV NDS " would be all the states that do not
allow for distillation of infinite pure entanglement (whatever it means). This
signifficantly increases the set of states that are interpreted as bound entangled.
222 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

As we shall see below this leads to more complicated issue. We will not
give definite answers here. However further methods of investigation will be
suggested.
Again, as in previous section, one of proposed definitions could be the
following:
A. The state (! represents ''fully CV free (nondistillable)" entanglement if
and only if it is possible (impossible) to distill nonzero amount of pure states
with infinite Schmidt rank from state {! .
Nonzero amount is here understood in sense of usual distillation yeld (i.
e. as a nonzero amount of pairs). Note that to qualify distilled entanglement
in finite dimensions the condition of asymptotic approaching the maximally
entangled state Iw+) = Jm l:~llei' ei) was required. It is known however
that there is no maximally entangled states of infinite Schmidt rank. Thus in
place of W+ one would have probably use some fixed pure state Woo having the
reduced density matrix non singular or at least of infinite rank (this is eqivalent
to the infinite Schmidt rank of woo).
Another interesting (weaker) defintion would be more in spirit of Ref. [6]
where increasing sequences of finite Schmidt rank were used. Namely one can
propose:
B. The state (! represents CV free (nondistillable) entangled if and only if
it is possible (impossible) to distill nonzero amount 'T/p > 0 of p-Schmidt rank
states with limsupp'T/p > 0 .
The main difficulty dealing with Schmidt rank in those definitions is that the
operational methods of its detection in context of CV are not enough developed.
For example it is not known whether the proposal B above is equivalent to
the following generalisation of the "two-qubit subspace"(see sec. II A):
the state is fully free (bound) entangled iff there is (no) n and the family of
bilocalfilters Ap 0 Bp such that the new n-copy states

(17.19)

violate the p-Schmidt rank test via positive map i. e. [ll 0 Ap] ((!~) is not positive
matrix. The map Ap(X) == Tr(X)J - (p _1)-1 X is p - I-positive but not
p-positive and was used to detect p-Schmidt rank of isotropic entangled states
[13].
Dealing with the state (17.19) is not easy because even in the case of finite
dimensions the possibility of asymptotically singular denominator in formulas
like (17.19) leads to suprising effects (see [15]). Nevertheless, after simplifi-
cation we shall utilise the above point of view. In particular, putting Ap 0 Bp
equal identity we shall generalise the results of section II.
 Bound entanglementfor continuous variables is a rare phenomenon 223

3.4 NDS FOR CV ARE NOWHERE DENSE:

GENERALISATION INVOLVING
SCHMIDT NUMBERS
Here, using the trace nonn topology from sec. II. B we shall prove the
stronger version of the main result of sect. IT. Suppose for moment that as
a "fully CV free entanglement" we shall treat only very special CV states.
Namely only those fonn which it is possible to produce p-Schmidt number
state (with p fixed but arbitrary high) from a single copy by means of special
LOCC protocol given in Appendix B. The protocol is a natural generalisation
of the protocol utilising reduciton criterion [14].
Those special states fonn the set, say D~ (1 stands for "single copy" and p
for Schmidt rank). This set is signifficantly smaller than the one fonned by the
classical definition of free entangled states (see sec. II. A.). If as "fully CV free
states" one treats the set D~ (which still seems to contain too much states, c. f.
discussion of sec. III.B, but this is for dydactic purposes) than one enlarges the
set of what is understood as "fully CV bound" or "fully CV nondistillable".
Below we shall see that though the latter is larger it is still nowhere dense.
To have it we need to prove that D~ is (i) open and (ii) dense in set of all density
operators. Any state g E D~ satisfies by the very definition (see Appendix B)
the inequality:

(wl[IT ® Ap](g)lw) = E < O. (17.20)

Now suppose that g' E B1/(g) = {g', IIg - fl'IIT < 1]}. Then
I(wl[rr ® Ap](g - g')lw)1 = l(wl(gA ® I - gA ® I
+(p _1)-I(g - g')lw)1 :::; ITr[gW(gA - gAJ ® III +
l(wl(P _1)-I(g - g')lw)1 :::; Tr[g~lgA - gAil +
(p -1)-II(wlg - g'lw)1 :::; 1](1 + (P _1)-1) (17.21)

where g~ is reduced density matrix of pure state Iw)(wl.

Where we have used two properties (i) II gA- gA liT:::; 1] because partial trace
is tracepreserving completely positive map which does not increase the trace
nonn IIAIIT; (ii) (¢IAI¢) :::; IIAII :::; IIAIIT for positive A where IIAII stands
for operator nonn as in sect. IT.B. Thus, for 1] < (1+()~1) 1), any g' E B1/(g)
satisfies (wi [IT ® Ap](g')lw) < 0 i. e. ergo belongs to D~. This implies thatthe
latter is open. Now to prove that D~ is dense one can repeat the argument of
w
sec. II.A with I N) E (HA e H f) ® (H B e H ff)
being maximally entangled
pure state of Schmidt rankp: Iw N) = )p L:f=llei, Ii) (the only difference is
that the resulting state should be shown to satisfy (17.20) which is easy to see).
Thus D~ is dense and open so its complement is nowhere dense in set of all
224 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

states which completes the proof. Remarkable that the line of the proof remains
completely correct for "uniform" assumption [i.e., (17.20) satisfied with one
€ for all natural p some l\]f) = 1\]f(P))] but than the assumption itself can be
easily shown to be false in the sense that no state can satisfy it.

4. CONCLUSIONS
In this paper we have considered non-distillable states for continuous vari-
ables. In the main part of the paper we have proven that the subset of non-
distillable states is nowhere dense in the set of all CV states. This is a much
stronger result than the recent one by Clifton and Halvorson [12], which prove
the same result for the set of separable states, since that one is contained in the
set of NDS. Moreover, our results imply that the subsets of BES and PPT states
are also nowhere dense. Thus, generic CV states are distillable. We have also
presented some examples of BES and discussed their genericity from the point
of view of CV and their Schmidt number. In the Appendix A we have presented
an evidence that all PPT BES in n ® n systems either have Schmidt number
smaller that 0 (1), or their partial transposes have this property. Finally we have
analysed the genericity of CV entanglement in context of Schmidt number. In
particular, we have studied the assumptions of the main theorem and proved
more general result that nowhere dense is the set of all states from which it is
impossible to produce p-Schmidt rank state from a single copy in some (well
defined) way. The latter involves single copy protocol (provided in Appendix
B) being a generalisation of that obtained with help of reduction cirterion.
By provinding some proposals of definitions of what can be treated as "fully
CV" we have shown that further investigation of genericity in context of CV is
desirable.
We thank Anna Sanpera, Dagmar BroS Geza Giedke and Otfried Giihne
for useful discussions. This work has been supported by the DFG (SFB 407
and Schwerpunkt "Quanteninformationsverarbeitung"), the Austrian Science
Foundation (SFB "control and measurement of coherent quantum systems"),
the European Union Programme "EQUIP" (IST-1999-11 053) and the Institute
for Quantum Information GmbH. Part of the work was completed at The
Erwin SchrOdinger International Institute for Mathematical Physics, during the
Program "Quantum Measurement and Information", Vienna 2000.

Appendix: Schmidt number of PPT BES for n ® n systems

In this Appendix we essentially repeat the arguments used in the Ref. [31] to
support the conjecture that in 3 ® 3 systems all PPT BES have Schmidt number
2. We consider now the n ® n case, with n large. Let r (p) denotes the rank fo
p; our aim is to present a strong evidence for the following conjecture:
 Bound entanglement/or continuous variables is a rare phenomenon 225

Conjecture.- All PPT entangled states in n ® n systems either have Schmidt

number of the order of 0(1) or their partial transposes have this property.
Note, that this conjecture concerns for instance projections of the PPT BES
(17.17) onto n ® n spaces. We observe that

• It is enough to show the conjecture for the, so called, edge states [22, 23,
24], i.e. the PPT states 8 such that there exist no product vector Ie, 1)
in their range, such that le*, 1) is in the range of the partially transposed
operator 8TA .
• Let r (p) denotes the rank of p. It is likely that it is enough to prove the
conjecture for the edge states of maximal ranks [22], i.e. those whose
ranks fulfill r(8) + r(8 TA ) = 2n 2 - 2n + 1. We expect that such states
are dense in the set of ~l edge states. To show the latter statement, we
consider an edge state 8 which does not have maximal ranks. We can
always add to it infinitesimal amount of projectors on product vectors
destroying the edge property. The resulting state p would have more
product states in its range, than the product states used to destroy the
edge property. Subtracting projector on product states different from
the latter ones, would typically allow to construct an ec!.ge state 8 with
maximal ranks, which would be infinitesimally close to 8 in any norm.
• Let R(A), K(A) denotes the range and kernel of A, respectively. The
canonical form of an non-decomposable entanglement witness that de-
tects the edge state 8 is ([23, 24], see also [32])

(17.A.1)

where the positive operators P, Q have their ranges R(P) = K(8),

R(Q) = K(8TA ), and E > 0 is sufficiently small so that for any product
vector (e, flWle, 1) ~ O.
• If we can show that for any edge state with maximal ranks and any
corresponding witness W detecting its entanglement, there exist a vector
I~S) of Schmidt number s such that (~SIWI~S) < 0, then we would
conclude that all edge states with maximal ranks, and thus all edge
states, and thus all PPT entangled state have the Schmidt number < s.
Equivalently, it is sufficient to show that (~SIW + EII~S) ~ o.

Let us therefore try to construct the desired vector I~S) of Schmidt number
s. In general such (unnormalized) vector will have a form
S

I~S) ex: Llilei,h), (l7.A.2)

i=l
226 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

where li are arbitrary complex coefficients for i = 1, ... ,s, and lei, h) are
linearly independent product vectors for i = 1, ... ,s. Note, that the vector
(17.A.2) depends on s complex parameters li for i = 1, ... ,s, whereas each of
the s vectors lei), Ih) depends themselves of n-l relevant complex parameters.
Let r(P) = kl' and r(Q) = 2n - 1 - kl. Since we want to prove the
conjecture either for the edge state 8, or for its partial transpose, without
loosing the generality, we may assume that kl 2: 1. We may then single
out one projector out of P, and write P = PI + Iw)(wl, where PI 2: 0,
r(PI) = r(P) - 1, and Iw) is in the range of P. We can choose then lei, h)
in such a way that Qlei,h) = 0, and Pllel,h) = 0. These are effectively
2n - 2 equations for vectors lei, h) which depend on 2n - 2 parameters, so
that we expect a finite, but quite large number of solutions (c.f. [22]). At
the same time, ('ljJsIQI'ljJS) will become a quadratic hennitian form of li'S with
vanishing diagonal elements. Such a hermitian form has typically more than
one dimensional subspace N of negative eigenvalues for large s. But, one has
to fulfill also the last equation implied by ('ljJslw) = 0; this limits the values
of li to a hyperplane, which should have at least one dimensional common
subspace with the subspace of negative eigenvalues N. This would prove that
either the Schmidt number of 8 or of 8T A is of the order of 1.
Note that for a given 8, if the presented construction can be shown to be
successful for every witness of 8, then it provides a sufficient condition for the
state 8 to have the Schmidt number smaller than s.

Appendix: Producing state of p-Schmidt rank form single CV

copy
In this Appendix we briefly show how to produce (by means of local opera-
tions and classical communication - LOCC) the p-Schmidt rank state from any
CV state violating the separability condition

(17.B.1)

Following Ref. [20] this is further generalisation of the distillation protocol of

Ref. [14] where the above criterion with p = 2 (called reduction criterion c. f.
[34]) has been used. The separability tests of the fonn (l7.B.l) (which can be
called p - I-reduction criteria) for p > 2 considered first in [13] are examples
of general positive maps separability tests (see [33]).
Consider some CV state [! and suppose that there exists Iw) such that (Wl[ll®
Ap]([!)lw) = E < 0. Then following arguments of Ref. [20] we get that there
must exist N such that for any m > N the new m ® m state [!m produced from
[! by projection onto the finitedimensional support of 1-l m ® 1-l m satisfies

(l7.B.2)
 Bound entanglement/or continuous variables is a rare phenomenon 227

Now instead of I'll) we put I'll') being a normalised projection of I'll) on support
of (lm. In this way we get the inequality identical to the one of [14] with the
only difference that p was equal 2 there. This allows to repeat the reasoning of
Ref. [14]: after the application of suitable local filtering and U ® U* twirling to
(lm one produces the m ® m isotropic state (lis = (1- q) ~ + q Iw+) (w + I with

the fidelity F == (W+I(lislw+) > p~l. But the latter implies (see [13]) that the
final state (produced from initial (l by means of local operations and classical
communication - LOCC) has the Schmidt number at least p. This concludes
the analysis. Note that further steps of recurrence protocol with generalised
XOR ([14]) can be applied.

Notes
1. Subsequently we shall denote by n (8) n states the states of quantum systems defined on the Hilbert
=
space 1{ en (8) en. The space will be sometimes called "n (8) n space".

References
[1] M. Horodecki, P. Horodecki and R. Horodecki, Phys. Rev. Lett. 80, 5239
(1998).
[2] C. H. Bennett, G. Brassard, S. Popescu, B. Schumacher, J. Smolin and W.
K. WooUers, Phys. Rev. Lett. 76, 722 (1996).
[3] P. Horodecki, M. Horodecki and R. Horodecki, Phys. Rev. Lett. 82, 1046
(1999).
[4] M. Murao, V. Vedral, Phys. Rev. Lett. 86, 352 (2001); P. W. Shor, J. A.
Smolin, and A. V. Thapliyal, quant-ph/0005U7. W. Diir, J. I. Cirac, and
R. Tarrach, Phys. Rev. Lett. 83, 3562 (1999); W. Diir and J. I. Cirac, Phys.
Rev. A 62, 22302 (2000).
[5] See S. Braunstein and J. Kimble, quant-ph/991 0010, and references
therein; for experiments see A. Furusawa et al. Science 282, 706 (1998).
[6] P. Horodecki and M. Lewenstein, Phys. Rev. Lett. 85,2657 (2000).
[7] R. F. Werner and M. M. Wolf, Phys. Rev. Lett. 86, 3658 (2001).
[8] K. Zyczkowski, P. Horodecki, A. Sanpera and M. Lewenstein, Phys. Rev.
A 58833 (1998); K. Zyczkowski, Phys. Rev. A 60 3496 (1999).
[9] G. Vidal, R. Tarrach, Phys. Rev. A 59141 (1999).
[10] S. L. Braunstein, C. M. Caves, R. Jozsa, N. Linden, S. Popescu and R.
Schack: Phys. Rev. Lett. 83, 1054 (1999); R. Schack and C. M. Caves,
Phys. Rev. A 60, 4354 (1999).
[11] N. Linden and S. Popescu, Phys. Rev. Lett. 87, 047901 (2001).
[12] R. Clifton and H. Halvorson, Phys. Rev. A 61,012108 (2000).
228 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[13] B. M. Terhal and P. Horodecki, Phys. Rev. A 61, 040301(R) (2000).

[14] M. Horodecki and P. Horodecki, Phys. Rev. A 59, 4026 (1999).
[15] M. Horodecki, P. Horodecki and R. Horodecki, Phys. Rev. A 60, 1888
(1999).
[16] R. V. Kadison and J. R. Ringrose, "Fundamentals of the Theory ofOper-
ator Algebras", Academic Press, (1983).
[17] A. Peres, Phys. Rev. Lett. 77, 1413 (1996).
[18] L.-M. Duan, G. Giedke, J.I. Cirac, and P. Zoller, Phys. Rev. Lett. 84, 2722
(2000).
[19] R. Simon, Phys. Rev. Lett. 842726 (2000).
[20] G. Giedke, ,L. M. Duan, J.I. Cirac, and P. Zoller, quant-phl0007061.
[21] M. Lewenstein and A. Sanpera, Phys. Rev. Lett. 80, 2261 (1998); A.
Sanpera, R. Tarrach and G. Vidal, Phys. Rev. A 58,826 (1998).
[22] B. Kraus, J. I. Cirac, S. Kamas, and M. Lewenstein, Phys. Rev. A 61,
062302 (2000); P. Horodecki, M. Lewenstein, G. Vidal, and J. I. Cirac,
Phys. Rev. A 62, 032310 (2000).
[23] M. Lewenstein, B. Kraus, J. I. Cirac, and P. Horodecki, Phys. Rev. 62,
052310 (2000).
[24] M. Lewenstein, B. Kraus, P. Horodecki, and J. I. Cirac, Phys. Rev. 63,
044304 (2001).
[25] G. Giedke, B. Kraus, M. Lewenstein and J. I. Cirac, Phys. Rev. A 64,
052303 (2001).
[26] G. Giedke, B. Kraus, M. Lewenstein and J. I. Cirac, Phys. Rev. Lett. 87,
167904 (2001).
[27] G. Giedke, L.-M. Duan, 1. I. Cirac and P. Zoller, Quant. Inf. Compo 1(3),
79 (2001).
[28] D.P. DiVincenzo, P. W. Shor, J. A. Smolin, B. Terhal and A. Tbapliyal,
Phys. Rev. A, 61 062312 (2000); W. Dilr, J. I. Cirac, M. Lewenstein and
D. Bruss, Phys. Rev. A 61062313 (2000).
[29] P. W. Shor, J. A. Smolin, and B. M. Terhal, Phys. Rev. Lett. 86, 2681
(2001).
[30] L. P. Hughston, R. Jozsa and W. W. Wooters, Phys. Lett. A 183,14 (1993).
[31J A. Sanpera, D. Bruss, M. Lewenstein, Phys. Rev. A 63, 050301 (R) (2001).
[32J B. Terhal, Lin. Algebr. Appl. 323, 61 (2000).
[33J M. Horodecki, P. Horodecki, R. Horodecki, Phys. Lett. A 223, 1 (1996);
[34J N. Cerf, C. Adami, and R. M. Gingrich, Phys. Rev. 60,898 (1999).
III
CONTINUOUS VARIABLE
OPTICAL-ATOMIC INTERFACING
Chapter 18

ATOMIC CONTINUOUS VARIABLE PROCESSING

AND LIGHT-ATOMS QUANTUM INTERFACE

Alex Kuzmich
Norman Bridge Laboratory of Physics 12-33
California Institute of Technology, California 91125, USA
[email protected]

Eugene S. Polzik
QUANTOP - Quantum Optics Center
Danish National Research Foundation
Institute of Physics and Astronomy, University of Aarhus, Denmark
[email protected]

Abstract In this Chapter methods for generation of squeezed and entangled states of atomic
ensembles are described along with the protocols for quantum state exchange
between light and atomic samples. Realization of these protocols provides the
means to store/retrieve quantum information transmitted by light in/from atomic
samples, which can be used for processing of this information. Polarization
variables (Stokes parameters) of a multi-photon light pulse and spin components
of a multi-atom atomic ensemble are the continuous variables employed in these
protocols. Two different methods for a quantum state exchange are analyzed:
(a) mapping of non-classical states of light onto atomic spins via complete
absorption of resonant light, (b) teleportation-like transfer of a quantum state via
a QND-type interaction between off-resonant light and atoms.

1. INTRODUCTION
The ability to implement quantum interfacing between light and matter is cru-
cial for many aspects of quantum information processing including distributed
computing, quantum networks, computational complexity, eavesdropping in
quantum cryptography, to name a few. Continuous field and atomic variables
have a dramatic practical advantage for this type of operations compared to
231
S.L. Braunstein andA.K. Pati (eds.), Quantum Information with Continuous Variables, 231-265.
© 2003 Kluwer Academic Publishers.
232 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

single-particle qubits, namely efficient quantum state exchange for the former
can be achieved without high-Q cavities necessary for operations with the latter.
Whereas efficient quantum state exchange between single photons and single
atoms requires cavity QED with strong coupling [1], efficient state exchange
between collective quantum variables of light and atoms can be achieved in
free space with very weak interaction on a single photon/single atom level.
"Strong" interaction in case of continuous variables for multiphoton pulses and
multi-atom samples reduces to high on-resonance optical density of the atomic
sample, which is easy to achieve experimentally.
An efficient quantum state interaction between light and atoms at the level
of continuous variables allows to carry out a number of quantum state/quantum
information processing protocols. In this chapter we will describe some of them
with the emphasis on experimental implementation. First we will define con-
tinuous variables for atomic spin polarized ensembles and polarized light and
introduce coherent, squeezed and entangled spin states. We will then consider
mapping of a quantum state of light onto atomic state via complete absorption
of resonant squeezed light. This method of mapping a quantum state of reso-
nant light onto atoms (Section 2) has been proposed in [2], further investigated
in [3,4], and experimentally implemented for generation of spin squeezing in
[5]. Next we will discuss an off-resonant dispersive interaction of light and
atomic ensembles. This approach proved to be especially successful in several
aspects. We will show how it allows for a quantum non-demolition (QND)
measurement of spin projection leading to generation of a spin squeezed state
[6]. An important problem of atomic teleportation for continuous variables will
be considered. Teleportation requires a resource of entanglement, more specif-
ically distant entangled atomic samples are needed. Whereas both discrete
and continuous variable entangled (EPR) states of light have been successfully
generated by several groups [7,8,9], entanglement of distant atomic systems
has been experimentally achieved only recently [10] with continuous variables
following the proposals [11, 12]. Its applications to atomic teleportation and
interspecies light-atoms teleportation will be considered.

2. CONTINUOUS QUANTUM VARIABLES FOR

POLARIZED LIGHT AND SPIN-POLARIZED
ATOMS
In this section we introduce continuous variables for light and atoms, which
will be used for all quantum protocols throughout the Chapter. Quantum state
of a polarized pulse of light will be described by the Stokes operators such
as Bx = ~ foT(a~(T)iL(T) + a~(T)a+(T))dT, By = 21 foL(a~(T)a_(T) -
a~a+(T))dT, Bz = ~ foL(a~(T)a+(T) - a~(T)a_(T))dT. a+(t) and a_(t)
are annihilation operators for two circular polarizations propagating along the
 Atomic continuous variable processing and light-atoms quantum interface 233

z axis, with commutation relations [ai(t), aj(t')] = OijO(t - t'), i,j = +,-.
The Stokes parameters obey the standard spin commutation relations,

(18.1)

In the following we will assume that light is polarized in the classical sense
along the x axis, i.e. that only Sx has a non-zero mean value. All the interactions
considered below introduce only minute changes in the mean polarization, and
therefore we substitute Sx with its classical mean value (Sx).
Analogous to polarized light, a quantum state of a collection of N spin 112
atoms can be described by its collective spin F == ~f=l F(k) with commuta-
tion relations

(18.2)

As with light, we will assume that atoms are spin polarized in the classical sense
along the x axis, and that this mean polarization does not change much. There-
fore the quantum states of polarization, both for light and atoms, considered
here are generated by small rotations of the collective spin (Stokes vector for
light) around its mean value. A formal analogy between the mUlti-particle spin
ensembles considered here and standard position/momentum-like continuous
variables in the phase plane comes from the commutation relations (18.1,18.2).
Under the assumption of small spin rotations, the right hand sides of these equa-
tions are constants. Dividing the equations by the absolute values of their right
hand sides, one obtains standard commutation relations of position/momentum
for normalized spin variables. The procedure of substituting small rotations on
the Bloch sphere with displacements on the plane is known [13] as contraction
of the group SU(2) to group U(l). For example, for small rotations around
the IF, - F) state (that is, the state with mean polarization along the z-axis),
this is achieved with the following correspondence between the generators of
the groups:
At
F+ -t a,
F_ -t a, (18.3)
(Fz + F) -t aAtAa,
(analogous relations hold for the Stokes vector of light, of course). This
contraction of the group SU(2) onto the group U(l) is at the heart of the
approaches to continuous quantum information processing that we describe
below. The spin formalism allows us to treat light and atoms on an equal
footing. In the following, when it is not specified otherwise, "spin" refers to
both light and atoms.
234 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

2.1 COHERENT SPIN STATES

For concreteness, let us choose the z-axis as the quantization axis and let us
write the state I~, O} (i.e. with all spins oriented along the x-axis) in the basis
of eigenstates of Fz:

(lS.4)

Here F == N /2. When N is a large number, it is possible to write this CSS in

the approximate form

(lS.5)

Commutation relations (1S.2) for the spin F have Heisenberg uncertainty rela-
tionship associated with them

(1S.6)

plus cyclic permutations. Here we define the variance of the operator A

by (~A)2 == (A2) - (A}2. The CSS (1S.5) satisfies the minimum uncer-
tainty product allowed by (1S.6), and so, is a minimum uncertainty state.
For example, it is straightforward to show that for the state in Eq.(lS.4)
((~Fz)2) = ((~Fy)2) = F/2, while (F:[;} = F, so that the equality in
(1S.6) holds. CSS can not exceed the standard quantum limit (SQL) of phase
measurement accuracy 8¢SQL given by l/m, where N is the number of
elementary 1/2 spins.

2.2 SQUEEZED SPIN STATES

A spin state F is called a Squeezed Spin State (SSS) if the squeezing param-
eter 'fJ defined by

(1S.7)

is less than 1 for some ((), ¢). Here Fl. is spin component in a direction
orthogonal to ((), ¢). The squeezing parameter 'fJ determines the accuracy of
phase measurement 8¢:

8¢=~. (lS.S)
 Atomic continuous variable processing and light-atoms quantum interface 235

It is possible to show that for a CSS the minimum value of 11 is 1. Moreover, it

has been shown [14] that there exist no unentangled spin states (in which the
collective spin state is the direct product of single-spin states) for which the
value of the squeezing parameter 11 is below unity. The only way to reduce 11
(and, as we see from Eq.(18.8), to enhance the accuracy of phase measurements)
is to introduce quantum-mechanical correlations between individual spins.
Many different entangled classes of states have been proposed theoretically
(see, for example, Refs. [15,16,17, 18] and references therein) that satisfy the
definition of SSS. In most practical situations of interest the SSS 's are entangled
rather weakly. Then a SSS can be written as (X « 1):

I ~2' 0 ,x) = (7rF)-1/4

. r1 3 2F
~
LJ
ex (_ m 2(1 + 3X2 F))
P 2F
IF m)
"
(18.9)
v + X m=-F

where the squeezing parameter 11 is related to X by 11 = 1/(1 + 3X2 F). Forthis

SSS Fz has reduced variance ((~Fz)2) = ~(1 + 3X2 F)-I, while the variance
of Fy is increased, ((~Fy)2) = ~(1 + 3X2 F). For X « 1 (Fx) ~ F, so that
this state is a minimum uncertainty one.

2.3 EPR SPIN STATES

Another important class of entangled spin states are EPR-correlated spin
states. An important feature of these states is that they describe two distant
spins that are correlated in a non-classical way. It is rather cumbersome to
describe this kind of entanglement in the Schrodinger picture that we used above
for CSSs and SSSs. Instead, in the Heisenberg picture, the spin components
satisfy the following relations

Fxl + Fx2 ..J2 exp( -r )Fxv,

FyI - Fy2 ..J2 exp( -r)Fyv, (18.10)

where Fxv , Fyv are x- and y- spin components of an auxiliary spin in a CSS
directed along the z-axis (we assume that all three spins have the same length
F). r is the parameter that characterizes the strength of the quantum correlations
between these two spatially separated spins.
For two spins with opposite classical orientation, (FzI ) = -(Fz2 ), the
necessary and sufficient condition for entanglement can be formulated in terms
of the variances of measured quantities [19] as

(18.11)
236 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

The interpretation of this condition follows from the fact that for both spins
in CSSs the following equality holds: ((Ll(Fx,y))2) = !(Fz ). Entanglement
between the elementary spins of the two samples is, therefore, according to the
above condition equivalent to the spin variances smaller than that for samples in
a CSSs, which is characterized by uncorrelated individual atoms. The entangled
state of this type is a two-mode squeezed state for the continuous spin variables.

3. LIGHT-ATOMS STATE TRANSFER VIA

RESONANT INTERACTION
One of the fundamental operations of the continuous quantum information
processing is the quantum mapping of a (short-lived) field mode state into a
corresponding quantum state of a collection of (long-lived) atomic spins and
backwards (such procedures are often called quantum memory writing and
reading). One of the ways to achieve such a mapping is by complete absorption
of e.-m. field by the atoms. In particular, absorption of squeezed states of light
gives rise to SSSs of atoms.
We consider free space illumination of a collection of atoms by a beam of
squeezed light. The discussion is limited to excited states of atoms, which are
not directly suitable for memory purposes. However, it is instructive to consider
this case anyway, for which 50% of the amount of squeezing of the exciting
optical field can be transferred onto spin squeezing of the excited atomic states.
The limitation on the degree of atomic spin squeezing is due to spontaneous
emission. Further development of this approach for ground state atoms and
Raman transitions has been proposed [3].

3.1 THEORY
Let us consider the propagation of quantum correlated e. -m. fields through an
atomic medium which fully absorbs (and spontaneously re-emits) the incoming
light field. In the process of propagation some quantum statistical properties
of the field are altered by the interaction with atoms. The atoms acquire
certain quantum features of the field via the absorption mechanism, and at
the same time, spontaneously emit photons. In the steady state the scattered
and the incoming light intensities are identical and they are proportional to
the total excited state population of the atomic medium. In the limit of weak
fields, the emission is also coherent; both the scattered field amplitude, and the
atomic dipole amplitude are proportional to the incoming field amplitude. This
suggests that certain collective atomic operators exhibit new and non-classical
noise properties due to the interaction with the fields. We consider a V-transition
with each arm 0 ++ 1(2) interacting with a separate quantum field, 121(2) (see
Fig. 18.1). The two excited states can be two magnetic sublevels (say, with
m = 1 and m = -1) in which case the spin will correspond to the orientation
 Atomic continuous variable processing and light-atoms quantum interface 237

Excited state
m=-l m=+l
cr+-
••••
• • ••
0-
~ m=O,
Atomic ground state
sample

Figure lB. 1 Spin squeezing of an excited state via complete absorption of squeezed light.
Linearly polarized squeezed light can be viewed as a superposition of quantum correlated
modes in right- and left-circular polarizations. When these modes excite the atomic state as
shown in the Figure, quantum correlations (squeezing) are partially transferred to the atomic
state. Spontaneous emission which prevents perfect transfer is also shown.

and/or alignment of the excited ensemble. Alternatively, the two states can be
two hyperfine sublevels. We treat the case when the 0 - 2 transition interacts
with a squeezed vacuum mode, and the 0-1 transition interacts with a coherent
field mode. To analyze the noise properties of the collective upper-states atomic
coherence FI2 we assume that the coherent amplitude a is much stronger than
the squeezed vacuum fluctuations, and subsequently that the coherent part of
FOi is much greater than the fluctuating quantity F02 . The mean spin is along
the z-direction and has a value equal to half of the total atomic population in
the excited state 11):

(F ) = (a~nt ain) (18.12)

z 2,'

(Fx) = (Fy) = O. The components where we shall look for squeezing are
Fx,y. Detailed calculations [2] result in the collective atomic operator FI2 at
frequency .6. being expressed simply in terms of the incident and the transmitted
field operators:

,-2.6.
a* . a2(in) (WI +.6.) +
,-2.6. d
a* . 2(in) (WI + .6.)08.13)

Here , is the rate of spontaneous decay, WI is the frequency of the atomic

transition, a-2 (in) (WI + .6.) is the annihilation operator describing the incident
238 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

squeezed vacuum field, d2 (in) (WI + b..) is the annihilation operator describing
the bath of the spontaneous decay modes.
Let us examine if the Fx,y are squeezed. We find from Eq.(18.13) that the
variances of the atomic spin components are linked to the quadrature phase
variances X;,y of the input field ih:

(18.14)

For the vacuum input field 0.2 the variances 4(X~,_) = 1 and the x, y spin
variances are equal to ~(Fz) as would be expected for the coherent spin state.
With broadband squeezed vacuum as input field 0.2, one of the variances van-
ishes, e.g. (X;) = 0, and we obtain the spin squeezed state with 50% degree
. • A2 _ 1 A
of squeezmg. (Fx) - 4: (Fz ).
In the example above spin squeezing is achieved in an off-diagonal element
of the excited state orientation. By slightly changing the geometry of the ex-
periment one can also achieve squeezing in the diagonal elements, i.e. in the
population difference between the two excited states. In order to achieve this
goal the coherent and squeezed vacuum fields with orthogonal polarizations
should be mixed at a polarizer to produce quantum correlated left- and right-
circularly polarized fields 0.1 and 0.2. With the average atomic spin oriented
along the direction of the coherent polarization x, either Fy or Fz compo-
nents can be squeezed by 50%, depending on the phase difference between
the squeezed vacuum and the coherent beam. The latter case ("longitudinal
spin squeezing") corresponds to sub-Poissonian fluctuations in the population
difference between the two excited states.

3.2 EXPERIMENT
The experimental realization of this proposal has been carried out with a
collection of cold Cesium atoms confined in a magneto-optical trap [5, 20].
Atoms are excited with a weak quantum pump (Fig.l8.2) from the ground state
to the excited state 6P3/2, F = 5 (hereby abbreviated as 6P). The pump is a
mixture of the x-polarized coherent beam and y -polarized squeezed vacuum
produced by a frequency tunable optical parametric oscillator below threshold.
A weak linearly polarized probe in the configuration shown in Fig.18.2 is
sensitive to the following three operators: Fz, F; - F;, and FxFy + FyFy
[21]. These are spin operators describing the entire 6P ensemble. They obey
the commutation relation ~ [Fz, FxFy + FyFx] = F; - F;. We consider these
three components as components of a quasi-spin operator. With the coherent
component of the excitation polarized along the x axis the only component
of the spin polarization with a non-zero mean is (F; - F;). The uncertainty
relation following from the commutation relation is: 8(Fz)8(FxFy + FyFx) ~
 Atomic continuous variable processing and light-atoms quantum inteiface 239

Lock-In (DC)
Lock-In (RF)
~

Cesium

6D
m
~~~~prObe
I,lt+--t'--·+-----, 6P312 F=5 -
852nm Quantum
6S F=4 pump
112

Figure 18.2 Outline of the experimental set-up.

I(F; - F;) I .Since the operators describe the whole ensemble, the right hand
side of this inequality is proportional to the number of atoms N in the 6P state.
Excitation of the atoms with coherent light will lead to the quantum noise of
the z component o(Fz)coh = V51(F; - FJ)I/(2F + 3) '" VN due to the
uncorrelated noise of individual atomic spins. Following the theory described
in the previous section, one should expect that injection of the squeezed e.-m.
vacuum into the other port of the polarizer PBS 1 (Fig.18.2) should lead to
squeezed o(Fz)sq < o(Fz)coh or anti squeezed o(Fz)antisq > o(Fz)coh spin
states depending on the phase of the squeezed vacuum field.
Turning now to the experiment, about 109 Cs atoms have been collected in a
magneto-optical trap. The spin noise ofthe 6P state has been detected using the
probe polarization noise technique described in detail in Ref.[22](Fig. 18.2).
The change in the probe differential photocurrent noise caused by atoms is

oi 2 (.6.) = -[1 - exp( -a~)l + sao exp( -2a~)(oF)2

(OF)2 is the atomic noise contribution depending on the geometry of the
experiment, a~ is the probe optical depth at detuning .6., s is the saturation
parameter of the probe light. The expression for Ji 2 is normalized to the probe
shot noise in the absence of atoms. The term in square brackets is the probe
shot noise reduction Ji~hot due to the absorption and the rest is the atomic noise
contribution of interest. Note that in the limit of small optical depth a~ this
240 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

expression is similar to the Eq.(18.17) for the polarization component of the

off-resonant probe described in detail below.
When atoms are excited in 6P state with linearly polarized coherent light,
a certain level of atomic noise (8F)2 corresponding to the coherent spin state
is measured. When squeezed light is used for excitation instead, a different
level of (8F)2 is generated. If this level is lower the spin state is squeezed.
The squeezed light after PBS 1 will have fluctuations in either the polarization
axis direction or in the ellipticity (depending on the phase) reduced below
the noise level of the coherent field (Standard Quantum Limit - SQL). For
the ellipticity-squeezed light, the fluctuations in the intensity difference be-
tween the 0'+ - and a- -polarized components of the pump field are reduced
below the SQL. This gives reduced fluctuations in the difference between the
number of atoms in the +m and -m Zeeman levels. This difference in popu-
lations is described quantitatively by the collective spin component Fz . Thus,
the ellipticity-squeezed light gives reduced fluctuations in Fz , whereas the
polarization-squeezed light is anti squeezed in ellipticity resulting in increased
Fz noise and reduced FxFy + FyFy noise. Squeezed vacuum with the central
frequency resonant with the pump transition is generated in the sub-threshold
optical parametric oscillator (OPO). The amount of squeezing is quantified by
the variance of the quadrature phase operator X; /2 of the e.-m. field polarized
along y and out-of-phase with the coherent component of the pump. For per-
fectly squeezed ellipticity X; /2 = 0 and for coherent fields/ordinary vacuum
4X; /2 = 1. In order to achieve the best mapping of quantum properties of
light onto atoms, the optical depth for the quantum pump is sustained at the
highest possible level, a pump ~ 4. We concentrate on the measurement of 8Fz
because the sensitivity of our polarization measurements to squeezing of the
conjugated variable 8(FxFy + FyFy) is found to be much smaller.
The experimental results presented in Fig. 18.3 are normalized to the atomic
spin noise corresponding to the SQL achieved with coherent excitation (hor-
izontal line). When the squeezed vacuum is out-of-phase with the coherent
component of the pump, 4X; /2 > 1. This phase corresponds to the quantum
noise of the spin component Fz above the coherent spin noise level. The ob-
served excess noise is plotted in Fig.18.3 as dots. The spectrum of the observed
anti squeezed excess spin noise can be written as 8i~ntisq (b.) - 8i~oh (b.) ex:
jJ2(b.) ((8Fz )2 - (8Fz)~Oh). The excess spin noise is expected to have the
spectral shape given by the square of the Doppler broadened dispersion profile
JJ2(b.). We introduce 'fI, the mapping-readout efficiency for quantum cor-
relations, by relating the observed spin noise to 4X; /2. With b.max being
the detuning giving the maximum excess noise we get: 8i~ntisq (b. max ) =
7J!1(1 + 'fI4X;/2)8i~oh (b. max). 'fI = 1 gives 50% of noise reduction for per-
 Atomic continuous variable processing and light-atoms quantum interface 241

';) 0,40
......
o
"0 0,30
'-'

..'
,,-....
8'.
'.
-0,04

-
$:I)
-0,06 Jg
L.--r-.....--.-----.---,--.--"'T""'=--.....---,--l-O,08 ~
-10 -5 0 5 10 '-'

Probe detuning [MHz]

Figure 18.3 Experimental data.

fect squeezing and is the theoretical maximum efficiency in a 3-level system

when spontaneous decay is taken into account, as discussed above. For a multi-
level system, such as Cesium ground state, the upper limit on rJ is lower [20].
The data in Fig.I8.3 corresponds to (4.5 ± 0.6) dB of the excess noise in the
antisqueezed quadrature of the pump giving 4X;/2 ::: 2.8. This corresponds
to the value of the mapping-readout efficiency rJ = 0.09 ± 0.02. The imperfect
efficiency is caused by many factors, including the admixture of the noise of
other spin polarization components than Fz , the multilevel effects, imperfect
polarizations, the residual magnetic field, imperfect overlap of the pump and
the probe, reabsorption of uncorrelated spontaneously emitted photons at 852
nm, and the finite bandwidth of the squeezed light.
With the squeezed vacuum in-phase we measure c5i~q (b. max ) . The average
squeezing of the pump light available at the trap site is (-1.8 ± 0.2) dB
corresponding to 4X;/2 ::: 0.65 . The expected quantum spin noise spectrum
for such pump and the efficiency rJ = 0.09 is plotted as a dotted line in
Fig. (18.3). The experimental data are plotted as triangles. All available
experimental data obtained in 9 runs each lasting between 4 and 11 hours is
shown. The long integration time is needed even with all the implemented
lock-in detection stages because the quantum spin noise reduction corresponds
to only ::: 2 . 10- 4 of the probe shot noise. Each point in the figure is the
average of one run with 13 to 45 individual measurements. Each of these
242 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

individual measurements consists of 6 min. of averaging with the squeezed

vacuum interacting with the atoms, and 6 min. of averaging with the squeezed
vacuum blocked.
We define the degree of the observed spin squeezing (quantum spin noise
reduction) by e = (8i~q - 8i~oh) / 8i~oh' The best experimental points in
Fig.18.3 are in reasonable agreement with the expected degree of the quantum
e
spin noise reduction which is -2.7%. A drift in the value of on the time
scale of hours within the range from 1% to -5% has been observed. We
attribute the drift to slow uncontrolled changes in the trap geometry, leading to
changes in local magnetic fields, density and polarizations. The best observed
e
value around ±6 MHz is = - (5 ± 1) % (4 hours average, 17 individual
measurements). The average value for all the 183 individual measurements at
e
±6 MHz is = -(1.4 ± 0.4)%.
Summarizing, the free space complete absorption approach does provide
mapping of a quantum state of light onto atomic state, albeit with relatively low
efficiency.

4. OFF·RESONANT ATOM·LIGHT INTERACTION

AS QUANTUM INTERFACE
Interaction of polarized light with spin-polarized atoms has been a subject of
intensive investigations in atomic physics for several decades (for a review, see
Ref. [23]). Polarization analysis oflight transmitted through an atomic medium
has been widely used for measurements of atomic orientation and alignment.
Off-resonant atom-photon interaction as a probe of the spin orientation for
atoms with total ground state electronic angular momentum equal Ti/2 has been
proposed in Ref. [24]. Recently it has been realized that the same interaction
can reveal not just the mean value of the atomic spin, but its quantum state as
well. As opposed to the classical case, in quantum world the back action of
light on atoms cannot be ignored. This back action leads to entanglement of
light and atoms, which in tum can be used to generate entanglement between
separate atomic samples and other quantum information operations on atoms,
as described below in this section.
The unitary time-evolution operator corresponding to the interaction with
light propagating along the z-axis has the following form (the derivation is
outlined in Appendix A):

(18.15)

where a is given by a = A{I:k) Ia v , a is the resonant absorption cross-

section for an unpolarized photon on an unpolarized atom of total spin F, A is
the area of the transverse cross section of the light beam, 'Y is the spontaneous
emission rate of the upper atomic level, !:l. is the detuning. The dynamic vector
 Atomic continuous variable processing and light-atoms quantum interface 243

polarizabilityav = ±1 for the Dl transition of alkali atoms, while a v = =f~ for

the D2 transition. Here the upper sign is for hyperfine sublevel with F = I + ~,
while the lower sign is for the F = I - ~ hyperfine sublevel, I is the value of the
nuclear spin. It is clear that the above Hamiltonian conserves the z-projection
of the atomic spin and therefore can be used for a QND measurement of the
spin as described below. This Hamiltonian can also serve as a resource for

sxj 1 ftx

Off-resonant :::I===:::J=t=======:::=:=::=_
<::'

linearly polarized
light Polarized Atoms

Figure J8.4 Outline of the atom-light QND interaction scheme.

a continuous XOR gate, one of the cornerstones for the quantum information
processing with continuous variables. To show this, let us derive equations of
motion for the field and atomic variables. Let the light pulse with the mean
polarization along the x-axis propagate along the z-axis through the atomic
sample also with the mean spin along the x-axis (Fig. 18.4). In the Heisenberg
picture transformations of the spin operators can be written as

- sin{aS
A

cos{aSz )
z) 0) (Fx
0
A

t:y
)
~n)

, (18.16)
o 1 Fz

and a similar expression with F and S interchanged. In the limit of small

rotations on the Bloch and Stokes spheres the relevant spin components can be
written as follows:
S(out)
y
""
""
s(in)
y
+ ap~n)(S~n))
z x
"" s~n) + an p(in)
""y 2z

p~out) ~ p~in) + (aN F)s~in), (18.17)

whereas the z-components of the spin are constants of motion. Here the
Stokes vector S of the light pulse has the length (Sin) = ~x, where n is the
244 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

number of photons in the pulse. As can be seen from the above equations, the
interaction Hamiltonian (1S.15) provides a continuous analog of XOR gate if
aJFnN/2 = 1 (F is the value of the spin of one atom). The rotations of
individual qunats are achieved for light with the help of A/2 and A/4 plates and
for atomic spins by applying dc magnetic fields.
Many algorithms of the field of discrete quantum information have their
continuous analogs if every qubit is replaced with a qunat, every XOR-gate
is replaced with a continuous XOR-gate, and every single qubit rotation is re-
placed with a corresponding rotation of the qunat in the quadrature plane [25].
Using collections of atomic spins and the field modes as qunats is a suitable ar-
chitecture for implementation of continuous quantum information algorithms.
Below we will show how using such philosophy one can, for example, achieve
exact mapping of quantum states between light and atomic spins, in both direc-
tions. But first we will describe the experiment demonstrating the feasibility
of the above Hamiltonian for continuous quantum information applications,
namely for generation of spin squeezed states of atoms.

4.1 SPIN SQUEEZING VIA QND MEASUREMENTS

The interaction described by Eq.(1S.15) is of the exact form required to
perform QND measurements of a spin projection [26,27]. Indeed, the essential
feature of a QND measurement [28] is that an operator of a signal particle A
is coupled to an operator iJ of a probe particle via an interaction Hamiltonian
iIi that commutes with A. By making a (destructive) measurement on the
operator 6 conjugate to iJ, one obtains information about A without disturbing
A. Another important requirement is that the free evolution Hamiltonian of the
signal particle commutes with A, so that the QND observable A is a constant
of motion.
The idea of the atom-light spin-QND interaction is outlined in Fig.lS.4.
Interaction (IS. 15) leads to rotation of the polarization of the field that is
proportional to ftz, entangling atoms and photons in the process. Subsequent
optical polarization measurements project atomic spins into a SSS, as first
proposed in Ref. [26].

4.1.1 Theory. The Heisenberg picture operators transform according to

the relations (1S.16). Now let us suppose that s~out) is measured. We assume
that initially both F and S are in coherent spin states along the x-axis, with
(F(in)) = ~ x, (s(in)) = ~x, where N and n are the numbers of elemen-
tary 1/2-spins comprising angular momenta F and S, respectively. We find
that (S~out)) = ~(sin(aftlin)))n, or, for aJ((t::..ftl in ))2) = (aV"N)/2 « 1,
(S~out)) :::: ~a(Flin) )n. The last approximate equality suggests assigning to
 Atomic continuous variable processing and light-atoms quantum interface 245

F z the "QND shift" of Pz ~

~ lS(out)
an y . Subtracting this operator from Fz(out)
we obtain the modified expression:

FI = F(out) _ ~S(out) (1S.1S)

z - z an y .

We find
N a
= -2 cos n -2'
A

(F)
x

N 1
-4 + -2na
- 2 (n(1 - cos N a) + (1 + cos N a))
N . a N-l a
- sm - cos - (1S.19)
a 2 2

For a and n such that N a 2 « n- 1 / 3 (we assume N, n » 1), we find (Fx) =

(N/2) exp[-~/2], ((f:..P~)2) = 1/(4~). The squeezing parameter as defined
by Eq.(1S.7) is then found to be

exp[~/2l
7] ~ JN~ , (1S.20)

where ~ = na 2 /4. The last expression for 7] is minimized when ~ = 1, and

we obtain 7]min ~ ve/JN ~ 1.7/JN, which shows that the scheme could
prepare nearly maximally-entangled (Heisenberg-limited) spin states. In the
spirit of the concept of QND measurement, the particular SSS is conditioned
(out) 2 (out)
on the outcome of the By measurement. Subtraction of an By in (IS.IS)
A A

takes care of the random variation from trial to trial of the direction of (F).

4.1.2 Experiment. The experimental implementation of spin squeezing

via QND measurement is reported in [6]. Let us consider a beam of atoms
of density p, spin-polarized along the x-axis, moving with speed v through
a light beam propagating along the z-axis. For simplicity we assume that
the atomic beam and the light beam have square profiles of the same size
d along the y-axis, while the atomic beam has width L along the z-axis.
Further, let us assume that the atomic spins are subjected to a time-dependent
sinusoidal magnetic field eyB cos(nt). The z-component of F(t) is given by
pz(t) = cos (¢(t)) pJin) (t) + sin (¢(t)) FJin\t). Here p(in) is the incident
collective spin, ¢(t) = ¢sin(nt) and we assume ¢ = (J-lB)/(lin) « 1 so that
Fz(t) ~ FJin) (t) + ¢ sin(nt)PJin) (t) (J-l is the atomic magnetic moment). The
QND-type atom-light interaction leads to polarization rotation of the probe light
by an amount which is proportional to Fz(t). Let us introduce the following
246 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

observable,

(18.21)

which corresponds to the output of a spectrum analyzer that is being fed

with the difference of photocurrents from the detectors Dl and D2, Sy(t) ==
~(at(t)av(t) - a1(t)ah(t)). Here T = liB, where B is the resolution band-
width of the spectrum analyzer. The accuracy of measurement of the amplitude
of spin rotation ¢ can be determined from equation

(18.22)

We obtain

(18.23)

where F(i) is the total spin of one atom (F(i) = 4 in our experiment), T == dlv,
N = pdvTL is the total number of atoms passing through the interaction region
during time T and P is the optical power of the probe beam. We obtain for the
second moment

(18.24)

where the first and second terms are due to (atomic) spin noise and photon shot
noise, respectively. Now we can calculate the measurement accuracy of the
spin rotation angle from Eq.(18.22) and we find

(18.25)

When (XT)~PB « 1, the photon shot noise corresponding to the last term
under the square root sign is much smaller than the atomic noise and can
be neglected. In this case, for low frequencies n « 1I T we find that the
measurement accuracy is given by the SQL, O¢SQL = 1/vF(i)N. From
Eq.(18.25) it follows that, as the frequency n of the applied spin rotation
increases, the phase uncertainty o¢ falls below the SQL. The limit on the
achievable uncertainty is due to the photon shot noise. If the collective atom-
photon coupling is high enough (through the use of either dense atomic samples
 Atomic continuous variable processing and light-atoms quantum inteiface 247

or optical cavities for the probing beam), the limit is detennined by the back-
action of the probe photons onto the collective atomic spin. The measurement
accuracy of the spin rotation angle is then the Heisenberg limit 1/N [26].
An outline of the experimental setup is shown in Fig. 18.5. A paraffin-
coated glass cell contains Cs atoms. Under the conditions of our experiment,
the measured lifetime of spin polarization in the cell is on the order of 1 s,
which means that the atoms undergo thousands of collisions with the walls
without being depolarized.

Figure 18.5 Outline of the experimental setup. Not shown are 3 pairs of coils to produce dc
magnetic fields and a coil producing rf magnetic field. See text for details.

Initially it is important to prepare the atomic sample in a CSS. There are two
important features of a CSS: (a) it is fully spin-polarized along some direction
(and therefore has maximum coherence) and (b) spin fluctuations of the sample
J
are at the shot-noise level given by F /2, where F is the value ofthe collective
spin.
In our experiment two 852 nm diode lasers DL1 and DL2 are used to spin-
polarize the atomic medium. Both light beams are circularly polarized with the
same helicity and propagate along the direction of the magnetic field of several
Gauss. The laser DL1 of about 3 mW average power (adjusted with a neutral
density filter) is tuned to the 681 / 2 , F = 3 --+ 6P3/2 transition(s), while the laser
DL2 ofless than 100 f-lW average power is tuned to the 681 / 2 , F = 4 --+ 6P3/2
transition(s). A small portion of the linearly polarized light from DL3 is split off
and used to monitor the degree of spin polarization. This probe beam propagates
through the cell along the direction of spin polarization. By scanning the
frequency across the Doppler profile, we measure the dispersive profile of the
polarization rotation angle and the Lorentzian profile for absorption. The major
248 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

part of the light from laser DL3 is used as our QND-probe (see below). We align
the magnetic field (and therefore also the spin polarization) orthogonally to the
probe beam with the help of the Faraday rotation signal in the dc channels of our
homodyne detectors on the QND-probe beam. The degree of spin-polarization
of the atoms is found to be greater than 95%.
The QND probe beam is detuned about three Doppler widths from the center
of the 681/2 , F = 4 -+ 6P3/2 transition(s). Al mW light beam with Gaussian
waist of 100 p,m is focused into the gas cell. The optical pumping beams and
the probe beam do not spatially overlap. The. probe beam is then refocused by
lens L2 onto a Glan-Thompson polarizing beam splitter (PBS) whose outputs
are measured by two (85% quantum efficiency) photodiodes D1 and D 2 • The
outputs of the photodiodes are amplified, and the difference signal is fed into
a spectrum analyzer 8 A, where noise spectra between 3 and 20 MHz are
recorded.
An important aspect of the spin squeezing experiment is the need to identify
the spin shot-noise level. By analogy with the use of thermal light for the
photon shot noise normalization, we make use of the unpolarized atoms as
benchmark for the spin shot noise. We have shown earlier in this Chapter, that
the spin noise of unpolarized atoms in a cell scales linearly with the number
of atoms, as expected for an uncorrelated sample of atoms. Because our Cs
atoms have total spin F = 4 instead of elementary I/2-spins, the spin noise
of an unpolarized collection of atoms is j(F + 1) times larger than the spin
noise of the CSS with an equal number of atoms. As we argued above, the spin
shot-noise can be measured at n = O. From the experimental point of view,
however, the difficulty of measuring it in this way is connected with the fact
that our probe laser has a large amount of excess amplitude and phase noise at
frequencies below 1 MHz. Because of this, we prefer to normalize with respect
to the shot noise, with the magnetic field applied in a direction perpendicular
to the QND-probe propagation. In this case the spin noise manifests itself at
the Larmor precession frequency WL, while the amount of the observed noise
is reduced to half. We find the peak shot atomic noise to be about four times
the photon shot noise.
To obtain the shot noise level for the CSS, we make use of the results of
the spin noise measurement with unpolarized atoms. We take into account the
factor of 2 (F + 1) /3 for the difference between the shot noises of an unpolarized
sample of atoms and the CSS, and also of the fact that after optical pumping
we have 16/9 times more atoms in the F=4 state than in the unpumped cell.
That was also confirmed experimentally from absorption measurements. The
spin noise level for the CSS obtained in this way is shown in Fig. 18.6 (broken
line).
Next, we reduce the magnetic field back to 2 Gs and apply a rf magnetic
field along the y-axis in order to produce a small rotation. Fig. 18.6 (solid
 Atomic continuous variable processing and light-atoms quantum interface 249

Cil
SOL
.2>
IJ)
4
Q)
IJ)
o 3
C
-0
Q)
.~ 2
ctl
E
"-
o 1
c
o ~~~~~~~~~~-L~~~~

8 10 12 14 16 18
rf frequency (MHz)

Figure 18.6 Measured spin noise spectrum. The dashed line indicates the SQL accuracy of
spin rotation measurement. The peak at 16 MHz is due to spin rotation by the applied rf coils
and is a demonstration of sub-shot noise atomic interferometry.

line) shows the spin noise spectrum observed under these conditions. The solid
line is the SQL of the spin projection measurement for a CSS when the shot
noise of the probe is taken into account. The peak at 16 MHz is the measured
spin rotation due to the applied rf field. This demonstrates the sub-shot noise
performance of the atomic spin interferometer.

4.2 QUANTUM MAPPING OF COLLECTIVE SPIN

ON POLARIZED LIGHT
Mapping of light onto atoms and visa versa using complete absorption of
light and interspecies teleportation is considered elsewhere in this Chapter.
Let us show that both of these goals can be achieved by using the atom-light
QND interaction of Eq.(l8.15). The method is similar to the interspecies
teleportation protocol, since it also requires an entangled (squeezed) input state
of the system that is being mapped on. For concreteness, let us consider
mapping the atomic spin F onto the Stokes vector S (although all of the
derivation below equally describes the reverse situation of mapping light onto
atoms simply by exchanging S and F). Let the Stokes vector S ofthe light pulse
be initially in a SSS, (Sin) ;::::: ~x, where n is the number of photons in the pulse,
and the y-component of the spin is squeezed, ((~S~in))2) «n/4. We further
assume n = F Nand a;
= 1. First, we direct the light pulse through the cell
250 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

along the z-axis. The Heisenberg picture transfonnations of the y-components

of the spins are given by Eqs. (18.17), whereas the z-components of the spins
are constants of motion. Next we exchange the z-and y-components:
S(out-l) s(out)
y z ,
S(out-l) = SA (out)
(18.26)
z Y •

As the next step, we pass the pulse through the cell along the y-axis. As a
result, we obtain
S(out-2) = S(out) = s(in)
z y y
+ p(in)
z' (18.27)

S(out-2) = S(out) _ F(out)

y z y
= s(in) _ (F(in)
z y
+ s(in))
z
= _F(in)
y' (18.28)

Since the light is initially in a squeezed state, with ((b.S~in))2) « n/4, we can
neglect the first tenn in the Eq.(18.27). Thus, we have achieved exact mapping
of atomic spin state onto polarized light.

4.3 ENTANGLEMENT OF TWO DISTANT ATOMIC

SAMPLES
Teleportation principle described elsewhere in this book requires an entan-
glement resource. More specifically, teleportation of atomic states or telepor-
tation of a state of light onto a state of an atomic sample requires two entangled
atomic samples as an initial resource.

4.3.1 Theory. Generation of the EPR-entangled atomic samples may be

utilized by a QND-type interaction with off-resonant light. It appears that the
off-resonant interaction described above is enough to produce an entangled
state of two separate atomic objects with just a coherent pulse of light as
initial resource. As shown in Ref. [12], when an off-resonant pulse classically
polarized along x is transmitted along z axis through two atomic samples with
opposite mean spins, Fxl = - F x2 , the light and atomic variables evolve as
s(out)
y
s{in)
y
+ an (p(in) + p(in)) S(out)
2 zl z2' z
= s(in)
z
(18.29)

p(out) F(in) + aN s(in) p(out) = p(in) _ aN s(in)

yl yl 2 z , y2 y2 2 z
pCout) p{in)
zl,2 zl,2

The first line describes the Faraday effect (polarization rotation of the probe),
whereas the second line shows the back action of light on atoms, i. e., spin
 Atomic continuous variable processing and light-atoms quantum inteiface 251

rotation due to the angular momentum of light. Therefore, the measurement

of s~out) reveals the value of F;~n) + F;~n) (provided the constant is large a;
enough, so that S~in) is relatively small) without changing this value. It follows
from the second and third lines that the total y projection for two samples stays
unchanged due to the back action cancellation. This is where the choice of
the opposite in direction and equal in size classical spin orientations for the
two samples is crucial. It is also assumed that the Siin) component of light is
not changed in propagation between the two samples, i.e., losses of light are
rather small. We refer to Ref. [12] for a detailed analysis of the role of losses.
The procedure can be repeated with another pulse of light measuring now the
sum of y components, F;~n) + F;~n), again in a non-demolition way, while at
the same time leaving the previously measured value of F;~n) + F;~n) intact.
As a result, the sum of the y components and the sum of the z components of
spins of the two samples are known exactly in the ideal case, and therefore the
two samples are entangled according to Eq.(IS.11), since the uncertainties on
the left-hand side become negligible. An important modification of the above
protocol is the addition of a magnetic field oriented along the x direction, which
allows to use a single entangling pulse to measure both y and z components at
the same time. The Larmor precession of the y, z components with a common
frequency S1 does not change their mutual orientation and size and therefore
does not affect the entanglement. Moreover, in the lab frame the spin state
is now encoded at the frequency S1, and, as usual, it is easier to reduce to the
quantum noise level an ae measurement than a de one. Measurements on the
light can be now conducted only around its spectral component S~out) (S1). By
choosing a suitable value of radio-frequency S1 we can reduce the probe noise
to the minimal level of the vacuum (shot) noise. In the presence of the magnetic
field along x the spin behavior is described by the following equations:
d d aN
dtFz(t) = S1Fy{t), dtFy{t) = -S1Fz(t) + TSz{t)
h h h h h

The Stokes parameters of light evolve according to Eq.(lS.29). Solving

Eq.(l8.29) we obtain:
S~out){t) = s~in){t) + a; ((F;~n) + F;~n)) cos{S1t)
+ (F;~n) + F;~n)) sin{S1t)}. (18.30)

The spin components Fi:;) are now defined in the frame rotating at the fre-
quency S1 around the magnetic field axis, x. Measuring the cos{S1t)j sin{S1t)
component of the photocurrent we perform a projection measurement of the
z / y component of the total spin for the two samples. This measurement entan-
gles two atomic ensembles. The degree of entanglement depends on the ratio
252 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

of the second "atomic" term in Eq.(18.30) to the Fourier component of the first
term, S~in) (t) cos(Ot) or S~in) (t) sin(Ot) . The latter is just the component of
the shot noise of the probe at the frequency O.

Oplical Enlangi ng Verifying

......
pumping pulse pulse

-
&~~. .~. .~.

B-field
~~.
i
Enlangnng and
• •
:
verifying pulses ______
•
: ' _ " - -_ _ _ _ _---.,
.... O.5ms-< Trme

Figure 18.7 Outline of the experimental set-up.

4.3.2 Experiment. Experimental implementation of the entanglement of

two atomic objects has been carried out with two gas samples each containing
approximately 10 12 Cesium atoms [10]. The schematic of the experimental
set-up is shown in Fig.18.7. The two cells are coated from inside with paraffin
coating which enhances the ground state coherence time up to 5 - 30m sec
depending on the density of atoms. The first and the second sample are initially
prepared in CSSs oriented along the magnetic field and against it, respectively.
This is achieved by optical pumping into the F = 4, mF = 4 and F = 4, mF =
-4, respectively for the two samples. The degree of orientation of around
97 - 98% brings us very close to a perfect CSS. Then the optical pumping is
switched off and a probe pulse with the duration of 0.5m sec is sent through both
samples. Its Stokes operator is measured using a polarizing beam splitter and
two balanced detectors. The differential photocurrent from the detectors is split
in two and its cos(Ot) and sin(Ot) power spectral components (S~~~s(O)) 2 and
(S~~fn (0)) 2 are measured with lock-in amplifiers. As follows from 18.30, the
total spectral variance at frequency 0 contains two contributions: one from the
shot noise of the probe light, another proportional to the atomic state variance:

6. 2 S2 = (S~~~s (0 )) 2 + (S~~fn (0 ) ) 2
8S; + '" { (F;in) + F;~n))2 + (F;in) + F;~n))2} . (18.31)
 Atomic continuous variable processing and light-atoms quantum interface 253

:::i 40 •
~
<l
•
of
en
::; 30
•
0..
Q)
..c
e
0..
Q)
:5 20
15
~
c:
CI'l
.~ 10
~
t5
Q)
0.. J x [1012]
U)
04-~--~~--~~--~~-,--~~--~-,
o 2 4 6
Collective spin of the atomic sample

Figure 18.8 Measured spectral variance for coherent spin state.

To establish the level of the measured quantity sbss

corresponding to the
CSS we have performed repeated measurements of .6. with freshly prepared
atomic samples in CSS. The results as a function of the mean collective atomic
spin Fx are shown in Fig.18.8. The value of .6. at Fx = 0 corresponds to the
shot noise of the probe. The linear growth of .6. with Fx is the characteristic
signature of the quantum (projection) spin noise. Classical fluctuations due to
technical noise of lasers, etc, result in a quadratic dependence, which can be
seen at higher values of Fx. The linear dependence together with nearly 100%
orientation of the sample allows us to take the linear fit to the data in Fig.18.8
as the CSS level corresponding to the right-hand side of the entanglement con-
dition (18.11). The coefficient K. = ~(0",n/4F A.6.)2 ~ 3 is estimated [12]
using 0" ~ >,2/27r as the resonant dipole cross section, , = 5M H z - the full
width of the optical transition, A = 2cm 2 - the probe beam cross section and
n;::::; 10 13 - the number of photons in the probe pulse with the power of 5mW.
With this estimate we obtain a resonable agreement with experimental results in
Fig.18.8. The measurement sequence aimed at the generation and verification
of the entanglement consists of the optical pumping pulses preparing samples
in a CSS, the entangling pulse preparing the samples in the entangled state
(pulse I) and the verifying pulse coming after the delay time and verifying the
entanglement (pulse II), Fig.18.7. These pulses have the same duration and
optical frequency as the probe pulse used for the CSS measurements. The pho-
254 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Collective spin of the atomic sample

Figure 18.9 Nonnalized spectral variance showing entangled state of the two cells.

tocurrents from the two pulses are subtracted electronically and the variance of
the difference, 81p R' is measured. The vanishing 81p R corresponds to two
repeated measurements on the total spin state of the two samples producing
the same results, i.e. to a perfect knowledge of both z and y total components
for the two samples, and therefore to a perfectly entangled state. In the experi-
ment the correlation between the entangling and verifying pulse measurements
is imperfect for several reasons. First, the two optical pulses possess quan-
tum(shot) noise which does not cancel out. Second, decoherence processes
change the spin state between the two measurements. The first imperfection
can be reduced by increasing the size of the atomic portion of 8 2 compared
to the shot noise of the probe. The results of measurements with the delay
of O.5m sec between preparation of the entangled state and its verification are
shown in Fig.18.9. The results are normalized to the CSS limit (the linear fit
in Fig.18.8). This limit thus corresponds to the unity level in Fig.18.9. The
raw experimental data for the entangled state are shown as stars. The values
below the unity level verify that the entangled state of the two atomic samples
has been generated and maintained for O.5m sec. For detailed derivation of
the degree of entanglement from the data in Fig.18.9 we refer to Ref. [10].
The degree of entanglement calculated operationally from the data without
additional assumptions is (35 ± 7)%. The degree of entanglement useful for
teleportation calculated using an additional, experimentally proven assumption
of the initially CSS for both samples is higher, (52±7)%. The predicted fidelity
 Atomic continuous variable processing and light-atoms quantum inteiface 255

of teleportation with such entangled samples is F = 55%, which is higher that

the classical limit of 50%. Higher degrees of atomic entanglement should be
possible with improved experimental conditions.

4.4 TELEPORTATION OF ATOMS ON LIGHT

Now we tum our attention to protocols of continuous quantum communica-
tion involving both atoms and light. Implementation of these protocols results
in transfer of the quantum state of one macroscopic spin onto the other one.
Shared entanglement of two auxiliary spins is necessary to achieve this goal.
It turns out that the EPR-entangled spins described by Eqs.(18.1O) are the right
kind of states for this task.
The e.m. field continuous quantum teleportation of Refs.[25, 30] relies on
the continuous entanglement of the EPR-type output of an optical parametric
oscillator (OPO) below threshold. The EPR-type correlations of the output
modes of the OPO and the EPR spin states are very closely related. In fact, the
most practical way to generate EPR spin state of light in the lab is to mix each
of the EPR modes (polarized, e.g., along the y-axis) with strong coherent fields
polarized along the x-axis, on polarizing beamsplitters. Then, for the Stokes
vectors of the output fields 8(1) and 8(2) we obtain:

(18.32)

where n is the number of photons in each of the coherent fields (for simplicity
assumed to be equal for both fields). That is, the spin components of the Stokes
vectors of the output states of light are given by the corresponding quadrature
operators of the OPO output. Since the latter obey relations of the form of
Eqs.(18.l0), we obtain relations Eqs.(18.1O) for the z- and y -spin components
of the output states of light.
As was pointed out in the original teleportation proposal [29], it is imperative
to be able to perform joint measurements on the quantum state to be teleported
and one of the EPR states. Off-resonant atom-photon interaction described
by Eq.(18.15) allows us to achieve this goal. The inter~ction leads to rotation
of the polarization of the field that is proportional to F z . Subsequent optical
polarization measurements provide the classical information in our protocols.
Let Alice pass her bright EPR beam 1 with the Stokes vector 8(1) through
the cell containing polarized atomic vapor along the z-axis. The unitary time
evolution operator of Eq.(l8.15) in the case of small spin fluctuations in F and
256 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

S(I)(out) :::::; s(1)(in) + anI p(in)

Y Y 2 Z ,

PJout) :::::; PJin) + (aN F)S~I)(in) . (18.33)

Memory -
Atoms

state
Actuator -
phase and amplitude
modulator

Figure lB. 10 Outline of the scheme to teleport atomic quantum state on light.

These equations for interaction of the collective atomic spin with light re-
semble the beam splitter relations used for teleportation of a mode of electro-
magnetic field in Refs.[25, 30]. An insightful analysis of why a linear device
of a beamsplitter allows to perform joint measurements necessary for quantum
teleportation has been given by Vaidman and Yoran [31].
The nonlinear atom-light interaction of Eq.(18.15) is analogous to the QND-
type interaction for the optical quadratures rv XaXb . Unlike the beam splitter,
which mixes both the X and the Y quadratures of the two input beams in
the same fashion, interaction rv XaXb mixes only one pair of quadratures (Ya
and Yb), leaving the other pair (Xa and Xb) unchanged. Relations (18.33)
correspond to a 50:50 beam splitter of the optical teleportation when

(18.34)

Subsequent (destructive) measurement of S~I)(out) provides Alice with half

of the classical information needed for implementation of "feed-forward" on
Bob's bright EPR beam to complete the quantum teleportation. The other
half must come from the measurement of PJout). Although it is possible
 Atomic continuous variable processing and light-atoms quantum interface 257

in principle to use a completely destructive measurement (e.g., by means of

photoionization), experimentally it may be more convenient to employ another
QND measurement using an auxiliary coherent pulse. It will be convenient to
rotate the spin by 7r /2 around the x-axis, so that propagating along the z-axis
pulse will measure PJout). The corresponding transformation is
S(coh) (out)
z

(18.35)

The first term is negligible if ncoh/nl » 1. After Alice sends the results of
(l)(out) (coh)(out)
measurements of Sy and Sz to Bob, reconstruction of F m in
A A A (. )

§ (2) is achieved by Bob rotating the latter around the z-axis by the angle (PI =
_ _2_S(1)(out)and around the y-axis by the angle ¢2 = _ 2y'nl S(coh)(out)
yfnln2 y n co hy'Ti2 z
. The rotations are equivalent to displacements because the angles are small.
We obtain using Eqs.(18.1O)

S(2)(out)
y

S(2)(out) (18.36)
z

The last terms in these expressions are due to the extra noise introduced by
the imperfect EPR spin state. This noise goes to zero as r goes to infinity.
The factor of square root of two in the last terms corresponds to the "quantum
duty" of the quantum teleportation [25], as seen explicitly in the case of zero
parametric gain, r = o. Eqs.(18.36) show that the Stokes vector §~ut is
identical to the initial collective spin vector of Alice's atoms when

(18.37)

The teleportation protocol described above may serve as a read-out for

quantum memory with atoms as a memory cell. On the other hand, if the
process of mapping of the Stokes vector §~ut onto another atomic spin [3]
follows the described atom-to-light teleportation, atom-to-atom teleportation is
achieved.

4.5 MAPPING OF AN UNKNOWN QUANTUM STATE

OF LIGHT ON ATOMS (QUANTUM MEMORY)
Suppose Alice and Bob have in their disposal two atomic ensembles entan-
gled as described in Section 3.3. Their mean polarizations are (PAx) = (PBx)
and the other two spin projections of the ensembles are entangled, so that
258 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

FAy - FEy = 0 and FAz + FEz = O. If Alice receives an unknown state of

light with a polarization state described by the non-zero mean Stokes parameter
Bux and zero-mean Buz, Buy, she can via a sequence of classical measurements
followed by a classical signal transmission map this state onto Bob's atomic
"memory cell" .
The protocol proceeds as follows. The unknown state of light is sent through
Alice's ensemble along the z axis and the B~y of the transmitted light is
measured by the detector D Al producing the signal

Alice's atomic state after this measurement becomes

FAy + SUz
'" I '" A/ " '"
FAz = FAz, FAy =

Next Alice's atomic state is rotated around the direction of the mean spin,
x , so that F~z -+ F~y and F~y -+ - F~z. A strong coherent beam with the
mean polarization along x axis is then sent through Alice's ensemble along the
z axis and its y Stokes parameter is measured by the detector D A2 with the
result

In the above equations n, nc are the mean photon numbers of the unknown
state of light and the strong coherent beam respectively, and we assume n / nc «
1.
The mapping of the unknown state of light onto Bob's ensemble is now
completed by displacing the z, y component of his spin by dAI, - :c dA2 to
obtain

F~z FEz + Buy + FAz = Buy,

F~y FEy - Buz - FAy = -Buz,
Thus, we have used entanglement between two macroscopic atomic spins to
map a polarization quantum state of a light pulse onto atoms, with potential for
long storing times.

4.6 COMMUNICATION VIA TELEPORTATION

BETWEEN ATOMIC SAMPLES
We now describe a protocol which performs the direct teleportation of the
quantum state of Alice's collective spin onto Bob's collection of atoms without
using light as an intermediate carrier of the quantum state. Suppose we have
 Atomic continuous variable processing and light-atoms quantum interface 259

two macroscopic spin systems (Alice's and Bob's) in the initial states given
by collective spin operators FA, FB with mean polarizations (FAx) = (FBx)
and with the other two projections FAy, FBy,FAz, FBz. We also have at our
disposal a source of the EPR spin states of light as described above. The
Stokes' vector of the Alice's EPR beam is rotated by 7r 12 around the x-axis
so that the Stokes operators are BAz = -BBy, BAy = -BBz. The protocol
begins with Alice sending her EPR beam along the z-axis and measuring its
y -Stokes parameter with the detector D AI, and Bob sending a coherent x-
polarized pulse B]3h containing ne photons along the y-axis and measuring its
z-Stokes parameter with the detector D BI. The resulting atomic states of Alice
and Bob are described by the following operators:
I ,.. .... ,

= = FAy + SAz,
A A ,..

FAz FAz, FAy (18.38)

In these equations we combined conditions nl = n2 = nand (18.34),(18.37)

to obtain unity coupling coefficients between F and Bprojections. The Stokes
parameters measured by detectors D Al and D B1 are

(18.39)
(18.40)

We assume nine « 1. While the above measurements are performed, the

second EPR beam sent by Alice to Bob begins its journey along the quantum
channel. Next Alice sends a coherent x-polarized probe containing ne photons
along y axis onto the detector D A2. The detector measures

(18.41)

Alice now sends D AI, .!!.

nc
D A2 to Bob along a classical channel. When Bob
receives the EPR beam from Alice he sends it along z axis onto the detector
D B2 which reads

+ FBz
A. " A/
DB2 = SBy (18.42)

After that the state of Bob's atoms is

(18.43)

To complete the teleportation Bob now rotates his atomic state. We use
displacements instead of rotations to simplify the expressions. Bob's state is
260 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

displaced along z by - D A2 - D B2 and along y by D Al - D BI. The final state

of his atoms, according to Eqs.(18.39,18.41,18.42,18.43), is described by

FBz
te1e =
-u
F Bz - DA2 - DB2 = -FAy
-
(18.44)
FBy
te1e =
~"
F By +DAI- DEl = FAz
A

and the teleportation of the unknown state of the Alice's collection of atomic
spins onto Bob's atoms is proven (within a rotation of 1r around the x-axis). In
the above equations we assumed perfect entanglement between the EPR beam
components (r -* 00 in Eqs.(18.1O». We have also used the fact that the
measured values FBz,By of operators FBz,By are obtained with the QND-type
Hamiltonian, and therefore FBz,By - FBz,By = 0 (same is true for 8 13;).

4.7 QUANTUM STATE SWAPPING BETWEEN TWO

ATOMIC SYSTEMS
We now describe a protocol which exchanges initial quantum states of two
collections of atomic spins. Suppose Alice's and Bob's spin samples are in the
initial state given operators FA) FB with mean polarizations (FAx) = (FBx).
We again have at our disposal the EPR source of light as described above, with
Stokes operators 8 z ) 8 y for one of the beams, and -8z ) 8 y for the other. Both
EPR beams are mixed with strong beams containing equal photon numbers,
nl = n2 = n, polarized along the x -axis in a way similar to the previous
section. One of the EPR beams is used for a joint measurement on the z spin
components of both samples by sending it through both of them along the z
-axis. The resulting states of atomic samples are:

A ;., I A.

FAz; FBz = FBz) (18.45)

,., ""'I " '"
FAy + Sz; F By = FBy + Sz;
and the state of the beam is

(18.46)

In the above equations we assumed conditions (18.34),(18.37) to be fulfilled.

Next, the second EPR beam shifted in phase by 1r /2 is sent along the y-axis
of the two atomic samples to perform a joint measurement on the y spin
components. The beam is transformed by the phase shift and the change of the
direction in the following way - 8z) - 8y -* 8;) 8~. The evolution operator in
this new coordinate system is (; = exp (-ia8~Fy). We obtain
 Atomic continuous variable processing and light-atoms quantum inteiface 261

F~z = FAz - s~ = FAz + Sy; fr~z = FBz + Sy, (18.47)

FAy FAy = FAy + Sz; FBy = FBy = FBy + Sz,
" " ,. , " " " " " , A. "

and

sr'z "/

-Sz = -Sz + FAy + FBy = FAy + FBy + SZ

A "/ ,.., A. .... ,..

(18.48)
sr'
y -
AI
-Sy = -Sy
A

Here we used Eqs. (18.45, 18.46). After interacting with the atoms the two
EPR beams are detected by photodetectors D 1 , D 2 • The Stokes parameter S~
(18.46) is measured for the first beam and the Stokes parameter S~' (18.48)
for the second. The results of the measurements are used to rotate the atomic
spins in order to achieve the teleportation. The projections F~z and F~z are
displaced by the value S~ obtained from D 1 , and the projections F~y and F~y
are displaced by the value S~/. The results are

.... 11 I ,..
ptele = Fz - Sy = -FBz, (18.49)
Az
FAy
te1e = fr; - S~' = -FBy.

Similarly for the other atomic system

FABz
swap
= - Az
FA
(18.50)
FAswap
By = -
FA
Ay
Eqs.(18.49),(18.50) prove that the initial collective quantum states of the two
samples have been exchanged.
A different method for atomic continuous variable teleportation, using co-
herent light and entangled atomic samples, was proposed in Ref. [12].

5. SUMMARY
One of the most striking features of quantum information processing with
continuous variables is the relative simplicity of realization of the light-atoms
quantum interface. As opposed to the case of discrete variables (single pho-
tons and atoms) such interface for continuous variables does not require strong
coupling via cavity QED. Free space interaction of light with atomic ensem-
bles provides enough coupling for quantum state exchange and entanglement
of continuous variables. We have presented two approaches to continuous
quantum information processing, one based on resonant interaction, the other
262 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

one relying on off-resonant atom-light interaction. Experiments within both

approaches, performed in the last three years, resulted in several advances in
the field of quantum information with continuous variables: mapping of non-
classical (squeezed) states of light onto collective atomic spin [5], preparation
of squeezed spin states of atoms via quantum non-demolition measurements
[6], demonstration of Einstein-Podolsky-Rosen entanglement of two distant
atomic samples [10], and, more recently, demonstration oflong-lived quantum
memory effect for non-classical states of light [32].
One can envision extending this work to more complex algorithms of con-
tinuous quantum information, such as, e.g. quantum search and quantum error
correction for continuous varIables [25,33] and quantum communication be-
tween distant atomic samples [11, 12, 34]. Another promising direction of
research is to implement real-time feedback on continuous atomic variables,
aimed, e.g., at continuous error correction in the system. Such continuous
(weak) quantum feedback can be often advantageous compared to instanta-
neous (strong) measurements [35].

Acknowledgments
We are indebted to many of our collaborators who contributed to the work
reviewed in this chapter, particularly to L. Mandel, N. P. Bigelow, I. Cirac,
L.-M. Duan, B. Julsgaard, J. Hald, A. Kozhekin, K.Mj1jlmer, J.L. Sj1jrensen, and
P. Zoller.

Appendix: A
Let us assume that we have a running-wave cavity with a single spatial mode
of the electromagnetic field interacting with a collection of N atoms. First,
we consider an idealized case of alkali-like atoms with zero nuclear spin. The
interaction Hamiltonian desribing the atoms-field systems can be written as
2 N
HI = Ii L L1 9 exp(i Wii +1 zJL)ai ah+1 + h.c., (l8.A.I)
.
~=1 JL=
C

where a+, a_ are annihilation operators for the right- and left-hand polarized
components of the field. 9 is the coupling constant, zJL is the position of the
p,-th atom in the sample. For large detuning ~ » 'Y the dynamics of the system
can be described by the effective interaction Hamiltonian (see, e.g. [36])

~ ~ g2 + ~ ~ ~ 1 ~
He!! = Ii ~ 1. ai aiJii = M2(SzJz + 4,'nN). (l8.A.2)
i=l 'lq +D
The second term is spin independent and can be omitted. He!! as it is written
above is not Hermitian. In order to make the Hamiltonian to be Hermitian, one
 Atomic continuous variable processing and light-atoms quantum interface 263

must add terms that describe interaction with the external system responsible
for relaxation (in our case, the bath of spontaneous emission modes of the atoms
serves as such external system). However, for our purposes, we can neglect this
issue and simply take the Hermitian part of the Hamiltonian (18.A.2). From
the experimental point of view, a very interesting situation is of free-space
atom light coupling. One can show [26, 12] that the effective Hamiltonian
still has the form of Eq. (18.A.2). The anti-Hermitian part of the Hamiltonian
(18.A.2), allows us to write down the expression for the coupling constant 9
through directly measurable quantities: photon-atom atomic cross-section (J"
and the transverse cross-sectional area of the light beam A. The on-resonance
absorption length la is given by ((J" N / AL ) -1. Using the equations of motion for
the electric field that Hamiltonian (18.A.2) results in, we obtain the following
expression for La:
C'Y
la = 2g2N
Comparing the two expressions, we find g2 = ((J"c'Y) / (2LA). If instead of the
Dl transition we consider the D2 one, we would find a similar result, with an
extra factor of ~ present in the expression for the Hamiltonian. The unitary
evolution (; operator is obtained by exponentiating the Hermitian part of the
Hamiltonian: (; = exp( -ifhL/c).
The isotopes of interest of the alkali atoms have non-zero nuclear spin.
In this case we need to express the "average" value of the z-component of
the electronic angular momentum through the total angular momentum of the
atom. The following relation can be obtained [37]:
javg = F F(F + 1) + 3/4 - J(J + 1) = ±F 1
Z Z 2F(F+1) z(I+1/2)
Combining the formulas, we arrive at the Eq.(18.15) for the unitary evolution
operator. It is interesting to note that a more rigorous derivation for atoms
possesing nuclear spin [38] gives the same result as the intuitive procedure of
averaging electronic spin.

References
[1] A. S. Parkins and H. J. Kimble, J. Opt. B 1, 496 (1999); S. Bose, P. L.
Knight, M. B. Plenio, and V. Vedral, Phys. Rev. Lett. 83, 5158 (1999).
[2] A. Kuzmich, K. Mjljlmer, andE. S. Polzik, Phys. Rev. Lett. 79, 4782 (1997).
[3] A. E. Kozhekin, K. Mjljlmer, E. S. Polzik, Phys. Rev. A 62, 033809 (2000).
[4] E. S. Polzik, Phys. Rev. A 59,4202 (1999).
[5] J. Hald, J. L. Sjljrensen, C. Schori, and E. S. Polzik, Phys. Rev. Lett. 83,
1319 (1999).
264 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[6] A. Kuzmich, L. Mandel, and N. P. Bigelow, Phys. Rev. Lett. 85, 1594
(2000).
[7] Z. Y. Ou, S. F. Pereira, 1. H. Kimble, and K. C. Peng, Phys. Rev. Lett. 68,
3663 (1992).
[8] Ch. Silberhorn, P. K. Lam, O. Weiss, F. Konig, N. Korolkova, and G.
Leuchs, Phys. Rev. Lett.86, 4267 (2001).
[9] C. Schori, J. L. Sj/lrensen, E. S. Polzik, quant-phl0205015.
[10] B. Julsgaard, A. Kozhekin, and E. S. Polzik, Nature 413,400 (2001).
[11] A. Kuzmich and E. S. Polzik, Phys. Rev. Lett. 85, 5639 (2000).
[12] Lu-Ming Duan, J.I. Cirac, P. Zoller, and E. S. Polzik, Phys. Rev. Lett. 85,
5643 (2000).
[13] F. T. Arecchi, E. Courtens, R. Gilmore, and H. Thomas, Phys. Rev. A 6,
2211 (1972).
[14] I. Cirac, unpublished.
[15] M. Kitagawa and M. Ueda, Phys. Rev. Lett. 67, 1852 (1991); Phys. Rev.
A 47, 5138 (1993).
[16] D. J. Wineland, J. J. Bollinger, W. M. !tano, and F. L. Moore, Phys. Rev.
A 46, R6797 (1992);
[17] D. J. Wineland, J. J. Bollinger, W. M. !tano, and D. J. Heinzen, Phys. Rev.
A 50, 67 (1994).
[18] G. S. Agarwal and R. R. Puri, Phys. Rev. A, 41,3782
[19] L.-M. Duan, G. Giedke, J. I. Cirac, P. Zoller, Phys. Rev. Lett. 84, 2722
(2000).
[20] J. Hald and E. S. Polzik. Special Issue: Quantum Coherence and Entan-
glement. Journal of Optics B: Quantum and Semiclassical Optics, 3, S83
(2001).
[21] F. Laloe, M. Leduc, and P. Miguzzi, J. Phys. 30, 277 (1969).
[22] J. L. S!Ilrensen, J. Hald, and E. S. Polzik, Phys. Rev. Lett. 80, 3487 (1998).
[23] W. Happer, Rev. Mod. Phys. 44, 169 (1972).
[24] W. Happer and B. S. Mathur, Phys. Rev. Lett. 18,577 (1967).
[25] S. L. Braunstein, Nature, 394, 47 (1998); S. Lloyd and S. L. Braunstein,
Phys. Rev. Lett. 82, 1784 (1999).
[26] A. Kuzmich, N. P. Bigelow, and L. Mandel, Europhys. Lett. A 42,481
(1998).
[27] K. M!Illmer, Eur. Phys. J. D 5, 301 (1999); Y. Takahashi, K. Honda, N.
Tanaka, K. Toyoda, K.Ishikava, and T. Yabuzaki, Phys. Rev. A 60, 4974
(1999).
 Atomic continuous variable processing and light-atoms quantum interface 265

[28] J. -Ph. Poizat, J. -F. Roch, and P. Grangier, Ann. Phys. Fr. 19,265 (1994).
[29] C. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. K.
Wooters, Phys. Rev. Lett. 70, 1895 (1993).
[30] A. Furusawa, J. L. Sj2!rensen, S. L. Braunstein, C. A. Fuchs, H. J. Kimble,
and E. S. Polzik, Science 282, 706 (1998).
[31] L. Vaidman and N. Yoran, Phys. Rev. A 59, 116 (1999).
[32] C. Schori, B. Julsgaard, J. L. Sj2!rensen, and E. S. Polzik, quant-
phl0203023.
[33] S. L. Braunstein, Phys. Rev. Lett. 80, 4084 (1998).
[34] P. van Loock and S. L. Braunstein, Phys. Rev. Lett. 84, 3482 (2000).
[35] C. Ahn, A. C. Doherty, and A. J. Landahl, quant-phlOl10111.
[36] M. Brune, S. Haroche, J. M. Raimond, L. Davidovich, and N. Zagury,
Phys. Rev. A 45,5193 (1992).
[37] L. D. Landau and E. M. Lifshitz, Quantum mechanics: non-relativistic
theory (Pergamon Press, New York, 1991).
[38] W. Happer and B. S. Mathur, Phys. Rev. 163, 12 (1965).
IV
LIMITS ON QUANTUM INFORMATION
AND CRYPTOGRAPHY
Chapter 19

LIMITATIONS ON DISCRETE QUANTUM

INFORMATION AND CRYPTOGRAPHY

Samuel L. Braunstein
Informatics, Bangor University, Bangor LL57 I UT, United Kingdom
[email protected]

Arun K. Pati
Institute of Physics, Bhubaneswar-75I005, Orissa, INDIA
Theoretical Physics Division, BARe, Mumbai, INDIA
[email protected]

Abstract In this chapter we briefly discuss some of the limitations to processing discrete
quantum information, such as no-cloning, no-complementing and no-deleting.
The no-cloning principle, in particular, has important practical implications; for
example, quantum cryptography uses it to guarantee security against eavesdrop-
pers.

1. INTRODUCTION
Quantum information differs from classical information in a variety of ways.
Knowing these differences is crucial to understanding the ultimate limits to our
ability to store, process and extract useful information from quantum states.
These are some of the primary tasks for any protocol in information theory.
Among the key differences between classical and quantum information are are
no-cloning [1,2], no-complementing [3] and no-deleting [4]. These limitations
may appear as hurdles, but with the right insight they can be turned into practical
applications. For example, the no-cloning principle lies at the heart of quantum
cryptography.

269
SL Braunstein andA.K. Pati (eds.), Quantum Information with Continuous Variables, 269-276.
© 2003 Kluwer Academic Publishers.
270 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Firstly, one must appreciate the fact that (or the lack of) knowledge about a
quantum state plays an important role. Suppose we are given an arbitrary qubit
I1/;) = ala) + (311) , (19.1)
where a and (3 are complex numbers. Because one can ignore the overall
phase factor we may always choose a to be real. Thus, an arbitrary qubit
may be represented by a point on a two-dimensional sphere with the two real
parameters Band ¢, where a = cos(B/2) and (3 = sin(B/2)exp(i¢) with
a S B S 11" and a S ¢ S 211".
If a qubit is prepared by a third party, then it may be completely unknown to
us. Our lack of knowledge about the preparation procedure translates into our
inability to pinpoint the state on this sphere. In principle, to precisely determine
such a completely unknown state might require a vast amount of information
corresponding to the many bits needed to specify the two parameters B and
¢ [5]. If we had prepared the qubit ourselves then we would have known its
location and hence we would not have lacked any information. In studying
the fundmentallimitations to processing quantum information this preparation
knowledge is important.
More generally, a state may be prepared from a limited alphabet of possible
choices. In particular, the state P is assumed to be selected from one of a set
of states {Pi} each of which occur with respective (and known) probabilities
Pi. If the alphabet (consisting of the set {Pi} and the associated probabilities
Pi) is known, then only the specific choice from this set remains unknown. For
example, for continuous variables, we often consider a restricted alphabet of
the set of coherent states for representing information.

2. THE NO-CLONING PRINCIPLE

In principle classical information can be copied perfectly. But what about
information stored in quantum states, such as in the polarization of a photon
or the spin of an electron? This interesting question was raised by Wootters,
Zuerk [1] and Dieks [2]. They discovered that the linearity of quantum theory
forbids the perfect copying of an unknown quantum state.
Suppose, we are asked to copy an unknown qubit. A cloning machine would
start with the original state, a second standard state which will be transformed
into the copy and possibly an ancilla state. Thus, the cloning machine will be
described by an operatorUc which mapsUc : til (2;/1i20ti3 -+ ti10ti20ti3,
with the three Hilbert spaces referring to original, copy and ancilla respectively.
For a given original state its action will be
(19.2)
where I~) is the standard (target) state, and IA) is the initial and IA'lj;) is the
final state of the ancilla.
 Limitations on discrete quantum information and cryptography 271

How would this machine act on orthogonal inputs 10) and II)? We would
expect them to transformation according to

10hlEhiAh -+ 10hlohlAo)s
11hlEhiAh -+ 11hllhlAlh . (19.3)
However, these transformations are enough to specify the actions of our
cloner machine on an arbitrary qubit I¢) as input. In particular, by the linearity
of quantum theory we have

l¢hIEhIA)s -+ alOhlOhlAo)s + ,811hllhl Alh

=1= [a210hI0)2 + ,8211hll)2
+a,8(IOhllh + 11hI Oh)]IA1jJh
l¢hl¢hIA1jJ)s. (19.4)

Thus, the cloning machine must fail for completely unknown states. If we
knew the state we could rotate it back to 10) or 11) and perform the cloning
operation.
For a restricted alphabet of quantum states such as two non-orthogonal
states from a set S = {¢d, k = 1,2, ... K, the proof of no-cloning theorem
follows just from unitarity of the evolution. Thus the deterministic cloning
of non-orthogonal states is impossible [8, 9, 10]. However, one can give up
either exactness of the clone or determinism of the process in the cloning
operation. It may be shown that approximate cloning is possible using a
unitary operation or that exact cloning is possible with a unitary operation plus
a suitable measurement outcome.
The possibility of producing approximate copies of a quantum state has been
considered by Buzek and Hillery [11]. Here, one demands that all inputs should
be copied equally well (so the performance should have a universal character)
and that the fidelity of cloning should be maximal. For example, the simplest
universal cloner of a qubit making 1 -+ 2 copies can attain an optimal [12]
fidelity F = 5/6. An explicit circuit for achieving this universal cloning has
been determined [13]. More generally, universal cloning has been considered
for 1 -+ M copies [14,15]. The fidelity goes as F1--tM = (2M + 1)/(3M) and
has been shown to be consistent with a no-superluminal-signaling condition
[16, 17]. Further, the fidelity of universal cloner for qudits (d-dimensional
generalizations ofqubits) for N -+ M copies goes as FN--tM(d) = [N(d-l)+
M(N + 1)]/[M(N +d)J. Notice that when d -+ 00 then FN--tM(OO) = N/M.
This was also obtained in Ref. [18] while discussing quantum information
distribution and continuous variable quantum cloning.
In addition, it is possible to design probabilistic cloning machines [19]. A
state selected from a set of non-orthogonal, linearly-independent states S =
{¢d, k = 1,2, ... ,K, can be exactly cloned with a finite probability of
272 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

success. For any two non-orthogonal pairs (i, j) the success probability is
!(Pi +Pj) ~ 1/(1 + I('l/Ji I'l/Jj) I)· Further, it was shown that linearly independent
quantum states can evolve into a linear superposition of multiple copies with a
branch for failure [20]. Probabilistic and deterministic cloning transformations
were shown to be special cases of this operation. A cloning machine that
interpolates between approximate and exact clones has also been proposed
[21].
The impossibility of cloning has been used to argue the impossibility of
completely determining the wavefunction of a single unknown quantum system
[9]. However, we might ask to what extent can one determine the state given a
finite number of copies. This is related to optimal state estimation. For the case
of qubits the best fidelity for recreating a state based on measurement results
from N copies [6] is F = (N + 1)/(N + 2). For ad-dimensional qudits this
optimal fidelity of state estimation [7] is F = (N + 1) / (N + d). Thus, given a
single qubit we can extract out only ~ 's worth of the fidelity. One can see that
if we are given an infinite number of copies will the state estimation fidelity
approaches unity.
Finally, one may ask whether it is possible to clone a quantum state by
supplementing extra information (either quantum or classical)? Classically we
can always determine the state, so no further information is necessary. Surpris-
ingly, however, for cloning a quantum system the supplementary information
must be sufficient to manufacture the clone! This provides us with a stronger
version of the no-cloning theorem [22].

3. NO·COMPLEMENTING PRINCIPLE
We know that a classical bit 0 can be complemented to 1 or vice-versa. In
the quantum world, if a qubit is in a computational state 10) or 11) it can also be
complemented, for example, by applying the Pauli operator ax. However, is it
possible to complement an arbitrary qubit? If we could complement a qubit by
a physical operation, then we would take I'I/J) to I~), where ('l/JI~) = O. So the
complementing operation should be defined by

alO) + {311) --+ a*ll) - {3*IO) . (19.5)

Note that this is an antiunitary operation. An anti-unitary operation though a

positive map is not a completely positive (CP) map. This means that it cannot
be implemented by any physical operation. This shows that it is impossible
to complement an unknown qubit [3]. Nevertheless, it is possible to design
an approximate and universal NOT (or complementing) gate. The fidelity
of producing the complemented qubit is exactly equal to the state estimation
fidelity of a single qubit, i.e., ~.
 Limitations on discrete quantum information and cryptography 273

4. NO-DELETION PRINCIPLE
Classically we may delete one copy against others, uncopying it in a perfectly
reversible manner. However, in the quantum world this is not the case. This
limitation complements no-cloning principle which says that given two (or
more) copies of an unknown quantum state one cannot delete a copy acting
jointly on both the copies [4].
If we could have a quantum deleting machine it would act on two initially
identical qubits in some unknown state I~) and an ancilla in some initial state
IA). This machine is supposed to delete one of the copies and replace it with
some standard qubit IL;). The quantum deleting therefore would yield

(19.6)

where IA1jJ) is the final state of the ancilla. The only solution to this equation is to
swap the second and third qubits. However, this obviously should be excluded
for any sensible definition of a deleting machine, since the information has only
been moved to another location.
Similar to the no-cloning theorem, deletion of a copy from two non-orthogonal
states is impossible [22]. Recently, it has been shown that probabilistic deletion
of copies of linearly independent quantum states is allowed [23, 24, 25]. How-
ever, it remains an open question how to design a universal and approximate
quantum deleting machine.
Next we discuss how these limitations can be turned around into positive
applications such as cryptography.

5. CRYPTOGRAPHY
Cryptography is the art of secretely sending messages from sender to receiver.
Usually the sender and receiver establish a key and using it they can encrypt
and decrypt a message. The security of a cryptic message thus depends on how
well the secrecy of the decryption key is maintained.
The first provably secure encryption scheme was invented by Vernam in
1917 and is ofen called the one-time pad. Here a secret key of random numbers
is added (using modulo arithmetic) to a simple numerical representation of the
message. For each number from the message a new number from the key must
be used. It can be shown that in the absence of knowledge of the key, the
encoded message carries no information about the original. A subtle point in
this result is that the key must not be reused, hence the name one-time pad.
Because of this the main problem with this encoding scheme is the requirement
for distributing a copy of the (long) key to the receiver without its falling into
hostile hands.
It should be noted that one way around the key-distribution problem has
been "solved" by public key cryptosystems. The most well-known is the RSA
274 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

scheme invented by Rivest, Shamir and Adelman [26]. Here the encoding
key is placed in the public domain, whereas the decoding key is kept secret.
Security relies on the computational difficulty of extracting the decoding key
from knowledge of the encoding key alone. It is generally accepted that any
procedure for determining the former relies on factoring the large numbers
appearing in the latter. Because factoring is apparently computationally in-
tractable, the RSA scheme is believed to provide suitable security. Of course,
all this would change if scalable quantum computers capable of running Shor's
algorithm became available.
Public key cryptography solves the key distribution problem at the expense
of absolute security. To guarantee absolute security we must go back to the
Vemam cipher and find a way of guaranteeing the security of distribution of
the key. Here is where quantum mechanics comes in. Quantum cryptography
is really a method of creating pairs of correlated keys between sender and
receiver, without the need for transporting these keys through the intervening
space. Quantum cryptography is sometimes more precisely called quantum
key distribution.
In quantum key distribution, the key is created through a protocol based on
the transfer of quantum states between sender and receiver (Alice and Bob,
respectively). Suppose an eavesdropper (Eve), who wishes to circumvent the
security of this protocol, couples an ancilla to the states transmitted between
Alice and Bob and evolves as

l1fJlhl A h --+ l1fJlhlAlh

11fJ2h IAh --+ 11fJ2h IA 2)2 . (19.7)

Her intention is to extract information about the states sent by measuring the
final state of the ancilla and distinguishing the outcomes. However, unitarity of
this coupling implies (1fJ111fJ2) = (1fJ111fJ2) (AIIA2). Thus, if the protocol utilizes
non-orthogonal states then the final states of the ancilla become identical. As a
consequence if Eve is to extract information, she can only do so by disturbing
the state transmitted. In essence, this is really nothing more than the no-cloning
principle at work.
For qubit-based quantum cryptography there have been two basic schemes.
Those involving the sending of states from non-orthogonal bases, such as the
original protocol Bennett and Brassard invented in 1984 (called BB84) [27],
and those relying on sharing entanglement between sender and receiver, such
as Ekert's scheme [28]. We shall not discuss the potential advantages of either
scheme here. Below we give a brief discussion to the BB84 protocol.
First Alice generates a random sequence of O's and 1 'So Then she randomly
chooses between two different bases to encode the information. In one basis,
she encodes 0 as 10) and 1 as 11). In the second bassis, she encodes 0 as ~ (10) +
 Limitations on discrete quantum information and cryptography 275

11)) and 1 as ~ (10) - 11)) . (Note that these two bases are incompatible.) At
the receiving end, Bob randomly chooses between these bases to make his
measurements. Only when Alice and Bob happen to be using compatible bases
will their bits be correlated. Next, Alice and Bob communicate over a public
channel. They reveal their choice of bases for each round, thus discovering
during which rounds their results should agree. Of these they make a small
selection to "sacrifice" in order to check the fidelity of this correlation. Any
deviation from the ideal is a signal of potential information gained by Eve.
When they are happy that Eve has been excluded, they may use the remaining
(unrevealed) bits from their compatible choice of bases. Thus, this protocol
has managed to create, in separated locations, pairs of random numbers - the
key.
There has been tremendous developments in the area of quantum cryptogra-
phy. For example, to exclude Eve, one may use privacy amplification or even
quantum privacy amplification in order to improve the robustness of the proto-
col [29]. Finally, an absolute prove of the security of BB84 has been given by
Shor and Preskill [30]. For a recent account on discrete quantum cryptography
see Ref. [31].

References
[1] W. K. wootters and W. H. Zurek, Nature (London) 299,802 (1982).
[2] D. Dieks, Phys. Lett. A 92,271 (1982).
[3] V. Buzek and M. Hillery, Phys. Rev. A 60, R2626 (1999).
[4] A. K. Pati and S. L. Braunstein, Nature 404, 164 (2000).
[5] R. Jozsa, in Geometric Issues in the Foundations of Science, Eds. S.
Huggett et aI, Oxford University Press, 1997.
[6] S. Massar and S. Popescu, Phys. Rev. Lett. 74, 1259 (1995).
[7] R. Derka, V. Buzek and A. Ekert, Phys. Rev. Lett. 80, 1571 (1997).
[8] H. P. Yuen, Phys. Lett. A 113, 405 (1986).
[9] G. M. D'Ariano and H. P. Yuen, Phys. Rev. Lett. 76, 2832 (1996).
[10] H. Barnum et aI, Phys. Rev. Lett.76, 2818 (1996).
[11] V. Buzek and M. H. Hillery, Phys. Rev.A54, 1844 (1996).
[12] D. Brub et aI, Phys. Rev. A 57, 2368 (1998).
[13] V. Buzek etaI, Phy. Rev. A. 56, 3446 (1997).
[14] N. Gisin and S. Massar, Phys. Rev. Lett. 79, 2153 (1997).
[15] M. Keyl and R. F. Werner, J. Math. Phys. 40, 3283 (1999).
[16] N. Gisin, Phys. Lett. A 242, 1 (1998).
[17] S. Ghosh, G. Kar and A. Roy, Phys. Lett. A 261, 17 (1999).
276 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[18] S. L. Braunstein, V. Buzek and M. Hillery, Phys. Rev. A 63, 052313

(2001).
[19] L. M. Duan and G. C. Guo, Phys. Rev. Lett. 80, 4999 (1998).
[20] A. K. Pati, Phys. Rev. Lett. 83, 2849 (1999).
[21] A. Cheftes and S. M. Barnett, Phys. Rev. A 60, 136 (1999).
[22] R. Jozsa, quant-phl0204153.
[23] Y. Feng, S. Zhang and M. Yim, Phys. Rev. A 65,042324 (2002).
[24] J. Feng, Y. F. Gao, J. S. Wang, and M. S. Zhan, Phys. Rev. A 65,052311
(2002).
[25] D. Qiu, Phys. Rev. A 65, 052303 (2002).
[26] R. Rivest, A. Shamir and L. Adelman, MIT Laboratory for Computer
Science, Technical Report, MITILCSffR-212 (January 1979).
[27] C. H. Bennett and G. Brassard, in Proc. IEEE Int. Conference on Com-
puters, Systems and Signal Processing (IEEE Press, Los Alamitos, Calif.
1984), p. 175.
[28] A. Ekert, Phys. Rev. Lett. 67, 661-663 (1991).
[29] D. Deutsch, A. Ekert, R. Jozsa, C. Macchiavello, S. Popescu and A.
Sanpera, Phys. Rev. Lett. 77, 288 (1996).
[30] P. W. Shor and J. Preskill, Phys. Rev. Lett. 85, 441 (2000).
[31] N. Gisin, G. Ribordy, W. Tittel and H. Zbinden, Rev. Mod. Phys. 74,145
(2002).
Chapter 20

QUANTUM CLONING
WITH CONTINUOUS VARIABLES

Nicolas J. Cerf
Ecole Poly technique, CP 165, Universite Libre de Bruxelles, 1050 Brussels, Belgium
[email protected]

1. INTRODUCTION
Quantum information theory has developed dramatically over the past de-
cade, driven by the prospects of quantum-enhanced communication and com-
putation systems. Among the most striking successes, one finds for example
the discovery of quantum factoring, quantum key distribution, or quantum tele-
portation. Most of these concepts were initially developed for discrete quantum
variables, in particular quantum bits, which have now become the symbol of
quantum information. Recently, however, a lot of attention has been devoted to
investigating the use of continuous-variable systems in quantum informational
or computational processes. Continuous-spectrum quantum variables, for ex-
ample the quadrature components of a light mode, may be easier to manipulate
than quantum bits. It is actually sufficient to process squeezed states of light
into linear optics circuits in order to perform various quantum information pro-
cesses over continuous variables [1]. As reported in the present book, variables
with a continuous spectrum have been shown to be useful to carry out quantum
teleportation, quantum entanglement purification, quantum error correction, or
even quantum computation.
In this Chapter, the issue of cloning a continuous-variable quantum system
will be analyzed, and a Gaussian cloning transformation will be introduced.
Cloning machines, that is, transformations that achieve the best approximate
copying of a quantum state compatible with the no-cloning theorem, have
been a fundamental research topic over the last five years (see e.g., [2] for an
overview). This question is of particular significance given the close connection
between quantum cloning and quantum cryptography: using an optimal cloner
generally makes it possible to obtain a tight bound on the best individual
eavesdropping strategy in a quantum cryptosystem. This provides a strong
277
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 277-293.
© 2003 Kluwer Academic Publishers.
278 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

incentive to investigating continuous-variable cloning in view of the recent

proposals for quantum key distribution relying on continuous (Gaussian) key
carriers [3,4].
Here, we will focus on a Gaussian cloning transformation, which copies
equally well any two canonically conjugate continuous variables such as the
two quadrature components of a light mode [5]. More precisely, it achieves the
optimal cloning of a continuous variable that satisfies the requirement of covari-
ance with respect to displacements and rotations in phase space. Consequently,
this cloner duplicates all coherent states with a same fidelity (F = 2/3). The
optical implementation of this cloner and its extension to N -to-M cloners will
also be discussed. Finally, the use of this cloner for the security assessment of
continuous-variable quantum key distribution schemes will be sketched.

2. LIMITS ON OPTIMAL CLONING

Let us start by stating the problem of continuous-variable cloning in physical
terms. Consider, as an example of canonically conjugate continuous variables,
the quadrature components of a light mode, denoted as x and p. This notation
reflects the fact that x and p behave just like the position and momentum of a
particle in a one-dimensional space, namely their commutator is [x,p] = i (we
put !i = 1 in this paper). If the wave function is a Dirac delta function-the
particle is fully localized in position space, then x can be measured exactly,
and several perfect copies of the system can be prepared. However, such
a cloning process fails to exactly copy non-localized states, e.g., momentum
states. Conversely, if the wave function is a plane wave with momentum p-the
particle is localized in momentum space, then p can be measured exactly and
one can again prepare several perfect copies of this plane wave. However, such
a "plane-wave cloner" is then unable to copy position states exactly. In short, it
is impossible to copy perfectly the eigenstates of two conjugate variables such
as x and p: this is essentially the content of the so-called no-cloning theorem
[6,7].
In the next Section, we will show that a cloning transformation can never-
theless be found that provides two copies of a continuous system, but at the
price of a non-unity cloning fidelity. In other words, the cloning machine yields
two imperfect copies of the system. Before describing this cloning machine in
detail, let us find a lower bound on the cloning-induced noise by exploiting a
connection with measurement theory. More specifically, we make use of the
fact that measuring x on one clone and p on the other clone cannot beat the
optimal joint measurement of x and p on the original system [8]. It is known
that such a joint measurement of a pair of conjugate observables on a single
quantum system obeys an inequality akin to the Heisenberg uncertainty relation
but with an extra contribution to the minimum variance [9]. Denoting by x and
 Quantum cloning with continuous variables 279

p the two quadratures of the input mode, and by X and P the corresponding
jointly measured output quadratures, we have
X = x+nx (20.1a)
P=p+np (20.1b)
where nx and np stand for the excess noise that we have on the measured
quadratures. Since we consider a joint measurement, the variables X and P
must commute: they can be viewed respectively as the x and p quadratures of
two distinct modes. Thus, we have
(20.2)
Assuming that the excess noises nx and np are independent of the input quadra-
tures, i.e., [x, n p] = [nx,p] = 0, we get [nx, n p] = -i, implying that nx and
np must obey an uncertainty relation. Specifically, any attempt to measure x
and p simultaneously on a quantum system is constrained by the inequality
(20.3)
where Llni: and Lln~ denote the variances of the excess noises originating from
the joint measurement device. If the variances of the x and p quadratures of
the input state are denoted by bx 2 and bp2, respectively, we thus have for the
variances of the measured values LlX2 = bx 2 + Llni: and LlP2 = bp2 + Lln~.
As a consequence, the Heisenberg uncertainty relation bx bp 2: 1/2 together
with inequality (20.3) implies the relation [9]
LlX LlP 2: 1 (20.4)

where we have used the inequality a 2 + b2 2: 2)a 2b2. Thus, the best possible
joint measurement of x and p with a same precision on both quadratures of a
coherent state (bx 2 = bp2 = 1/2) gives
LlX2 = LlP2 =1 (20.5)
Compared with the vacuum noise, we note that the joint measurement of x and
p effects an additional noise of minimum variance 112, so that the measured
values suffer twice the vacuum noise.
Inequality (20.3) immediately translates into a lower bound on the cloning-
induced noise variance [8]. If we assume that the device that is used in order
to perform the joint measurement of x and p is actually a cloning machine
followed by two measuring apparatuses (x being measured on one clone and
p on the other clone), we conclude that the variance of the noise added by this
cloning machine cannot be lower than 112 in order to comply with Eq. (20.3),
that is
(20.6)
280 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

(We require here the same noise level on x and p.) This can also be shown
explicitly by writing the canonical transformation of the c10ner [10]. Denoting
by Xa(b) and Pa(b) the two quadratures ofthe output mode a (resp. b), we have

Xa = x+nx,a (20.7a)
Pa =p+np,a (20.7b)
Xb = x+nx,b (20.7c)
Pb =p+np,b (20.7d)

where x and p are the two quadratures of the input mode and nx/p,a/b stand for
the excess noises. Since the clones are carried by different modes (a and b), we
have [Xa, Pb] = [Xb, Pal = O. Assuming, as before, that the excess noises are
independent of the input mode, we get [nx,a, np,b] = [nx,b, np,a] = -i. This
gives rise to two no-cloning uncertainty relations

~nx,a ~np,b 2: 1/2 (20.8a)

~nx,b ~np,a 2: 1/2 (20.8b)

which constrain the excess noise variances ~n;/p,a/b of the two clones [5, 10].
Consequently, if the cloning process induces a small position (momentum)
error on the first copy, then the second copy is necessarily affected by a large
momentum (position) error. The Gaussian cloner we will discuss in the next
Session saturates these inequalities and is symmetric in a and b (and in x and
p):

(20.9)

To simplify the notation, we will denote this cloning-induced excess noise

variance as (]"2 in the following.

3. GAUSSIAN CLONING TRANSFORMATION

We will define a class of cloning machines that yield two imperfect copies
of a continuous-variable system, the underlying cloning transformation being
covariant with respect to displacements in phase space (x, p). By this, we
mean that any two input states that are related by a displacement result in
copies that are related in the same way; hence, the resulting cloning fidelity
is invariant under displacements in phase space. Specifically, let us seek for a
displacement-covariant transformation which duplicates with a same fidelity all
coherent states I'ljJ). Thus, if two input states are identical up to a displacement
iJ (x' ,p') = e -ix' fi eip' x, then their respective copies should be identical up to
the same displacement. Denoting by 11. the Hilbert space corresponding to a
single system, cloning can be defined as a completely-positive trace-preserving
 Quantum cloning with continuous variables 281
r--·----·--·------------f-·----·-·-l

10>

10>
. :
!. -.-----.-----..-.--------------.. ----.---~
ancilla

Pteplll'8llOn Clllnmg TrDn~ronnDbon

Figure 20.1 Quantum circuit for the continuous-variable cloning transformation. It consists of
four C-NOT gates preceeded by a preparation stage. Here, the ancillae are prepared in the state
given by Eg. (20.19). See [5, 12].

linear map C : 1£ -+ 1l®2 : 11/1) (1/11 -+ C(I1/1) (1/11) such that

C [b(x',p') 11/1) (1/11 bt(x',p')]

= b(x',p')®2 C(I1/1) (1/11) b t (x',p')®2 (20.10)

for all displacements b(x',p').

As shown in [5], this cloning map can be achieved via a unitary transfofI!1a-
tion U acting on three modes: the input mode (variable 1) supplemented with
two auxiliary modes, the blank copy (variable 2) and an ancilla (variable 3).

!i:
The two auxiliary variables must be initially prepared in the joint state

Ixh,3 = dxdp f(x,p) Iw(x, -p)h,3 (20.11)

where f(x,p) is an (arbitrary) complex amplitude function, and

Iw(x,p)) = 1
v27r
to=
1 00

-00
dx' e~Px
.
Ix')lx' + x)
I
(20.12)

are the EPR states (the maximally-entangled states of two continuous variables).
The cloning transformation is defined as

(20.13)

where Xk (fik) is the position (momentum) operator for variable k. As shown

in Fig. 20.1, this can be interpreted as a sequence of four continuous-variable
controlled-NoT (C-NOT) gates, each being defined as the unitary transformation
e-ixkiJI with k (l) referring to the control (target) variable [11].
Remarkably, Eq. (20.13) coincides with the discrete C-NOT gate sequence
that achieves the qubit cloning transformation [13], up to a sign ambiguity
282 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

originating from the fact that a continuous C-NOT gate is not equal to its

!I:
inverse. After applying U to the state 11/Ih Ixh,3, we get the joint state

dx dp f(x,p) b(x,p) 11/Ih Iw(x, -p)h,3 (20.14)

where variables 1 and 2 are taken as the two outputs of the cloner (clones a
and b), while variable 3 (the ancilla) must simply be traced over. This is a
peculiar state in that it can be reexpressed in a similar form by exchanging the

!I:
two clones, namely

dx dp g(x,p) b(x,p) 11/1)2 Iw(x, -p)h,3 (20.15)

II:
with

g(x,p) = ;7r dx' dp' ei(px'-xp') f(x',p') (20.16)

being the two-dimensional Fourier transform of f (x, p). The reSUlting state of

!I:
the individual clones can then be written as

!I:
Pa = dxdp If(x,p)1 2 b(x,p) 11/1) (1/11 bt(x,p) (20.17a)

Pb = dxdp Ig(x,p)1 2 b(x,p) 11/1) (1/11 bt(x,p) (20.17b)

which is consistent with tracing Eq. (20.10) over anyone of the clones. Thus,
the clones are affected by position and momentum errors that are distributed
according to If (x, p) 12 and I9 (x, p) 12. A central point here is that interchanging
the two clones amounts to substitute the function f with its two-dimensional
Fourier transform g. This property is crucial as it ensures that the two copies
suffer from complementary position and momentum errors. Indeed, one can

II:
check [5] that the four excess noise variances defined as

L1n;,a = dxdp x 2 If(x,p)1 2 , (20. 18a)

L1n~,a= II:
!I:
dxdpp 2 If(x,p)1 2 (20. 18b)

L1n;,b = dxdp x 2 Ig(x,p)1 2 , (20.18c)

L1n~,b= II: dxdpp 2 Ig(x,p)1 2 (20.18d)

obey the no-cloning inequalities (20.8a) and (20.8b). (Here, we assume that
the first-order moments of If(x,p)1 2 and Ig(x,p)1 2 vanish, that is, the clones
are not biased.)
 Quantum cloning with continuous variables 283

Within this class of cloning machines parametrized by f (x, p), a particularly

simple rotation-covariant cloner can be found that provides two identical copies
of a continuous system with the same error distribution in position and mo-
mentum. It corresponds to the choice f(x,p) = g(x,p) = e-(x 2+p2)/2 / Vi.
This cloner is named "Gaussian" as it effects Gaussian-distributed position-
and momentum-errors on the input mode: the excess noise on both clones is
distributed as e-(x2+p2) /7[, that is, as a bi-variate rotational-invariant Gaussian
of variance a 2 = 1/2. This cloner is optimal, as it satisfies Eq. (20.9). Here,

IX)2,3 = i:
the two auxiliary variables must be prepared in the state

);;! dy dz e-(y2+z2)/2 Iyh Iy + Z)3 (20.19)

which is simply the product vacuum state 10hiOh processed by a C-NOT gate
e-iX2P3. The resulting transformation effected by U on an input position state
Ix) is thus given by
Ix}t Ixh,3
Ix + y}t Ix + Z)2 Ix + y + zh (20.20)

where the three variables denote the two clones and the ancilla, respectively.
For an arbitrary input state 11/;), it is readily checked that this transformation
outputs two clones whose individual states are Gaussian distributed with a
variance a 2 = 1/2, namely

Pa = Pb = ..!.
7[
j.J-oo
(Xl dx dp e-(x 2+p2) b(x,p) I1/;) (1/;1 bt(x,p), (20.21)

In particular, if the input is a coherent state la) with a = (x +ip)/..;2, it is easy

to calculate the fidelity ofthis cloner by using l(ala/)1 2 = exp( -Ia - a / 12 ):
1 2
F = (aIPa(b) la) = 1 + a 2 = 3" (20.22)

This cloning fidelity does not depend on a, so this Gaussian cloner copies all
coherent states with the same fidelity 2/3. It can be viewed as the continuous
counterpart of the universal qubit cloner [13], as its cloning fidelity is invariant
under rotations in phase space. The physical origin of the cloning noise be-
comes, however, much more evident in the case of continuous variables: the
Gaussian noise that affects the clones can simply be traced back to the Gaussian
wave function of the two ancillary modes, see (20.19). This suggests that the
noise that inevitably arises when cloning is intrinsically linked to the vacuum
fluctuations of the auxiliary modes.
284 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Note finally that this formalism can easily be extended to the cloning of
squeezed state instead of coherent states [5]. One simply unsqueeze the state
before cloning and then squeeze the clones again. For any value of the squeezing
parameter r, one can then define a Gaussian cloner that copies with fidelity
2/3 all squeezed states of which the same quadrature is squeezed by the same
amount r. In contrast, cloning these squeezed states using the rotation-covariant
cloner defined above results in a fidelity that decreases as r increases.

4. OPTICAL IMPLEMENTATION
It is very instructive to write the cloning transformation in the Heisenberg
picture, that is, following the evolution of the annihilation operators associated
with the modes that are involved. Again, mode 1 denotes the input mode, and
modes 2 and 3 the ancillary modes. Mode l' and 2' stand for the two clones,
while 3' is the ancilla that is traced over after cloning. Here, aj = (x j +ipj) / v'2
stands for the annihilation operator for mode j. We require that the cloning
transformation conserves the mean values, i.e., (a~) = (a~) = (al), so that
the clones are centered on the original coherent state. We also require that the
cloning transformation is covariant under rotations in phase space. It is shown
in [14] that the optimal transformation satisfying these requirements is

, a2 a1
al al+-+- (20.23a)
v'2 v'2
a2 a3t
a'2 al--+- (20.23b)
v'2 v'2
a'3 at + v'2 a 3 (20.23c)

where mode 1 is initially prepared in an arbitrary coherent state la), with

a = (x + ip)/v'2, while modes 2 and 3 are prepared in the vacuum state. This
transformation clearly satisfies the commutation rules [a~, aj] = 8i ,j and yields
the correct mean values (x,p) for the two clones (modes l' and 2'). Also,
one can easily check that the quadrature variances of the clones are equal to
twice the vacuum noise, in accordance with the cloning excess noise variance
(]"2 = 1/2. This transformation actually coincides with the Gaussian cloner
introduced in the previous Section. Interestingly, we note here that the state
in which the ancilla 3 is left after cloning is centered on (x, -p), that is the
phase-conjugated state la*). This means that, in analogy with the universal
qubit cloner, the Gaussian cloner generates an "anticlone" (or time-reversed
state) together with the two clones.
As suggested by the above transformation, a possible optical implementation
of this Gaussian cloner consists in processing the input mode al into a linear
 Quantum cloning with continuous variables 285

a'
Vacuum 2

j
Pump

"""'"
'3

al
. Linear
Ampli
a out
/
/
BS
a'

(G=2)

a'
3
AnclUa

Figure 20.2 Implementation of a Gaussian doner using a phase-insensitive linear amplifier

and a 50:50 beam-splitter CBS). See [12].

phase-insensitive amplifier [15] of gain G = 2:

(20.24)

with mode 3 denoting the idler mode. This amplifier is limited by the quantum
noise so it naturally leads to an optimal cloner. A gain G = 2 is needed since
the cloner doubles the energy by creating two clones with the same energy as
the input state. One then produces these two clones simply by processing the
output signal of the amplifier through a 50:50 phase-free beam splitter,

(20.25)

as shown in Fig. 20.2. The rotation covariance of the resulting cloner is

ensured by the fact that the amplifier and the beam splitter are phase-insensitive.
Actually, combining Eqs. (20.24) and (20.25) results in the same canonical
transformation as above, so this optical setup indeed implements the optimal
Gaussian cloner. It is readily checked that this setup leads to an equal x- and
p-error variance of 1/2 for both clones.

5. GAUSSIAN CLONERS WITH MULTIPLE INPUTS

AND OUTPUTS
Let us now consider the general problem of optimal N -+ M cloning,
extending what was done in [16] for the case of quantum bits. Consider a
Gaussian transformation which, from N (~ 1) identical replicas of an original
input state, produces M (~ 2) output copies whose individual states are again
given by an expression similar to Eq. (20.21) but with an error variance 0'2 N,M.
(For the 1 -+ 2 Gaussian cloner above, we had 0'2 1 ,2 = 1/2.) Using an
286 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

argument based on the concatenation of cloners, it is possible to derive a lower

bound on 0"2 N ,M, that is [8]
211
O"NM>---
, -N M'
(20.26)

so that the corresponding cloning fidelity for coherent states satisfies

F MN (20.27)
N,M~MN+M_N
The proof is connected to quantum state estimation theory, the key idea being
that cloning should not be a way of circumventing the noise limitation encoun-
tered in any measuring process. More specifically, concatenating a N -+ M
cloner with a M -+ L cloner results in a N -+ L cloner that cannot be better that
the optimal N -+ L cloner. We then make use of the fact that the excess noise
variance of this N -+ L cloner simply is the sum of the excess noise variances
of the two component cloners [8]. Denoting by O"Jv Mthe excess noise variance
of the optimal N -+ M cloner, we get the inequ~lity O"Jv,L ~ O"Jv,M + O"k,L.
In particular, if L -+ 00, we have

(20.28)

Since the limit of cloning with an infinite number of clones corresponds to a

measurement, Eq. (20.28) simply implies that cloning the N replicas before
measuring the M resulting clones does not provide a mean to enhance the
accuracy of a direct measurement of the N replicas. This limit is useful because
the joint measurement of x and p on N identical replicas of a coherent state is
known to give a minimum noise variance O"Jv 00 = liN. This, combined with
Eq. (20.28), gives the minimum noise varianc~ induced by cloning, Eq. (20.26),
along with the corresponding cloning fidelity, Eq. (20.27). Note that these
bounds can also be derived when N = 1 using techniques similar to the ones
used for describing quantum nondemolition measurements. This was done in
a paper establishing a link between cloning and teleportation for continuous
variables [10]: for the 1 -+ 2 cloner, the teleportation fidelity must exceed
F 1 ,2 = 2/3 in order to guarantee that the teleported state is of better quality
than the state kept by the emitter.
Just like for the 1 -+ 2 cloner, the bounds Eqs. (20.26) and (20.27) can
be attained by a transformation whose implementation requires only a phase-
insensitive linear amplifier and beam splitters [14, 17]. Loosely speaking, the
procedure consists in concentrating the N input modes into a single mode by
use of a network of beam splitters, then in amplifying the resulting mode and
distributing the output mode of the amplifier into M modes through a second
network of beam-splitters. A convenient way to achieve these concentration
and distribution stages is provided by networks of beam splitters that realize
 Quantum cloning with continuous variables 287

a Discrete Fourier Transform (DFT). Cloning is then achieved by the follow-

ing three-step procedure (see Fig. 20.3). First step: the N input modes are
concentrated into a single mode through a DFT (acting on N modes):
N-l
ak = ~L exp(ikl27rIN) ai, (20.29)
vN 1=0

with k = 0 ... N - 1. This operation concentrates the energy of the N

input modes al into one single mode a~ (hereafter renamed ao) and leaves the
remaining N - 1 modes (ai ... a~_l) in the vacuum state. Second step: the
mode ao is amplified with a linear amplifier of gain G = MIN. This results in

fM ao+JM
VN N
-lat
z'
(20.30a)

J~ -1 ab + az · I¥r (20.30b)

Third step: amplitude distribution by performing a DFT (acting on M modes)

between the mode a~ and M - 1 blank modes in the vacuum state:
M-l
a% = ~ L exp(ikl27rIM) a;, (20.31)
vM 1=0

with k = 0 ... M - 1, and a~ = ai for i = N ... M - 1. The DFT now

distributes the energy contained in the output of the amplifier among the M
output clones.
It is readily checked that this procedure meets the requirements we put on the
N -+ M cloner, and is optimal. Indeed the quadrature variance of the M output
modes gives 1/2+ liN -11M, implying that the cloning-induced excess noise
variance is liN -11M. Furthermore, the transformation is rotation covariant
since the amplifier and the beam splitters are phase insensitive. In conclusion,
we see that the optimal N -+ M cloning transformation can be implemented
using only passive elements except for a single linear amplifier.
The above cloning transformation can be extended even further by consider-
ing a generalized cloner that produces M clones from N replicas of a coherent
state and N' replicas of its complex conjugate [18]. It is again universal over
the set of coherent states in the sense that the cloning fidelities are invariant
for all input coherent states. Interestingly, it can be shown that supplement-
ing the N input states 1"p)®N with N' phase-conjugated input states 1"p*)®N'
can, under certain circumstances, provide clones with a higher fidelity than
the above N + N' -+ M cloner. Note that, together with the M clones, this
phase-conjugate input cloner also yields M' anticlones (approximate copies
288 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

BlankM-N
aM_I
-
,
. OulpUIM-1

,
aN
Blank I OulpulN
aN_I
-
D
,
loputN-I OutputN-I

I
! a, C :

.
loput I Oulpull
a0
lopulO Output 0
Linear
Ancillu
a, '-- AmpJi -

1
Figure 20.3 Implementation of an N -t M continuous-variable cloning machine based on a
phase-insensitive linear amplifier. Here, C stands for the amplitude concentration stage while
D refers to amplitude distribution. Both can be realized using a network of beam-splitters that
achieve a DFf. See [14].

of I~*)) at no cost, with N - N' = M - M'. The advantage of having

phase-conjugated inputs for a continuous-variable cloner actually also has a
counterpart in the context of qubit cloners. Indeed, motivated by this finding
on continuous-variable cloners, an optimal universal cloning transformation
was recently derived that produces M copies of an unknown pair of orthogonal
qubits [19]. For M > 6, the cloning fidelity for a pair of orthogonal qubits can
be shown to be higher than that of the optimal cloning of a pair of identical
qubits. This is a first example of a quantum informational process that was
initially described for continuous-variable systems and only later on extended
back to quantum bits.

6. EAVESDROPPING IN CONTINUOUS· VARIABLE

QUANTUM CRYPTOGRAPHY
As mentioned above, quantum cloning can be viewed as an individual eaves-
dropping strategy in continuous-variable quantum cryptography. Consider a
quantum key distribution scheme in which the key is encoded into the displace-
ment of a coherent or a squeezed state that is drawn from a Gaussian distribution
[3,4]. In the continuous-variable protocol defined in [3], which we will analyze
here, squeezed states need to be used. The emitter (Alice) prepares a squeezed
state for which the quadrature that is squeezed, x or p, is chosen at random,
and then displaces it by .o(r, 0) or .0(0, r) depending on x or p is squeezed.
Here, r is drawn from a Gaussian distribution, and constitutes a continuous key
element. The receiver (Bob) then measures either the x- or p-quadrature of the
state he received, this choice being again random. After Bob's measurement,
 Quantum cloning with continuous variables 289

Alice reveals the quadrature she squeezed (and displaced) and Bob rejects the
cases where he measured the wrong quadrature, this discussion being made
over an authenticated public channel (this procedure is known as sifting). The
subset of states that are accepted by Bob then constitutes a Gaussian raw key
(correlated Gaussian data at Alice's and Bob's side). Indeed, denoting as v
the variance of the quadrature that is squeezed by Alice, Bob gets for his mea-
sured quadrature an outcome r' that is Gaussian distributed around r with a
variance v (assuming for the moment that the quantum channel is perfect and
that there is no eavesdropping). If the variance of the random displacements r
imposed by Alice is noted V, then this raw key shared by Alice and Bob can be
viewed as resulting from a Gaussian additive-noise channel characterized by a
signal-to-noise ratio of Vlv.
The maximum amount of shared key bits that can be extracted from this
Gaussian raw key can be analyzed by applying some standard notions of Shan-
non theory for continuous channels (see e.g. [20]). Consider a discrete-time
continuous channel that adds a Gaussian noise of variance v to the signal. If
the input r of the channel is a Gaussian signal of variance V, the uncertainty on
r can be measured by its Shannon entropy h(r) = 2- 1 10g2 (21f e V) bits. Con-
ditionally on r, the output r' is distributed as a Gaussian of variance v, so that
the entropy of r' conditionally on r becomes h(r'lr) = 2- 1 10g2 (21f e v) bits.
Now, the overall distribution of r' is of course the convolution of these two
distributions, i. e., a Gaussian of variance V + v, so that the output entropy
is h(r') = 2- 1 10g2 (21fe (V + v)) bits. According to Shannon theory, the
information processed through this noisy channel r --+ r' can be expressed as
the amount by which the uncertainty on r' is reduced by knowing r, that is

I (bits) = h(r') - h(r'lr) = ~ log2 (1 + :) (20.32)

where Vlv is the signal-to-noise ratio. This is Shannon's famous formula for
the capacity of a Gaussian additive-noise channel. It is worth noticing that this
capacity is achieved in the case where the input is distributed as a Gaussian,
which is precisely the case under consideration here.
In the protocol analyzed in [3], the variances v and V are related by the
constraint that Alice's choice of encoding the key into either x or p should be
invisible to a potential eavesdropper. In the first case, Alice applies a Gaussian-
distributed displacement D(r, 0) on a squeezed state whose x quadrature has
a variance v, so that the quadratures x and p of this Gaussian mixture have a
variance V + v and 1/(4v), respectively. In the second case, Alice applies a
displacement D(O, r) on a squeezed state in p, resulting in a Gaussian mixture
with variances 1I (4v) and V + v for x and p. These two Gaussian mixtures are
required to be indistinguishable, which simply translates into the requirement
290 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

that they have the same x variance and p variance:

1
V+v=- (20.33)
4v
This gives for the information

I = log2 1/2)
( --:;;- (20.34)

which measures the maximum number of key bits that can be extracted asymp-
totically (at the limit of long sequences) per use of the channel. (The factor
1/2 here is just the vacuum noise, so we see that this protocol requires squeez-
ing, that is, v < 1/2.) The actual methods that may be used to discretize the
Gaussian raw key and correct the resulting errors so as to extract a common bit
string are known as reconciliation protocols [21].
Let us now consider the information that is transmitted in the presence of
an eavesdropper. We assume that the eavesdropper (Eve) processes each key
element into a Gaussian cloning machine, keeps one clone, and sends the other
one to Bob. Once the quadrature that contains the key (x or p) is revealed by
Alice and Bob, Eve properly measures her clone. Clearly, Eve needs to use
an asymmetric version of the Gaussian cloner described above as she must be
able to tune the information she gains, and therefore the disturbance she effects
in the transmission. (A possible implementation of this asymmetric Gaussian
cloner is discussed in [17].) Thus, Eve adds some extra noise on the quadrature
encoding the key, which results in a reduced signal-to-noise ratio on Alice-Bob
channel. Remember here, that the quality of the two clones obey a no-cloning
uncertainty relation akin to the Heisenberg relation, implying that the product
of the x-error variance on the first clone times the p-error variance on the second
one remains bounded by (1/2)2; see Eqs. (20.8a) and (20.8b). In particular, if
x and p are treated symmetrically, we have

(20.35)

This translates into a balance between the signal-to-noise ratio in Alice-Bob

channel V/(v + .6.n~) and that in Alice-Eve channel V/(v + .6.n~). This
latter channel is also a Gaussian channel so it can be treated similarly. Using
Eq. (20.33), we can write the information processed respectively in Alice-Bob
and Alice-Eve channels as

(20. 36a)

(20.36b)
 Quantum cloning with continuous variables 291

which gives

(20.37)

One can then show that lAB + I AE - I :::; 0 by checking that the quantity inside
the logarithm is less or equal to one. This simplifies to the condition

(20.38)

which is indeed true as a consequence of Eq. (20.35) and v < 1/2. Con-
sequently, we have proven that, in this quantum cryptographic protocol, the
no-cloning uncertainty relation translates into an information exclusion princi-
ple [3]

(20.39)

In other words, the information I AE gained by Eve is upper bounded by the

defect of information at Bob's side, I - lAB, which implies that the secu-
rity is guaranteed if lAB ;::: l/2 (since Bob then has an advantage over Eve,
lAB;::: lAE)' Note that the bound in Eq. (20.39) is saturated by the asymmet-
ric Gaussian cloner discussed above, which strongly suggests that this is the
optimal individual attack (this actually can be proven rigorously). In practice,
Alice and Bob can estimate the potentially eavesdropped information in the
following way. Alice discloses the values r she sent for a random subset of the
raw key. Then, Bob compares them to the values r' he received, in order to
estimate the variance of the distribution of the differences r' - r, i. e., the excess
noise variance Lln1. This is sufficient to estimate lAB, and, via Eq. (20.39),
an upper bound on I AE.
An extended continuous-variable quantum key distribution protocol relying
on Gaussian key carriers has recently been proposed in [4], where coherent
states may be used instead of squeezed states. The encoding then consists
in imposing a displacement D(x,p) onto the vacuum state with x and p be-
ing drawn from a bi-variate Gaussian distribution. Here, the choice of the
quadrature is made by Bob, who decides to measure x or p at random and
then discloses his choice on the public channel. The corresponding value of
Alice's displacement (x or p) together with Bob's measured outcome again
can be viewed as resulting from a Gaussian channel, so the above information-
theoretic treatment can be extended. In particular, one can calculate lAB and
lAE in the case of an individual attack based on asymmetric Gaussian cloners.
The security analysis of this coherent-state protocol is beyond the scope of the
present paper (see [4]).
292 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Acknowledgments
I would like to thank: S. L. Braunstein, S.lblisdir, P. van Loock, S. Massar,
and G. Van Assche for their contribution to the work reported on here.

References
[1] S. L. Braunstein. Quantum error correction for communication with linear
optics. Nature 394, 47 (1998).
[2] S. L. Braunstein, V. Buzek, and M. Hillery. Quantum-information distribu-
tors: Quantum network for symmetric and asymmetric cloning in arbitrary
dimension and continuous limit. Phys. Rev. A 63, 052313 (2001).
[3] N. J. Cerf, M. Levy, and G. Van Assche. Quantum distribution of Gaussian
keys using squeezed states. Phys. Rev. A 63, 052311 (2001).
[4] F. Grosshans and P. Grangier. Continuous variable quantum cryptography
using coherent states. Phys. Rev. Lett. 88, 057902 (2002).
[5] N. J. Cerf, A. Ipe, and X. Rottenberg. Cloning of continuous quantum
variables. Phys. Rev. Lett. 85, 1754 (2000).
[6] W. K. Wootters and W. H. Zurek. A single quantum cannot be cloned.
Nature 299, 802 (1982).
[7] D. Dieks. Communication by EPR devices. Phys. Lett. A 92, 271 (1982).
[8] N. J. Cerf and S.lblisdir. Optimal N -to-M cloning of conjugate quantum
variables. Phys. Rev. A 62, 040301 (2000).
[9] E. Arthurs and J. L. Kelly, Jr. On the simultaneous measurement of a pair
of conjugate observables. Bell Syst. Tech. J. 44, 725 (1965).
[10] F. Grosshans and P. Grangier. Quantum cloning and teleportation criteria
for continuous quantum variables. Phys. Rev. A 64, 010301 (2001).
[11] S. L. Braunstein. Error correction for continuous variables. Phys. Rev.
Lett. 80, 4084 (1998).
[12] N. J. Cerf and S. Iblisdir. Universal copying of coherent states: a Gaus-
sian cloning machine. In Quantum Communication, Computing, and
Measurement 3, (Kluwer Academic, New York, 2001), pp. 11-14.
[13] V. Buzek and M. Hillery. Quantum copying: Beyond the no-cloning
theorem. Phys. Rev. A 54, 1844 (1996).
[14] S. L. Braunstein, N. J. Cerf, S. Iblisdir, P. van Loock, and S. Massar. Op-
timal cloning of coherent states with a linear amplifier and beam splitters.
Phys. Rev. Lett. 86,4438 (2001).
 Quantum cloning with continuous variables 293

[15] c. M. Caves. Quantum limits on noise in linear amplifiers. Phys. Rev. D

26, 1817 (1982).
[16] N. Gisin and S. Massar. Optimal quantum cloning machines. Phys. Rev.
Lett. 79,2153 (1997).
[17] J. Fiurasek. Optical implementation of continuous-variable quantum
cloning machines. Phys. Rev. Lett. 86,4942 (2001).
[18] N. J. Cerf and S .lblisdir. Quantum cloning machines with phase-conjugate
input modes. Phys. Rev. Lett. 87,247903 (2001).
[19] J. Fiurasek, S. Iblisdir, S. Massar, and N. J. Cerf. Quantum cloning of
orthogonal qubits. Phys. Rev. A 65, 040302(R) (2002).
[20] T. M. Cover and J. A. Thomas. Elements of Information Theory. Wiley
& Sons, New York, 1991.
[21] N. J. Cerf, S.lblisdir, and G. Van Assche. Cloning and cryptography with
quantum continuous variables. Eur. Phys. 1. D 18, 211 (2002).
Chapter 21

QUANTUM KEY DISTRIBUTION

WITH CONTINUOUS VARIABLES IN OPTICS

T. C. Ralph
Department of Physics, Centre for Quantum Computer Technology,
University of Queensland, St Lucia 4072, QLD, Australia
[email protected]

Abstract We discuss a quantum key distribution scheme in which small phase and ampli-
tude modulations of quantum limited, CW light beams carry the key information.
We identify universal constraints on the level of shared information between the
intended receiver (Bob) and any eavesdropper (Eve) and use this to make a
general evaluation of the security and efficiency of the scheme.

1. INTRODUCTION
The distribution of random number keys for cryptographic purposes can
be made secure by using the fundamental properties of quantum mechanics
to ensure that any interception of the key information can be detected. This
was first discussed for discrete systems in Refs. [1, 2, 3]. Experimental
demonstrations have been carried out using low photon number, optical sources
[4,5].
The basic mechanism used in quantum cryptographic schemes is the fact that
the act of measurement in quantum mechanics inevitably disturbs the system.
This measurement back-action exists for both discrete and continuous quan-
tum mechanical variables. Thus it is natural to ask if quantum cryptographic
schemes based on continuous variables are possible. There are a number of
practical disadvantages with discrete quantum cryptographic schemes, mainly
associated with the lack of true single photon sources. Also it is of fundamental
interest to quantum information research to investigate links between discrete
variable, single photon phenomena and continuous variable, multi-photon ef-
fects. This has motivated a consideration of quantum cryptographic schemes
using multi-photon light modes [6, 7,8,9, to, 11, 12, 13, 14].
295
S.L. Braunstein andA.K. Pati (eds.), Quantum Information with Continuous Variables, 295-316.
© 2003 Kluwer Academic Publishers.
296 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Most of these schemes use squeezed light [15] in their protocols, either by
producing entanglement from the squeezing [11, 9, 10] or using the squeezing
directly [8, 12]. In contrast to these, schemes based on coherent states have also
been discussed [7, 14]. The signals from which the key material is obtained
are encoded in various ways in the different schemes.
The question of optimum protocols and eavesdropper strategies has been
studied in detail for the single quanta case [16], leading to general proofs
of security for discrete systems [17, 18]. A general proof of the optimum
eavesdropper strategy for individual attacks in a simple continuous variable
scheme was presented in Ref. [11]. Physical implementations saturating this
optimum strategy were discussed in Refs [11, 13, 14]. A general proof of
absolute security for a more sophisticated scheme was presented in Ref. [12].
In this chapter we will analyse in some detail quantum key distribution proto-
cols based on the optical coherent state and squeezed state schemes introduced
in Refs. [7, 11]. Our emphasis will be on specific implementations that Alice
and Bob might use rather than general limits. The particular implementations
have been chosen mostly for their simplicity rather than their optimality. Eve
on the other hand is always assumed to be employing the optimum eavesdrop-
ping strategies allowed by quantum physics [19]. We estimate the efficiency of
the two schemes and hence secure key transmission rates under conditions of
negligible and non-negligible losses.
In Section I we review the encoding of information on light with small am-
plitude and phase modulations and introduce a particular encoding scheme. In
Section II we find the minimum disturbance that an optimum eavesdropping
scheme will introduce. The coherent state cryptographic scheme is introduced
in Section III and the minimum error rates that an optimum Eve will intro-
duce are calculated. In Section IV the concepts of mutual information, data
reconciliation and privacy amplification are introduced and specific examples
are applied to the coherent state scheme. The security and efficiency of the
scheme are evaluated. The squeezed state cryptographic scheme is introduced,
analysed and evaluated in Section V. In section VI we discuss a physical im-
plementation of the optimal eavesdropper strategy and we conclude in Section
VII.

2. ENCODING INFORMATION WITH SMALL

AMPLITUDE AND PHASE MODULATIONS
One way of encoding information on a light beam is by imposing small
modulations of the phase or amplitude of the beam at some radio frequency (rt)
with respect to the main optical frequency. We suppose that these signals are
imposed at an rf sufficiently large that technical noise can be ignored and so our
measurement precision is limited only by quantum noise. Typically frequencies
 Quantum key distribution with continuous variables in optics 297

in excess of about a MHz will suffice. That quantum mechanics must impose
limits in this situation is because the amplitude and and phase quadrature
amplitudes of the beam are the analogues of position and momentum variables.
Hence they are continuous, non-commuting variables that exhibit uncertainty
relations.
We can represent our light field via

a.(t) = a + 8a(t) + 8s(t) (21.1)

where a. is a bosonic annihilation operator which we have decomposed into a

steady state part, the coherent amplitude, a, treated classically, and two time
varying parts: the quantum fluctuations, modelled by the operator 8a.(t); and
the classical modulation, modelled by 8s(t). If we take the phase of a real then
the amplitude fluctuations, X+, and the phase fluctuations, X- , are given by

X+ 8a.(t)t + 8s(t)* + 8a.(t) + 8s(t)

X- = i(8a.(t)t + 8s(t)* - 8a(t) - 8s(t)) (21.2)

Homodyne detection using a local oscillator with a coherent amplitude much

larger than that of the signal beam can be used to measure the fluctuations.
Spectral analysis then extracts the fluctuation power at a particular rf, w, such
that

(21.3)

and

(21.4)

where Vn+ (Vn-) is the amplitude (phase) quantum noise power whilst V/ CV~-)
is the amplitude (phase) signal power. The tilde indicates a Fourier transform.
The amount of information that can be carried on a Gaussian, additive noise,
communication channel, such as we will consider here, depends on the signal
to noise [20]. For a fixed bandwidth, any reduction in the signal to noise will
inevitably lead to increased errors in the transmission. In our cryptographic
scheme signals will be encoded on both quadratures but read out from only
one, randomly chosen. This will force any eavesdroppers to monitor both the
amplitude and phase quadratures simultaneously. For these non-commuting
observables the information that can be obtained in this way is strictly limited by
the generalized uncertainty principle for simultaneous measurements [21,22].
We will discuss this principle in detail in the next section. Here let us consider
a simple example. Suppose we try to observe both quadratures by dividing the
beam in two at a 50:50 beamsplitter and detecting the amplitude quadrature of
298 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

one beam and the phase quadrature of the other. Originally the signal to noises
are given by

(SIN)± = ~: (21.5)
Vn
However the signal to noises detected after the beamsplitter are

(SIN)±' = ( Vn±
szm Vn± + vJ )SIN± = T±SIN± (21.6)

where we define T+ (T-), the amplitude (phase) signal transfer coefficient, as

the ratio of signal to noise out to signal to noise in. The quantum noise which
is inevitably added through the empty beam splitter port is V~. The spectral
powers are normalized to the quantum noise limit (QNL) such that a coherent
beam has Vn± = 1. Normally the partition noise will also be at this limit
(V~ = 1). For a c1assicallight field, i.e. where Vn± > > 1 the penalty will be
negligible. However for a coherent beam a halving of the signal to noise for
both quadratures is unavoidable.
One specific encoding scheme is that of binary pulse code modulation. The
data is encoded as a train of Fourier transform limited pulses with average
power Va. A pulse on represents a "I", a pulse off represents a "0". For such
bandwidth limited transmission the bit error rate or error probability (8) and
the signal to noise (SIN) are related by [23]

8 = 1 1~
-erfc- -SIN (21.7)
222
Suppose our signal to noise is initially 13 dB. From Eq.21.7 direct detection of
a single quadrature will retrieve its pulse train with a bit error rate of 1%. If
the beam is in a coherent state and we simultaneously detect both quadratures
then Eq.21.6 tells us that the signal to noise is halved. Eq.7 then predicts the
error rate will rise to 5 %.
Alternatively, and more efficiently, signals can be encoded as coherent sig-
nals. Given the ability to phase lock to the signal frequency a mathematical
equivalence exists between the situation of signal side bands on a QNL back-
ground and that of a DC coherent state of the same amplitude. We can thus
represent a coherent signal, of amplitude a by the state ket la}. If the infor-
mation is sent as a Gaussian distribution of such states, then in principle the
Shannon channel capacity [20]

(21.8)

can be achieved by a suitable encoding. By using Gaussian distributions of

coherent states on both quadratures (ie of purely real and imaginary amplitudes)
 Quantum key distribution with continuous variables in optics 299

a more efficient continuous variable quantum key distribution protocol can

be constructed [14]. A similar encoding, but based on squeezed states was
introduced earlier [13]. Never-the-Iess qualitatively similar results are obtained
from an analysis of the binary pulse encoding, which we will pursue here for
its simplicity.

3. OPTIMUM EAVESDROPPER STRATEGY

In this section we will use the generalized uncertainty principle to identify
the minimum disturbance allowed by quantum mechanics to the information
Bob receives given a particular level of interception by Eve. The idea is shown
schematically in Fig. 21.1. A single quantum limited beam is sent from Alice to
Bob. Eve makes some unspecified interception of the beam enroute. Bob and
Eve obtain some measurement results. We will show that quantum mechanics
sets unambiguous limits on the level of quantum noise that must appear in
Bob and Eve's results. In the following sections we will apply these results to
specific quantum cryptographic systems.

vI + vI OR Vi + Vi

vI
Vi

Figure 21.1 Schematic of general set-up. Alice sends information encoded in the amplitude
(VI) and phase (Vi) spectra. Bob makes measurements of either the amplitude or phase
quadrature. Some additional noise is present on his measurements, vi. Eve does not know
which quadrature Bob will measure thus she needs to be able to extract information about both
quadratures from her intercepted material. This leads to strict bound on the allowed values of
the additional noise which must appear on her measurements (Vi)

A more general statement of the generalized uncertainty principle [22] re-

quires that for any simultaneous measurements of conjugate quadrature ampli-
tudes

v+v-
M M-
>1 (21.9)

where V~ are the measurement penalties for the amplitude (+) and phase
( -) quadratures, normalized to the amplification gain between the system
observables and the measuring apparatus. For example suppose an attempt to
measure the amplitude quadrature variance of a system vt
returned the result
300 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

G l Vk+ + G2 VJ where VJ represents noise. Then we would have vAi =

(Gd Gt) VJ. Eq.21.6 follows directly from Eq.21.9 for ideal simultaneous
measurements. Let us investigate what general restrictions this places on
the information that Eve can intercept and the subsequent corruption of Bob's
signal. Firstly Eve's measurements will inevitably carry measurement penalties
vi constrained by

v+v- >1
E E - (21.10)

Now suppose Bob makes an ideal (no noise added) amplitude measurement
on the beam he receives. In order to satisfy Eq.21.9 it must be true that the
noise penalty carried on the amplitude quadrature of this beam Vi
due to Eve's
intervention, is sufficiently large such that

v+v,-
B E -
>1 (21.11)

Similarly, Bob can also choose to make ideal measurements of the phase
quadrature so we must also have

y+V-
E B
>
-
1 (21.12)

Eqs.21. 10,21. 11,21. 12 set strict quantum mechanical limits on the minimum
disturbance Eve can cause to Bob's information given a particular maximum
quality of the information she receives. This applies regardless of the method
she uses to eavesdrop. Note that quantum memory does not negate the above
results provided we insist that Alice and Bob do not exchange any potentially
revealing classical information until Alice is sure that Bob has received and
measured her signals.
These relations could form the basis of a security analysis of any continuous
variable quantum cryptographic scheme in which a single quantum beam is
exchanged. However the ramifications of a particular level of disturbance will
vary for different schemes. In the following section we will analyse the security
of a very simple scheme based on the exchange of a beam in a coherent state.

4. COHERENT STATE QUANTUM CRYPTOGRAPHY

Consider the set up depicted in Fig. 21.2. A possible protocol is as follows.
Alice generates two independent random strings of numbers and encodes one
on the phase quadrature, and the other on the amplitude quadrature of a bright
coherent beam. The amplitude and phase signals are imposed at the same
frequency with equal power. Bob uses homodyne detection to detect either
the amplitude or phase quadrature of the beam when he receives it. He swaps
randomly which quadrature he detects. On a public line Bob then tells Alice
at which quadrature he was looking, at any particular time. They pick some
 Quantum key distribution with continuous variables in optics 301

subset of Bob's data to be the test and the rest to be the key. For example,
they may pick the amplitude quadrature as the test signal. They would then
compare results for the times that Bob was looking at the amplitude quadrature.
If Bob's results agreed with what Alice sent, to within some acceptable error
rate, they would consider the transmission secure. They would then use the
undisclosed phase quadrature signals, sent whilst Bob was observing the phase
quadrature, to create their key. By randomly swapping which quadrature is key
and which is test throughout the data comparison an increased error rate on
either quadrature will immediately be obvious.

random number
generators
homodyne detection
randomly measuring
either phase or
amplitude signal power

I------;~ - - - - - ------i.~D
coherent AM PM
source
ALICE BOB

Figure 21.2 Schematic of coherent light cryptographic set-up. AM is an amplitude modulator

whilst PM is a phase modulator.

Before making a general analysis of security let us first consider some specific
strategies an eavesdropper could adopt. Eve could guess which quadrature Bob
is going to measure and measure it herself. She could then reproduce the digital
signal of that quadrature and impress it on another coherent beam which she
would send on to Bob. She would learn nothing about the other quadrature
through her measurement and would have to guess her own random string of
numbers to place on it. When Eve guesses the right quadrature to measure Bob
and Alice will be none the wiser, however, on average 50% of the time Eve
will guess wrong. Then Bob will receive a random string from Eve unrelated
to the one sent by Alice. These will agree only 50% ofthe time. Thus Bob and
Alice would see a 25% bit error rate in the test transmission if Eve was using
this strategy. This is analogous to the result for single quanta schemes in which
this type of strategy is the most readily available. Another single measurement
strategy Eve could use is to do homodyne detection at a quadrature angle
half-way between phase and amplitude. This fails because the signals become
mixed. Thus Eve can tell when both signals are 0 or both are 1 but she cannot
tell the difference between 1,0 and 0,1. This again leads to a 25% bit error rate.
However, for bright beams it is possible to make simultaneous measurements
of the quadratures, with the caveat that there will be some loss of information.
So a second strategy that Eve could follow would be to split the beam in half,
302 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

measure both quadratures and impose the information obtained on the respective
quadratures of another coherent beam which she sends to Bob. How well will
this strategy work? We performed this calculation at the end of section I using
Eq.21.7. The halving of signal to noise imposed by the 50:50 beamsplitter
means the information Eve intercepts and subsequently passes on to Bob will
have an error probability of 5% (for the particular case of bandwidth limited
binary pulse code modulation). This is clearly a superior strategy and would
be less easily detected. Further more Eve could adopt a third strategy of only
intercepting a small amount of the beam and doing simultaneous detection on
it. For example, by intercepting 16% of the beam, Eve could gain information
about both quadratures with an error rate of 25% whilst Bob and Alice would
observe only a small increase of their error rate to 1.7%. In other words Eve
could obtain about the same amount of information about the key that she could
obtain using the "guessing" strategy, whilst being more difficult to detect.
Now let us analyze this coherent state scheme using Eqs.21.1O,21. 11,21. 12.
We choose to couch our evaluation in terms of bit error rates because they
represent an unambiguous, directly observable measure of the extent to which
Eve can intercept information and the resulting corruption of Bob's information.
This connection will be developed in Section IV. Depending on the particular
technique Eve uses Bob and Alice may be able to gain additional evidence for
Eve's presence by making a more detailed comparison ofthe sent and received
signals. This can only increase the security of the system. By considering a
general limit on error rates we can find a minimum guaranteed security against
eavesdropping regardless of the technique Eve employs.
The signal transfer coefficients for Bob and Eve will be given by

T+ =
(SjN)tve Vi!
E
(SjN)in Vi! + vt
T- (SjN);ve Vi;;:
E = =
(SjN)in Vi;;: + Vi
T+ (SjN)tb Vi!
B
(SjN) in Vi! + V;
T- =
(SjN)"bob Vi;;: (21.13)
B
(SjN)in Vi;;: + Vi
Substituting Eqs.21.13 into Eqs.21.1O,21.11,21.12 and using the factthat Vi! =
1 we find

Ti +TE < 1
Ti +T"B < 1
Tit +TE < 1 (21.14)
 Quantum key distribution with continuous variables in optics 303

Eqs.21.14 clearly show that any attempt by Eve to get a good signal to noise
on one quadrature (e.g. Ti -+ 1) results not only in a poor signal to noise in
her information of the other quadrature (e.g. Ti -+ 0) but also a poor signal to
noise for Bob on that quadrature (e.g. Tii -+ 0), making her presence obvious.
This is the general limit of the guessing strategy presented in the last section
and leads to the same error rates.
Because of the symmetry of Bob's readout technique Eve's best approach
is a symmetric attack on both quadratures. Eqs.21.14 then reduces to two
equations

(21.15)

If Eve extracts her maximum allowable signal to noise transfer, T~ = 0.5, then
ideally Bob suffers the same penalty Ti= 0.5. This is the general limit of the
second strategy of the previous section. The same reduction in Bob's signal
to noise occurs as in the specific implementation thus this implementation can
be identified as an optimum eavesdropper strategy for obtaining maximum
simultaneous information about both quadratures.
Eve's best strategy is to intercept only as much information as she can
without being detected. The system will be secure if that level of information
can be made negligible. Suppose, as in the last section, Eve only intercepts
a signal transfer of T~ = .08. From Eq.21.15 this means Bob can receive
at most a signal transfer of Ti = .92. This is greater than the result for the
specific implementation discussed in the last section, thus that implementation
is not an optimum eavesdropper strategy. Using the optimum eavesdropper
strategy the error rates for the specific encoding scheme discussed in the last
section will be: if Eve intercepts information with an error probability of 25%,
then the minimum error rate in Bob's information will be 1.4%.
In Fig. 21.3 we represent the general situation by plotting the minimum
error rate Bob and Alice can observe against the error rate in Eve's intercepted
information using Eq.21.7 and 21.15. An error rate of 50% (i.e. completely
random) represents no information about the data. Two traces are shown,
representing different initial signal to noises in Alice's data. This graph shows
that in principle any incursion by Eve will result in some increase in Bob and
Alice's error rate. However one could also argue that any finite resolution in
Bob and Alice's determination of their error rate will allow Eve to do better
than the random result. In order to assess whether this system can be made
secure we need to introduce the concepts of mutual information and privacy
amplification.
304 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

12
10
Bob's 8
Error
6
Probability
% 4
(a)
2

10 20 30 40 50

Eve's Error
Probability %

Figure 21.3 Minimum allowable error probabilities in the data of Bob and Eve are plotted for
two signal to noise levels of Alice's beam. Trace (a) is for a signal to noise of 13dB whilst trace
(b) is for a signal to noise of lOdB.

5. MUTUAL INFORMATION AND PRIVACY

AMPLIFICATION
The mutual information of party 1 and party 2 is the information overlap
between the data possessed by the two parties. The binary entropy of party 1 's
data, x, is given by

(21.16)

where Px and 1 - Px are the probabilities of the two outcomes. Similarly party
2' s data, y, has binary entropy

(21.17)

The joint entropy of the two data strings is then given by

H{x, y) = -L.. x,yPx,ylog2Px,y (21.18)

with Px,y the joint probabilities. The mutual information is defined

H{x : y) = H{x) + H{y) - H{x, y) (21.19)

If the two data strings x and yare random then H{x) = H{y) = 1. Suppose
the error probability between the data strings is 8, then the joint probabilities
are given by Po,o = PI,1 = 1 - 8 and PO,1 = PI,O = 8. Thus we find

H{x : y) = 1 + 8log28 + (I - 8)log2{1 - 8) (21.20)

 Quantum key distribution with continuous variables in optics 305

Suppose A is Alice's data string, B is Bob's data string and E is Eve's data
string. Maurer has shown [24] that provided H(A : B) > H(A : E) then it
is in principle possible for Alice and Bob to extract a secret key from the data.
Eve's mutual information with this secret key can be made arbitrarily small.
From Eq.21.20 we see that this condition will be satisfied provided Bob's error
rate is less than Eve's. From Fig. 21.3 we see that provided Alice and Bob's
error rate does not exceed 5% for case (a) or 12% for case (b) then secret key
generation is in principle possible. In the following we will look at a simple
specific example of a secret key generation protocol and evaluate its efficiency.
Because of the transmission errors (and possibly the actions of Eve) Alice
and Bob won't share the same data string. However techniques exist for data
reconciliation which allow Alice and Bob to select with high probability a subset
of their data which is error free, whilst giving Eve minimal extra knowledge.
As a simple example Alice and Bob could perform a parity check on randomly
chosen pairs of bits. If the error rate between Bob and Alice is low then the
probability of both bits being wrong is very low. Thus discarding all pairs which
fail the parity check will lead to a big reduction in errors in the shared data
whilst not revealing the values of the individual bits to Eve. A series of parity
checks will lead with high probability to zero errors. Eve can also remove the
pairs that Bob removes and in a worse case scenario may remove up to the same
number of errors as Bob. But if Eve initially had significantly more errors than
Bob then she will still have significant errors after the reconciliation, whilst
Bob and Alice will have virtually none. The data string length will be reduced
by a factor of approximately 1 - 2BB, where BB is Bob's error probability.
In order to reduce Eve's mutual information to a negligible amount the
technique of privacy amplification is employed [25]. This involves the random
hashing or block coding of the reconciled key into a shorter key. As a simple
example Alice and Bob could randomly pick data strings of length n from
the reconciled key and form a new key from the sum, modulo 2, of each n
unit block. It is important that the privacy amplification is "orthogonal" to
the reconciliation protocol. That is none of the pairs used in the parity checks
should appear together in the privacy amplification blocks. The length of the
new key will be reduced by a factor of 1/ n. The error probability in the new
key will be given by

B - En / 2 n! (1 B)n-2k B2k (21.21)

pa - k=O (2k)!(n - 2k)! -

where B is the error probability of the original string. If B ~ 0, as for Bob and
Alice, then this process introduces virtually no errors. But when Eve copies
this process her errors will be "amplified", hopefully te the point where her
mutual information is negligible. Some caution is required in evaluating Eve's
mutual information now. Just as Bob and Alice were able to select a sub-set
306 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

of results they knew were correct in the reconciliation process, so Eve can also
obtain a (smaller) subset of results for which she has greater confidence. We
make the worst case assumption that after privacy amplification Eve is left with
some small probability, Pr, of possessing certain bits that she knows are right,
and a large probability, 1- Pr, of possessing bits which are completely random.
In such a situation it is appropriate to set Eve's mutual information as

H (A : E) = Pr = 1 - 2Bpae (21.22)

where Bpae is Eve's average error probability, as given by Eq.21.21.

Let us now apply these techniques to the continuous variable protocol of the
previous section to evaluate its security. After Bob has received all the data
from Alice he tells her at which quadrature he was looking at any particular
time and Alice sorts out her sent data accordingly. They then compare a
randomly chosen sub-section of their data (approximately half) and determine
the error rate. For the example in the previous section they expect a base error
rate of 1%. Let them reject the data and start again if they detect an error
rate of ~ 2%. To be cautious, let us assume that in fact the error rate could
have been as high as 2.5%. For sufficiently long data strings there will be
negligible probability of this error rate being exceeded in the undisclosed data
[26]. From Fig. 21.3 (trace (a» we can read off that Eve's error rate must be
~ 10.5%. Applying our simple information reconciliation protocol Bob and
Alice's error probability can be reduced to virtually zero whilst Eve's error rate
is ~8%. We now apply privacy amplification. Fig. 21.4 (trace (a» shows Eve's
mutual information as a function of the block length, n. Clearly Eve's mutual
information is decreasing exponentially as a function of block length. This is
the signature of a secure system. A linear expenditure of resources results in
an exponentially small mutual information with Eve. In fact Bob and Alice can
do better by using a smaller initial signal strength. If Alice reduces the size
of the signal she sends to about half that of the previous example (now with
a signal to noise of about 1OdB) Bob's base error rate will rise to 5%. They
set their error threshold at 6%. To be cautious we assume the error rate could
be as high as 6.5%. From Fig. 21.3 (trace (b» we find that Eve's error rate
must be ~26%. After reconciliation Eve's error rate must still be ~19.5%.
Fig. 21.4 (trace(b» plots Eve's mutual information as a function of the block
length for this situation showing a more rapid decay. This is approximately the
optimum signal strength. However Alice and Bob may also seek to improve
the efficiency of the system by employing more sophisticated reconciliation
and privacy amplification protocols.
To this point we have assumed that the transmission line between Alice and
Bob is lossless. In practice this will not be true. If we make no constraints on
Eve's technical abilities then we must assume that all lost light has fallen into
her hands [27]. Thus we must calculate Eve's potential mutual information
 Quantum key distribution with continuous variables in optics 307

Mutual
Information

20 25 30

Block Length

Figure 21.4 Decay of Eve mutual information as a function of the block length, n, in Alice and
Bob's privacy amplification protocol is plotted for two signal to noise levels of Alice's beam.
Trace (a) is for a signal to noise of 13dB whilst trace (b) is for a signal to noise of IOdB. The
solid traces are exponential fits.

from Bob's error rate as if there was no loss, but we must set our error threshold
quite high because the losses will drive up Bob's errors. In this simple approach
it is clear that loss of 50% or more can not be tolerated because Eve's and Bob's
error rates become equal at this point. Indeed as losses approach 50% the
expenditure of resources by Bob and Alice needed to reconcile and privacy
amplify will increase rapidly.
Let us estimate by what factor the length of the final secure key would be
reduced over the length of the original string sent by Alice in a system with
25% loss. Consider an original signal to noise of about lOdB, leading to a base
error rate with 25% loss of 7.7%. Setting as before our maximum error rate
1.5% above the base rate at 9.3% we can bound Eve's error rate at ~ 16.3%.
Bob and Alice sacrifice half their data in this step. Reconciliation will reduce
Bob and Alice's data string by a factor of 0.81 and leave Eve with an error
rate ~7%. If we require that Eve's mutual information be ~0.001 for the
transmission to be considered secure then we find a block length of n = 46 is
required in the privacy amplification step. Thus the secure key will be reduced
by a factor of 0.5 x 0.81 x 0.02 = 0.01. (A similar estimate for the optimum
no loss case gives a reduction factor of 0.025) Data transmission via rf signals
is a mature technology and bit transmission rates of 100 MHz would seem
quite reasonable. Thus secure key transmission rates of a a MHz would seem
practical under these conditions. This is about three orders of magnitude better
than what is presently achievable with single photon schemes. On the other
hand single quanta schemes can tolerate much higher losses [4].
The loss problem can be circumvented by using postselection techniques [28]
or reverse reconcilliation [29]. Both these techniques exploit the asymmetry
in the correlations between Eve and Alice's data and Eve and Bob's data, ie
308 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

the fact that H(B : E) < H(A : E). Using these techniques there is no in
principle loss bound, just as in the discrete case.

6. SQUEEZED STATE QUANTUM CRYPTOGRAPHY

The preceding discussion has shown that a cryptographic scheme based on
coherent light can produce secure keys with an efficiency of at least 1/40 -t
1/100. We now consider whether squeezed light can offer improved efficiency.

ALICE BOB
,----------1 r--------i
I
---~:
+ - : I I
t---c>l'--'-----.----- - - - HD I
I amplitude I
sum or I
phase I
difference I

Figure 21.5 Schematic of squeezed light cryptographic set-up. Sqza and sqzb are phase locked
squeezed light sources. Rna and Rnb are independent random number sources. Bs and pbs
are non-polarizing and polarizing beamsplitters respectively. Half-wave plates to rotate the
polarizations are indicated by >../2 and optical amplification by A. The 7r /2 phase shift is also
indicated. HD stands for homodyne detection system.

The set-up is shown in Fig. 21.5. Once again Alice encodes her number
strings digitally, but now she impresses them on the amplitude quadratures
of two, phase locked, amplitude squeezed beams, a and b, one on each. A
7r /2 phase shift is imposed on beam b and then they are mixed on a 50:50
beam splitter. The resulting output modes, c and d, are given by

c /{(a + ib)

d = /{(a-ib) (21.23)

These beams are now in an entangled state which will exhibit Einstein, Podol-
sky, Rosen (EPR) type correlations [30, 32]. Negligible information about the
signals can be extracted from the beams individually because the large fluctua-
tions of the anti-squeezed quadratures are now mixed with the signal carrying
squeezed quadratures. One of the beams, say c, is transmitted to Bob. The
other beam, d, Alice retains and uses homodyne detection to measure either
its amplitude or phase fluctuations, with respect to a local oscillator in phase
with the original beams a and b. She randomly swaps which quadrature she
measures, and stores the results. Bob, upon receiving beam c, also randomly
 Quantum key distribution with continuous variables in optics 309

chooses to measure either its amplitude or phase quadrature and stores his
results. After the transmission is complete Alice sends the results of her mea-
surements on beam d to Bob on an open channel. About half the time Alice
will have measured a different quadrature to Bob in a particular time window.
Bob discards these results. The rest of the data corresponds to times when
they both measured the same quadratures. If they both measured the amplitude
quadratures of each beam Bob adds them together, in which case he can obtain
the power spectrum

V+ = < I(ct + c) + (dt + d)12 >

Vs,a + Vn~a (21.24)

where the tilde indicate Fourier transforms. Thus he obtains the data string
impressed on beam a, Vs,a, imposed on the sub-QNL noise floor of beam a,
Vn\ . Alternatively if they both measured the phase quadratures of each beam,
Bob subtracts them, in which case he can obtain the power spectrum

V- < I(ct - c) - (dt - d)12 >

Vs,b + Vn~b (21.25)

i.e. he obtains the data string impressed on beam b, Vs,b, imposed on the sub-
QNL noise floor of beam b, Vn\ . Thus the signals lie on conjugate quadratures
but both have sub-QNL noise fl'oors. This is the hallmark of the EPR correlation
[33]. As for the coherent state case Alice and Bob now compare some sub-set
of their shared data and check for errors. If the error rate is sufficiently low they
deem their transmission secure and use reconciliation and privacy amplification
on the undisclosed sub-set of their data to produce a secure key.
Consider now eavesdropper strategies. Eve must intercept beam c if she is
to extract any useful information about the signals from the classical channel
(containing Alice's measurements of beam d) sent later. She can adopt the
guessing strategy by detecting a particular quadrature of beam c and then using
a similar apparatus to Alice's to re-send the beam and a corresponding classical
channel later. As before she will only guess correctly what Bob will measure
half the time thus introducing a BER of 25%. Instead she may try simultaneous
detection of both quadratures of beam c. As in the coherent case the noise she
introduces into her own measurement (Vi') and that she introduces into Bob's
(Vf) are in general limited according to Eqs.21.l0,21.11 and 21.12. However
now the consequences of these noise limits on the signal to noise transfers that
Eve and Bob can obtain behave quite differently because the signals they are
trying to extract lie on sub-QNL backgrounds. The maximum signal transfer
310 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

coefficients that Eve can extract are given by

TE+ --
2Vn~a Vn~b + vt (Vn~a + Vn~b)
(Vi + 2Vn~a) vn~b
(21.26)

Similarly Bob's are

T+
B
2Vn~a vn~b + VI (Vn~a + Vn~b)
(Vi + 2Vn~a)Vn~b
TB = (21.27)

To achieve maximum security we require that the anti-squeezed quadratures of

the beams have large excess noise. This could easily be arranged experimen-
tally. The maximum signal transfer coefficients (Eq.21.26 and Eq.21.27) then
reduce to

Vn~a +O.5Vt
Vn\ ,
T-
E (21.28)

and similarly Bob's are

T+
B
Vn~a + O.5VI
V/b,
TB = (21.29)

For the squeezed noise floors the same (Vn~a = Vn~b = Vn ) we find the signal
transfers are restricted via
2 1 1
411: ( - - 1)(- - 1) > 1 (21.30)
n T+ T- -
E E

2 1 1
411: ( - - 1)(- - 1) > 1 (21.31)
n T+ T- -
E B
 Quantum key distribution with continuous variables in optics 311

2 1 1
4Vn (T+ - l)(T- - 1) ~ 1 (21.32)
B E
It is straightforward to show that a symmetric attack on both quadratures is
Eve's best strategy as it leads to a minimum disturbance in both her and Bob's
measurements. Using this symmetry to simplify Eq.21.30 leads to the following
general restriction on the signal transfer Eve can obtain:

T± < 2Vn (21.33)

E - 2Vn + 1
Once the squeezing exceeds 3 dB (Vn = 0.5) the signal to noise that Eve
can obtain simultaneously is reduced below that for the coherent state scheme.
In the limit of very strong squeezing (Vn -+ 0) Eve can extract virtually
no information simultaneously. Similarly Bob's signal transfer is restricted
according to:

(21.34)

If squeezing is strong then almost any level of interception by Eve will result
in very poor signal transfer to Bob. In Fig. 21.6 we show plots of error rates
of Bob versus minimum error rates of Eve for various levels of squeezing.
In comparison with the coherent scheme (Fig. 21.3) it can be seen that larger
disturbances are caused in Bob's information for the same quality of Eve's
interception. As a numerical example consider the specific encoding scheme
of section I and suppose the squeezing is 10 dB (Vn = 0.1). Assuming no loss
and using the same assumptions as those used to evaluate the coherent scheme
in the last section we find that a secure key of length 0.07 times the original
data string length can be generated. That is an efficiency of about 1/14, to be
compared to the coherent case of 1/40, a clear improvement.
As for the coherent scheme losses of 50% or more cannot be tolerated in
this simple approach however the more sophisticated protocols mentioned in
the coherent state context can also be implemented with squeezing to achieve
high loss operation.

7. TELEPORTATION AS AN OPTIMUM
EAVESDROPPER STRATEGY
It is interesting to consider what physical techniques Eve could use to realize
the optimal attack strategy we have assumed her capable of throughout this
discussion. Firstly she would need to replace the lossy transmission line that
Bob and Alice are using with her own transmission line of negligible loss. Given
that Bob and Alice will presumably employ the most efficient transmission line
312 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

20
90%

Bob's 15
Error
Probability 10 75%
%
5

10 20 30 40 50
Eve's Error
Probability %

Figure 21.6 Minimum allowable error probabilities in the data of Bob and Eve are plotted for
various levels of squeezing.

they can obtain, Eve's job is not trivial. A conceptually simple strategy for
Eve is a completely passive intervention; also known as the beam splitter attack
[14]. That is she takes only the lost light. This she holds in "quantum memory"
until Bob has revealed the bases in which his measurements were made. At
this point she then measures her held portion of the beam in the relevant bases.
In the presence of transmission efficiency TJ Bob's noise penalty will be

± I-TJ
VB = - - (21.35)
TJ
whilst if Eve obtains all the lost light she will have a noise penalty

(21.36)

thus saturating the inequalities of Eq.21.11 and 21.12.

In a low loss situation Eve needs to make an active interception of the beam.
An optimum technique in this situation is for Eve to make an optimal clone
of the beam which she again holds in quantum memory until Bob reveals his
bases. This technique also saturates the inequalities of Eq.21.11 and 21.12
[13].
The above technique requires quantum memory. Eve can also use contin-
uous variable teleportation [35, 36, 37] in the following way as an optimum
eavesdropper strategy without requiring quantum memory.
Quantum teleportation uses shared entanglement to convert quantum in-
formation into classical information and then back again (see Fig. 21.6). In
particular continuous variable teleportation uses 2-mode squeezed light as its
 Quantum key distribution with continuous variables in optics 313

entanglement resource. In the limit of very strong squeezing no information

about the teleported system can be extracted from the classical channel but a
perfect reproduction of the quantum system can be retrieved. On the other
hand with lower levels of squeezing some information about the system can be
obtained from the the classical channel but at the expense of a less than perfect
reproduction. We show in the following that under particular operating condi-
tions the disturbance in the teleported state is precisely the minimum required
by the generalized uncertainty principle, given the quality of information that
can be extracted from the classical channel. Teleportation thus constitutes an
optimum eavesdropper strategy.
Eve's strategy would be to send the field she intercepts from Alice through a
teleporter, adjusted such that she can read some information out of the classical
channel, but still reconstruct the field sufficiently well such that Bob and Alice
don't see a large error rate. The classical channel of a lossless continuous
variable teleporter can be written [11, 32]

(21.37)

where lin is the annihilation operator of the input to the teleporter and = 31
VGV1 + VG - 1v~ is the annihilation operator for one of the entangled beams.
The Vi are the vacuum mode inputs to the squeezers, G is the parametric gain
of the squeezers and K > > 1 is the measurement amplification factor. Being a
classical channel simultaneous measurements of both quadratures can be made
without additional penalty thus immediately Eve's measurement penalty is

Vf = 2G-l (21.38)

For no squeezing (G = 1) Vf = 1, the minimum possible for simultaneous

detection of both quadratures (see Eq.21.1O). For large squeezing (G > > 1)
vf become very large and Eve can obtain little information from the classical
channel.
The output of the teleporter is given by
A A At A
fout )..fin + i1 - i2
)..lin + ()"Va - vG -1)vi + (Va - )"vG - I)V2(21.39)

where ).. is the gain of the teleporter and 32 = VGV2 + VG - 1 is the vt

annihilation operator for the other entangled beam. Thus Bob's measurement
penalty for ideal measurements of either of the quadratures is

(21.40)
314 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Eve

from Alice Fc to Bob

/
"

Entanglement V sq

Figure 21.7 Schematic of teleportation being used as an optimum eavesdropper strategy.

If Eve operates the teleporter with gain [38]

1 + ~~
Aopt = 1 _ V2 (21.41)
sq

where Vsq = (va - y'G-=1) 2 , then Bob's noise penalty is

± 1
VB (Aopt) = 2G _ 1 (21.42)

and so Eve causes the minimum allowable disturbance, i.e. vfv: = 1.

8. CONCLUSION
In this chapter we have investigated continuous variable quantum cryptog-
raphy as it could be realized in optics by analysing the security and efficiency
of specific implementations of two systems based on coherent and squeezed
state light respectively. An Eve employing an optimal eavesdropper attack is
assumed throughout. Possible optimal attack strategies that Eve could employ
are outlined.
We find that the coherent scheme can be made secure, but is not very efficient.
None-the-Iess, given the maturity of optical communication technology based
on rf modulation, this system may prove competitive with discrete schemes.
The squeezed state scheme can also be made secure and in principle is more
efficient than the coherent state system.
We have looked at simple protocols throughout this analysis which we
hope clearly illustrate the basic principles. However we have noted that more
sophisticated encoding, reconciliation and privacy amplification techniques
would lead to significant improvements in performance.
 Quantum key distribution with continuous variables in optics 315

Acknowledgements
We thank Michael Nielsen, Christine Silberhorn and Philippe Grangier for
useful discussions. This work was supported by the Australian Research Coun-
cil.

References
[1] S. Wiesner, Sigact News, 15, 78 (1983), C. H. Bennett and G. Brassard,
Proc. IEEE Int. Conf. on Computers, Systems and Signal Processing (Ban-
galore), 175 (1984).
[2] C. H. Bennett, Phys. Rev. Lett.68, 3121 (1992).
[3] A. K. Ekart, Phys. Rev. Lett.67, 661 (1991).
[4] W. T. Buttler et al, Phys. Rev. A57, 2379 (1998).
[5] H. Zbinden et aI, Appl.Phys.B 67, 743 (1998).
[6] Y. Mu et aI, Opt.Comm. 123, 344 (1996).
[7] T. C. Ralph, Phys. Rev. A61 010303(R) (1999).
[8] M. Hillery, Phys. Rev. A61 022309 (2000).
[9] M. D. Reid, Phys. Rev. A62 062308 (2000).
[10] Ch .Silberhorn, N .Korolkova and G .Leuchs, Phys. Rev. Lett.88, 167902
(2002).
[11] T. C. Ralph, Phys. Rev. A62 062306 (2000).
[12] D. Gottesman and J. Preskill, Phys. Rev. A63, 022309 (2001).
[13] N. J. Cerf, M. Levy, G. Van Assche, Phys. Rev. A63, 052311 (2001).
[14] F. Grosshans and P. Grangier, Phys. Rev. Lett.88 057902 (2002).
[15] D. F. Walls and G. J. Milburn, Quantum Optics (Springer-Verlag, Berlin,
1994).
[16] C. A. Fuchs and A. Peres, Phys. Rev. A53, 2038 (1996), c. A. Fuchs,
N. Gisin, R. B. Griffiths, C. -So Niu and A. Peres, Phys. Rev. A56, 1163
(1997), I. Cirac and N. Gisin, Phys.Lett.A 229, 1 (1997).
[17] D. Mayers, Advances in Cryptology, Proceedings of Crypto '96, 343
(Springer-Verlag, 1996).
[18] H.-K. Lo and H. F. Chau, Science 283,2050 (1999).
[19] We only explicitly consider individual eavesdropper attacks here.
[20] C. E. Shannon, Bell System Tech. J. 27, 623 (1948).
[21] Y. Yamamoto and H. A. Haus, Rev.Mod.Phys., 58, 1001 (1986).
[22] E. Arthurs and M. S. Goodman, Phys. Rev. Lett.60, 2447 (1988).
316 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[23] A. Yariv, Optical Electronics in Modern Communications (Oxford Uni-

versity Press, 5th Edition, New York 1997).
[24] U. M. Maurer, IEEE Trans.lnf.Theo. 39, 1733 (1993).
[25] C. H. Bennett, G. Brassard, C. Crepeau and U. M. Maurer, IEEE
Trans.lnf.Theo. 41, 1915 (1995).
[26] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum
Information (Cambridge University Press, Cambridge 2000).
[27] This is a stronger assumption about Eve's capabilities than was used in
assessing the effect of losses in Ref.[IO].
[28] Ch. Silberhorn, T. C. Ralph, N. Liitkenhaus, G. Leuchs, quant-phl0204064
(2002).
[29] F. Grosshans, P. Grangier, quant-phl0204127 (2002).
[30] A. Einstein, B. Podolsky and N. Rosen, Phys. Rev. 47, 777 (1935).
[31] G. Yeoman and S. M. Barnett, Journal Mod. Opt. 40, 1497 (1993).
[32] T. C. Ralph and P. K. Lam, Phys. Rev. Lett.81, 5668 (1998).
[33] Z. Y. Ou, S. F. Pereira, H. J. Kimble, and K. C. Peng, Phys. Rev. Lett.68,
3663 (1992).
[34] S. M. Barnett and S. J. D. Phoenix, Phil.Trans.R.Soc.Lond.A 354, 793
(1996).
[35] L. Vaidman, Phys. Rev. A49, 1473 (1994).
[36] S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett.80, 869 (1998).
[37] A Furusawa, J L Sorensen, S L Braunstein, C A Fuchs, H J Kimble and
E S Polzik, Science, 282, 706 (1998).
[38] The gain condition Aopt corresponds to the point of maximum signal
transfer on the T-V graph of Reference [32]
Chapter 22

SECURE QUANTUM KEY DISTRIBUTION

USING SQUEEZED STATES

Daniel Gottesman
EECS: Computer Science Div., University of California, Berkeley, CA 94720, USA
[email protected]

John Preskill
California Institute of Technology, Pasadena, CA 91125, USA
[email protected]

Abstract We prove the security of a quantum key distribution scheme based on transmission
of squeezed quantum states of a harmonic oscillator. Our proof employs quantum
error-correcting codes that encode a finite-dimensional quantum system in the
infinite-dimensional Hilbert space of an oscillator, and protect against errors that
shift the canonical variables p and q. If the noise in the quantum channel is weak,
squeezing signal states by 2.51 dB (a squeeze factor e T = 1.34) is sufficient
in principle to ensure the security of a protocol that is suitably enhanced by
classical error correction and privacy amplification. Secure key distribution can
be achieved over distances comparable to the attenuation length of the quantum
channel.

1. INTRODUCTION
Two of the most important ideas to emerge from recent studies of quantum
information are the concepts of quantum error correction and quantum,key
distribution. Quantum error correction allows us to protect unknown quantum
states from the ravages of the environment. Quantum key distribution allows
us to conceal our private discourse from potential eavesdroppers.
In fact these two concepts are more closely related than is commonly appre-
ciated. A quantum error correction protocol must be able to reverse the effects
of both bit flip errors, which reflect the polarization state of a qubit about the
x-axis, and phase errors, which reflect the polarization about the z-axis. By
317
S.L. Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 317-356.
© 2003 Kluwer Academic Publishers.
318 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

reversing both types of errors, the protocol removes any entanglement between
the protected state and the environment, thus restoring the purity of the state.
In a quantum key distribution protocol, two communicating parties verify
that qubits polarized along both the x-axis and the z-axis can be transmitted
with an acceptably small probability of error. An eavesdropper who monitors
the x-polarized qubits would necessarily disturb the z-polarized qubits, while
an eavesdropper who monitors the z-polarized qubits would necessarily disturb
the x-polarized qubits. Therefore, a successful verification test can show that
the communication is reasonably private, and the privacy can then be amplified
via classical protocols.
In quantum key distribution, the eavesdropper collects information by entan-
gling her probe with the transmitted qubits. Thus both error correction and key
distribution share the goal of protecting quantum states against entanglement
with the outside world.
Recently, this analogy between quantum error correction and quantum key
distribution has been sharpened into a precise connection, and used as the basis
of a new proof of security against all possible eavesdropping strategies [18].
Earlier proofs of security (first by Mayers [13, 14], and later by Biham et
al. [3]) made no explicit reference to quantum error correction; nevertheless,
the connection between quantum error correction and quantum key distribution
is a powerful tool, enabling us to invoke the sophisticated formalism of quantum
error-correcting codes in an analysis of the security of quantum key distribution
protocols.
Also recently, new quantum error-correcting codes have been proposed that
encode a finite-dimensional quantum system in the infinite-dimensional Hilbert
space of a quantum system described by continuous variables [8]. In this paper,
we will apply these new codes to the analysis of the security of quantum key
distribution protocols. By this method, we prove the security of a protocol that
is based on the transmission of squeezed quantum states of an oscillator. The
protocol is secure against all eavesdropping strategies allowed by the principles
of quantum mechanics.
In our protocol, the sending party, Alice, chooses at random to send either a
state with a well defined position q or momentum p. Then Alice chooses a value
of q or p by sampling a probability distribution, prepares a narrow wave packet
centered at that value, and sends the wave packet to the receiving party, Bob.
Bob decides at random to measure either q or p. Through public discussion,
Alice and Bob discard their data for the cases in which Bob measured in a
different basis than Alice used for her preparation, and retain the rest. To
correct for possible errors, which could be due to eavesdropping, to noise
in the channel, or to intrinsic imperfections in Alice's preparation and Bob's
measurement, Alice and Bob apply a classical error correction and privacy
 Secure quantum key distribution using squeezed states 319

amplification scheme, extracting from the raw data for n oscillators a number
k < n of key bits.
Alice and Bob also sacrifice some of their data to perform a verification test to
detect potential eavesdroppers. When verification succeeds, the probability is
exponentially small in n that any eavesdropper has more than an exponentially
small amount of information about the key. Intuitively, this protocol is secure
because an eavesdropper who monitors the observable q necessarily causes a
detectable disturbance of the complementary observable p (and vice versa).
Since preparing squeezed states is technically challenging, it is important
to know how much squeezing is needed to ensure the security of the protocol.
The answer depends on how heavily the wave packets are damaged during
transmission. When the noise in the channel is weak, we show that it suffices in
principle for the squeezed state to have a width smaller by the factor e- r = .749
than the natural width of a coherent state (corresponding to an improvement
by 2.51 dB in the noise power for the squeezed observable, relative to vacuum
noise). It is also important to know that security can be maintained under
realistic assumptions about the noise and loss in the channel. Our proof of se-
curity applies if the protocol is imperfectly implemented, and shows that secure
key distribution can be achieved over distances comparable to the attenuation
length of the channel. Squeezed-state key distribution protocols may have some
practical advantages over single-qubit protocols, in that neither single-photon
sources nor very efficient photodetectors are needed.
Key distribution protocols using continuous variable quantum systems have
been described previously by others [16, 9, 17], but ours is the first complete
discussion of error correction and privacy amplification, and the first proof of
security against arbitrary attacks.
In §2. we review continuous variable quantum error-correcting codes [8] and
in §3. we review the argument [18] exploiting quantum error-correcting codes
to demonstrate the security of the BB84 quantum key distribution scheme [1].
This argument is extended to apply to continuous variable key distribution
schemes in §4. and §5. Estimates of how much squeezing is required to ensure
security of the protocol are presented in §6. The effects on security of losses
due to photon absorption are analyzed in §7., and §8. contains conclusions.

2. CODES FOR CONTINUOUS QUANTUM

VARIABLES
We begin by describing codes for continuous quantum variables [8]. The
two-dimensional Hilbert space of an encoded qubit embedded in the infinite-
dimensional Hilbert space of a system described by canonical variables q and
p (satisfying [q,p] = i) can be characterized as the simultaneous eigenspace of
320 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

the two commuting operators

Sq = ei (2yfii')q, Sp = e- i (2y0i')p , (22.1)

the code's "stabilizer generators." If the eigenvalues are Sq = Sp = 1, then the

allowed values of q and p in the code space are integer mUltiples of ...(ii, and
the codewords are invariant under shifts in q or p by integer multiples of 2...(ii.
Thus an orthogonal basis for the encoded qubit can be chosen as
00

10) ex: L
8=-00
Iq = (28) . Vi)

00

ex: L
8=-00
Ip = 8· Vi),
00

11) ex: L
8=-00
Iq = (28 + 1) . ../if)

L
00

ex: (-1)8Ip = 8· ../if) . (22.2)

8=-00

The operators

e- i (y0i')p , (22.3)

commute with the stabilizer generators and so preserve the code subspace; they
act on the basis eq. (22.1) according to

Z: 10) ---+ 10) , 11) ---+ -11) ,

X: 10) ---+ 11) , 11) ---+ 10) . (22.4)

This code is designed to protect against errors that induce shifts in the values
of q and p. To correct such errors, we measure the values of the stabilizer
generators to determine the values of q and p modulo y7f, and then apply a
shift transformation to adjust q and p to the nearest integer multiples of ../if. If
the errors induce shifts 6.q, 6.p that satisfy

l6.ql < ../if/2, l6.pl < ../if/2, (22.5)

then the encoded state can be perfectly restored.

A code that protects against shifts is obtained for any choice of the eigen-
values of the stabilizer generators. The code with

S p -- e-2rrirjJp
, (22.6)
 Secure quantum key distribution using squeezed states 321

can be obtained from the <pq = <pp = 0 code by applying the phase space
translation operator

(22.7)

the angular variables <pq and <pp E (-1/2,1/2] denote the allowed values of
q/ ...;rr
and p / ...;rr
modulo an integer. In this code space, the encoded operations
Z and X (which square to the identity) can be chosen to be
(22.8)

The code with stabilizereq. (22.1) can be generalized in a variety of ways [8].
For example, we can increase the dimension of the protected code space, and
we can modify the code to protect against shifts that are asymmetric in q and
in p. If we choose the stabilizer to be

exp [i(v'2nd). (q/a)] ,

exp [-i( v'2nd) . (pa)] , (22.9)

where d is a positive integer and a is a positive real number, then the code has
dimension d and protects against shifts that satisfy

I~ql <"2. Vd:'

a fj:;
I~pl<-·
1
2a
fin- .
d
(22.10)

The codewords eq. (22.1) are nonnormalizable states, infinitely "squeezed"

in q and p. In practice, we must always work with normalizable finitely
squeezed states. For example, a Gaussian approximation 16} to the ideal
codeword Ii)} of the d = 2, a = 1 code, characterized by squeezing parameters
~q, ~p « 1, is

16} ::::::
(4)
:;
1/4
1 00
-00 dq Iq} e-HA~)q2
00

x L e- ~(q_2sy'1r)2 / A~

::::::
s=-oo

- 1
'Jr1/4
1-00
00
dp Ip} e - 21(A2)
q P
2

x L00
e-~(p-sy'1r)2/A~ j (22.11)
s=-oo

the approximate codeword 16} can be obtained by subjecting Ii)} to shifts in q

and p governed by Gaussian distributions with widths ~q and ~p respectively.
322 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

If b. q and b. p are small, then in principle these shifts can be corrected with high
probability: e.g, for b. q = b.. p == b., the probability that a shift in q or p causes
an uncorrectable error is no worse than the probability that the size of the shift
exceeds ...fif/2, or

Error Proh < _2_

J 1f b. 2
1v:rr
00

/2
dq e- q2 /6. 2

2b..
< -exp(-1f/4b. 2 ). (22.12)
1f
For the d = 2 code with ai-I, this same estimate of the error probability
applies if we rescale the widths appropriately,
b. q = b.. . a , b.. p = b../ a . (22.13)
We can concatenate a shift-resistant code with an [[n, k, d]] stabilizer quan-
tum code. That is, first we encode (say) a qubit in each of n oscillators; then k
better protected qubits are embedded in the block of n. If the typical shifts are
small, then the qubit error rate will be small in each of the n oscillators, and
the error rate in the k protected qubits will be much smaller. The quantum key
distribution protocols that we propose are based on such concatenated codes.
We note quantum codes for continuous quantum variables with an infinite-
dimensional code space were described earlier by Braunstein [5], and by Lloyd
and Slotine [11]. Entanglement distillation protocols for continuous variable
systems have also been proposed [15, 7].

3. QUANTUM KEY DISTRIBUTION AND QUANTUM

ERROR-CORRECTING CODES
Now let's recall the connection between stabilizer quantum codes and quan-
tum key distribution schemes [18].
We say that a protocol for quantum key distribution is secure if (1) the
eavesdropper Eve is unable to collect a significant amount of information about
the key without being detected, (2) the communicating parties Alice and Bob
receive the same key bits with high probability, and (3) the key generated is
essentially random. Then if the key is intercepted, Alice and Bob will know
it is unsafe to use the key and can make further attempts to establish a secure
key. If eavesdropping is not detected, the key can be safely used as a one-time
pad for encoding and decoding. 1
Establishing that a protocol is secure is tricky, because there inevitably will
be some noise in the quantum channel used to distribute the key, and the effects
of eavesdropping could be confused with the effects of the noise. Hence the
protocol must incorporate error correction to establish a shared key despite the
noise, and privacy amplification to control the amount of information about the
key that can be collected by the eavesdropper.
 Secure quantum key distribution using squeezed states 323

In the case of the BB84 key distribution invented by Bennett and Bras-
sard [1], the necessary error correction and privacy amplification are entirely
classical. Nevertheless, the formalism of quantum error correction can be use-
fully invoked to show that the error correction and privacy amplification work
effectively [18]. The key point is that if Alice and Bob carry out the BB84
protocol, we can show that the eavesdropper is no better off than if they had
executed a protocol that applies quantum error correction to the transmitted
quantum states. Appealing to the observation that Alice and Bob could have
applied quantum error correction (even though they didn't really apply it), we
place limits on what Eve can know about the key.

3.1 ENTANGLEMENT DISTILLATION

First we will describe a key distribution protocol that uses a quantum error-
correcting code to purify entanglement, and will explain why the protocol is
secure. The connection between quantum error correction and entanglement
purification was first emphasized by Bennett et al. [2]; our proof of security
follows a proof by Lo and Chau [12] for a similar key distribution protocol.
Later, following Shor and Preskill [18], we will see how the entanglement-
purification protocol is related to the BB84 protocol.
A stabilizer code can be used as the basis of an entanglement-purification
protocol with one-way classical communication [2, 12]. Two parties, both
equipped with quantum computers, can use this protocol to extract from their
initial shared supply of noisy Bell pairs a smaller number of Bell pairs with
very high fidelity. These purified Bell pairs can then be employed for EPR
quantum key distribution. Because the distilled pairs are very nearly pure, the
quantum state of the pairs has negligible entanglement with the quantum state
of the probe of any potential eavesdropper; therefore no measurement of the
probe can reveal any useful information about the secret key.
Let's examine the distillation protocol in greater detail. Suppose that Alice
and Bob start out with n shared EPR pairs. Ideally, these pairs should be in the
state

(22.14)

where 14>+) is the Bell state (100) + Ill))/Y2; however, the pairs are noisy,
approximating 1<I?(n)) with imperfect fidelity. They wish to extract k < n pairs
that are less noisy.
For this purpose, they have agreed in advance to use a particular [[n, k, dJ] sta-
bilizer code. The code space can be characterized as a simultaneous eigenspace
of a set of mutually commuting stabilizer generators {Mi' i = 1, 2, ... , n - k}.
Each Mi is a "Pauli operator," a tensor product of n single-qubit operators where
324 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

each single-qubit operator is one of {I, X, Y, Z} defined by

I=(~ ~), X=(~ ~)

0
Y = ( i
-i)
0 ' Z = (~ ~1) (22.15)

The operations {Xa, Za, a = 1,2, ... , k} acting on the encoded qubits are
Pauli operators that commute with all of the Mi.
The Bell state 11>+) is the simultaneous eigenstate with eigenvalue one of
the two commuting operators X A (8) X Band Z A (8) Z B (where subscripts A and
B indicate whether the operator acts on Alice's or Bob's qUbit). Thus the state
1<I>(n)) is the simultaneous eigenstate with eigenvalue one of the commuting
operators

Mi,A(8)Mi,B, i= 1,2, ... ,n - k ,

Xa,A (8)Xa,B, a = 1,2, ... , k,
Za,A (8) Za,B, a = 1,2, ... , k. (22.16)

Now suppose that Alice and Bob both measure the n - k commuting Mi'S. If
the state they measure is precisely 1<I>(n)), then Alice and Bob obtain identical
measurement outcomes. Furthermore, since their measurements do not disturb
the encoded operations Xa and Za, their measurement would prepare the
encoded state 1<I>(k)) == I<i>+)®k, the encoded state with

Xa,A (8) Xa,B = Za,A (8) Za,B = 1 ,

a = 1,2, ... , k , (22.17)

in the code subspace with the specified values of Mi = ±l.

However, since the initial pairs are noisy, Alice's and Bob's measurement
of the Mi'S need not match perfectly; they should apply error correction to
improve the fidelity of their encoded pairs. Thus Alice broadcasts the values
of the Mi,A'S that she obtained in her measurements. Comparing to his own
measurements, Bob computes the relative syndrome Mi,A . Mi,B. From this
relative syndrome, he infers what recovery operation he should apply to his
qubits to ensure that the Mi,B'S match the Mi,A'S, and he performs this opera-
tion. Now Alice and Bob are in possession of k encoded pairs with improved
fidelity.
These encoded pairs can be used for EPR key distribution. For each a =
1,2, ... , k, Alice and Bob measure Za, obtaining outcomes that are essentially
random and agree with high probability. These outcomes are their shared
private key.
 Secure quantum key distribution using squeezed states 325

3.2 VERIFICATION
If the initial pairs are too noisy, either because of the intervention of an
eavesdropper or for other reasons, then the purification protocol might not
succeed. Alice and Bob need to sacrifice some of their EPR pairs to verify that
purification is likely to work. If verification fails, they can abort the protocol.
Under what conditions will purification succeed? If their pairs were perfect,
each would be in the state Iif>+), the simultaneous eigenstate with eigenvalue one
of the two commuting observables X ® X and Z ® Z. Suppose for a moment,
that each of the pairs is a simultaneous eigenstate of these observables (a Bell
state), but not necessarily with the right eigenvalues: in fact no more than tx
of the n pairs have X ® X = -1, and no more than tz of the n pairs have
Z ® Z = -1. Then, if Alice and Bob use a stabilizer code that can correct
up to t z bit flip errors and up to t x phase errors, the purification protocol will
work perfectly - it will yield the encoded state 1<I>(k)) = I¢+)®k with fidelity
F=1.
Now, the initial n pairs might not all be in Bell states. But suppose that Alice
and Bob were able to perform a Bell measurement on each pair, projecting it
onto a simultaneous eigenstate of X ® X and Z ® Z. Of course, since Alice and
Bob are far apart from one another, they cannot really do this Bell measurement.
But let's nevertheless imagine that they first perform a Bell measurement on
each pair, and then proceed with the purification protocol. Purification works
if the Bell measurement yields no more than tx pairs with X ® X = -1 and
no more than t z pairs with Z ® Z = -1. Therefore, ifthe initial state of the
n pairs has the property that Bell measurement applied to all the pairs will,
with very high probability, produce pairs with no more than t z bit flip errors
and no more than t x phase errors, then we are assured that Bell measurement
followed by purification will produce a very high fidelity approximation to the
encoded state 1<I>(k)).
But what if Alice and Bob execute the purification protocol without first per-
forming the Bell measurement? We know that the purification works perfectly
applied to the space 'Hgood spanned by Bell pairs that differ from Iif>+)®n by
no more than t z bit flip errors and no more than t x phase errors. Let II denote
the projection onto 'Hgood . Then if the protocol is applied to an initial density
operator p of the n pairs, the final density operator p' approximates 1<I>(k)) with
fidelity

(22.18)

Therefore, the fidelity is at least as large as the probability that t z or fewer

bit flip errors and tx or fewer phase errors would have been found if Bell
measurement had been performed on all n pairs.
326 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

To derive the inequality eq. (22.18), we represent p as a pure state IW)SE

of the n pairs (the "system" S) and an ancilla (the "environment" E, which
might be under Eve's control). The recovery superoperator can be represented
as a unitary operator USR that is applied to S and an auxiliary system (the
"reservoir" R) that serves as a repository for the entropy drawn from the pairs
by error correction. Denote the initial pure state of the reservoir by 10) R. Then
the state of system, environment, and reservoir to which the recovery operation
is applied can be resolved into a "good" component

(22.19)

and an orthogonal component

(22.20)

Since the states IWgood)SER and IWbad)SER are orthogonal, the unitary re-
covery operation USR ® IE maps them to states IW~ood)SER and IW~ad)SER
that are also orthogonal to one another. Furthermore, since recovery works
perfectly on the space 1-lgood' we have
, - (k)
IWgood)SER = leI> )s ® ljunk)ER, (22.21)

where the state Ijunk) ER of environment and reservoir has norm

ER(junkljunk)ER = SER(W~oodlw~ood)SER
= SER(WgoodIWgood)SER = tr(ITp) . (22.22)

Thus the fidelity of the recovered state can be expressed as

F SER(W'I (1~(k))S s(~(k)l) ® IERlw')SER

= SER(W~oodl (1~(k))S s(~(k)l) ® IERlw~ood)SER

+ SER(W~adl (I~(k))s s(~(k)l) ® IERlw~ad)SER

tr(ITp) + (~(k)IP~adl~(k)) ~ tr(ITp) , (22.23)

where

(22.24)

eq. (22.18) then follows. The key point is that, because of eq. (22.21), and
because IW~ood)SER and IW~ad)SER are orthogonal, there is no "good-bad"
cross term in eq. (22.22).
Our arguments so far show that Alice and Bob can be assured that entangle-
ment purification will work very well if they know that it is highly unlikely that
 Secure quantum key distribution using squeezed states 327

more than t z bit flip errors or more than t x phase errors would have been found
if they had projected their pairs onto the Bell basis. While they have no way
of directly checking whether this condition is satisfied, they can conduct a test
that, if successful, will provide them with high statistical confidence. We must
now suppose that Alice and Bob start out with more than n pairs; to be definite,
suppose they have about 2n to start, and that they are willing to sacrifice about
half of them to conduct their verification test. Alice randomly decides which
pairs are for verification (the "check pairs") and which are for key distribution
(the "key pairs"), and for each of her check qubits, she randomly decides to
measure either X or Z. Then Alice publicly announces which are the check
pairs, whether she measured X or Z on her half of each check pair, and the
results of those measurements (in addition to the results of her measurements
of the stabilizer generators).
Upon hearing of Alice's choices, Bob measures X or Z on his half of each of
the check pairs; thus Alice and Bob are able to measure X (8) X on about half of
their check pairs, and they measure Z (8) Z on the remaining check pairs. Now
since the check pairs were randomly chosen, the eavesdropper Eve has no way
of knowing which are the check pairs, and she can't treat them any differently
than the key pairs; hence the measured error rate found for the check pairs will
be representative of the error rate that would have been found on the key pairs if
Alice and Bob had projected the key pairs onto the Bell basis. Therefore, Alice
and Bob can use their check data and classical sampling theory to estimate how
many bit flip and phase errors would have been expected if they had measured
the key pairs.
For example, in a sample of N pairs, suppose that if Alice and Bob both
measured Z for all the pairs, a fraction p of their measurements would disagree,
indicating bit flip errors. Then if they randomly sample M < N of the pairs,
the probability distribution for the number M (p - c:) of errors found would be2
P(c:) <exp(-Mc: 2 /2p(1-p)). (22.25)
If Alice and Bob have no a priori knowledge of the value of p, then by Bayes'
theorem, the conditional probability that the total number of errors in the
population is pN, given that there are pzM errors in the sample, is the same as
the probability that there are pzM errors in the sample given that there are pN
errors in the total popUlation. Writing p = pz + c:, the number of errors on the
N-MuntestedpairsisNp-Mpz = (N-M)pz+Nc: = (N-M)·(pz+c:'),
where c:' = N c: / (N - M). Expressing P (c:) in terms of c:' we find

, ( M(N - M)2c:,2)
P(c:) < exp - 2N2 pz (1 - pz) , (22.26)

a bound on the probability that the fraction of the untested pairs with errors is
larger than pz + c:'. In particular, ifthey test about M = n/2 pairs for bit flip
328 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

errors out of a total of about N = n + n/2 pairs, the probability that a fraction
p z + c' of the remaining N - M = n pairs have bit flip errors is

P{c') < exp ( -nc,2 /9pz{1 - pz)) . (22.27)

A similar argument applies to the probability of phase errors. We conclude that

by conducting the verification test, Alice and Bob can be very confident that, if
they had measured Z ® Z (or X ® X) on the n key pairs, no more than (p z +c')n
(or (px + c')n) errors would have been found. By choosing a quantum error-
correcting code that can correct this many errors with high probability, they
can be confident that the encoded state they prepare approximates I~(k)) with
fidelity exponentially close to one.
It is important to emphasize that this argument requires no assumption about
how the errors on different pairs may be correlated with one another. Rather
the argument is applied to a hypothetical situation in which the value of Z ® Z
(or X ® X) already has been measured and recorded for all of the check pairs
and all of the key pairs. Sampling theory is then used to address the question:
how reliably does a "poll" of M bits randomly chosen from among N allow us
to predict the behavior of the rest of the population. Classical sampling theory
can be applied to the values of both Z ® Z and X ® X for the key pairs, because
the operators commute and so are simultaneously measurable in principle [12].
Furthermore, if the state of the encoded pairs that Alice and Bob use for
key distribution is exponentially close to being a pure state, it follows from
Holevo's theorem that Eve's mutual information with the distributed key is
exponentially small [12, 18]. In the worst case, the imperfect fidelity of Alice's
and Bob's pairs is entirely due to Eve's intervention; then the complete state
consisting of the pairs and Eve's probe is pure, and the Von Neumann entropy
S{PE) == -tr PE 10gpE of the state PE of the probe equals the entropy of the
state PAB of the pairs. By extracting a key from their pairs, Alice and Bob in
effect prepare a state for Eve governed by an ensemble with density matrix PE.
According to Holevo's theorem, the mutual information J(AB; E) of this state
preparation with any measurement that Eve can carry out on her probe satisfies

J(AB; E) ::; S(PE) = S(PAB) , (22.28)

and since PAB is very nearly pure, S(PAB) and J(AB; E) are very close to zero.
Specifically, if the fidelity of PAB is F = 1 - 8, then the largest eigenvalue of
PAB is at least 1 - 8. For a system with dimension D, the density matrix with
largest eigenvalue 1 - 8 that has the maximal Von Neumann entropy is

Pmax = diag ( 1 - 8, D ~ 1 ' D ~ 1 ' ... , D ~ 1) . (22.29)

 Secure quantum key distribution using squeezed states 329

for which
S(Pmax) = -(1 - 6) log2(1- 6) - 610g2(6/(D - 1))

= 6· (10~ 2 + log2(D - 1) -log2 6) + 0(62) . (22.30)

Taking D = 22k (the total dimension of Alice's and Bob's code spaces), we
conclude that

S(PAB) :S 6· CO:e 2 + 2k + log2(1/6)) + 0(62) . (22.31)

Finally, we have shown that if the verification test succeeds, then with proba-
bilityexponentially close to one (the probability that the error rate inferred from
the check sample is not seriously misleading), Eve's mutual information with
the key is exponentially small (because the state of the key bits approximates
I<i>(k)) with fidelity exponentially close to one). This proof of security applies
to any conceivable eavesdropping strategy adopted by Eve.
The proof relies on the ability of quantum error-correcting codes to reverse
the errors caused by interactions between the key pairs and Eve's probe. Hence
it may seem odd that the proof works for arbitrary attacks by Eve, since
quantum error correction works effectively only for a restricted class of error
superoperators. Specifically, the error superoperator acting on a block of n
qubits can be expanded in terms of a basis of "Pauli error operators," where in
each term of the expansion bit flip errors and/or phase errors are inflicted on
specified qubits within the block. The encoded quantum information is well
protected only if the error superoperator has nearly all of its support on Pauli
operators that can be corrected by the code, e.g., those with no more than tz
bit flip errors and t x phase errors.
If Eve's probe interacts collectively with many qubits, it may cause more
bit flip or phase errors than the code can correct. But the crucial point is that,
with high probability, an attack that causes many errors on the key bits will
also cause many errors on the check bits, and Alice and Bob will detect Eve's
presence.

3.3 REDUCTION TO THE BB84 PROTOCOL

Since the entanglement distillation protocol requires only one-way classical
communication, this protocol is actually equivalent to one in which Alice,
rather than preparing Bell pairs and sending half of each pair to Bob, instead
prepares an encoded quantum state that she sends to Bob. Using a set of
stabilizer generators on which she and Bob have agreed in advance, Alice
chooses a random eigenvalue for each stabilizer generator Mi; then employing
the corresponding [[n, k, d]] quantum code, she prepares one of 2k mutually
orthogonal codewords.
330 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Alice also decides at random which of her qubits will be used for key
distribution and which will be used for verification. For each of the check
bits, she decides at random whether to send an X eigenstate (with random
eigenvalue) or a Z eigenstate (with random eigenvalue).
Bob receives the qubits sent by Alice, carefully deposits them in his quantum
memory, and publicly announces that the qubits have been received. Alice then
publicly reveals which qubits were used for the key, and which qubits are the
check qubits. She announces the stabilizer eigenValues that she chose to encode
her state, and for each check qubit, she announces whether it was prepared as
an X or Z eigenstate, and with what eigenvalue.
Once Bob learns which qubits carry the encoded key information, he mea-
sures the stabilizer operators and compares his results with Alice's to obtain
a relative error syndrome. He then performs error recovery and measures the
encoded state to decipher the key.
Bob also measures the check qubits and compares the outcomes to the values
announced by Alice, to obtain an estimate of the error rate. If the error rate is
low enough, error recovery applied to the encoded key bits will succeed with
high probability, and Alice and Bob can be confident in the security of the key.
If the error rate is too high, Bob informs Alice and they abort the protocol.
As described so far, the protocol requires that Alice and Bob have quantum
memories and quantum computers that are used to store the qubits, measure
stabilizer generators, and correct errors. But ifthey use a stabilizer code of the
CSS (Calderbank-Shor-Steane) type [6, 19], then the protocol can be simplified
further. The crucial property of the CSS codes is that there is a clean separation
between the syndrome information needed to correct bit flip errors and the
syndrome information needed to correct phase errors.
A CSS quantum stabilizer code is associated with a classical binary linear
code CIon n bits, and a subcode C2 C C 1 . Let HI denote the parity check
matrix of C 1 and H2 the generator matrix for the code C2 (and hence the parity
check matrix of the dual code Cf). The stabilizer generators of the code are
of two types. Associated with the i th row of the matrix HI is a "Z -generator,"
the tensor product of I's and Z's

(22.32)

and associated with the ith row of H2 is an "X -generator," the tensor product
of I's and X's

(22.33)

Since HI has n - kl rows, where kl = dim(Cd, and H2 has k2 rows, where

k2 = dim( C 2) there are all together n - kl + k2 stabilizer generators, and the
dimension of the code space (the number of encoded qubits) is k = kl - k2 .
 Secure quantum key distribution using squeezed states 331

From measurements of the Z generators, bit flip errors can be diagnosed, and
from measurement of the X generators, phase errors can be diagnosed.
The elements of a basis for the code space with eigenvalues of stabilizer
generators

MZ,i = (_1)8 i , MX,i = (_1)ti (22.34)

are in one-to-one correspondence with the k cosets of O2 in 0 1 ; they can be

chosen as

(22.35)

here v E 0 1 is a representative of a C2 coset, and x, z are n-bit strings satisfying

(22.36)

Thus, to distribute the key, Alice chooses x and z at random, encodes one of
the 11fJ (v )) X,z 's, and sends the state to Bob. After Bob confirms receipt, Alice
broadcasts the values of x and z. Bob compares Alice's values to his own
measurements of the stabilizer generators to infer a relative syndrome, and
he performs error correction. Then Bob measures Z of each of his n qubits,
obtaining a bit string v + w + x. Finally, he subtracts x and applies H2 to
compute H 2 v, from which he can infer the coset represented by v and hence
the key.
Now notice that Bob extracts the encoded key information by measuring Z
of each of the qubits that Alice sends. Thus Bob can correctly decipher the
key information by correcting any bit flip errors that occur during transmission.
Bob does not need to correct phase errors, and therefore he has no use for the
phase syndrome information; hence there is no need for Alice to send it.
Without in any way weakening the effectiveness of the protocol, Alice can
prepare the encoded state 11fJ (v ) ) x ,z, but discard her value of z, rather then
transmitting it; thus we can consider the state sent by Alice to be averaged over
the value of z. Averaging over the phase (-1 y.w destroys the coherence of
the sum over w E O2 in 11fJ(v))x,z; in effect, then, Alice is preparing n qubits
as Z eigenstates, in the state Iv + w + x), sending the state to Bob, and later
broadcasting the value of x. We can just as well say that Alice sends a random
string u, and later broadcasts the value of u + v. Bob receives u + e (where e
has support on the bits that flip due to errors) extracts v + e, corrects it to the
nearest 01 codeword, and infers the key, the coset v + O2 •
Alice and Bob can carry out this protocol even if Bob has no quantum
memory. Alice decides at random to prepare her qubits as X or Z eigenstates,
with random eigenvalues, and Bob decides at random to measure in the X
or Z basis. After public discussion, Alice and Bob discard the results in the
332 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

cases where they used different bases and retain the results where they used
the same basis. Thus the protocol we have described is just the BB84 protocol
invented by Bennett and Brassard [1], accompanied by classical error correction
(adjusting v + e to a C1 codeword) and privacy amplification (extracting the
coset v + C2).
What error rate is acceptable? In a random CSS code, about half of the
n - k generators correct bit flips, and about half correct phase flips. Suppose
that the verification test finds that bit flip errors (ZA ® ZB = -1) occur with
probability pz and phase errors (XA ® XB = -1) occur with probability px.
Classical coding theory shows that a random CSS code can correct the bit flips
with high probability if the number of typical errors on n bits is much smaller
than the number of possible bit flip error syndromes, which holds provided that

( n ) 2-(n-k)/2 rv 2nH2 (pz)-(n-k)/2 « 1 (22.37)

npz '

where H2(X) = -x 10g2 X - (I-x) 10g2(1- x) is the binary entropy function.

Similarly, the phase errors can be corrected with high probability provided the
same relation holds with pz replaced by Px. Therefore, asymptotically as
n -+ 00, secure key bits can be extracted from transmitted key bits at any rate
R satisfying

k
R= - < 1-2H2(pz) ,
n
k
R =- < 1 - 2H2(PX) . (22.38)
n
This upper bound on R crosses zero at pz (or px) = .1100. We conclude that
secure key distribution is possible if p x,z < 11 %.
The random coding argument applies if the errors in the key qubits are
randomly distributed. To assure that this is so, we can direct Alice to perfonn
a random pennutation of the qubits before sending them to Bob. After Bob
confinns receipt, Alice can broadcast the pennutation she perfonned, and Bob
can invert it.
Again, the essence of this argument is that the amount of infonnation that an
eavesdropper could acquire is limited by how successfully we could have carried
out quantum error correction if we had chosen to - and that this relation holds
irrespective of whether we really implemented the quantum error correction or
not.
Other proofs of the security of the BB84 protocol have been presented [13, 3],
which don't make direct use of this connection with quantum error-correcting
codes. However, these proofs do use classical error correction and privacy
amplification, and they implicitly exploit the structure of the CSS codes.
 Secure quantum key distribution using squeezed states 333

3.4 IMPERFECT SOURCES

Our objective in this paper is to analyze the security of key distribution
schemes that use systems described by continuous quantum variables. The anal-
ysis will follow the strategy we have just outlined, in which an entanglement-
purification protocol is reduced to a protocol that does not require the distribu-
tion of entanglement. But first we need to discuss a more general version of
the argument.
In the entanglement-purification protocol, whose reduction to the BB84 pro-
tocol we have just described, there is an implicit limitation on the eavesdropper's
activity. We have assumed that Alice prepares perfect entangled pairs in the
state 1</>+), and then sends half of each pair to Bob. Eve has been permitted to
tamper with the qubits that are sent to Bob in any way she chooses, but she has
not been allowed any contact with Alice's qubits. Therefore, if we imagine that
Alice measures her qubits before sending to Bob, we obtain a BB84 protocol
in which Alice is equipped with a perfect source of polarized qubits. When she
sends a Z eigenstate, the decision to emit a 10) or a 11) is perfectly random, and
the state emerges from her source with perfect fidelity. Similarly, when she
sends an X eigenstate, the decision to send I±) == (10) ± 11) )/v'2 is perfectly
random, and the state is prepared with perfect fidelity. Furthermore, Eve has
no knowledge of what Alice's source does, other than what she is able to infer
by probing the qubits as they travel to Bob.
Security can be maintained in a more general scenario. In the entanglement-
purification protocol, we can allow Eve access to Alice's qubits. As long
as Eve has no way of knowing which pairs Alice and Bob will select for
their verification test, and no way of knowing whether the check pairs will be
measured in the Z or X basis, then the protocol still works: eavesdropping can
be detected irrespective of whether Eve probes Alice's qubits, Bob's qubits, or
both.
Now if we imagine that Alice measures her qubits before sending to Bob, we
obtain a BB84-like protocol in which Alice's source is imperfect and/or Eve is
able to collect some information about how Alice's source behaves. Our proof
that the BB84-like protocol is secure still works as before. However the proof
applies only to a restricted type of source - it must be possible to simulate
Alice's source exactly by measuring half of a two-qubit state.
To be concrete, consider the following special case, which will suffice for
our purposes: Alice has many identical copies of the two-qubit state PAB. To
prepare a "Z-state" she measures qubit A in the basis {IO)A, 11)A}. Thus she
334 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

sends to Bob one of the two states

A(OIPAB 10) A
Po =
tr(A(OlpABIO)A) ,
A(IIPABll)A
PI = (22.39)
tr (A(IIPABll)A) ,

chosen with respective probabilities

Prob(O) = tr(A(OlpABIO)A) '

Prob(l) = tr(A(1IPABll)A). (22.40)

Similarly, to prepare an X -state she measures in the basis {I +), I-)}, sending
one of
A(+lpABI+)A
P+ tr (A(+lpABI+)A) ,
A(-lpABI-)A
P- = tr (A(-lpABI-)A) ,
(22.41)

chosen with respective probabilities

Prob(+) tr(A(+lpABI+)A) ,
Prob(-) tr (A(-lpABI-)A) . (22.42)

Unless the state PAB is precisely the pure state Icf;+), Alice's source isn't doing
exactly what it is supposed to do. Depending on how PAB is chosen, the source
might be biased; for example it might send Po with higher probability than Pl.
And the states Po and PI need not be the perfectly prepared 10) and 11) that the
protocol calls for.
Now suppose that Alice's source always emits one of the states Po , PI, P+, P_,
and that after the qubits emerge from the source, Eve is free to probe them any
way she pleases. Even though Alice's source is flawed, Alice and Bob can
perform verification, error correction, and privacy amplification just as in the
BB84 protocol. To verify, Bob measures Z or X, as before; if he measures Z,
say, they check to see whether Bob's outcome 10) or 11) agrees with whether
Alice sent Po or PI (even though the state that Alice sent may not have been a
Z eigenstate). Thereby, Alice and Bob estimate error rates p z and p x. If both
error rates are below 11 %, then the protocol is secure.
We emphasize again that the security criterion p x, P z < 11 % applies not
to all sources, but only to the restricted class of imperfect sources that can
be simulated by measuring half of a (possible noisy) entangled state. To give
an extreme example of a type of source to which the security proof does not
apply, suppose that Alice always sends the Z -state 10) or the X -state 1+).
 Secure quantum key distribution using squeezed states 335

Clearly the key distribution protocol will fail, even if Bob's bits always agree
with Alice's! Indeed, a source with these properties cannot be obtained by
measuring half of any two-qubit state PAB. Rather, if the source is obtained by
such a measurement, then a heavy bias when we send a Z-state would require
that the error probability be large when we send an X -state.

4. DISTRIBUTING A KEY BIT WITH CONTINUOUS

VARIABLES
Now let's consider how the above ideas can be applied to continuous variable
systems. We will first describe how in principle Alice and Bob can extract good
encoded pairs of qubits from noisy EPR pairs. However, the distillation protocol
requires them to make measurements that are difficult in practice. Then we
will see how key distribution that invokes (difficult) entanglement distillation
can be reduced to key distribution based on (easier) preparation, transmission,
and detection of squeezed states.
Suppose that Alice and Bob share pairs of oscillators. Ideally each pair
has been prepared in an EPR state, a simultaneous eigenstate (let's say with
eigenvalue 0) of qA - qB and PA + PB. Now suppose that Alice measures
the two commuting stabilizer generators defined in eq. (22.1), obtaining the
outcomes
S q,A -- e27riI/Jq A
" S p,A -- e- 27ri I/Jp,A
, (22.43)
or
qA = <Pq,A·.Ji (mod.Ji) ,
PA = <Pp,A·.Ji (mod .Ji) . (22.44)
Now, the initial state was an eigenstate with eigenvalue one of the operators
Sq,A ® s;,1 and Sp,A ® Sp,B. The observables measured by Alice commute
with these, and so preserve their eigenvalues. Thus if the initial EPR state of
the oscillators were perfect, Alice's measurement would also prepare for Bob
a simultaneous eigenstate of the stabilizer generators with
Sq,B = e 27ri I/Jq,B = e 27ri I/Jq,A ,

Sp,B == e- 27ri I/Jp,B = e 27ri I/Jp,A , (22.45)

or
qB = qA (mod.Ji) ,
PB = -PA (mod.Ji) . (22.46)
Similarly, the initial state was an eigenstate with eigenvalue one of the observ-
abIes
(22.47)
336 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

which also commute with the stabilizer generators that Alice measured. Thus
Alice's measurement has prepared an encoded Bell pair in the code space
labeled by (cpq, cpp), the state

14>+)AB = ~ (IO)AIO)B + II)AII)B) . (22.48)

Of course the initial EPR pair shared by Alice and Bob might be imperfect,
and then the encoded state produced by Alice's measurement will also have
errors. But if the EPR pair is not too noisy, they can correct the errors with high
probability. Alice broadcasts her measured values of the stabilizer generators
to Bob; Bob also measures the stabilizer generators and compares his values to
those reported by Alice, obtaining a relative syndrome

(22.49)

That is, the relative syndrome determines the value of qA - qB (mod Vi), and
PA + PB (mod y'i). Using this information, Bob can shift his oscillator's q and
P (by an amount between -Vi/2 and Vi/2) to adjust qA - qB (mod Vi), and
PA + PB (mod y'i) both to zero. The result is that Alice and Bob now share a
bipartite state in the code subspace labeled by (cpq, cpp).
If the initial noisy EPR state differs from the ideal EPR state only by relative
shifts of Bob's oscillator relative to Alice's that satisfy I~ql, l~pl < Vi/2,
then the shifts will be corrected perfectly. And if larger shifts are highly
unlikely, then Alice and Bob will obtain a state that approximates the desired
encoded Bell pair 14>+) with good fidelity. This procedure is a "distillation"
protocol in that Alice and Bob start out with a noisy entangled state in a tensor
product of infinite dimensional Hilbert spaces, and "distill" from it a far cleaner
entangled state in a tensor product of two-dimensional subspaces.
Once Alice and Bob have distilled an encoded Bell pair, they can use it to
generate a key bit, via the usual EPR key distribution protocol: Alice decides
at random to measure either X or Z, and then publicly reveals what she chose
to measure but not the measurement outcome. Bob then measures the same
observable and obtains the same outcome - that outcome is the shared key bit.
How do they measure X or Z? If Alice (say) wishes to measure Z, she
can measure q, and then subtract cPq from the outcome. The value of Z is
determined by whether the result is an even (Z = 1) or an odd (Z = -1)
multiple of Vi. Similarly, if Alice wants to measure X, she measures P and
subtracts CPP - The value of X is determined by whether the result is an even
(X = 1) or an odd (X = -1) multiple of Vi.
Imperfections in the initial EPR pairs are inescapable not just because of
experimental realities, but also because the ideal EPR pairs are unphysical
nonnormalizable states. Likewise, the stabilizer operators cannot even in prin-
ciple be measured with arbitrary precision (the result would be an infinite bit
 Secure quantum key distribution using squeezed states 337

string), but only to some finite m-bit accuracy. Still, if the EPR pairs have rea-
sonably good fidelity, and the measurements have reasonably good resolution,
entanglement purification will be successful.
To summarize, Alice and Bob can generate a shared bit by using the contin-
uous variable code for entanglement purification, carrying out this protocol:

Key distribution with entanglement purification

1: Alice prepares (a good approximation to) an EPR state of two oscillators,
a simultaneous eigenstate of qA - qB = 0 = PA + PB, and sends one of
the oscillators to Bob.
2: After Bob confirms receipt, Alice and Bob each measure (to m bits of
accuracy) the two commuting stabilizer generators of the code, ei (2y0i')q
and e- i (2y0i')p. (Equivalently, they each measure the value of q and P
modulo .,fii.) Alice broadcasts her result to Bob, and Bob applies shifts
in q and P to his oscillator, so that his values of q and P modulo .,fii
now agree with Alice's (to m-bit accuracy). Thus, Alice and Bob have
prepared (a very good approximation to) a Bell state I¢+) of two qubits
encoded in one of the simultaneous eigenspaces of the two stabilizer
operators.
3: Alice decides at random to measure one of the encoded operators X or
Z; then she announces what she chose to measure, but not the outcome.
Bob measures the same observable; the result is the shared bit that they
have generated.
Now notice that, except for Bob's confirmation that he received the states,
this protocol requires only one-way classical communication from Alice to
Bob. Alice does not need to receive any information from Bob before she
measures her stabilizer operators or before she measures the encoded operation
X or Z. Therefore, the protocol works just as well if Alice measures her
oscillator before sending the other one to Bob. Equivalently, she prepares an
encoded state, adopting randomly selected values of the stabilizer generators.
She also decides at random whether the encoded state will be an X eigenstate
or a Z eigenstate, and whether the eigenvalue will be +1 or -1.
Again, since the codewords are unphysical nonnormalizable states, Alice
can't really prepare a perfectly encoded state; she must settle for a "good
enough" approximate codeword.
In summary, we can replace the entanglement-purification protocol with this
equivalent protocol:

Key distribution with encoded qubits

1: Alice chooses random values (to m bits of accuracy) for the stabilizer gen-
erators ei (2y0i')q and e- i (2y0i')p, chooses a random bit to decide whether
338 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

to encode a Z eigenstate or an X eigenstate, and chooses another random

bit to decide whether the eigenvalue will be ± 1. She then prepares (a
good approximation to) the encoded eigenstate of the chosen operator
with the chosen eigenvalue in the chosen code, and sends it to Bob.
2: After Bob confirms receipt, Alice broadcasts the stabilizer eigenvalues
and whether she encoded a Z or an X.
3: Bob measures q or p. He subtracts from his outcome the value modulo
..,fi determined by Alice's announced value of the stabilizer generator,
and corrects the result to the nearest integer multiple of ...(if. He extracts
a bit determined by whether the multiple of ..,fi is even or odd; this is
the shared bit that they have generated.

To carry out this protocol, Alice requires sophisticated tools that enable her to
prepare the approximate codewords, and Bob needs a quantum memory to store
the state that he receives until he hears Alice's classical broadcast. However,
we can reduce the protocol to one that is much less technically demanding.
When Bob extracts the key bit by measuring (say) q, he needs Alice's value
of q modulo ..,fi, but he does not need her value of the other stabilizer generator.
Therefore, there is no need for Alice to send it; surely, the eavesdropper will
be no better off if Alice sends less classical information. If she doesn't send
the value of Sp, then we can consider the protocol averaged over the unknown
value of this generator. Formally, for perfect (nonnormalizable) codewords the
density matrix describing the state that is accessible to a potential eavesdropper
then has a definite value of Sq but is averaged over all possible values of Sp - it is
a (nonnormalizable) equally weighted superposition of all position eigenstates
with a specified value of q mod ...(if; e.g. in the case where Alice prepares a Z
eigenstate, we have

p(¢q, Z = 1)
ex L Iq = (28 + ¢q)v-rr)(q = (28 + ¢q)v-rrl ,
s
p(¢q, Z = -1)
ex L Iq = (28 + 1 + ¢q)v-rr)(q = (28 + 1 + ¢q)J7f1 .
s
(22.50)

Averaged over ¢q as well, Alice is sending a random position eigenstate.

Likewise, in the case where Alice prepares an X eigenstate, she sends a random
momentum eigenstate.
Therefore, the protocol in which Alice prepares encoded qubits can be
replaced by a protocol that is simpler to execute but is no less effective and no
less secure. Instead of bothering to prepare the encoded qubit, she just decides
 Secure quantum key distribution using squeezed states 339

at random to send either a q or p eigenstate, with a random eigenvalue. If Bob

had a quantum memory, he could store the state, and wait to hear from Alice
whether the state she sent was a q or p eigenstate; then he could measure that
observable. Subtracting <pqVi (or <pp..ji) from his measurement outcome, he
would obtain an even or odd multiple of ..ji.
But Bob does not really need the quantum memory. As in the BB84 protocol,
it suffices for Bob to decide at random to measure either q or p, and then
publicly compare his basis with Alice's. They discard the results where they
used different bases and retain the others.
A problem with this procedure is that the position and momentum eigenstates
are unphysical nonnormalizable states, and the probability distribution that
Alice samples to decide on what value of q or p to send is also nonnormalizable.
For it to implementable, we need to modify the procedure so that Alice sends
narrow q or p wave packets, and chooses the position of the center of the wave
packet by sampling a broad but normalizable distribution.
Therefore, Alice and Bob can adopt the following protocol:

Key distribution with squeezed states

1: Alice chooses a random bit to decide whether to send a state squeezed

in q or in p. She samples a (discrete approximation to) a probability
distribution Ppos (q) or Pmom (P) to choose a value of q or p, and then
sends to Bob a narrow wave packet centered at that value.
2: Bob receives the state and decides at random to measure either q or p.
3: After Bob confirms receipt, Alice and Bob broadcast whether they sent
or measured in the q or p basis. If they used different bases, they discard
their results. If they used the same basis, they retain the result and
proceed to Step 4.
4: Alice broadcasts the value that she sent, modulo Vi (to m-bit accuracy).
Bob subtracts Alice's value from what he measured, and corrects to the
nearest integer mUltiple of Vi. He and Alice extract their shared bit
according to whether this integer is even or odd.

5. A SECURE PROTOCOL USING CONTINUOUS

VARIABLES
Now we are ready to combine the protocol of §3. with the protocol of §4.. The
result is a protocol based on concatenating the continuous variable code with
an [[n, k, dl] binary CSS code. The concatenated code embeds a k-dimensional
Hilbert space in the infinite-dimensional Hilbert space of n oscillators.
Again, we first imagine that Alice and Bob carry out an entanglement distil-
lation protocol. They start out sharing n pairs of oscillators, each in a (noisy)
EPR state. By measuring the stabilizer generators of the concatenated code,
340 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

they distill k encoded Bell pairs of much better fidelity, and then generate a key
by measuring the encoded Bell pairs.
By once again following the chain of reductions recounted in §3. and §4.,
we arrive at an equivalent protocol involving transmission of squeezed states.
The complete protocol, including verification, error correction, and privacy
amplification, becomes:

Continuous-variable QKD

1: Alice has (4 + 8)n oscillators. For each oscillator, Alice decides at

random to prepare either a state squeezed in q or a state squeezed in p.
The position of the squeezed state is determined by sampling (a discrete
approximation to) a probability distribution Ppos(q) or Pmom(P). Alice
then sends the oscillators to Bob.
2: Bob receives the (4 + 8)n oscillators, measuring each in the q or p basis
at random.
3: Bob confirms that the oscillators have been received, and then Alice
announces whether each oscillator was squeezed in q or in p.
4: Alice and Bob discard the results in the cases where Bob measured in a
different basis than Alice used in her preparation. With high probability,
there are at least 2n measured values left (if not, abort the protocol).
Alice decides randomly on a set of 2n values to use for the protocol, and
chooses at random n of these to be check values.
5: For a112n measured values, Alice announces the value of q or p modulo
v'1f (to m bits of accuracy).
6: Bob subtracts the corresponding number announced by Alice from each
of his measured values, and then corrects the result to the nearest integer
multiple of fi. Bob and Alice now extract bit values determined by
whether the multiple of fi is even or odd.
7: Alice and Bob announce the values of their check bits. If too few of the
check bits agree, they abort the protocol.
8: Alice announces u + v, where u is the string consisting of the remaining
non-check bits, and v is a random codeword in C 1 .
9: Bob subtracts u + v from his code qubits, u + e, and corrects the result,
v + e, to a codeword in C 1 . With high probability, Bob recovers v.
10: Alice and Bob use the C 2 coset v + C 2 as the key.

Here, to be specific, we have instructed Alice and Bob to sacrifice n check

bits for each n bits that are used for key distribution. They might instead use
fewer or more, depending on how stringent a bound on the eavesdropper's
mutual information they require.
The check bits provide Alice and Bob with estimates of the bit error rates p z
(respectively px) when states squeezed in q (respectively p) are transmitted.
 Secure quantum key distribution using squeezed states 341

Our analysis of the BB84 protocol indicates that the squeezed state protocol
is secure provided that p z and p x are both below 11 %, and assuming that
Alice and Bob scramble and unscramble the oscillators (by applying a random
permutation and its inverse).
However, as noted in §3.4, the proof and the security criterion Pz, Px < 11%
apply only if Alice's source can be simulated by measuring half of an entangled
state of two oscillators. In particular, we may imagine that Alice has many
pairs of oscillators identically prepared in the state PAB, and that she prepares
the state that she sends to Bob by measuring oscillator A. When she measures
in the q basis, she sends the state

(22.51)

with probability

(22.52)

and when she measures in the p basis, she sends the state

(22.53)

with probability

(22.54)

Thus, the states that Alice sends need not be perfect position or momentum
eigenstates for the proof of security to work, and Alice's source might even
have a bias so that the raw key bit carried by an oscillator is more likely to be
a 0 than a 1. Still, for a source of this type, if Alice and Bob verify that the
error rate for the raw key bits is below 11 % in both bases, then the protocol is
provably secure. We will discuss examples in §6. and §7.
Intuitively, the squeezed state protocol is secure because the eavesdropper
cannot monitor the value of q (or p) transmitted without introducing a detectable
disturbance in the complementary observable p (or q). As shown in Fig. 22.1,
the Wigner functions of the signal states squeezed in p and in q overlap, so that
the states cannot be reliably distinguished.

6. GAUSSIAN STATES
Perfectly squeezed states (position or momentum eigenstates) are unphysical
nonnormalizable states, so the protocol will actually be carried out with imper-
fectly squeezed states. Furthermore, engineering a source that produces highly
squeezed states would be quite technically demanding. How much squeezing
342 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

r------+q

Figure 22.1 One-sigma contours of the Wigner functions for typical squeezed states used in
the quantum key distribution protocol, with squeeze factor Li = e- r = 1/2. The signal states
squeezed in p and in q overlap with one another, preventing Eve from learning about one without
disturbing the other.

is really needed for the protocol to be secure? A related question is, how must
we choose the probability distributions Ppos (q) and Pmom (p) that govern the
center of the squeezed state?
We will analyze the most favorable case, in which the squeezed states are
Gaussian wave packets and the probability distributions are also Gaussian. We
will begin again with a description of how the code is used for entanglement
purification, but where Alice and Bob start with many copies of a Gaussian
entangled pair of oscillators that is an approximate eigenstate of qA - qB and
PA + PB · If we imagine that Alice measures half of each pair before she
sends the other half to Bob, then we obtain a protocol in which Alice sends
imperfectly squeezed states governed by a particular probability distribution.
The initial Gaussian entangled state of the two oscillators is

(22.55)
 Secure quantum key distribution using squeezed states 343

where ~2 is real and positive. Since I'¢'(~))AB is actually invariant under

(22.56)

we may assume without loss of generality (changing the sign of the position and
momentum of Bob's oscillator if necessary), that 0 < ~2 ~ 2. In the limiting
case ~2 = 2, I'¢'(~))AB becomes the product of two oscillator vacuum states.
For ~ 2 < 2, it is an entangled state. The amount of entanglement shared
between the oscillators, in "ebits," is defined as

(22.57)

(the Von Neumann entropy of Alice's density matrix PA = trBI'¢'(~)) (1/!(~)I),

and can be expressed as [20]

E(~) = (cosh2 r) 10g2 (cosh2 r)

(sinh2 r) log2(sinh 2 r) , (22.58)

where

(22.59)

In this entangled state, if Alice measures the position of her oscillator and
obtains the outcome qA, she prepares for Bob the Gaussian state

1,¢,(qA))B = (1fA~)1/4! dqB

x exp ( -~(qB - qBO)2 / A2) 19B) , (22.60)

where

(22.61)

and
- 2 ~2
~ = 1 + !~4 . (22.62)
4

The probability distribution for the outcome of Alice's measurement can be

expressed as

(22.63)
344 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

and we can easily see from eq. (22.54) that if Alice and Bob both measure q,
then the difference of their outcomes is governed by the probability distribution

Prob(qA - qB) = ~ exp[-(qA - qB)2 1Ll 2] . (22.64)

V7rLl2
Similar formulas apply if Alice and Bob measure p.
Suppose that Alice and Bob try to distill one good qubit from the imperfect
entangled state 1'l/J(Ll))AB. They both measure the stabilizer generators, that
is, the values of q and p modulo..Jff. Alice broadcasts her values, and Bob
adjusts his values so that they agree with Alice's; thereby they obtain a pair of
encoded qubits, which would have been in the state l¢j+) if the initial pair of
oscillators had been a perfect EPR pair (Ll 2 = 0). Then if Alice and Bob were
to proceed to perform a complete Bell measurement on their encoded qubit
pair, the probability pz that they would find Z ® Z = -1 is no worse than the
probability that, if qA and qB were measured, the results would differ by more
than Vir12, or

(22.65)

and similarly for px (the probability that X ® X = -1). For the values of Ll
that are typically of interest (e.g. Ll < 1), the error probability is dominated by
values of qA - qB (or PA + PB) lying in the range [..Jff/2, 3Vir/2], so that the
estimate of the error probability can be sharpened to

2 J3..;7i /2 2 b. 2
pz,px rv - - dq e- q / • (22.66)
V7rLl2 ..;7i/2
After error correction and measurement in the encoded Bell basis, the initial
bipartite pure state of two oscillators, with entanglement E given by eq. (22.57)
and (22.59), is reduced to a bipartite mixed state, diagonal in the encoded Bell
basis, with fidelity F = (1 - Pz ) (1 - Px); this encoded state has entanglement
of formation [2]

(22.67)

(where H2 is the binary entropy function).

If Alice and Bob have a large number n of oscillators in the state 1'l/J(Ll)) AB,
they can carry out an entanglement distillation protocol based on the concatena-
tion of the single-oscillator code with a binary CSS code, and they will be able to
 Secure quantum key distribution using squeezed states 345

distill qubits of arbitrarily good fidelity at a finite asymptotic rate provided that
pz and Px are both below 11 %; from eq. (22.66) we find that this condition is
satisfied for ~ < .784 (which should be compared with the value ~ = v'2 cor-
responding to a product of two oscillators each in its vacuum state). Thus secure
EPR key distribution is possible in principle with two-mode squeezed states
provided that the squeeze parameter r satisfies r > -loge(.784/v'2) = .590;
from eq. (22.57) and (22.67), ~ = .784 corresponds to E = 1.19 ebits carried
by each oscillator pair, which is reduced by error correction and encoded Bell
measurement to E = .450 ebits carried by each of the encoded Bell pairs.
Now consider the reduction of this entanglement distillation protocol to a
protocol in which Alice prepares a squeezed state and sends it to Bob. In the
squeezed-state scheme, Alice sends the state 11/J(qA)} with probability P(qA).
The width A of the state that Alice sends is related to the parameter ~ appearing
in the estimated error probability according to

(22.68)

The state Alice sends is centered not at qA but at qBO = qA . (1 - A4)1/2.

Nevertheless, in the squeezed state protocol that we obtain as a reduction of
the entanglement distillation protocol, it is qA rather than qBO that Alice uses
to extract a key bit, and whose value modulo .Ji she reports to Bob. The
error probability that is required to be below 11 % to ensure security is the
probability that error correction adjusts Bob's measurement outcome to a value
that differs from qA (not qBO) by an odd multiple of.Ji. As we have noted,
this error probability is below 11 % for ~ < .784, which (from eq. (22.62»
corresponds to A < .749; this value should be compared to the value A = 1 for
an oscillator in its vacuum state. Thus, secure squeezed-state key distribution
is possible in principle using single-mode squeezed states, provided that the
squeeze parameter r defined by Li = e- r satisfies r > -10ge(.749) = .289.
When interpreted as suppression, relative to vacuum noise, of the quantum noise
afflicting the squeezed observable, this amount of squeezing can be expressed
as 10 . 10glO ( A-2) = 2.51 dB.
The error rate is below 1% for A < .483 (~ < .486), and drops precipitously
for more highly squeezed states, e.g., to below 10- 6 for A '" ~ < .256. For
example, if the noise in the channel is weak, Alice and Bob can use the Gaussian
squeezed state protocol with Li '" 1/2 (see Fig. 22.2) to generate a shared bit
via the q or p channel with an error rate ('" 1.2%) comfortably below 11%;
thus the protocol is secure if augmented with classical binary error correction
and privacy amplification.
Of course, if the channel noise is significant, there will be a more stringent
limit on the required squeezing. Many kinds of noise (for instance, absorption
of photons in an optical fiber) will cause a degradation of the squeezing factor.
346 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

, ,
, ,
,,
,,
,,
.,
,,

q or p
-4 4

Figure 22.2 Probability distributions for the squeezed quantum key distribution protocol, with
squeeze factor Ii = 1/2. The dotted line is the probability distribution P (a Gaussian with
variance (1/21i 2) . (1 -Ii 4) that Alice samples to determine the center of the squeezed signal
that she sends. The solid lines are the probability distributions in position or momentum of the
squeezed states (Gaussians with variance Ii 2 /2, shown with a different vertical scale than P)
centered at -,ji, 0, and ,ji. The intrinsic error probability due to imperfect squeezing (prior
to binary error correction and privacy amplification) is 1.2%.

If this is the only consequence of the noise, the squeezing exiting the channel
should still satisfy !:::. < .784 for the protocol to be secure, as we discuss in
more detail in §7. Otherwise, the errors due to imperfect squeezing must be
added to errors from other causes to determine the overall error rate.
So far we have described the case where the p states and the q states are
squeezed by equal amounts. The protocol works just as well in the case of
unequal squeezing, if we adjust the error correction procedure accordingly.
Consider carrying out the entanglement distillation using the code with general
parameter Q rather than Q = 1. The error rates are unaffected if the squeezing
in q and p is suitably rescaled, so that the width of the q and p states becomes

!:::.q = !:::. . Q , !:::.p = !:::./ Q • (22.69)

In this modified protocol, Alice broadcasts the value of q modulo ..,fir. Q or the
value of p modulo ..,fir/ Q. Bob subtracts the value broadcast by Alice from his
own measurement outcome, and then adjusts the difference he obtains to the
nearest multiple of ..,fir . Q or ..,fir/ Q. The key bit is determined by whether the
multiple of..,fir . Q, or ..,fir/ Q, is even or odd.
Thus, for example, the error rate sustained due to imperfect squeezing will
have the same (acceptably small) value irrespective of whether Alice sends
states with !:::.q = !:::.p = 1/2, or !:::.q = 1 and !:::.p = 1/4; Alice can afford to
send coherent states about half the time if she increases the squeezing of her
other transmissions by a compensating amount.
 Secure quantum key distribution using squeezed states 347

Can we devise a secure quantum key distribution scheme in which Alice

always sends coherent states? To obtain, as a reduction of an entanglement
distillation protocol, a protocol in which coherent states (Li = 1) are always
transmitted, we must consider the case b. 2 = 2. But in that case, the initial
state of Alice's and Bob's oscillators is a product state. Bob's value of q or pis
completely uncorrelated with Alice's, and the protocol obviously won't work.
This observation does not exclude secure quantum key distribution schemes
using coherent states, but if they exist another method would be needed to
prove the security of such schemes.
In general, the source that we obtain by measuring half of the entangled
pair is biased. If b. is not small compared to .Jif, then Alice is significantly
more likely to generate a 0 than a 1 as her raw key bit. But as we have already
discussed in §3.4, after error correction and privacy amplification, the protocol
is secure if p x and p z are both less than 11 %. This result follows because the
squeezed state protocol is obtained as a reduction of an entanglement distillation
protocol.

7. LOSSES AND OTHER IMPERFECTIONS

The ideal BB84 quantum key distribution protocol is provably secure. But
in practical settings, the protocol cannot be implemented perfectly, and the
imperfections can compromise its security. (See [4] for a recent discussion.)
For example, if the transmitted qubit is a photon polarization state carried by an
optical fiber, losses in the fiber, detector inefficiencies, and dark counts in the
detector all can impose serious limitations. In particular, if the photons travel
a distance large compared to the attenuation length of the fiber, then detection
events will be dominated by dark counts, leading to an unacceptably large error
rate.
Furthermore, most present-day implementations of quantum cryptography
use, not single photon pulses, but weak coherent pulses; usually the source
"emits" the vacuum state, occasionally it emits a single photon, and with non-
negligible probability it emits two or more photons. Quantum key distribution
with weak coherent pulses is vulnerable to a "photon number splitting" attack,
in which the eavesdropper diverts extra photons, and acquires complete infor-
mation about their polarization without producing any detectable disturbance.
A weaker pulse is less susceptible to photon number splitting, but increases the
risk that the detector will be swamped by dark counts.
From a practical standpoint, quantum key distribution with squeezed states
may not necessarily be better than BB84, but it is certainly different. Alice re-
quires a source that produces a specified squeezed state on demand; fortunately,
the amount of squeezing needed to ensure the security of the protocol is rela-
tively modest. Bob uses homodyne detection to measure a specified quadrature
348 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

amplitude; this measurement may be less sensitive to detector defects than the
single-photon measurement required in BB84.
But, as in the BB84 protocol, losses due to the absorption of photons in the
channel will enhance the error rate in squeezed-state quantum key distribution,
and so will limit the distance over which secure key exchange is possible. We
study this effect by modeling the loss as a damping channel described by the
master equation

p=r(apat-~atap-~pata) j (22.70)

here p is the density operator of the oscillator, a is the annihilation operator,

and r is the decay rate. Eq. (22.70) implies that

(22.71)

where

(O)t = tr (Op(t)) (22.72)

denotes the expectation value of the operator 0 at time t. Integrating, we find

(22.73)

and so, by expanding in power series,

(22.74)

where j is an analytic function, and: j : denotes normal ordering (that is, in

: j(a t , a) :, all at's are placed to the left of all a's).
In particular, by normal ordering and applying eq. (22.74), we find

(22.75)

where q = (a + at) /.../2 is the position operator. A similar formula applies

to the momentum operator or any other quadrature amplitude. Eq. (22.75)
shows that if the initial state at t = 0 is Gaussian (q is governed by a Gaussian
probability distribution), then so is the final state at t = T [10]. The mean (q)
and variance b..q2 of the initial and final distributions are related by

(22.76)

Now let's revisit the analysis of §6., taking into account the effects of losses.
We imagine that Alice prepares entangled pairs of oscillators in the state (22.54),
 Secure quantum key distribution using squeezed states 349

and sends one oscillator to Bob through the lossy channel; then they perform
entanglement purification. This protocol reduces to one in which Alice prepares
a squeezed state that is transmitted to Bob. In the squeezed-state protocol, Alice
decides what squeezed state to send by sampling the probability distribution
P(qA) given in eq. (22.63); if she chooses the value qA, then she prepares and
sends the state 1'I/J(qA)} in eq. (22.59). When it enters the channel, this state is
governed by the probability distribution

(22.77)

and when Bob receives the state this distribution has, according to eq. (22.76),
evolved to

(22.78)

where

q~o = ~qBO == ~(1- Li4)1/2qA ,

b.,2 = eLi2 + (1 - e) . (22.79)

By integrating over qA in P'(qA, qB) = P'(qBlqA) . P(qA), we can obtain the

final marginal distribution for the difference qA - qB:

which generalizes eq. (22.68). We can express the damping factor ~ as

~ = e-K.d/2 , (22.81)

where d is the length of the channel and 1\:-1 is its attenuation length (typically
of the order of 10 km in an optical fiber).
The protocol is secure if the error rate in both bases is below 11%; as in
§6., this condition is satisfied for b.~ < .784. Thus we can calculate, as a
function of the initial squeezing parameter Li, the maximum distance d max that
the signal states can be transmitted without compromising the security of the
protocol.
For Li « 1, we find

d max = (1.57) . b. + O(b. ) .

- -2
I\: (22.82)
350 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Figure 22.3 The effect of channel losses on the security of quantum key distribution using
squeezed states. The maximum length r;,d max of the channel (in units of the attenuation length)
is plotted as a function of the width t. of the squeezed state that enters the channel. For a longer
channel, the error rate due to losses is too large and the proof of security breaks down. The
curve labeled "with amplification" applies to the protocol in which the signal is amplified prior
to detection in order to compensate for the losses; the curve labeled "without amplification"
applies to the protocol in which the signal is not amplified.

Thus, the more highly squeezed the input signal, the less we can tolerate the
losses in the channel. This feature, which sounds surprising on first hearing,
arises because the amount of squeezing is linked with the size of the range
in qA that Alice samples. Errors are not unlikely if losses cause the value
of qB to decay by an amount comparable to yI1r /2. In our protocol, if the
squeezed states have a small width Li, then the typical states prepared by Alice
are centered at a large value qA "'-' Li -1; therefore, a small fractional decay can
cause an error.
On the other hand, even without losses, Alice needs to send states with
.6. < .749 to attain a low enough error rate, and as Li approaches .749 from
below, again only a small loss is required to push the error probability over
11 %. Thus there is an intermediate value of .6. that optimizes the value of d max ,
as shown in Fig. 22.3. This optimal distance,

K, dmax,opt ~ .367 , (22.S3)

is attained for Li "'-' .426.

Our analysis so far applies if Alice and Bob have no prior knowledge about
the properties of the channel. But if the loss e
= e- Kd is known accurately,
they might achieve a lower error rate if Bob compensates for the loss by
multiplying his measurement outcome by e-
1 before proceeding with error
 Secure quantum key distribution using squeezed states 351

correction and privacy amplification. This amplification of the signal by Bob

is entirely classical, but to analyze the security in this case, we may consider an
entanglement purification scenario in which Bob applies a quantum amplifier
to the signal before measuring. Since the quantum amplifier (which amplifies
all quadrature amplitudes, not just the one that Bob measures) is noisier, the
protocol will be no less secure if Bob uses a classical amplifier rather than a
quantum one.
So now we consider whether entanglement purification will succeed, where
the channel acting on Bob's oscillator in each EPR pair consists of transmission
through the lossy fiber followed by processing in Bob's amplifier. If the error
rate is low enough, the key will be secure even if the amplifier, as well as the
optical fiber, are under Eve's control.
Bob's linear amplifier can be modeled by a master equation like eq. (22.70),
but with a and at interchanged, and where r is now interpreted as a rate of gain.
The solution is similar to eq. (22.74), except the normal ordering is replaced
by anti-normal ordering (all a's are placed to the left of all at's), and with e
e-
replaced by the gain 2 = erT ~ 1. We conclude that the amplifier transforms
a Gaussian input state to a Gaussian output state, and that the mean (q) and
variance b..q2 of the Gaussian position distribution are modified according to

(q) -t e- (q) ,
1

b..q2 -t e- 2b..q2 +! (e- 2 -1) (22.84)

2
Other quadrature amplitudes are transformed similarly.
e
Now suppose that a damping channel with loss is followed by an amplifier
with gain e- 2 . Then the mean of the position distribution is left unchanged,
but the variance evolves as

b..q2 -t e- 2 (eb.. q2 + ~ (1- e)) + ~ (e- 2 -1)

= b..q2 + (e- 2 - 1) . (22.85)

For this channel, the probability distribution governing qA - qB is again a

Gaussian as in eq. (22.79), but now its width is determined by

(22.86)

Error rates in the q and p bases are below 11 %, and the protocol is provably
secure, for (b..~) amp < .784.
By solving (b..~)amp = .784. we can find the maximum distance d (where
e- 2 = eK.d) for which our proof of security holds; the result is plotted in
352 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Fig. 22.3. When the squeezed input is narrow, is. < < 1, the solution becomes
e = exp (11; dmax ) = 1.307 + O(~- 2 ) ,
-2 _
(22.87)

or

11; dmax ;:::;:j .268 . (22.88)

Comparing the two curves in Fig. 22.3, we see that the protocol with am-
plification remains secure out to longer distances than the protocol without
amplification, if the input is highly squeezed. In that case, the error rate in the
protocol without amplification is dominated by the decay of the signal, which
can be corrected by the amplifier. But if the input is less highly squeezed, then
the protocol without amplification remains secure to longer distances. In that
case, the nonzero width of the signal state contributes significantly to the error
rate; the amplifier noise broadens the state further.
With more sophisticated protocols that incorporate some form of quantum
error correction, continuous-variable quantum key distribution can be extended
to longer distances. For example, if Alice and Bob share some noisy pairs
of oscillators, they can purify the entanglement using protocols that require
two-way classical communication [15, 7]. After pairs with improved fidelity
are distilled, Alice, by measuring a quadrature amplitude in her laboratory,
prepares a squeezed state in Bob's; the key bits can be extracted using the
same error correction and privacy amplification schemes that we have already
described.
Our proof of security applies to the case where squeezed states are carried
by a lossy channel (assuming a low enough error rate), because this scenario
can be obtained as a reduction of a protocol in which Alice and Bob apply
entanglement distillation to noisy entangled pairs of oscillators that they share.
More generally, the proof applies to any imperfections that can be accurately
modeled as a quantum operation that acts on the shared pairs before Alice and
Bob measure them. As one example, suppose that when Alice prepares the
squeezed state, it is not really the q or p squeezed state that the protocol calls
for, but is instead slightly rotated in the quadrature plane. And suppose that
when Bob performs his homodyne measurement, he does not really measure
q or p, but actually measures a slightly rotated quadrature amplitude. In the
entanglement-distillation scenario, the imperfection of Alice's preparation can
be modeled as a superoperator that acts on her oscillator before she makes a
perfect quadrature measurement, and the misalignment of Bob's measurement
can likewise be modeled by a superoperator acting on his oscillator before
he makes a perfect quadrature measurement. Therefore, the squeezed state
protocol with this type of imperfect preparation and measurement is secure, as
long as the error rate is below 11 % in both bases. Of course, this error rate
 Secure quantum key distribution using squeezed states 353

includes both errors caused by the channel and errors due to the imperfection
of the preparation and measurement.
We also recall that in the protocols of §5., Alice's preparation and Bob's
measurement were performed to m bits of accuracy. In the entanglement
distillation scenario, this finite resolution can likewise be well modeled by a
quantum operation that shifts the oscillators by an amount of order 2- m before
Alice and Bob perform their measurements. Thus the proof applies, with the
finite resolution included among the effects contributing to the permissible
11 % error rate. The finite accuracy causes trouble only when Alice's and Bob's
results lie a distance apart that is within about 2- m of ..j7r/2; thus, just a few
bits of accuracy should be enough to make this additional source of error quite
small.

8. CONCLUSIONS
We have described a secure protocol for quantum key distribution based on
the transmission of squeezed states of a harmonic oscillator. Conceptually,
our protocol resembles the BB84 protocol, in which single qubit states are
transmitted. The BB84 protocol is secure because monitoring the observable
Z causes a detectable disturbance in the observable X, and vice versa. The
squeezed state protocol is secure because monitoring the observable q causes a
detectable disturbance in the observable p, and vice versa. Security is ensured
even if the adversary uses the most general eavesdropping strategies allowed
by the principles of quantum mechanics.
In secure versions of the BB84 scheme, Alice's source should emit single-
photons that Bob detects. Since the preparation of single-photon states is
difficult, and photon detectors are inefficient, at least in some settings the
squeezed-state protocol may have practical advantages, perhaps including a
higher rate of key production. Squeezing is also technically challenging, but
the amount of squeezing required to ensure security is relatively modest.
The protocol we have described in detail uses each transmitted oscillator to
carry one raw key bit. An obvious generalization is a protocol based on the code
with stabilizer generators given in eq. (22.8), which encodes ad-dimensional
protected Hilbert space in each oscillator. Then a secure key can be generated
more efficiently, but more squeezing is required to achieve an acceptable error
rate.
Our protocols, including their classical error correction and privacy amplifi-
cation, are based on CSS codes: each of the stabilizer generators is either of the
"q"-type (the exponential of a linear combination of n q's) or of the "p-type"
(the exponential of a linear combination of n p's). The particular CSS codes
that we have described in detail belong to a restricted class: they are concate-
nated codes such that each oscillator encodes a single qubit, and then a block of
354 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

those single-oscillator qubits are assembled to encode k better protected qubits

using a binary [[n, k, d]] stabilizer code. There are more general CSS codes
that embed k protected qubits in the Hilbert space of n oscillators but do not
have this concatenated structure [8]; secure key distribution protocols can be
based on these too. The quantum part of the protocol is still the same, but
the error correction and privacy amplification make use of more sophisticated
close packings of spheres in n dimensions.
We analyzed a version of the protocol in which Alice prepares Gaussian
squeezed states governed by a Gaussian probability distribution. The states,
and the probability distribution that Alice samples, need not be Gaussian for
the protocol to be secure. However, for other types of states and probability
distributions, the error rates might have to be smaller to ensure the security of
the protocol.
Our proof of security applies to a protocol in which the squeezed states prop-
agate through a lossy channel, over a distance comparable to the attentuation
length of the channel. To extend continuous-variable quantum key distribution
to much larger distances, quantum error correction or entanglement distillation
should be invoked.
Strictly speaking, the security proof we have presented applies if Alice's
state preparation (including the probability distribution that she samples) can
be exactly realized by measuring half of an imperfectly entangled state of
two oscillators. The protocol remains secure if Alice's source can be well
approximated in this way. Our proof does not work if Alice occasionally
sends two identically prepared oscillators when she means to send just one; the
eavesdropper can steal the extra copy, and then the privacy amplification is not
guaranteed to reduce the eavesdropper's information to an exponentially small
amount.

Acknowledgments
We thank Andrew Doherty, Steven van Enk, Jim Harrington, Jeff Kimble,
and especially Hoi-Kwong Lo for useful discussions and comments. This
work has been supported in part by the Department of Energy under Grant No.
DE-FG03-92-ER40701, and by DARPA through the Quantum Information and
Computation (QUIC) project administered by the Army Research Office under
Grant No. DAAH04-96-1-0386. Some of this work was done at the Aspen
Center for Physics. This paper first appeared in Physical Review A
 Secure quantum key distribution using squeezed states 355

Notes
I. We implicitly assume that Eve uses a strategy that passes the verification test with nonnegligible
probability, so that the rate of key generation is not exponentially small. If, for example, Eve were to
intercept all qubits sent by Alice and resend them to Bob, then she would almost certainly be detected, and
key bits would not be likely to be generated. But in the rare event that she is not detected and some key bits
are generated, Eve would know a lot about them.
2. This bound is not tight. It applies if the sample of M pairs is chosen from the population of N with
replacement. In fact the sample is chosen without replacement, which suppresses the fluctuations. A better
bound was quoted by Shor and Preskill [18].

References
[1] Bennett, C. H., and G. Brassard (1984), "Quantum cryptography: Public-
key distribution and coin tossing," in Proceedings of IEEE International
Conference on Computers, Systems and Signal Processing (Bangalore,
India), pp. 175-179; C. H. Bennett and G. Brassard (1985), "Quantum
public key distribution," IBM Technical Disclosure Bulletin 28, 3153-
3163.
[2] Bennett, C. H., D. P. DiVincenzo, J. A. Smolin and W. K. Wootters (1996),
"Mixed state entanglement and quantum error correction," Phys. Rev. A
54,3824-3851, quant-phl9604024.
[3] Biham, E., M. Boyer, P. O. Boykin, T. Mor and V. Roychowdhury (2000),
"A proof of the security of quantum key distribution," in Proceedings of
the Thirty-Second Annual ACM Symposium on Theory of Computing, pp
715-724. New York: ACM Press, quant-phl9912053.
[4] Brassard, G., N. Liitkenhaus, T. Mor, and B. C. Sanders (2000), "Lim-
itations on practical quantum cryptography," Phys. Rev. Lett. 85, 1330,
quant-phl9911054.
[5] Braunstein, S. (1998), "Error correction for continuous quantum vari-
ables," Phys. Rev. Lett. 80, 4084, quant-phl9711049.
[6] Calderbank, A. R., and P. W. Shor (1996), "Good quantum error correcting
codes exist," Phys. Rev. A 54, 1098-1105, quant-phl9512032.
[7] Duan, L. M., G. Giedke, J. I. Cirac, and P. Zoller (1999), "Entangle-
ment purification of Gaussian continuous variable quantum states," quant-
ph/9912017; L. M. Duan, G. Giedke, J. I. Cirac, and P. Zoller (2000),
"Physical implementation for entanglement purification of Gaussian con-
tinuous variable quantum systems," quant-phl0003116.
[8] Gottesman, D., A. Kitaev, and J. Preskill (2000), "Encoding a qudit in an
oscillator," quant-phl0008040.
[9] Hillery, M. (1999), "Quantum cryptography with squeezed states,"
quant-phl9909006.
356 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[10] Holevo, A. S. (1998), "Sending quantum information with Gaussian

states," quant-phl9809022.
[11] Lloyd, S., and J. E. Slotine (1998), ''Analog quantum error correction,"
Phys. Rev. Lett. 80, 4088, quant-phl971l021.
[12] Lo, H.-K., and H. F. Chau (1999), "Unconditional security of quantum
key distribution over arbitrarily long distances," Science 283, 2050-2056,
quant-phl9803006.
[13] Mayers, D. (1996), "Quantum key distribution and string oblivious trans-
fer in noisy channels," Advances in Cryptology-Proceedings of Crypto
'96, pp. 343-357. New York: Springer-Verlag
[14] Mayers, D. (1998), "Unconditional security in quantum cryptography," J.
Assoc. Comput. Mach. (to be published), quant-phl9802025.
[15] Parker, S., S. Bose, and M. B. Plenio (2000), "Entanglement quantification
and purification in continuous variable systems," Phys. Rev. A 61, 32305,
quant-phl9906098.
[16] Ralph, T. c. (1999), "Continuous variable quantum cryptography,"
quant-phl9907073; "Security of continuous variable quantum cryptog-
raphy," quant-phl0007024.
[17] Reid, M. D. (1999), "Quantum cryptography using continuous variable
Einstein-Podolsky-Rosen correlations and quadrature phase amplitude
measurements," quant-phl9909030.
[18] Shor, P. W., and J. Preskill (2000), "Simple proof of security of the BB84
quantum key distribution protocol," Phys. Rev. Lett. 85, 441-444, quant-
phl0003004.
[19] Steane, A. M. (1996), "Multiple particle interference and error correction,"
Proc. Roy. Soc. Lond. A 452, 2551-2577, quant-phl9601029.
[20] van Enk, S. J. (1999), ''A discrete formulation of teleportation of contin-
uous variables," Phys. Rev. A 60,5095, quant-phl9905081.
Chapter 23

EXPERIMENTAL DEMONSTRATION OF
QUANTUM DENSE CODING AND
QUANTUM CRYPTOGRAPHY WITH
CONTINUOUS VARIABLES

Kunchi Peng, Qing Pan, Jing Zhang and Changde Xie

The State Key LAboratory of Quantum Optics and Quantum Optics Devices
Institute of Opto-Electronics
Shanxi University
Taiyuan 030006
P.R. China

Abstract In this paper we will present the experimental demonstrations of quantum dense
coding and quantum cryptography using continuous electromagnetic field with
Einstein-Podolsky-Rosen(EPR) correlations. The bright EPR optical beams
with the quantum correlations between the amplitude and phase quadratures
are produced from a nondegenerate optical parametric amplifier. The direct
detection technology of the Bell-state is ultilized in the measurements of the
quantum correlations and the signals modulated on the quadratures instead of
usual homodyne detection. Usability of experimentally accessible squeezed-state
entanglements, high efficiencies of bit transmission and information detection,
relatively straightforward systems and operating procedures, and security directly
provided by quantum correlations make the presented schemes valuable to be
applied to the developing quantum information science.

1. INTRODUCTION
Recently, more and more investigation interests in quantum information
science have focus on exploiting the quantum system which possesses contin-
uous spectra [1-19]. The successful experiments on the quantum teleportation
[20] and quantum dense coding [21] using EPR correlations of continuous
electromagnetic fields provided possible technologies for quantum information
processing based on continuous variables. High bit transmission rates and

357
S.L Braunstein andA.K. Pati (eds.), Quantum Information with Continuous Variables, 357-378.
© 2003 Kluwer Academic Publishers.
358 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

high detection efficiency are the key advantages of continuous variable systems
better than discrete systems.
In this paper we will present the experimental schemes and results on con-
tinuous variable quantum dense coding and quantum cryptography. The ex-
perimental technologies for the generation of EPR correlated beams and the
detection of the Bell-state will be described firstly.

2. GENERATION OF BRIGHT EPR BEAMS

It has been theoretically and experimentally demonstrated that EPR beams
with correlated amplitude quadratures and anticorrelated phase quadratures or
correlated phase quadratures and anticorrelated amplitude quadratures may be
produced by a continuous nondegenerate optical parametric amplifier(NOPA)
operating at amplification or deamplification [21-26]. For the former, the
variances of the difference of amplitude quadratures (6(Xl - X2)2) and the
sum of phase quadratures (O(YI + Y2)2) are both smaller than the shot noise
limit ( SNL ) defined by the vacuum fluctuation. For the later, there are inverse
correlations between the quadratures, that is

and (23.1)

r-------l Frequency lock system ~---1

Figure 23.1 Experimental setup for the generation of EPR beams: YAP-Nd:YAI0 3 , F.R-
Faraday Rotator, EOM-Electric Optical modulator, PBS-polarizing beam splitter.

The schematic of experimental setup for the generation of EPR beams is

shown in Fig. 23.1. The entangled EPR beam is generated from a NOPA
consisting of an a-cut type II KTP (KTiOP04, Potassium Titanyl Phosphate)
 Experimental demonstration of dense coding and quantum cryptography 359

crystal (10mm long), the front face of which is coated to be used as the input
coupler( the transmission>95% at 540nm wavelength and 0.5% at lOS0nm)
and the other face is coated with the dual-band antireflection at both 540nm
and lOS0nm, as well as a concave mirror of 50mm-curvature radius, which is
used as the output coupler of EPR beam at lOSOnm (the transmission of 5%
at lOS0 and high reflectivity at 540nm). The output coupler is mounted on a
piezoelectric transducer to lock actively the cavity length on resonance with the
injected seed wave at lOS0nm using the FM sideband technique. By fine tuning
the crystal temperature the birefringence between signal and idler waves in KTP
is compensated and the simultaneous resonance in the cavity is reached. The
process of adjusting temperature to meet double resonance can be monitored
with an oscilloscope during scanning the length of cavity. Once the double
resonance is completed the NOPA is locked on the frequency of the injected seed
wave [25]. The measured finesse, the free spectral range, and the line-width of
the parametric oscillator are 110, 2.SG, and 26 MHz, respectively. The pump
source of NOPA is a home-made all-solid-state intracavity frequency-doubled
and frequency-stabilized CW ring Nd:YAP (Nd:YAI03, Yttrium-Aluminum-
Perovskite) laser [27]. The output second-harmonic wave at 540nm and the
leaking fundamental wave at 1080nm from the laser serve as the pump light and
the seed wave respectively. The laser-diode pumped all-solid-state laser and the
semi-monolithic F-P configuration of the parametric cavity ensure the stability
of system, so the frequency and phase of light waves can be well-locked during
the experiments.
The NOPA is pumped by the harmonic wave at 540nm, that is controlled
just below the oscillation threshold of the NOPA, and the polarization of that
is along the b axis of the KTP crystal. Due to the large transmission (>95%)
of input coupler at 540nm, the pump field only passes the cavity twice without
resonating. After the seed beam at 1080nm polarized at 45° relative to the b
axis of the KTP crystal is injected into the cavity, it is decomposed to signal
and idler seed waves with identical intensity and the orthogonal polarization
along the b and c axes, respectively, which correspond to the vertical and
horizontal polarization. The temperature of KTP crystal placed in a special
designed oven is actively controlled around the temperature for achieving type
II noncritical phase matching (63°C) with a broad full width of about 30°C
[25]. An electronics feedback circuit is employed to stabilize actively the
temperature of crystal to a few mK.
Locking the relative phase between the pump laser and the injected seed wave
of NOPA to 2mr or (2n+ 1)-11", where n is integer, to enforce the NOPA operating
at amplification or deamplification, the entangled EPR beam with the quantum
correlation of amplitude(or phase) quadratures and the quantum anticorrelation
of phase(or amplitude) quadratures was generated [21,25]. The two halves of
the EPR beams are just the signal and idler modes of the subharmonic wave field
360 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

produced from type II parametric down conversion, thus they have orthogonal
polarizations originally. The output signal and idler beams are separated by
a polarizing-bearn-splitter (PBS2) to be a pair of EPR beams, (X1,Y1) and
(X2, Y 2), which possess the feature of nonlocal quantum entanglement of the
amplitude and phase quadratures.

3. DIRECT MEASUREMENT OF BELL-STATES

The homodyne detection is extensively applied in phase-sensitive measure-
ments of electromagnetic field quadratures and has been successfully used in
the quantum teleportation experiment of continuous veriables [20]. However,
for performing the homodyne detection the local oscillator(LO) is needed and
the well spatial and temporal mode matching between the LO and the detected
beam is required, which make some troubles to experimental implementation.
For passing by the troubles we proposed a direct detection scheme of Bell-
state [28] and applied it to accomplish the quantum dense coding [21] and the
quantum cryptography [29] with continuous variables.

RF
Splitter

DJ
Ya-Yb
d
RF
Splitter
b D2

a rc/2 phase shifter

Figure 23.2 Direct measurement of the Bell state. BS: 50% beam splitter. DI and D2:
detectors.

Fig. 23.2 is the diagram of the direct detection system. Two bright coherent
beams with identical intensities are expressed by the annihilation operators a
and b. A phase shift of 7r/2 is imposed on beam a, and then the beams are
mixed on a 50% beam splitter. The resulting output beams c and d are given by:

d= V;(a-ib). (23.2)
 Experimental demonstration of dense coding and quantum cryptography 361

We define upper-case operators in the rotating frame about the center frequency
wo,

O(t) = o(t)eiwot , (23.3)

with 0 = [a, b, c, d] and 0 = [A, B, C,D]. By the Fourier transformation we
have

0(0) = ~/ dtO(t)e- int . (23.4)

The fields are now described as functions of the modulation frequency O. Thus
we can consider any field as a carrier 0(0) oscillating at frequency Wo with an
average value equal to the steady state field, surrounded by "noise side-bands"
0(0) oscillating at frequency Wo ± 0 with zero average values. The amplitude
and phase quadrature can be written as

Xo(O) = 0(0) + 0+(-0); Yo(O) = ~[O(O) -

~
0+( -0)], (23.5)

The two bright output beams can be directly detected by Dl and D2. The
discussed photocurrents are normalized with the average value of the field.
The normalized output photocurrents spectra are given by

1
= "2 (Xa(O) + Ya(O) - Yb(O) + Xb(O)) (23.6)
1
"2(Xa(O) - Ya(O) + Yb(O) + Xb(O))
The Eq. (23.6) shows that the photocurrent of the each arm of the 50% beam-
splitter consists of two parts, part one (term 1 and term 4) is self terms of
two input fields Xa(O) and Xb(O) at the beamsplitter, the part 2 comprises
the interference terms (2 and 3) including phase quadratures Ya(O) and Yb(O)
deriving from the 7f /2 phase shift. Comparing ic and ict
in Eq. (23.6), it is
obvious that the self terms of two arms are correlated (or in phase), and the
interference terms are anticorrelated (or out of phase). Each of photocurrents
is divided into two parts through the RF power splitter. The sum and difference
of the divided photocurrents are

1 ~ ~ 1
= v'2(i c (O) + id(O)) = v'2(Xa(O) + Xb(O)) (23.7)

~(ic(O) - id(O)) = ~(Ya(O) - Yb(O))

362 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

We can see that the sum i+ of photocurrents of two arms c, d only leaves
the self tenns which include the amplitude quadrature of the signal field a
and b; the difference photocurrent i_ leaves the interference tenns, which
gives the infonnation of their phase quadratures. Thus a combined Bell-
state measurement of beams and is achieved with this simple self-homodyne
detector. In our dense coding and cryptography experiments, the detected
a
fields and b are the signal and idler modes, (X 1,Y1) and (X2,Y2), of the
output field from the NOPA operating at deamplification, so there are quantum
anticorrelation and correlation between their amplitude and phase quadratures
(see the inequalities (23.1) ).
The correlations measured by self-homodyne detector between the quad-
rature-phase amplitudes of the two halves of EPR beam are show in Fig. 23.3.
Both variances of (8(X1 + X2)2) (Fig. 23.3(a» and (8(Y1 - Y2)2) (Fig.23.3
(b» measured directly are -4dB below that of the SNL (considering the electron-
ics noise that is 8dB below the SNL, the actual fluctuation should be -5.4dB be-
low that of the SNL). The product of the correspondent conditional variances of
the EPR beam is (8(X1 + X2)2) (8(Y1 - y 2)2)=0.63. The bright EPR beam
of -70j.tmW(the correspondent photon-number flow is about 3.8 x 1014 8- 1 )
was obtained at the following operation parameters ofNOPA: the pump power
150mW just below the power of the oscillation threshold of 175mW and the
polarization of that was along the b axis of the KTP crystal. The power of the
injected seed wave was 10mW before entering the input coupler of the cavity
and that was polarized at 45° relative to the b axis.

4. QUANTUM DENSE CODING

The experimental diagram of the quantum dense coding is shown in Fig. 23.4.
The two halves ( 1 and 2 ) of the EPR entangled beam were distributed to the
sender Alice and the receiver Bob, respectively. At Alice the classical amplitude
and phase signals were modulated on the one half of EPR beam (beam 1),
which led to a displacement signal sent via the quantum channel. Since each
half of EPR beam had huge noise individually, for perfect EPR entanglement
(8(Xl(2))2) -+ 00 and (8(Yl(2))2) -+ 00, the signal to noise ratios in the beam
1 for both amplitude and phase signals tended to zero, so no one other than Bob
can gain any signal infonnation from the modulated beam under ideal condition.
At Bob, the signals were decoded with the aid of the other half of the EPR
beam (beam 2), which was combined with the modulated first half ofEPR beam
(beam 1) on a 50% beam-splitter, in our experiment which consisted of two
polarized-beam splitters and a half-wave-plate [30], and, before combination,
a 1f/2 phase shift was imposed between them according to the requirement of
direct measurement of Bell-State [28]. The bright outcomes from two ports
of the beam splitter were directly detected by a pair of photodiodes Dl and D2
 Experimental demonstration of dense coding and quantum cryptography 363

-90

-92

t~
-94 <O(X t + X2)2>

j "l'f~N\-V\J"\Ji'..v\ ",vJ\\fl;V'.;'!,;VV'~V/,1t~\:".\{V\"v;y~/;f'i".;"\
-96

-98 +--~--'r--~---r-~--r--~--r-~-..,
time (a)
-86

~oo~
I~2
51 -94
'(5
Z

time (b)

Figure 23.3 Spectral densities of photocurrent fluctuations (c5(Xl + X2)2) (a) and (c5(Yl +
Y2?) (b), SNL-the Shot Noise Limit. Acquisition parameters: radio frequency (rf)
!1127r=2MHz, resolution bandwidth t..!1127r=30KHz, Video bandwidth O.1KHz, the electronics
noise is 8dB below the SNL,

(ETX500 InGaAs), and then each photocurrent of Dl and D2 was divide into
two parts through the power splitters_ The sum and difference of the divided
photocurrent were nothing else but the transmitted amplitude and phase signals
from Alice to Bob, which are expressed by [28]
364 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

(23.S)

where X s (n) and Y s (n) are the modulated amplitude and phase signals on the
first half of EPR beam at the sender Alice. With perfect EPR entangled beam
(X1 (n) + X2(n)) -+ 0 and (Y1 (n) - Y2 (n)) -+ 0, Eqs. (23.7) and (23.S) are
simplified:

(23.9)

(23.10)

It means that under ideal conditions, the signals Xs(n) and Ys(n) encoded on
the amplitude quadrature and phase quadrature of beam 1 are simultaneously
retrieved at the receiver Bob without any error. In general, the both encoded
signals will be recovered with a sensitivity beyond that of the SNL when the
beam 1 and beam 2 are quantum entangled. In fact, the sum of the amplitude
quadratures between the two halves ofEPR beam commutes with the difference
of its phase quadrature, therefore the detected variances in them can be below
that of the SNL simultaneously, and not violate the uncertainty principle [24].
Fig. 23.5 shows the directly measured amplitude [Fig. 23.5(a)] and phase
[Fig. 23.5(b)] signals at Bob, which are the signals of 2MHz modulated on
the first half of EPR beam (beam 1) by the amplitude and phase modulators
at Alice. It can be seen that the original signals are retrieved with the high
signal to noise ratio of -4dB and -3.6dB beyond that of the SNL under the
help of the other half of EPR beam (beam 2)(accounting for the electronics
noise of -SdB below the SNL, the actual value should be -5.4dB and -4.SdB
respectively) . Compared with the previously completed sub shot noise limit
optical measurements and quantum non-demolition measurements for a signal,
in which the squeezed state light have been applied, our experiments have
achieved the simultaneous measurements of two signals modulated on the
amplitude and phase quardratures respectively with the precision beyond that
of the SNL by means of exploiting the EPR entanglement. Although when
sending and modulating two quadratures of a coherent beam a factor of two in
channel capacity may also be gained, the signal-to-noise ratios of measurements
are not able to breakthrough the SNL.
As well-discussed in Ref. [15], in which a signal is transmitted via two
quantum channels of EPR pair, in our scheme the individual signal channel
 Experimental demonstration of dense coding and quantum cryptography 365

:
540nrn

Laser 1080nm

AM PM

Figure 23.4 Schematic ofthe experimental apparatus for dense coding for continuous variables.
Two bits of classical information X sand Y s are encoded on the amplitude and phase quadratures
of a half of EPR beam (beam 1) at Alice, then are decoded by the other half of EPR beam (beam
2) at Bob.

has a high degree of immunity to unauthorized interception since very low

signal-to-noise ratios. Fig. 23.6 demonstrates the security of the signal channel
against eavesdropping, where the trace 23.6(a) is the fluctuation spectrum of
the first half of EPR beam 1 with the modulated signals measured individually
at Bob while the other half of EPR beam is not applied, which is -4.4dB (after
the correction to the electronics noise floor, it should actually be -5.4dB) above
the SNL. It is obvious that no signal can be extracted since the signals are
totally submerged in the large noise background.

5. QUANTUM CRYPTOGRAPHY
In recent years the cryptographic schemes employing continuous coherent
and nonclassical light fields have been suggested [6-19]. A lot of interest has
arisen in continuous variable quantum cryptography (CVQC) with EPR beams
due to that the experimental demonstrations of quantum teleportation [20]
and quantum dense coding [21] for continuous variables (CV). J.Kimble and
his colleagues completed the quantum communication of dual channels with
correlated nonclassical states of light [15]. However in this case, an eavesdrop-
per (Eve) could simultaneously access the signal and idler beams without the
366 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

I~+---~-.--~---.------,---~-.
10 1.5 2C 15
Frequoncy(MHz}
(a)
·M

I ~; +---~--.-------.---~--.---~~
I .U 1~ 2.U :.1.1,)I.'
(b) Frequency(MHz)

Figure 23.5 Measured amplitude (a) and phase signal (b) at Bob, when EPR beam 1 is phase
and amplitude modulated at 2M Hz at Alice. SNL-the Shot Noise Limit (black line). Acquisition
parameter: measured frequency range LOMHz-3.0MHz, resolution bandwidth 30KHz, video
bandwidth 0.1KHz, the electronics noise is 8dB below the SNL.

knowledge of the legitimate receiver. The first experimental demonstration on

CVQC has been achieved by Peng's group at very recent date [29].
In Ref. [29], a scheme using the non local correlation of bright EPR beams
to implement QC is presented, in which the source generating EPR beams is
placed inside the receiver, only one of the EPR correlated beams (signal beam)
is sent to Alice and other one (idler beam) is retained by Bob. The security of the
presented system is directly provided by EPR correlations between amplitude
and phase quadratures of continuous bright optical beams and no-cloning of
quantum fluctuations. Any random choice on measurements of quadratures are
not needed and the sender encodes the key string with predetermined binary
 Experimental demonstration of dense coding and quantum cryptography 367

-80

10 1.5 2.0 2.5 3-'

Frequency(MHz)

Figure 23.6 Spectral density of photo current fluctuations of beam 1 with the modulation signals
(trace(a», the modulated signals are submerged in the noise background. SNL-the Shot Noise
Limit (trace(b» Acquisition parameter: measured frequency range l.OMHz-3.0MHz, resolution
bandwidth 30KHz, vides bandwidth O.IKHz; the electronics noise is 5.6dB below the SNL.

bits [13]. All transmitted bits are used for constituting the key string without
bit rejection, thus the transmission efficiency of 100% may be achieved in
principle.
The quadrature amplitudes of the two output field modes from a NOPA
operating at deamplification are [22,23,26]:

Xl [XOI cosh(r) - X 02 sinh(r)] (23.11)

YI [YOI cosh(r) + Y02 sinh(r)]
X2 [X02 cosh(r) - X OI sinh(r)]
Y2 [Y02 cosh(r) + YOI sinh(r)]
where Xo I , X02 and YOl, Y0 2 are the amplitude and phase quadratures of input
signal and idler modes of the NOPA respectively, r (0 ~ r ~ +00) is the
correlation parameter between modes 1 and 2 which depends on the strength
and the time of parametric interaction, r = 0 without correlation, r > 0 with
partial correlation, r -+ 00 with perfect correlation. From Eqs. (23.1) the
normalized fluctuation variances of the amplitude quadratures for the output
modes 1 and 2 are calculated:

(23.12)
368 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

The quantum correlation variances between the quadratures are

(23.13)

Here we have assumed that the two modes are totally balanced during the
process of measurements and this requirement is easily achieved in the exper-
iments. Generally, r is a function of the noise frequencies. The modulated
signals at given radio frequencies (rf) can be considered as a noise 8Xs(O)
and 8Ys(f!). When the powers of 8Xs(O) and 8Ys(f!) are smaller than the
original quantum noise power of the signal beam (Eqs. (23.12» and larger
than the correlation noise power (Eqs. (23.13», the signals are submerged in
the noise background of the signal beam ( Xl, YI ) and may be decoded by
the correspondent idler beam ( X 2, Y2 ) of the EPR beams. Thus, the strength
(8 2 Xs) of the modulated signals should satisfy the following inequalities:

2e- 2r (r!) < (23.14)

2e- 2r (r!) <

Fig. 23.7 shows the functions of the variances (8 2 (XI)) = (8 2 (YI)) (curve I)
and (8 2(Xl + X 2)) = (8 2(YI - Y2 )) (curve II) versus the correlation param-
eters . At the cross point A of the curves 1 and II, r = rl = 0.275, the variance
is 2.4dB below the normalized SNL of EPR beams (line ii) and when r > rl the
variances of signal beam (curve I) are higher than the correlation fluctuations
(curve II). The line of variance=1 (line i) stands for the SNL of the noise power
of signal beam which is 3dB below that of both EPR correlated beams (line ii).
The larger the correlation parameter is, the higher the noise power of the signal
beam is and the lower the correlation variance is. For a perfect transmission
line of noiseless the correlation parameter of the NOPA should be larger than
rl = 0.275 at least for accomplishing the quantum cryptography. The quantum
correlation of -2.4dB is a kind of boundary where the security becomes more
favorable due to that the interception of Eve becomes harder. The height of the
modulated signals should be between curve (I) and curve (II) for a given value
r > rl.
At Bob station, the signals 8Xs(f!) and 8Ys(f!) modulated on the amplitude
and phase quadratures of signal beam ( Xl, YI ) are decoded by the retained
idler beam by means of the direct detection of photocurrents and two rf beam
splitters. The normalized photocurrents measured with the positive (+) and
negative (-) power combiners are :
 Experimental demonstration of dense coding and quantum cryptography 369

Variance

r
0.2 rl r2 0.4 0.6 0.8

Figure 23.7 The variances of W(Xl)) = (c\'2(YI)) (curve I) and W(X 1 +X2)) = W(Yl -
Y 2 )) versus the correlation parameter r. The line (i) and (ii) are the SNL of signal beam and
EPR beams, respectively. At the point A (r = r' ~ 0.275) the variance is ....... 2.4dBm below the
SNL of EPR beams (ii).

1
yI2{[X1 (O) + X2(O)] + Xs(O) * BV} (23.15)
1 --
yI2{[Yl(O) - Y2(O] + Ys(O) * BV}

Where BV and BV stand for the bit values 1 and O. BV is the NOT of BV,
that means when BV equals "I" the BV must be "O",vice versa.
Fig. 23.8 is the schematic of the quantum cryptography system. The EPR
source, NOPA, is set inside the Bob receiving station. The one of the bright
EPR correlated beams, the signal beam ( XI, Yl), is sent to the Alice sending
station where Alice encodes the transmitted information on the amplitude and
phase quadratures by the choice of the modulation types with the binary bit
values, for example the modulated amplitude signals stands for "I" and the
phase signals for "0". Then the encoded signal beam is transmitted back to
Bob where the information is decoded by the retained other one of the EPR
correlated beams.
370 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Figure 23.8 The schematic of the quantum cryptography using the EPR beams, AM-amplitude
modulator, PM-phase modulator, RNG-random number generator, SA-spectrum analyzer, PS-
phase shifter, BS-beamsplitter of 50%, PBS-polarization beamsplitter, NOPA-nondegenerate
optical parametric amplifier.

A secretly encoded and transmitted binary bit string is summarized in Ta-

ble 23.1. At the time points tI, t2 ... t6, Alice encodes the signal beam and sends
it back to Bob, then Bob simultaneously measures the amplitude and phase
modulation signals. The measured amplitude signals attl, t4, t5 stand for "I",
the phase singles at t2, t3, t6 for "0". The upper traces are the spectral densities
of noise power of the SNL for the EPR beams which is used to evaluate the
quantum correlations. The middle traces are the noise spectra of signal beam in
which the modulated signals are submerged in the noise background totally and
the lower traces are the variances (8 2(Xl + X 2)) and (8 2(YI - Y2 )) , which
are 3.8dBm below the noise level of the SNL of the EPR beams, and the secret
signals modulated at 2MHz emerge from the squeezed noise backgrounds. A
secret key string characterized by the binary bit values 100110 is obtained. The
acquisition parameters of the noise spectral densities in Table 23.1 are: the
measured frequency range 1.0-3.0MHz, resolution bandwidth 30KHz, video
bandwidth 0.1 KHz. The electronic noise, which has not been given in the noise
spectra of Table 23.1, is about 8dBm below that of the SNL of the EPR beams.
In experiments, we should pay attention on matching the optical distances
between the signal and idler beams to reach optimum correlation and detection
efficiency. Especially for long distance transmission one must add optical
retarder and attenuator in the idler beam retained by Bob to make its travel
distance from EPR source to the beam splitter of the detection system and the
 Experimental demonstration of dense coding and quantum cryptography 371

T T. Ts

Alice PM PM AM AM PM

AQ

Bob

PQ

CD 3.8 3.8 3.8 3.8 3.8 3.8

(dBm)

Key o o o

Table 23.1 The generation of the predetermined secret key string, AQ-Amplitude modulation
signals, PQ-Phase modulation signals, CD-The correlation degrees between signal and idler
beams measured by Bob. Key-The secret keys, upper traces-The SNL of EPR beams, middle
traces-the noise spectra of signal beam, lower traces-the noise spectra of (8 2(Xl + X 2)) and
(8 2(Yl - Y2)).

detected intensity approximately equal to that of the signal beam. Recently,

the schemes on storing quantum entanglement in atomic spins system or in
cavity QED have been presented [31,32], that probably can provide a practical
technique for retaining an EPR component in Bob.
The intensity fluctuations of signal and idler fields of the bright EPR beams
generated by the NOPA are quantum-correlated at any instant and nobody can
make or copy a light field which has the identical quantum fluctuations. The
basis of the security of the presented protocol is the quantum correlation. In
experiment, the SNR(signal to noise ratio) of the decoded signals depends on
the original correlation of EPR beams and the losses of a given transmission
line, which is independent of Alice's bit value. Any decrease of the measured
SNR and the correlation degree relative to the predicted results alerts Bob to
372 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

the additional loss caused by a partial tapping of the signal channel, perhaps
by Eve. If Eve wants to perform the quantum-nondemonlition-measurment
(QND) on one of the amplitude-phase quadratures of the signal beam, the
other quadrature must be disturbed and the disturbance must be reflected on
the measured results of Bob. The deviation would indicate the possibility that
Eve have performed a QND. More quantitative analyses on protecting against
the optical tap attack of Eve have been presented in the previous publications
[11-15] and can be used in the discussion to the presented protocol.
For the ideal case of the EPR beams with perfect quantum correlation, the
SNR on the signal channel goes to zero due to the fact that the quantum noise
of the signal beam goes to infinite [15,28]. Thus, the attacking from optical tap
or the optical QND on the signal beam has no possibility to extract information
hidden in the infinite quantum noise without the help of the other one of the EPR
correlated beams. However, for the imperfect correlations the signal channel
has large but finite quantum noise background. In this case, the eavesdroppers
can enhance the measurement SNR by narrowing the resolution bandwidth
(RBW) of the spectral analyzer (SA) according to [33]:

<i~)
SNR = (iJv) = 'fJls/BFo (23.16)

where <i~) and <iJv) are the mean square signal and noise photocurrents
respectively, Is is the photon number flux of the signal beam, "I is the quantum
efficiency of the detection system and Fo is the Fano factor of the signal beam,
Fo -+ 00 for perfect correlated EPR beams and Fo = 1 without quantum
correlation(coherent state light). For the given "I, Is and F o, the usable RBW
of the measurement is limited by the requirement of SNR;::: 1,i.e.

B0 "lIs
<
- Fo
- (23.17)

It means that the duration of measurement should be larger than BOI. For
secure communication Alice should encode the message with a bit rate larger
than Bo decided by Eq. (23.17), so that Eve can not intercept information
due to having no enough interval to accomplish the measurement. In our
experiment when B = 30kHz, the signals have been totally submerged in the
noise background (the middle traces in Table 23.1), thus the condition is easily
satisfied in the practical optical communication.
A possible eavesdropping scheme using fake EPR beams is shown in Fig. 23.9.
Eve intercepts totally the signal beam (Xl, Yl ) and transmits a fake signal beam
(Xf, Y{) produced by a fake EPR source (EPR2) in her station to Alice. Alice
has no ability to recognize the fake beam. She modulates the information on it
and then sends out as usual. Eve intercepts the beam with messages again and
 Experimental demonstration of dense coding and quantum cryptography 373

Bob

Detector 1 t====1

Eve

(X/,Yj)

(Xt',Y/)

Alice

Figure 23.9 Diagram of eavesdropping Scheme using fake EPR beams, EPR1, EPR2-Sources
of EPR beams, AM, PM-Amplitude and phase modulators, SA-spectral analyzer, OSl, OS2-
Oscilloscopes, Detector 1, 2-Detection systems of Bell-state, Do-photoelectric detector, RNG-
random number generator.

decodes the infonnation with the other one of the fake EPR correlated beams
(X~, Yf) retained by her. At the same time she modulates the real signal beam
(XI'yI) according to the intercepted bit values and sends it back to Bob who
also does not know that the infonnation has been intercepted. To reveal this
type of quantum interception Alice may randomly block the signal beam with
374 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

a photoelectric detector (Do) connected to an oscilloscope at some appointed

time points tk for a short time interval Ll.tk during the duration of transmission.
The shape of intensity fluctuation of the signal beam at the blocked moment
can be recorded as a function of time by an oscilloscope (081 ), In Bob station,
the photocurrent of the sum of the amplitude quadratures (d 2(XI + X 2)) is
split to two parts by a power splitter, one is sent to a spectrum analyzer (SA)
for the noise spectrum measurement and other one to an oscilloscope (OS2)
for recording the fluctuation of photocurrents. If there is no Eve between Al-
ice and Bob, while the signal beam is blocked by Do the oscilloscope 082
in Bob station records the function of intensity fluctuation of the idler beam
( d2 X 2 (t)) which is anticorrelated with (d 2 XI (t)) recorded by 081 at same
time and the correlation extent only depends on the quality of the EPR source
(EPRl) and the losses of transmission line. Fig. 23.10 shows the intensity fluc-
tuation shapes of (d 2 X 1 (t)) (tracel) and (d 2 X 2 (t)) (trace2) simultaneously
recorded by 081 and 082 respectively. The partial anticorrelation between the
shapes of the fluctuations is very obvious and the measured root-mean-square
(rms) voltages for the sum and difference of traces land 2 are 568.64JL V and
655.29JL V respectively. For two uncorrelated beams with same intensity, the
rms voltages for sum and difference photocurrents should be equal, thus the dif-
ference between rms voltages also shows the presence of quantum correlation.
If Eve interrupts the real signal beam, the intensity fluctuations of (d 2 X~ (t) )
recorded by Alice at the blocked moment are totally not correlated with that
recorded by 082 at same time. After a set of communication is finished, Alice
publicly sends the wave shapes of intensity fluctuation and the rms voltages
recorded by 081 to Bob, where Bob compares with that recorded by himself
with 082 at same time points. The absence of anticorrelation between the two
sets of fluctuation shapes and the same rms voltages for the sum and difference
indicate the presence of Eve. In this case the transmitted secret key string
should be abandoned. If the fluctuation shapes recorded by Bob and Alice are
quantum anticorrelated in some extent determined by the EPR source (EPRl)
of Bob and losses of transmission line, we may deem that the communication
is secure. During the duration of the signal transmission the more the times of
blocking the signal beam are, the higher the degree of security is, but the longer
the time on the transmission line occupied by the measurement for security is.
Therefore we have to do a trade-off between the signal transmission and the
security measurement in practical communication.
The intervention strategies would be various and the discussion on the com-
plete security lies beyond the scope of the present paper. We consider that the
presented protocol is a worthy candidate for investigating the quantum cryp-
tography because of its good security directly provided by quantum correlation
and no-cloning of quantum fluctuation.
 Experimental demonstration of dense coding and quantum cryptography 375

I
965-. 56-18~
-965.5r 184 ••

Figure 23.10 The fluctuation shapes of signal (I) and idler (2) beams recorded by the oscillo-
scopes OSl and OS2 at Alice and Bob.

6. CONCLUSION
We have experimentally demonstrated the quantum dense coding and quan-
tum cryptography with continuous variables. Due to using the bright EPR
correlated beams, the transmitted signals are directly modulated on the am-
plitude and phase quadratures of signal beam with the amplitude and phase
modulators compatible with that used in the present optical communication
system operating at very high rates. The application of the direct detection
technology of the Bell-state simplifies the measurement system and the align-
ing procedures, and further improves the detection efficiency. Available high
bits transmission rates and high detection efficiency are favorable features of the
presented schemes. Besides, we have proved that the experimentally achiev-
able quantum correlation and squeezing level may be used for performing the
quantum communication. Unlike most of proposed protocols for quantum
cryptography based on BB84 [34], our scheme does not require any randomly
chooses of measuring components, so there is no the usual 50% bit rejec-
tion. The predetermined secret key string is directly modulated on the optical
beam(the carrier of signals) just like that used in traditional optical communi-
cation systems. The partial compatibility with classical optical communication
376 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

technology and the simplicity of encoding and decoding procedures would be

welcome by technicians working for communications.

These works were funded by the National Fundamental Research Program

(No.2001CB309304) and the National Nature Science Foundation (No.698370
10,60178012)

References
[1] Braunstein, S. L. (1998) Error correction for continuous quantum vari-
ables. Phys. Rev. lett. 80,4084-4087.
[2] Braunstein, S. L. (1998) Quantum error correction for communication
with linear optics. Nature (London) 394. 47-49
[3] Braunstrin, S. L. & Kimble, H. J. (2000) Dense coding for continuous
variables. Phys. Rev. A 61, 042302
[4] Ban, M. (1999) Quantum dense coding via a two-mode squeezed-vacuum
state. J. opt. B: Quantum Semic1ass. Opt. 1 L9-Lll
[5] Loock, P. Van &Braunstein, S. L. (2000) Uconditional entanglement
swapping for continuous variables. Phys. Rev. A 61, 10302(R)
[6] Gottesman, D., & Preskill, J. (2001) Secure quantum key distribution
using squeezed states. Phys. Rev. A 63, 022309
[7] Hillery, M., (2000) Quantum cryptography with squeezed states. Phys.
Rev. A 61, 022309
[8] Cerf, N. J., Levy, M., & Assche, G. v., (2000) Quantum distribution of
Gaussian keys using squeezed states. Phys. Rev. A 63,052311
[9] Cerf, N. J., Iblisdir, S., & Assche, G. V., (2001) Cloning AND
Cryptography with Quantum Continuous Variables. quant-phlO107077
[Eur.Phys.J.D(to be published)]
[10] Assche G. V. et al., (2001) Reconciliation of a Quantum-Distributed Gaus-
sian Key. cS.CRlO107030 (to be published)
[11] Ralph, T. C. (2000) Continuous variable quantum cryptography. Phys.
Rev. A 61, 010303(R)
[12] Ralph, T. c., (2000) Security of continuous-variable quantum cryptogra-
phy. Phys. Rev. A 62, 062306
[13] Reid, M. D., (2000) Quantum cryptography with a predetermined key,
using continuous-variable Einstein-Podolsky-Rosen correlations. Phys.
Rev. A 62, 062308
[14] Silberhom, c., Korolkova, N., & Leuchs, G., (2001) Quantum key distri-
bution with bright entangled beams. quant-phlO109009
 Experimental demonstration of dense coding and quantum cryptography 377

[15] Pereira, S. E, Ou, Z. Y., & Kimble, H. J., (2000) Quantum communication
with correlated nonclassical states. Phys. Rev. A 62, 042311; Kimble, H.
J., Ou, Z. Y., & Pereira, S. E, Method and Apparatus for Quantum Com-
munication Employing Nonclassical Correlations of Quadrature-Phase
Amplitudes. u.s. Patent No. 5,339,182, Issued 8/16/94.
[16] Bencheikh, K. et al., (2001) Quantum key distribution with continuous
variables. J. Mod. Opt. 48, 1903
[17] Lorenz, S. et al., (2001) Squeezed light from microstructured fibres: to-
wards free space quantum cryptography. quant-phlOl09018
[18] Navez, P. et aI., (2001) A "quantum public key" based cryptographic
scheme for continuous variables. quant-phlOlO1113
[19] Grosshans, E, & Grangier, P., (2002) Continuous Variable Quantum Cryp-
tography Using Coherent States. Phys. Rev. Lett. 88,057902
[20] Furusawa, A. et al. (1998) Unconditional quantum teleportation. Science
282, 706-709
[21] Li, X. Y. et ai. Quantum Dense Coding Exploiting a Bright Einstein-
Podolsky-Rosen Beam. Phys. Rev. Lett. 88,047904 (2002)
[22] Reid, M. D., & Drammond, P. D. (1988) Quantum Correlations of Phase
in Nondegenerate Parametric Oscillation. Phys. Rev. Lett. 60, 2731-2733
[23] Reid, M. D. (1989) Demonstration of the Einstein-Podolsky-Rosen para-
dox using nondegenerate parametric amplification. Phys. Rev. A 40,913
[24] Ou, Z. Y.,Pereira, S. E, & Kimble, H. J. (1992) Realization of the Einstein-
Podolsky-Rosen paradox for continuous variables in nondegenerate opti-
cal parametric amplifier. Appi. Phys. B 55, 265
[25] Zhang, Y. et al. (2000) Experimental generation of bright two-mode
quadrature squeezed light from a narrow-band nondegenerate optical para-
metric amplifier. Phys. Rev. A 62,023813
[26] Zhang, Y., Su, H., Xie, C. D. & Peng, K. C. (1999) Quantum variances
and squeezing of output field from NOPA. Phys. Lett. A 259, 171
[27] Li,X. Y., Pan, Q., Jing, J. T., Xie, C. D., & Peng, K. C.,(2001) LD pumped
intracavity frequency-doubled and frequency-stabilized Nd:YAP/KTP
laser with 1.1 w output at 540nm, Optics Communications, (01) 01685-6
[28] Zhang, J. & Peng, K. C. (2000) Quantum teleportation and dense coding
by means of bright amplitude-squeezed light and direct measurement of
a Bell state. Phys. Rev. A 62, 064302
[29] Jing, J.,Pan, Q., Xie, C. D. & Peng, K. C. (2002) Quantum Cryptography
Using Einstein-Podolsky-Rosen Correlations of Continuous Variables.
quant-phl0204111
378 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[30] Pan, Q., Zhang, Y., Zhang, T. c., Xie, C. D. & Peng, K. c., Experimental
inverstigation of intensity fifference squeezing using Nd: YAP laser as
pump source, J. Phys. D: Appl. Phys.30(l997) 1588-1590
[31] Julsgaard, B., Kozhekin, A., & Polzik, E. S. (2001) Experimental long-
lived entanglement of two macroscopic objects. Nature 413, 400-403
[32] Parkins, A. S. & Kimble, R. J. (2000) Position-momentum Einstein-
Podolsky-Rosen state of distantly separated trapped atoms. Phys. Rev. A
61,052104
[33] Li, Y. Q., Lynam, P., Xiao, M., & Edwards, P. J. Sub-Shot-Noise laser
Doppler Anemometry with Amplitude-Squeezed Light. Phys. Rev. Lett.
78, 3105 (1997)
[34] Bennett, C. R., & Brassard, G. Quantum Cryptography: Public Key Dis-
tribution and Coin Tossing. Proc.IEEE Int. Conf. On Computers, Systems
and Signal Processing(Bangalore), 175-179 (IEEE, New York,1984)
Chapter 24

QUANTUM SOLITONS IN OPTICAL FIBRES:

BASIC REQUISITES FOR EXPERIMENTAL
QUANTUM COMMUNICATION

G. Leuchs, Ch. Silberhorn, F. Konig, P. K. Lam, A. Sizmann, N. Korolkova

Zentrumfiir Moderne Optik,
Universitiit Erlangen - Niirnberg, Germany
[email protected]
www.zemo.org

Abstract Continuous variable quantum entanglement emerges from nonlinear interactions

of. fibre optical solitons in combination with some linear operation. We de-
scribe the detection and characterization of bright EPR-entanglement and QND-
entanglement produced in this way and discuss the prospects of bright-beam-
based quantum communication.

1. INTRODUCTION
Non-local properties of continuous variables have been the focus of the
historic discussion which today is considered the starting point of quantum in-
formation [1]. At the time, however, the disputants did not have any application
in mind. They were rather struggling for the proper interpretation of quantum
theory. Einstein, Podolsky, and Rosen called the scenario they described [2]
a Gedanken-experiment for a good reason. Nobody knew how to create con-
tinuous variable quantum systems in an entangled state which could be used
in the laboratory to test for the non-locality of quantum theory. Then in 1951,
Bohm reformulated the EPR-Gedanken-experiment in terms of discrete spin
variables [3] laying the foundation for the laboratory studies to come. The in-
centive for more refined experiments rose when Bell derived a general criterion
based on which the non-locality of quantum theory could be put to a stringent
test [4]. The experiments by Clauser et al. [5] and Aspect et al. [6] which
followed all aimed at a better understanding of the foundations of quantum
379
S.L Braunstein and A.K. Pati (eds.), Quantum Information with Continuous Variables, 379-421.
© 2003 Kluwer Academic Publishers.
380 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

theory. One reason for their success was that dichotomic quantum systems can
be prepared experimentally in a state of close-to-perfect entanglement. And
even without any entanglement a discrete quantum system has the additional
advantage that the detector sends 'on-off' messages conveniently providing a
built-in discriminator function. It is therefore not surprising (in retrospect) that
when proposing quantum teleportation and quantum key distribution Bennett et
al. [7] used the example of the discrete quantum variables which had proven to
be so successful. The numerous experiments which followed were pioneered
by Zeilinger et al. [9]. Soon cryptography will be the first real world appli-
cation of quantum information [8]. However, the production process for the
entangled photon pairs used in these experiments is spontaneous parametric
down conversion and is therefore probabilistic. You never know ahead of time
at which point the experiment will be successful. For cryptographic key distri-
bution the probabilistic production of single photons may be less of a concern
but for other cases such as teleportation the concern seems more serious. Along
with the probabilistic production, the signal rate is fairly low.
Ou et al. [10] were the first to demonstrate that entanglement can be pro-
duced experimentally also with the continuous variable field quadratures of
two different modes of light. Ou et al. used an optical parametric amplifier
to produce entangled light beams at a very low light level. Such continuous
variable entanglement can be seen to be quite complementary to entangled pairs
of photons. Continuous variable entanglement is less perfect and more fragile
with respect to attenuation but it is not probabilistic. It is there at your disposal.
This and the much higher transmission rates may well be favourable in some
applications. Recently bright light entanglement was achieved using the in-
terference scheme [11] and using a nondegenerate optical parametric amplifier
[12], an additional advantage being the availability of efficient direct detection.
Furthennore, nonlinear coupling between bright light fields has been demon-
strated to lead to quantum correlation and entanglement [13, 14, 15] which has
not yet been achieved for single interacting photons. It should be noted here
that future developments of cavity quantum electrodynamics may lead to deter-
ministic sources of single photons [16, 17] and maybe also of entangled photon
pairs. Commercial tools for fibre-optic test and measurement are beginning to
exploit the ultrasensitivity of photon counting techniques at telecommunication
wavelength [18, 19], paving the way towards photon-counting communication
technology.
In the following sections of this chapter we discuss the steps towards ex-
perimental quantum information processing with an example of experiments
using fibre optical soliton pulses. The nonlinear interaction of solitons in a
fibre combined with linear optical elements such as beam splitters is enough
to assemble non-trivial basic building blocks for continuous variable quantum
communication.
 Quantum solitons: towards experimental quantum communication 381

2. CONTINUOUS VARIABLE ENTANGLEMENT:

AN EXPERIMENTAL CLASSIFICATION
Implementation of inter-channel coupling. One of the most attractive fea-
tures of continuous variable quantum information is the possibility of informa-
tion processing exclusively with linear inter-channel operations [20]. The basic
example is the simple and elegant way to generate continuous variable entan-
glement by overlapping two squeezed beams on a beam splitter [11, 21, 22, 23].
Closely related is the entanglement produced at the output of the non-degenerate
type II optical parametric oscillator (OPO) [10, 12,24], which, however, uses
a nonlinear interaction between two channels.
Single squeezed beams for entanglement generation can be produced in
different nonlinear interactions within a single channel, e.g. type I OPO [22],
or Kerr medium [11]. These beams are then used as a resource to create
entanglement by their linear interference at a beam splitter. The entanglement,
therefore, emerges from a linear interaction between two non-linear channels.
Soliton pulses in silica fibres allow for very efficient implementation of this
scheme exploiting the Kerr nonlinearity of an optical fibre to generate long-
term stable and easy to handle amplitude squeezing. The entanglement, which
emerges in this case, is connected to the non-local correlations between quantum
uncertainties of the conjugate variables in two spatially separated intense beams.
It is also possible to employ a non-linear interaction between two channels
to create entanglement. Apart from the self-phase modulation effect used
for squeezing generation, the Kerr nonlinearity also results in cross-phase
modulation if optical fields of different polarizations or different frequencies
are involved. For instance, back-action evading measurement schemes with
colliding solitons are based on the cross-Kerr-effect [25]. The non-linear Kerr
interaction between two channels entangles two pulses, the signal pulse and
the probe pulse, and hence a QND measurement of the signal photon number
can be performed. In one scheme the non-linear inter-channel interaction leads
to a non-local quantum correlation between the photon-number of the signal
pulse, ns and a phase shift of the probe pulse, fl</Jp, [25, 26]. A non-local
quantum correlation between the signal photon number ns and a frequency
shift fljp of the probe emerges in another scheme [14, 15]. In terms of
observables it corresponds to a correlation between the photon number (ns) and
the momentum Pp of the probe pulse. Applying a spectral filtering technique,
this correlation can be transferred into a photon number - photon number
correlation between the signal (ns) and the spectrally filtered probe pulses
(np ). Continuous variable entanglement generated in QND interaction can
also be employed for quantum communication purposes. Recently, a quantum
teleportation scheme was proposed using quantum nondemolition technique
382 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[27], where the QND interaction of coherent input fields was discussed as well
as QND coupling of squeezed beams aiming at improved teleportation quality.

Continuous variable EPR-paradox. EPR-entanglement and ideas of QND-

measurement [27, 28, 29] are closely related. A clear insight in this tight
connection can be found in the characterization of quantum correlations.
The quality of the quantum correlations between the signal As and probe
Bp variables at the output of the QND-device is associated with the quality
of the state preparation. An experimental measure for this is the variance
in the signal output given a measured value for the probe field, e.g. the
variance of the difference of the measured values of the two correlated variables.
It is the variance of the relevant conditional probability distribution and is
conventionally referred to as the conditional variance [28]:

(24.1)

with

(24.2)

where V(A) denotes a variance (A2)_(A)2 of an observable A and C 2 (As, Bp)

is the correlation coefficient. The variance V (As ,SN) corresponds to the shot
noise, i.e. vacuum noise level which marks a boundary between classical and
quantum regime. Equivalently Vcond(As /Bp) can be expressed in terms of the
normalized variance of As + gBp for an optimal gain g:

(24.3)

In a QND regime the conditional variance (24.3) should be nonclassical reveal-

ing a degree of quantum correlations which is better than the vacuum noise
level of a signal beam:

(24.4)

The notion of continuous EPR-like correlations of the amplitude and phase

quadratures is related to the demonstration of the EPR-paradox for continuous
variables introduced by Reid and Drummond in 1988 [24]. If two spatially
separated beams are correlated in their amplitude Xj = a} + aj and phase Yj =
i(a} - aj) quadratures, it is possible to infer e.g. X 2 by a measurement of Xl
 Quantum solitons: towards experimental quantum communication 383

or Y2 by YI . The quality of this continuous variable entanglement is restricted

by finite squeezing values so that there will always be a limit in this inference.
This was originally referred to as an inference error Vinf (Xj ), Vinf ("fj) [24].
To minimize the error, a variable gain 9 [24, 25, 28] is introduced:

(24.5)

The upper signs are used if the amplitudes are anti-correlated and the phases
are correlated and the lower signs are for the reversed situation.
Note, that for the optimal gains gx = l;r, gy = g?t the inference errors
(24.5) in the derivation of the EPR-paradox are equivalent to the conditional
variances (24.1,24.3):

(24.6)

Consider now the quantum noise limit for the measurement of conjugate
quadratures of a single field which is given by the Heisenberg relation:

V(X)V(Y) ~ 1. (24.7)

The continuous variable EPR-paradox is related to an apparent violation of

this fundamental limit [24]. In our notation (24.6), the demonstration of the
EPR-paradox requires simultaneous non-classical values of the respective con-
ditional variances:

(24.8)

(24.9)

This specifies the ability to infer" at a distance" either of the two non-commuting
signal observables with a precision below the shot noise level of the signal
beam [24], (24.4). Condition (24.9) defines the EPR boundary for a non-
local quantum correlation and provides a sufficient condition for a state to be
entangled. Hence, if the quantum correlation is stronger than a certain threshold
one refers to EPR-entanglement and this is closely related to the idea of a QND
measurement.
384 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Non-separabilty criterion for continuous variables. It is important to de-

fine accurately the relation between notions "correlated systems" and "en-
tangled systems". Which kind of correlations implies entanglement? In
the paragraph above we have defined EPR-correlations in terms of the EPR-
Gedankenexperiment introduced for amplitude and phase quadrature operators
by Reid [24]. This represents a sufficient entanglement criterion. Are there
other ways which allow one to draw the line between classical correlations and
entanglement for continuous variable systems more precisely? The conven-
tional notion of entanglement for pure states is intrinsically related to quantum
inseparability which means that in general it is not possible to assign a single
state vector to either of two systems which have interacted in the past [9]. For
mixed states it was first formulated by Werner [30]: "A state of a composite
quantum system is called classically correlated if it can be approximated by con-
vex combinations of product states, and Einstein-Podolsky-Rosen correlated
otherwise" .
Hence, a separable state of a joint system described by a density matrix p
can be written as:

(24.10)

where Pil, Pi2 denote density matrices of two subsystems 1 and 2 (see also
[30]). In contrast, an entangled state is a non-separable quantum state of a
system which cannot be represented as a convex sum of product states of two
subsystems. For discrete variables, i.e. for the Hilbert space of 2 x 2 and
2 x 3 dimensions, a necessary and sufficient criterion for separability is the
requirement that the partial transpose pT of the density matrix is positive which
is known as the Peres-Horodecki criterion.
Recently, the Peres-Horodecki criterion was extended to higher dimensions
of the Hilbert space, i.e. for continuous variables, using two different ap-
proaches (for details see [31, 32] and chapters 13 and 14 of the present book).
The analysis of Simon [31] is accomplished using the formalism of the phase
space Wigner function W(x,p) [21, 33]. It is based on the recognition that
the partial transpose pT is equivalent to a mirror reflection in phase space.
If p is separable, its Wigner distribution necessarily goes over into a Wigner
distribution under the phase space mirror reflection [31]:

W(x,p) ---+ W(x, -p). (24.11)

Here x, p are the canonical variables position and momentum. This recognition
(24.11) leads to uncertainty principles to be obeyed by all separable states. For
all bipartite Gaussian states, the continuous variable Peres-Horodecki criterion
can then be written, which turns out to be a necessary and sufficient condition
for separability [31]. The criterion of Simon corresponds to the necessary and
 Quantum solitons: towards experimental quantum communication 385

sufficient form of the criterion derived by Duan et al [32]. The rest of this
section is devoted to this latter criterion first in its sufficient and then in its
necessary and sufficient form as it allows one to use the conventional tools
of experimental quantum optics and is more convenient for the experimental
verification of entanglement of intense beams.
Duan et al [32] are deriving the non-separability criterion in the spirit of
two-mode squeezing [33]:
(24.12)
They introduce the EPR-like operators in the form of joint variables:

AliA
U= a Xl + -alAX2, AliA
V= a PI - -alAP2; (24.13)
[xj,lh] = i8j k (j, k = 1,2) (24.14)
where a is an arbitrary non-zero real number. Then the total variance of u
and v is derived under the assumption that (24.10) holds. The lower bound
for this total variance is calculated which represents a separability condition.
A sufficient criterion for non-separability of an arbitrary two-mode state is
obtained whenever the following inequality is fulfilled [32]:

V(u)p + V(v)p < a2 + 12 • (24.15)

a
For a bipartite Gaussian state this inequality can also give a necessary and
sufficient condition for non-separability. For this purpose both beams should
be transformed into a canonical form by performing local linear Bogolubov
operations on them and by using an optimal gain (see [32] and next subsection
for details). The criterion of Duan et al [32] (24.15) and the equivalent to
its necessary and sufficient form criterion of Simon [31] have an advantage
for experimental quantum communication as they can be expressed in terms
of measurable quantities. In Eqn. (24.15) the variances V(U), V(v) are
presented without any additional normalization. For the analytical analysis the
normalization is not needed: the reference level is the noise variance of the
coherent beam or equivalently of the coherent vacuum which is equal to 1/2
in the notation of (24.13-24.15). However, in an experiment the respective
reference is represented by the shot noise level and has to be always explicitly
determined in each measurement run. To account for this, we rewrite the
Peres-Horodecki non-separability criterion (24.15) for continuous variables
quadratureamplitudesXj = a}+aj, Yj = i (a} - aj) with [Xj, Yk] = 2i8jk,
j, k = 1, 2 of two subsystem, beams 1 and 2. We introduce the joint variables
X = Xl +gX2 and Y = 1\ - gY2 oftwo bright beams analogously to (24.13),
where 9 is a variable gain like in the description of continuous variable EPR-
paradox presented above. The quadrature correlations between two beams can
386 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

be characterized by the normalized squeezing variances v;,~ (.X), v;,~ CY) which
are known from the context of two-mode squeezing [33]:

(24.16)

(24.17)

The field modes are denoted by the respective subscripts and SN denotes the
shot noise limit of the respective beam and gx = gy = 9 by the argument of
symmetry. Apart for a difference in the normalization, the variances in (24.15)
correspond to the squeezing variances in (24.16, 24.17). The joint variables
(24.13) for the optical beams are expressed in terms of amplitude and phase
quadratures.
To rewrite the non-separability criterion (24.15) in terms of measured vari-
ances of quadrature operators, let us consider the amplitude quadratures Xj
first. If we are dealing with bright beams, we can write:

n·"'n·+a·8X·
J'" J J J (24.18)

where aj » 1 is a strong classical coherent amplitude, a = (a); 8aj is

an operator with zero mean which describes the quantum uncertainty and
nj is a photon number operator of the respective beam with mean value nj.
The expression for nj is written using the linearization approach, omitting
terms in the second order of 8Xj (8Xj = 8a} + 8aj) and higher. The direct
photo detection in each beam 1 and 2 yields photo currents containing the
measurement results of nj which are then used to determine the relevant shot
noise reference and the combined variance of two beams (see (24.18)). The
shot noise variance is related to the strong coherent amplitude and equals:

(24.19)

The variance of the combined variables read out by means of photo detection
is given by (24.18):

V(ih + 9 n2) = V(a1 8X1 + 9 a2 8X2) (24.20)

To facilitate the comparison with (24.15), it can be expressed as

(24.21)
 Quantum solitons: towards experimental quantum communication 387

The variance Yst (.X) corresponding to that of the joint variables of (24.13)
[32] is then obtained in the form:

V+(X) = V(nl + 9 n2) = V ( fj£ oXl + .flf OX2 ) (24.22)

sq VSN(nl + 9 n2) ..!!L
g0l2
+~
011

For two beams 1 and 2 of equal intensity it takes a simple form discussed by
Reid in 1989 while relating their EPR criterion [24] for continuous variables
amplitude and phase to two-mode squeezing:

(24.23)

Compare (24.22) to (24.13, 24.15):

V+(X) = 2 V(U)
sq a2 + Q;21 for a= ff£
-
90.2
1
(24.24)

where the factor 2 comes from different normalizations used: [Xj,Pk] = iOjk
in (24.13) and [Xj, Yk] = 2iojk for quadrature operators, i.e. V(Xj) =
V ( ~ ). The latter normalization for quadratures is convenient in the current
context while it corresponds to the Heisenberg uncertainty relation bounded by
unity and the two-mode squeezing condition of Yst(X) < 1. Analogously,

V-(Y) = 2 V(V)
sq a2 + Q;21 for a= ff£
-.
90.2
1
(24.25)

Compiling (24.16, 24.17,24.24, 24.25, 24.15), we can write the non-separability

criterion [31, 32] in terms of measurable squeezing variances of two bright
beams:

(24.26)

This is a sufficient criterion for non-separability for any two-mode bipartite

state (24.15) put in the form suitable for the experiments where the observables
are continuous variables, amplitude and phase quadrature operators. A state
obeying (24.26) we refer to as a two-mode non-separable state.

Necessary and sufficient criterion, covariance matrix, correlation matrix

and separability criterion for all bipartite Gaussian state. For a certain
class of Gaussian two-mode states, inequality (24.26) can give a necessary
and sufficient condition for a state being entangled [32, 35]. This is, for
388 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

example, the case for bright entangled beams generated in our experiment
[11] (Sec. 3.) and it corresponds to a particular choice of the Xj, Yj basis
ensuring maximal available correlations and to the optimized gain gOpt [32].
In general, a Gaussian state of n modes is completely characterized by a
covariance matrix [34] or a correlation matrix [35], the elements of which
are measurable quantities, covariances of Gaussian probability distributions
describing the correlations between all relevant conjugate variables in and
between all involved modes of a Gaussian bipartite state. Separability and
the positivity of the partial transpose of any Gaussian bipartite state can be
characterized in terms of the correlation matrix [34, 35]. The necessary and
sufficient condition for separability [32, 35] and distillability criterion [35] for
all Gaussian bipartite states have been recently derived using this formalism
(see also chapters 13-15 ofthis book). The experimental determination of the
correlation matrix for bright beams is rather evolved as it requires the use of
strong, phase-matched local oscillators or other techniques for tomographic
measurements of all quadrature operators. In our current experiments, we thus
restrict ourselves to the entanglement characterization using Eq. (24.26) [32].
However, the correlation matrix is an important tool for describing Gaussian
states. The development of experimental methods to record it are in progress.

Entanglement quantification and types of entanglement. We aim at en-

tanglement applications in quantum communication with bright beams. For
this purpose the following paragraph establishes entanglement measures which
are reliably observable in an experiment. We use the notions of two-mode
squeezing [33], of non-separability [31, 32] and of the continuous variable
EPR paradox [24] to define experimental criteria which are well-suited for the
evaluation of continuous entanglement produced in the linear interaction of
squeezed fields, in any non-linear interaction, or in QND-interactions. In what
follows, the optimal gain is taken to be l;r = gr;r = 9 assuming a symmetry
between the beams in the quantum uncertainties inherent to entangled states
and a symmetry in optical powers [24, 11].

1. Squeezed-state entanglement. Squeezed-state-entanglement (SSE in Ta-

ble 24.1) defines a non-separable [30, 31, 32] two-mode state which satisfies
the conditions

~~(X) = ~(Xl ± 9 ~2) < 1, (24.27)

V(X1,SN + gX2,SN)
~~cY) = ~CYl =f 9 ~2) <1 (24.28)
V(Y1,SN + gY2,SN)
for the variances (24.16), (24.17) of conjugate variables with the field modes
denoted by the respective subscripts.
 Quantum solitons: towards experimental quantum communication 389

Table 24.1 Types of continuous variable non-local correlations.

Generating Inter-channel Correlated Correlation

process coupling variables type

OPOtype II; non-linear 8X1 ex: 8X2 EPR

Ou et al. [10]. 8Yl ex: - 8Y2 and/or SSE

OPOtype I; linear 8Xl ex: 6X2 EPR

Furusawa et al. [22]. 6Y1 ex: - 6Y2 and/or SSE

Kerr nonlinearity in fibre; linear 8Xl ex: - 6X2 EPR

Silberhom et al. [11]. 6Y1 ex: 6Y2 and/or SSE

QND (phase shift); D..cpp ex: ns QND;

Friberg et al. [26]. non-linear D..cps ex: np one-wayEPR

QND (spectral shift); non-linear !:lip ex: ns (pp ex: ns) QND
Konig et al. [14]. !:lis ex: np (Ps ex: np)

QND (spectral filtering); non-linear 6np ex: 6n s QND

Konig et al. [14].

QND (squeezed light beam

splitter v(x~n) < 1); linear Xp ex: XS QND
Bruckmeier et al. [37].

To test for squeezed-state entanglement one has to determine experimentally

the variances (24.16), (24.17) of normalized sum and difference photo currents
of the fields 1 and 2 for both amplitude and phase quadratures. For instance,
if sum squeezing for the amplitude quadrature is given by ~t (.X) < 1 and
difference squeezing for the phase by ~q CY) < 1 then this implies squeezed-
state entanglement of the fields 1 and 2. The output state has the nature of the
two-mode squeezed state [33], hence its name "squeezed-state entanglement".
Note, that this definition of squeezed-state entanglement refers to the quan-
tum properties of the output fields and can be verified experimentally in a
straightforward measurement. In the literature, "squeezed-state entanglement"
390 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Table 24.2 Various boundaries for non-separable states.

Relevant criterion Physical meaning

non-separability criterion:

always sufficient [32] and for a

certain class of Gaussian states

also necessary [31, 32, 35]

±
3. ~q (X) Vs~ (Y) < 4 a sufficient criterion
A A

for non-separability [36]

demonstration of the EPR

Gedankenexperiment [24]

squeezed-state entanglement

EPR entanglement

is sometimes related to the way how entanglement is produced, namely by

the interference of two squeezed beams. This is difficult to quantify exper-
imentally. We prefer the definition based on the output properties as this is
essentially what counts for further applications.

2. EPR-entanglement. The EPR-entanglement (EPR in Table 24.1) defines

a non-separable [30, 31, 32] two-mode state which obeys the inequality (24.9)
and which satisfies both conditions (24.8) for the conditional variances of
conjugate variables.
The demonstration of the EPR-entanglement implies the experimental ob-
servation that the two conditional variances are both well below the quantum
 Quantum solitons: towards experimental quantum communication 391

limit:

(24.29)

(24.30)

To test experimentally for EPR entanglement one has to measure the sum
and difference photo currents, but a different normalization has to be used
compared to (24.16), (24.17). The sum and difference photo currents for both
amplitude and phase quadratures are normalized to the shot noise variance of
the beam described by the first variable in the argument of Vcond.

Requiring quantumness for both conjugate variables. The feasibility of

EPR entanglement applications in quantum communication and computation
purposes requires:

where the sign "«" emphasizes the ability to reliably differentiate between the
measured correlation level and the quantum/classical boundary V(X, YSN).
The limit of maximally entangled beams would imply

(24.32)

The closer the continuous entanglement is to the maximal entanglement (24.32),

the better is the efficiency and the quality of the information processing based
upon it. However, this limit (24.32) is never achieved for continuous variables
because it would require an infinite degree of squeezing and therefore an infinite
energy content in the system.
The non-separability condition (24.26) is very close to our definition of
squeezed-state entanglement. Note, however, that the Peres-Horodecki crite-
rion is not requesting both variances of conjugate variables to drop below the
quantum limit. Analogously, the demonstration of the EPR paradox also does
not demand each variance of the product to be non-classical. These conditions
set limits on the sum or product of two variances.
In contrast, the definitions of squeezed-state and EPR-entanglement impose
more stringent requirements on the degree of a quantum correlation which
are needed for an experimental implementation of quantum communication
protocols in accordance with (24.31,24.32). Both conjugate variables have to
exhibit a quantum correlation to guarantee a secure quantum key distribution.
A quantum correlation of both conjugate variables is also preferable for the
reconstruction of an unknown state in quantum teleportation.
392 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

oj b)

o 2

Figure 24.1 Different boundaries for the continuous variable entanglement and respective
entanglement regions in terms of conditional and squeezing variances: (a) Demonstration of the
EPR-Gedankenexperiment [24] (dark grey region below the curve) and EPR-entanglement (light
grey square); (a) Non-separable quantum states [32, 31] (dark grey region below the line) and
squeezed-state entanglement (light grey square). The gain is taken to be gx = gv = 9 = 1 to
facilitate the comparison. Note, however, that this gain corresponds to the optimal one only on
the diagonal of the squares where Vcond(5XlIJX2)=Vcond(5i\15Y2) and v;.t(5X)=v;.~(5Y).
Thus all the depicted boundaries are sufficient conditions for a quantum state to be entangled.

The discussed limits are summarized in Table 24.2. The mutual relations
between these differerit boundaries is illustrated in Fig. 24.1. Figure 24.1
assumes l;r= gfft = g, as well as the whole discussion in this paragraph
does. Note that for entangled beams asymmetric in terms of uncertainties,
different optimal gains might be required for different quadratures. Moreover,
to be able to verify the non-separability condition 24.26 in its necessary and
sufficient form, local linear unitary Bogoliubov operations should be applied
to the two-mode quantum state in this case, i.e. local squeezing transformation
together with some rotations [32].
The boundaries (24.27), (24.28), (24.29), (24.30) seem to be the most con-
servative and reliable experimental criteria for the evaluation of the continuous
variable entanglement for quantum communication purposes.

3. QND-entanglement. The QND-entanglement (QND in Table 24.1) de-

fines a non-separable two-mode state [30, 31, 32] which obeys at least one of
the inequalities of the type of (24.9) for a QND variable As of the signal and
an observable Bp of the probe:

V± (A IE ) = V(As:= g .8p ) < 1 . (24.33)

cond s P V(A)
s,SN

Here, the conditions

(24.34)
 Quantum solitons: towards experimental quantum communication 393

have to be satisfied for the relevant signal variable( s) where iI = iI0 + iII is the
total Hamiltonian of the measurement interaction (for explanations and details
about a QND measurement see part 3 of this chapter and references therein). An
example of such entanglement and its application in quantum communication is
the quantum teleportation scheme based on QND-entanglement between field
quadratures of bright light [27]. Further examples are considered thoroughly in
part 3, where the QND-entanglement emerging in soliton collisions in different
schemes is presented.
In analogy to the definitions of EPR versus squeezed-state entanglement,
consider the situation when the condition (24.8) is replaced by (24.27), (24.28):

(24.35)

That would imply a quantum correlation between signal and probe variables,
however, without necessarily the possibility to infer the signal variable As with
a precision below the quantum noise limit by a measurement of the probe Sp.
In this case the generated entanglement might be too weak for obtaining a QND
readout.
The interesting question which still has to be answered is how the EPR-
entanglement and the QND-entanglement are related to each other (see Fig.24.2).
One can introduce two sub-classes of the EPR-entanglement important for un-
derstanding EPR and QND properties. Two-way EPR-entanglement of some
conjugate pairs of variables Xl, Y1 ; X 2, Y2 obeys not only the EPR conditions
stated as:

(24.36)

Squeezed-state
entanglement

one-way EPR-entanglement

Figure 24.2 Relation between the different types of continuous variable entanglement.
394 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Signal Probe

measurement action
"QND
observable"
"Probe
observable"

back action

Figure 24.3 QND interaction as one-way EPR-entanglement. Solid and dashed lines corre-
spond to two possible pairs of correlated variables in Eq. 24.38,24.39.

or

(24.37)

but also the same relations under exchange of indices ("second way"). The
schemes for continuous entanglement generation involving OPO [10, 24] and
interference of squeezed beams [11,22] all are examples of two-way entangle-
ment.
For one-way EPR-entanglement holds (note the order of the indices I):

(24.38)

or

(24.39)

Though QND-entanglement does not mean EPR entanglement per definition,

in many schemes the situation does appear symmetrical for signal and probe
pulse and the EPR condition (24.38, 24.39) is satisfied. EPR-entanglement
generated in QND interactions always belongs to the sub-class one-way EPR
entanglement. A fundamental reason for that is illustrated in diagram of Fig.
24.3. The signal acts on the probe system, modifying the probe observable
in accordance with the QND observable. Each measurement imposes an un-
avoidable back action on the signal system. The peculiar feature of the QND
interaction is that the QND observable evades the back action ("back action
evading measurement of a signal ") channeling the back action of the probe into
a signal variable conjugate to the QND one (Fig. 24.3). For details see part 3 of
this chapter devoted to the QND measurements. Here we briefly discuss the ex-
ample of a QND scheme with colliding optical fibre solitons [14,25,26]. There
the EPR-criterion is satisfied: the phase of the probe pulse is entangled with the
photon number of the signal and its photon number is in tum entangled with
 Quantum solitons: towards experimental quantum communication 395

the signal phase (see Table 1). However, there is a certain asymmetry which
comes from the QND-requirements set on the signal variable. It means, one
can conclude on the signal photon number from the measurement of the probe
phase. It is also possible to infer the information of the probe phase from the
measurement of the signal photon number but it is not possible to do that with a
precision below the vacuum noise level. That would contradict the conditions
(24.34) set on the QND-variable of the signal. Thus this QND entanglement
represents the one-way EPR entanglement between conjugate variables.
For the aim of the QND measurement itself, however, it is not important
whether the EPR condition is satisfied or not. What is of interest is the entan-
glement between the relevant variable of the signal and the observable of the
probe. Therefore, in many QND experiments the behaviour of the conjugate
pair of the variables was not at all addressed. Are there always two entangled
pairs of conjugate variables arising in a QND interaction? Note, that even
if there appears to be only one entangled pair, e.g. Xl ex X2 but YI is not
quantum correlated with Y2 , the situation may change by moving into the other
basis, for example, XI cos () + YI sin () and X2 cos <p + Y2 sin <p. In general, the
quantum state can be non-separable in only one variable (in each subsystem)
and in this sense it can be entangled in one variable. This can be readily seen
from the non-separability criterion (24.26) which can be satisfied if only one
of the conditional variances obeys the condition (24.33) for EPR-like quantum
correlations.
An example of a QND scheme where there seems to be only one entangled
pair of variables, i.e. which is not an EPR-entanglement, is a QND measurement
of an amplitude quadrature of the signal field via an interference on a beam
splitter with an amplitude-squeezed probe field [37]. In the limit of perfect
squeezing and assuming a 50/50 beam splitter, the correlation coefficients for
both quadratures are equal to unity, x~ut ex x;n=out and Ysout ex y.;ut, but the
corresponding conditional variances are given by:
v;conds
(xin=outlxout)
p
-+ 0 , v;cond (y'0utlyout)
p s -+ 1 (24.40)

thus implying the QND-entanglement (24.26, 24.33) but satisfying neither of

the EPR conditions (Tab. 24.1).
Figure 24.2 illustrates, how the different entanglement types defined in this
paragraph relate to each other. Table 24.1 completes this comparison with
examples of various entanglement schemes.
The quality of EPR-entanglement is a critical parameter in quantum com-
munication applications. In the case of continuous variables there are inherent
advantages and disadvantages in this respect, as discussed in the introduc-
tion. The main advantages are the high efficiency of sources and detectors
and the controllable generation process. The main disadvantages are fragility
against losses and finite degree of entanglement. In what follows, we attempt
396 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

to review the bright entanglement schemes which can deliver a high degree of
entanglement, making full use of the advantages discussed in the introduction
and exhibiting experimental feasibility with the prospects for technological
applications.

3. BRIGHT EPR-ENTANGLEMENT WITH

SQUEEZED FIBRE OPTICAL SOLITONS
The generation of entanglement with continuous variables typically uses op-
tical parametric down conversion in a below threshold OPO. This process cre-
ates two vacuum states with quantum correlated amplitude and anti-correlated
phase quadratures or vice versa [10, 22]. However, because of this entanglement
generation being phase sensitive, one has to apply additional more complicated
phase locking techniques to the dark output beams to obtain a long-term stable
source for EPR-entanglement, which is costly in terms of squeezing. A new
promising scheme for the generation of EPR-entangled beams employs the
Kerr-nonlinearity of an optical fibre to produce two bright amplitude squeezed
pulsed light fields. These fields are made to interfere at a beam splitter with
the phase adjusted to obtain two bright output beams of equal power. In this
way EPR-entanglement with anti-correlated amplitude and correlated phase
quadrature is established between these beams.

3.1 ENTANGLEMENT SCHEME

The scheme, on which the experiment is based, utilizes the superposition of
two independently squeezed bright light fields to create quantum correlations
between the two output ports [23]. For this purpose the optical phases of the
incoming fields are chosen such that the initial fields overlap at the outputs with
a phase difference of 90 o. Fig.24.4 illustrates this interference graphically
in a phase space representation. The ellipses and circles depict the quantum
uncertainties of the optical fields (i. e. the half-height contours of Wigner
functions) and the sticks relate to the strong classical amplitudes. The uncer-
tainty areas of the input fields are spanned by orthogonal phasors in amplitude
and phase direction each contributing respectively to the uncertainties at the
outputs. EPR-entanglement is obtained only if quantum correlations for two
conjugate quadratures, like amplitude and phase in the described experiment,
can be proven (see Eq. (24.29, 24.30». The anti-squeezed phase quadratures
of the inputs dominate the uncertainty areas at the outputs and are responsi-
ble for the output correlations. The squeezed amplitude quadratures introduce
the errors in the emerging correlations. The 90 0 phase difference between
the interfering fields in Fig. 24.4 guarantees the most precise mapping of the
uncertainties of the output beams with minimum errors, which yields the best
achievable quantum correlations.
 Quantum solitons: towards experimental quantum communication 397

Y
4
=ex + 8a
A 1\
A

Figure 24.4 Generation of EPR entanglement: interference of two amplitude squeezed beams
where the beam splitter is taken to be symmetric (t = 1/V2. r = i/V2).

In Fig. 24.4 this mapping is illustrated with the help of the input phasors in
phase direction (large arrows) spanning the uncertainty regions at the outputs.
The classical fields add vectorially and form a new amplitude direction 45°
to the input fields. The projections of the phase uncertainties of the incident
fields onto the new amplitude direction of the output beams show strict anti-
correlations, those onto the new phase direction strict correlations. This would
correspond to perfect entanglement. However, the finite amplitude uncertainties
of the squeezed input beams reduce the correlations as indicated by the small
light circles at the outputs. According to this the sum variance of the two
entangled output beams, reflecting the amplitude uncertainties, is decreasing
linearly with the squeezing at the inputs. Respectively, the difference variance
for the phase uncertainties is growing linearly. The phase diagrams in Fig 24.4
describe the interference of two single mode fields. The reasoning leading to
entanglement also holds in the more complex case of squeezed multi-mode
fields such as amplitude squeezed pulses [23].

3.2 "TWO-IN-ONE" SQUEEZER FOR

SQUEEZED-STATE-ENTANGLEMENT
In the experiment the first task is to produce two independently squeezed
beams, which are also optically coherent. An asymmetric fibre Sagnac inter-
ferometer [38, 39, 40] can be used to provide directly detectable amplitude
398 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

squeezing (photon-number squeezing) of a single beam. This squeezing is

highly robust and can be long term stable. The Sagnac interferometer consists
of a highly asymmetric beam splitter with a splitting ratio around 9011 0 and a
polarization maintaining fibre. Polarization maintaining fibres are birefringent
with two distinct orthogonal main axes, along which light travels at different
speeds. The cross-talk between the corresponding polarizations is negligible
in such fibres. Therefore it is possible to build up a "two-in-one squeezer"
out of one Sagnac interferometer by adjusting the polarization of the incoming
beam to 45° with respect to the slow and fast fibre axes [41]. The energy is
then equally distributed between the two corresponding polarizations and two
squeezed output beams of orthogonal polarization can be obtained. They can
be separated by a polarizing beam splitter.
Fig 24.5 shows the associated experimental setup. A mode-locked CrH: YAG

VA
c:::J---88-b . . ., :. . . . ,.: . .
').J2 ••••::.

(.:~.:~:. :. ::~o
IL---f

Cr:YAG
laser

.....~....
squeezed
beam #2

Figure 24.5 Schematic of the experimental setup of a two-in-one-squeezer for the generation
of two squeezed beams. VA: variable attenuater, ),,/2: half-wave plate, G: gradient index lens,
PBS: polarizing beam splitter, 90110: beam splitter with 90% reflectivity. Insert shows the
polarization direction of the beam at the input of the fibre with respect to the main axes.

laser serves as a source for linearly polarized hyperbolic-secant (sech) shaped

optical pulses. After a variable attenuation and the proper orientation of their
polarizations by a half wave plate the pulses are launched into the Sagnac in-
terferometer. In order to run the fibre Sagnac interferometer on both axes the
splitting ratio has to be approximately the same for all polarizations. There-
fore the angle of incidence at the asymmetric beam splitter is chosen to be
sufficiently close to 0° to ensure that the mirror reflectivities are the same for
both polarization. Thus a strong and a weak pulse counter-propagate through
the Sagnac loop for each polarization. Due to the different propagation speeds
along the fibre axes each of the two co-propagating pulses of the incoming
beam will separate quickly after only a few centimeters inside the fibre. This
 Quantum solitons: towards experimental quantum communication 399

ensures the independence of the squeezing of the two output beams, which was
checked experimentally.
The Kerr nonlinearity induces an intensity dependent phase shift during the
propagation of the pulses along the fibre. This in tum provides a relative phase
shift between the strong and weak counter-propagating pulses having the same
polarization and it influences the quantum characteristics of the strong pulse. In
a single mode model the circular shaped phase space uncertainty of a coherent
light field is formed into an ellipse by the Kerr effectKerr,effect during the
propagation along the fibre [42, 43]. However, in order to be able to detect
the amplitude-squeezing in direct detection, the uncertainty ellipse has to be
realigned. The interference of the strong pulse with the weak one at the output
of a fibre Sagnac interferometer implements this re-orientation due to the Kerr-
induced relative phase shift between the strong and weak counter-propagating
pulses [42]. Still, the intensity dependent phase of the interference has to be
matched and hence the directly detectable amplitude squeezing depends on the
pulse energy. To detect the squeezing, the light is equally distributed to a pair of
balanced photo diodes of high efficiency and the generated photo currents are
added and subtracted. The noise variances of the sum and difference signals
give the noise of the output beam and the corresponding shot noise level.
Fig. 24.6 shows the results for the output ports of the described "two-in-one"
squeezer. The noise powers in Fig. 24.6 are recorded as a function of the input
energy and the grey areas indicate energy ranges where squeezing occurs.

~.

I'
~'
"2 .
I
1/. · \
I/ ~
":. \
E E
co ~ . III

... if
~ ~
....
Q)
·70 Q) -70 _ 1_-.-,
~ . : ~ ,..._--r
\j'\j
~.

, i

1\ ;1
c.. Co
, ..."
.. L - - - " r Q)

V\J
Q)
III III
'0
c: -75 'g ·75
,
;' , I
80
40 60 eo 100 120 140 160 ~~~0~4~0~6~0~e~0~1~00~1~~~1~~~1~6~O~
20
input energy [pJ] input energy [pJ]

Figure 24.6 Noise powers of the outputs of a "two-in-one" squeezer as a function of the input
energy; doted line: noise of the output beams; solid line: corresponding shot noise levels as
derived from the difference signal and verified by attenuation measurements. Inserts show the
normalized variances of the respective traces.
400 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

3.3 GENERATION OF EPR CORRELATIONS BY

LINEAR INTERFERENCE
In order to obtain the interference of the two squeezed beams emerging
from the Sagnac interferometer one has to have temporal, spatial, polarization
and spectral overlaps of the outcoming pulses. Therefore the traveling time
difference of the pulses inside the fibre is compensated for by an appropriate
delay line. A half wave plate corrects the polarization and the phase is controlled
by using the DC output signals of the detectors. In the following again only
the single mode picture of Fig. 24.4 is used to model the conditions of the
superposition of the independently squeezed light fields at the beam splitter.
However, because of the beam splitter being a linear element, it is possible to
extrapolate all conclusions directly from the single-mode situation to the multi-
mode one by the superposition principle. In this consideration the Raman effect
is ignored, which introduces a spectral shift of the sub-picosecond pulses and
decreases the spectral overlap of the pulses during the interference.
To evaluate the EPR-entanglement and its dependence on squeezing, the
noise variance of the input and output beams are calculated using the ansatz
of a strong coherent amplitude and the linearized treatment of the uncertainies.
a
The annihilation operator of the field is given by = a + 80, with a being
the coherent amplitude of a classical mean field and the operator 80, describing
the quantum uncertainty. To model a superposition on the beam splitter, the
classical parts of the input fields are both taken to be real with no relative
phase difference. The beam splitter in Fig. 24.4 introduces a phase shift of
90° for each reflection, whereas the transmitted beams do not experience any
phase shift. This gives the output fields a = r A + tB and b = tA + r B
with reflection r = i ~ and transmission t = ~ coefficients corresponding
t<,? the situation of Fig. 24.4. If the amplitude quadrature is denoted by
Xa = at(t) + a(t) and the phase quadrature by Ya = i(at(t) - a(t)) the
spectral variances of the beams V(Xa) = V(Xa(w)) = (18a(w) + 8O,t(w)j2)
and V(Ya) = V(Ya(w)) = (18a(w) - 8a t (w)j2) can be calculated via Fourier
transformation. In the experiment the amplitude anti-correlations of the output
beams are demonstrated by directly detecting the sum and difference photo
currents at the two output ports. Fig 24.7 shows the expected noise variances of
the output beams (normalized to their respective shot noise level) as a function
of the squeezing of the input beams calculated for minimum uncertainty input
states.
The experimental data for the amplitude anti-correlation are recorded similar
to the squeezing results, where the noise powers of the different traces are
plotted versus the input power of the Sagnac interferometer [11]. In Fig. 24.8,
traces 1 and 2 indicate the noise powers of the sum and the difference of the
two output beams. Traces 3 and 4 show the noise of the individual beams, and
 Quantum solitons: towards experimental quantum communication 401

iii' 10
3:!..
CI)
o
~ 5
.s:;
<II
>
Q>
Ul
'0
c:
"C
.~ ·S
"iij
E
o .io,_ _ _-=-__ ~ __ __;_--_..,~-__""':'
c: o 6

normalized squeezing [dB]

Figure 24.7 Calculated noise variances of the output beams at the beam splitter versus the
squeezing of the input beams for minimum uncertainty states.

~I

I:
10
g.•
0
c .•
·50
E' .55
i
~ .60 i 1
{ ............. .1. .......... ....
~ ·65 i;."/
0.·70 SN
CI>
<I)
'0 ·75
1"' ................,
c
·80
20 40 60 80 100 120 140 160
input energy [pJ]

Figure 24.8 Experimental data: noise measurements of the output beams at the beam splitter;
insert shows normalized variance of the sum of the output beams demonstrating amplitude
quantum correlation

the trace labeled SN gives the shot noise level, which corresponds to the optical
power of the combined output beams. The grey areas mark energy regions,
where the fields entering the input ports of the beam splitter are amplitude-
squeezed (see Figs. 24.7 and 24.8).
The calculations (see Fig. 24.7) predict no correlations for coherent states
with shot noise limited uncertainty for amplitude and phase. In the experimental
data of Fig. 24.6 and Fig. 24.8 these states should be found at the boundaries of
402 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

the grey areas, which correspond to 0 dB input squeezing. However, in Fig. 24.8
the noise levels of the sum, difference and single beam signals (traces 1 - 4) at
odB input squeezing differ significantly from each other indicating still existing
classical anti-correlation of the amplitude quadratures. This phenomenon can
be understood by additional noise in the phase quadrature of the input beams
due to the interaction of the photon pulse with thermal phonons in the fibre,
which leads to classical amplitude anti-correlation in addition to the quantum
effects. This classical anti-correlation can result in high correlation coefficients,
but the conditional variance as the figure of merit for the quantum correlation
will never drop below unity. The enlarged phase noise is present for all energies
and explains the high noise powers of the individual output beams as well as
the noise trace for the difference photo current. In contrast, calculations for
the sum photo current show, that the noise variance of the sum is not affected
by the additional phase noise. The noise power of the sum photo currents was
observed to drop up to 4.0 ± 0.2 dB below the shot noise level, the quantum
limit for the anti-correlation. As expected, this quantum anti-correlation is
limited by the squeezing of the input beams, as seen by the comparison of Fig.
24.6 and Fig. 24.7.
Simultaneous squeezing of the input beams is also recorded for higher ener-
gies. However, in that region the sum variance of the amplitudes lies above the
quantum limit. There are two possible explanations. Firstly, the interference
was optimized for lower energies and, secondly, the detection of the quantum
amplitude anti-correlation is very sensitive to the balance of the photo-detector
pair. In the ideal situation traces 3 and 4 for the single beams in Fig. 24.8
should coincide for all energies, but they separate for the high energy range.
This can be attributed to an imbalance of the photo detectors which where
calibrated for the lower energy band around 100 pJ.
For the complete determination of the EPR-entanglement in addition the
quantum phase correlation has to be investigated. In order to directly measure
the phase quadrature variances, strong optical local oscillators with matched
spectra and wavefronts are required. To circumvent the experimental difficulties
involved when applying this technique to bright light pulses an interferometric
measurement scheme for the characterization of the state was invented.
In Fig. 24.9 the source of the EPR-entanglement is treated as a black box
with two output beams. These beams are made to interfere at yet another SO/50
beam splitter with the relative phase of the beams being scanned (Fig. 24.10).
To prove the non-separability of the state it is now sufficient to record the
amplitude noise variance in addition to the shot noise of one of the resulting
output beams c and d. Calculations show, that for a suitable interference
phase the reduced amplitude noise of a single output beam is equivalent to
the measurement of i: (V(Xa + Xb) + VCYa - Yb)). The shot noise levels
 Quantum solitons: towards experimental quantum communication 403

balanced
Scan of detector
Interference
Phase e

Figure 24.9 Schematic for indirect measurement for the phase quadrature correlation

V(X1,SN + X2 ,SN) and V(Y1,SN + Y2 ,SN) are both equivalent to twice the shot
noise ~ (Xc/ d,S N ) of cor d. Thus the normalized amplitude noise of the beams
cand d is given by

That means the described setup permits the direct detection of the Peres-
Horodecki non-separability criterion for continuous variables (see Eq.24.26):
+
Vsq (X) + Vs~ (Y) < 2.
A A

In the experiment a value of vst(X) + Vsq (Y) = 0.80 ± 0.03 was observed,
demonstrating the high non-separability of the state. Fig.24.10 represents the
corresponding experimental data.
a
Provided the variance of the sum of the amplitude of beams and bis already
known, the quantum phase correlation can be inferred from the measured
amplitude noise of c or d. A quantum phase correlation of up to 4.0 ± 0.4 dB
was verified from the data of Fig. 24.10.
Note, that all indicated experimental values describe the noise statistics of
the light without correction for linear losses. The photo-detectors showed a
detection efficiency of around 92 ± 5% and the noise powers were recorded
for a detection frequency of 10 MHz with a resolution bandwidth of 300 kHz.
The measured data are corrected for the electronic noise of the photo-detectors
and the spectrum analysers. For quantum information applications the dark
noise can limit the availability of the entanglement. If one corrects only the
shot noise level for electronic noise to evaluate the quantum limit and for an
uncorrected signal noise, the value of 1.12 ± 0.06 < 2 was observed in the
experiment for the Peres-Horodecki criterion.
404 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

...... - + channel
E
m
......
"0
...CD -6
~
0
a..
CD
III
'0
- shot noise level =
Z quantum limit

Interference Phase e [0]

Figure 24.10 Experimental data: Amplitude noise variances for the indirect detection of the
phase correlation. For these data the initial squeezed light beams forming the entangled pair
interfere with a phase difference of ¢ = 30°. For ¢ = 90° phase difference the noise variance
detected in "+" channel will be below shot noise for all phase angles (J (see Fig. 24.9).

3.4 DEMONSTRATION OF THE EPR-PARADOX AND

QUANTUM COMMUNICATION
After determining the quantum correlations of the two conjugate variables,
amplitude and phase, in addition to the Peres-Horodecki criterion the figures
of merit for the demonstration of EPR-entanglement can be calculated (see
Eqn. (24.8, 24.9 and 24.29, 24.30». For the sum of the amplitude quadra-
tures squeezing of up to 4.0 ± 0.2 dB was observed. For the squeezed-state
entanglement (24.27, 24.16) this gives V+(.X) = OAO±0.02 < 1. This corre-
sponds to a quantum anti-correlation with a non-classical conditional variance
V (JXdJXb)
~ ~
ofVcond(XaIXb) = vex ) = 0.80±0.03 < 1. Due to the enlarged phase
a,SN

noise of the input fields in this case the optimal gain for (24.8) was lIt = 1.
Similarly, for the phase quadrature difference squeezing (24.28, 24.17) of also
up to 4.0 ± 004 dB was associated with V- CY) = 0040 ± 0.04 < 1. This
indicates a quantum correlation with the same non-classical conditional vari-
ance of Vcond(YalYb) = V~~-J~b)
a,SN
= 0.80 ± 0.07 < 1 (the optimal gain is
again g?Jt = 1) . Thus the EPR-paradox for continuous variables (24.9) was
demonstrated:

(24.41)
 Quantum solitons: towards experimental quantum communication 405

with a quantum limit of

(24.42)

Equation (24.41) gives an appropriate measure of the EPR-entanglement

from the point view of both, fundamental research and experimental applica-
tions. In recent years, the study of quantum correlations in quantum optics
has experienced a resurgence of interest focusing nowadays more and more
on applications in optical transmission systems and signal processing. Soli-
ton squeezing discussed above is being tested for applicability in fibre-optical
communication systems [43]. The upward trend towards novel optical tech-
nologies has led to new paradigms also for the research in the field of quan-
tum entanglement. The utilization of EPR correlations is evolving from the
purely fundamental tests of the validity of quantum mechanics to applica-
tions in quantum information processing and communication. These set also
other requirements for the correlation measures with a particular emphasis on
quantifying the efficiency of the resulting applications, e.g. in optical and
quantum communication. Such a "pragmatic" approach leads to defining the
quality of correlations by means of signal-to-noise ratios, efficiency of noise
reduction in optical signal detection, fidelity of state reconstruction in quantum
teleportation, or efficiency and security of quantum key distribution systems.
The measures (24.27-24.35) which we have discussed here lie also within this
trend. This constitutes a new link between the fundamental and applied aspects
of continuous variable quantum correlation.

4. QND EXPERIMENTS WITH OPTICAL SOLITONS

'Quantum nondemolition' (QND) devices are among the basic building
blocks of quantum information processing, because of their wide-ranging ap-
plications. QND measurements were proposed for entanglement purification
for continuous variable entanglement [44], for an eavesdropping attack in the
context of quantum cryptography [45, 46], for a quantum optical bus sys-
tem [47] and for monitoring in quantum computing [48]. Many applications
for a back-action evading (BAE) measurement can be derived because it is a
paradigm for a quantum mechanical measurement.
The concept of a QND measurement is briefly recalled in section 4.1. In
section 4.2 it is described how the interaction of solitons leads to QND measure-
ments. Section 4.3 explains the QND detection scheme with collision induced
phase shifts together with the experiments. Finally, in section 4.4 the novel
frequency shift QND detection scheme and recent experiments are reviewed.
406 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

4.1 CONCEPT OF A QND MEASUREMENT IN

QUANTUM OPTICS
Originally the concept of QND detection was introduced by Braginsky and
Vorontsov [49] for the quantum mechanical measurement of very small deflec-
tions of a massive mechanical resonator for gravitational wave detection (see
also [29]). Later this idea was extended from mechanical to optical harmonic
oscillators in quantum optics to detect a quadrature component Xs of a light
field [50, 51, 52].
A QND measurement involves two quantum mechanical systems described
by the conjugate pairs Xs, -Vs for the signal system Sand Xp , ~ for the probe
system P, respectively. The observable to be measured, Xs , is referred to as
'signal'. The state of the system S + P prior to the measurement is given by
I~S) ® I~P), where I~S) and I~P) are the respective state vectors of Sand
P. In a first step the signal system S is coupled to the probe system P to
establish a correlation between Sand P, such that Xs can be inferred from a
direct measurement on P. Through this correlation S and P are entangled and
the system S + P is in a non-separable state. Finally, the state P is directly
and destructively measured to infer Xs , and thus to reduce the signal state to
an eigenstate of Xs.
The coupling between S and P is described by the total Hamiltonian H =
Ho + HI composed of the interaction part HI and the free evolution part Ho
of the individual systems S and P. The measurement H and the QND variable
Xs must satisfy certain conditions to allow for a QND measurement [51,53].
A QND variable Xs is required to be a constant of motion ([Ho, Xs] = 0 and
[HI, Xs] = 0)1 for the unperturbed system described by Ho as well as the
i!lteractio~ HI,: The QND meaAsurement is acc?mplished throu~h a coupling to
Yp ,i.e. [HI, Yp] =1= 0, where HI depends on Xs. In this way Yp is affected by
Xs.
4.2 QND INTERACTIONS OF OPTICAL SOLITONS
Solitons are exact, analytial, and robust solutions of the nonlinear Schrodinger
equation, describing pulse propagation in a fibre with group-velocity dispersion
and Kerr-effect nonlinearity [55]. In the classical model, solitons are stationary
solutions. These particle-like wavepackets recover their shape, velocity and
energy after a collision.
In quantum field theory, coherent solitons are non-stationary solutions. The
soliton mode is described by the four soliton parameters photon number n,
phase <p, momentum p and position x [56]. As a QND observable the photon
number is most suitable, since it is practically preserved in the evolution of a
soliton in a fibre 2 [43, 57, 58]. This is verified using the Hamilton operator in
 Quantum solitons: towards experimental quantum communication 407

.9 ~ 0.9 ~
.3
#
"
Q) 0.3
~
"
Q)

-0.3~ -0.3 ~
.~ .~
-3.2 -1 .6
-0.9 Q
F 0 16 -0.9 Q
0.4
reqUency ITHz 3.2

Figure 24.11 Time and frequency domain interaction of two fundamental solitons. The in-
tensity of the field is displayed in a reference frame moving with the center of mass of the
solitons.

the fibre [51, 59]:

(24.43)

it can immediately be seen that the signal photon number is preserved, i.e.
[Hs, ns] = O. If the photon numbers are as large as 108 , a linearized description
of the field fluctuations can be used [60]. In terms of quadratures the photon
number or amplitude quadrature can then be written as Xs = ns/(2) (ns))
and the conjugate phase quadrature as Ys = J(ns)~s.3
The Kerr effect couples two copropagating and interacting modes in a fibre
by an intensity dependent phase modulation (XPM). In soliton QND schemes
XPM coupling is achieved by a soliton interaction, i.e. the collision of a signal
with a probe soliton. This is depicted in Fig. 24.11 in the time and in the
frequency domain. The colliding pair is described by an unbound solution
of the nonlinear SchrOdinger equation. Maximum pulse overlap occurs in the
center of the collision at z = O. Before the pulses begin to overlap they
propagate with unaltered shape as expected from single solitons. The collision
transiently disturbs the envelopes of the pulses, which retrieve their pulse
envelope, energy, and momentum after the interaction. Therefore only two of
the four soliton parameters, the photon number and momentum, are conserved.
The photon number preservation during the interaction can be verified with
the Hamiltonian of Eq. 24.43, [HI, ns ] = O. From the pulse trajectories an
acceleration of the pulses towards each other is visible, causing a subsisting
shift in phase and position, which depends on the intensity of the other soliton.
The probe soliton phase shift is a possible readout for the signal amplitude,
because HI is a function of ns and [HI, Yp] # 0 ([yp, np] # 0) [25,63] . For
a small phase shift, Yp represents the probe phase shift, which can be read
out by an interference with an additional phase reference soliton. This type of
QND interaction is explained in section 4.3. Transient changes in the solitons
408 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

during the collision such as the change in relative velocity of the pulses can
be used for a QND readout as well: Because of the chromatic dispersion of
the fibre a relative group-velocity increase corresponds to a transient increase
in the spectral separation of the pulses (Fig.24.11). Note, that in the center
of the collision the spectral density at line center approaches zero. This holds
for any relative velocity and phase of the solitons provided they have equal
amplitudes. Detecting the spectral shift of the probe pulse serves as a QND
readout of the amplitude of the signal soliton, as explained in section 4.4. To
date, these two are the only schemes for a QND measurement with solitons in
fibres. XPM provides the cross talk mechanism to write the signal intensity
information into the probe system without degrading the signal-to-noise ratio
of the signal. Thus XPM generates the desired entanglement of the signal and
the probe system. The back-action noise introduced by the measurement is fed
into the signal phase and position only (Fig. 24.3).
QND measurements based on the Kerr effect were among the first proposals
for QND measurements in optics [50,51,52]. The first QND experiment was
realized using the Kerr effect in fibres [64]. Fibre-optical experiments aiming
at back-action evading measurements of a quantum-limited signal intensity
are listed in Table 24.3. The experiments used shot-noise limited intensity

Table 24.3 Experiments aiming at fibre QND detection of a shot noise limited signal beam.
Pump light denotes the signaUprobe power or pulse energy and center wavelength. The corre-
lation coefficient C is the fraction of probe readout photocurrent rms noise due to signal shot
noise. Leff is the effective interaction length, different from fibre length due to resonator finesse
[65] or pulse walk-off [14,26,66].

Pump light C ¢XPM Fibre Leff Experimental Ref.

(rad) length configuration
phase modulated CW, 0.37 no 114m 114m linear travel- [64];
130 mW@676nm/ data 2K ling wave [67]
60 mW@647nm
CW, 15 mW@676nm/ 0.26 a no 13 m 100m resonant [65]
16 mW@647nm data ring
solitons, 2.6 ps, 15 pJ 0.39 1.22 400m 110m linear travel- [26]
@1460.7 nm/3.6 ps, ling wave
6 pJ@1455 nm
solitons, 200 fs, 40 pJ 0.27 0.71 6.3m 0.9m linear travel- [14]
@1495 nm/200 fs, ling wave
40 pJ@151O nm

a Results for temporarily optimum polarization conditions.

 Quantum solitons: towards experimental quantum communication 409

fluctuations as a quantum signal (quantum-modulation or quantum-information

transfer). A dual successive back-action evading measurement based on the
experiment of Ref.[26] was reported as well [66]. The QND experiments of
Refs. [26,64,65,67] were limited by background phase noise contaminating
the probe phase readout for the signal photon number. The latest experiment
[14] was limited by the internal noise structure of the probe solitons (see
section 4.4). The general state of the art in QND detection is a conditional
noise reduction of the shot-noise limited signal of 3.5 dB(55%) with cold
atoms in a trap [68]. A similar result was obtained with an OPA system [69].

4.3 PHASE·SHIFT QND DETECTION

This section discusses the QND detection using the probe phase shift and the
optimum field quadrature for a readout. The only experiment so far using phase
detection is reviewed. The section summarizes the corresponding discussion
given by Sizmann and Leuchs [43]. The phase quadrature of the probe soliton is
changed by the collision with the signal soliton and provides information on the
signal amplitude quadrature. Investigating small fluctuations in the quadrature
field, the input-output relations are:

(24.44)

Here 'Y is the QND gain and a the self phase modulation (SPM) coefficient.
Note that both depend on the interaction lengths. The probe readout quadrature
oY.;ut contains the signal oX;n. The inevitable back-action noise due to the
measurement is channeled into the quadrature Ys, conjugate to Xs. However,
one typically finds that a > 'Y and the probe observable is contaminated by
SPMnoise.
The corresponding fibre QND scheme is shown in Fig. 24.12. In order
to evaluate the performance of a certain QND device, transfer coefficients
are practical quantities for the determination of the signal degradation and
information transfer to the probe observable [28, 70, 71]. They are measureable
quantities even in a single stage BAE measurement. In the case of uncorrelated
input states, i.e. (x;nYsin)sym = 0, the transfer coefficients can be derived
directly from the correlation coefficients [28]:

(24.45)
410 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

0.9 ~

"
c:8 c:If
~
-0.3.fY -0.3 .fY
.f!} .f!}
-3.2 -1 .6
-0.9 Q -0.9 Q
0.4 F 0 16
reqUency ITHz 3.2

Figure 24.12 QND scheme using XPM induced phase changes in optical fibres. The signal-
intensity dependent refractive index changes the phase of a probe pulse and vice versa. A QND
measurement of the signal amplitude can be performed via a probe phase detection, leaving the
signal amplitude unperturbed (after Sizmann and Leuchs [43]).

Here sin, sout and pout refer to the observables in a general QND measure-
ment. The inequality
(24.46)
characterizes a quantum measurement, in which the signal information is noise-
lesslyamplified. Using Eq. 24.44 we can calculate the transfer coefficients for
a coherent signal and a coherent probe input (V (x~n) = V (x~n) = V cv;n) =
1):

Without self-phase modulation acting on the probe (0" = 0) a nearby perfect

QND measurement can be performed by increasing the probe power ("( -+ 00).
In this case the 'macroscopic' phase change ofthe probe due to the signal dom-
inates over the small phase uncertainty of the probe itself. In practice, however,
0" 1= 0 and self-phase modulation (SPM) phase noise increases proportional to
probe power. Photon-number fluctuations of the coherent probe input feed into
phase fluctuations via SPM. This way the probe transfer coefficient is degraded,
since 0" > "(.
To eliminate the SPM noise in Eq. 24.47 for a noiseless QND experiment the
probe observable can be changed to a linear combination of phase and amplitude
 Quantum solitons: towards experimental quantum communication 411

quadratures [72, 73]. The probe output fluctuations are then measured in the
quadrature Bp:
8B;ut('Ij;) = 8X;ut cos('Ij;) + 8y;ut sin('Ij;)
and for optimized angle 'lj;o [43]:

8B;ut('Ij;o) = (1 + a 2fl/2 (8y;n + ,8X!n). (24.48)

The probe observable, 8B~ut then contains only the probe input phase quadra-
ture noise and the QND copy of the signal. In analogy to Eq.24.47 the sum of
the transfer coefficients is:
,2
T s +Tp = 1 + - 12' (24.49)
+,
This semiclassical analysis shows that an SPM-noise-free QND measurement
is possible when using a combined amplitude and phase detection.

Experiment with colliding solitons using phase shift detection. In what

follows, the experiment with colliding solitons is reviewed which uses phase
detection of the probe pulse [26], see also Table 24.3.
In this QND measurement of the signal amplitude one detects the phase
change of the probe soliton using a reference soliton which does not interact
with the signal soliton, as is shown in Fig. 24.13. A detailed study of this
soliton-collision interferometer was performed [74] and with this experiment,
Friberg, Machida and Yamamoto [26] realized the first QND measurement with
optical solitons in fibres. Signal (TFWHM = 2.6 ps), probe and reference (each
3.6 ps) pulses were spectrally filtered out from a single pulse of higher energy,
which was SPM-broadened to up to 10 nm spectral width in an additional
fibre [26]. The timing of the pulses was adjusted such that a complete collision
between signal and probe occurred in the 400 m long fibre leaving the reference
pulse unaffected. GAWBS 4 noise was eliminated by a differential technique.
The probe soliton followed the reference soliton with a delay of only 30 ps so
that both pulses experience the same GAWBS phase fluctuations [76]. On the
detection side the signal was separated from probe and reference pulses by a
grating and the QND signal was extracted in a Mach-Zehnder interferometer
(Fig. 24.13).
The experiment showed a 0.25 dB noise reduction below the combined noise
level due to the correlation between signal amplitude and probe phase, where
the signal photon number was shot-noise limited. The total transfer Ts + Tp
is larger than unity (quantum domain) or just below unity (classical domain)
depending on whether or not the losses for the signal are included [26, 43].
An advantage of soliton QND experiments is the straight forward extension
to repeated BAE detection. In a double-collision experiment [26, 66] it took
 0
412 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

Negative Dispersion

+
Probe Reference Optical Fiber,

"
400m

-----J\~A~/\~O
From Two-Color "
O~
Soliton Source Signal

Signal Soliton Shot-Noise

Measurement \ -E: ::·1 r{}-·..
{ - , ,....., L .. _.. _~

<j -;- . _. _. _..

~
• I
i
1:
_...... _.... _
OIIIIIIIICE
+ .......
..
r"'!i :r"~! r"'1
....................................
r"~
• i : ! ............ .
i
Ta L.; t_J t.; l.. J
RF Spectrum Analyzer Delay Line, 30 m

Figure 24.13 Outline of a soliton QND measurement. After the probe-signal collision in the
fibre, the phase difference between probe and reference is a readout of the signal soliton photon
number. For phase detection, probe and reference interfere with a 11"/2 relative phase delay in
addition to the group delay (after Friberg et al. [26]).

only a different pulse sequence and a longer fibre to realize two collisions. In
the experiment it was shown that probe and reference, when both colliding
with the signal, experience the same XPM to within 13 dB. However, repeated
BAE detection requires individual readouts of the two probe pulses of the
two collisions, which can be implemented in this scheme with two additional
solitons which do not collide with the signal soliton.
The soliton QND experiments discussed so far were limited by detection
efficiency and by SPM noise of the probe system. The two limitations were
inherent in the Mach-Zehnder and the pulse delay detection scheme. The SPM
noise limitation leads to a predicted conditional variance which is at best 0.8,
corresponding to Ts + Tp = 1.2 in this case [72]. To get closer to an ideal
soliton QND experiment the probe would have to be detected with a linear
combination of amplitude and phase quadratures [72, 77, 78]. A different
detection scheme [73] could reduce both classical GAWBS noise and quantum
SPM noise simultaneously.
 Quantum solitons: towards experimental quantum communication 413

4.4 FREQUENCY-SHIFT QND DETECTION

This section is about the fibre-optical QND scheme using frequency shift
detection by spectral filtering. In a first paragraph the historical evolution and
the mechanism of the novel approach is considered. Then the experimental
realization is reviewed.
Despite the stationary properties of solitons used in the phase shift detecting
QND scheme, a coherent quantum soliton propagating in a fibre is nonstation-
ary, i.e. the quantum fluctuations of the field do change. The coherent soliton
can be described as a superposition of Fock state solitons, each propagating
stationary and experiencing different nonlinear phase shifts (SPM). This led
on the one hand to soliton quadrature squeezing experiments [79, 80] and on
the other hand to the squeezed state generation by spectral filtering [81]. In
further investigations of the latter with shorter fundamental solitons, the mul-
tim ode quantum correlation structure of a soliton was discovered which was
found to be responsible for the observed squeezing [82,83]. Largely enhanced
nonclassical correlations compared to the fundamental soliton were predicted
in the case of bound higher order solitons, because of the interaction of its
different soliton components [84]. These spectral correlations between soliton
components are used in the novel QND detection scheme.
The soliton collision (see Fig. 24.11) corresponds to an unbound second
order soliton where the soliton components are the colliding signal and probe
pulses. The spectral shift of the two pulses shown in Fig.24.11 is due to XPM.
If the pulse spectra do not overlap, the shift mainly depends on the respective
amplitudes of the pulses and the relative group velocity V = Vs - vp. Thus,
the frequency shift of one pulse is a measure of the photon number of the other
pulse: b..fp <X ns/IVI. Experimentally, the frequency shift is measured by
detecting the part of the probe pulse, which is transmitted through a spectral
low- or high-pass filter. The QND conditions (see section 4.1) for the signal
photon number conservation are valid for this scheme as well. The interaction
depends again on the signal photon number ns. The probe observable is the
probe momentumpp , which is proportional to vp. The change in probe velocity
in the interaction is related to [H[,pp] =I O.
Compared to the phase shift QND scheme this scheme provides the following
advantages: It creates directly detectable photon number entanglement, it is
not affected by thermal or quantum phase diffusion noise (SPM), and it has a
higher energy efficiency because no third pulse is required as a phase reference.
On the other hand, a drawback of this scheme is a limited QND performance
found in the simulation of the experiments [14].

Frequency shift QND experiment. The experimental realization with pulses

in the sub-ps regime is shown in Fig. 24.14. The laser source is a modelocked
414 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

-eco
.~ ~.,.:-+--:.>4.J.L...,..~""-l
c
OJ
"0

1i!
uOJ
C.
en

14SO 1550

Figure 24.14 The QND experiment using frequency shift detection (left). Experimental spectra
(right) (a) of two colliding pulses in the center of the collision and (b) of the pulses without a
collision.

solid-state laser emitting 150-fs pulses with 163 MHz repetition rate at the
wavelength of 1.5 p,m. Each laser pulse is separated spatially by a dichroic
mirror into two shot noise limited and uncorrelated signal and probe pulses,
each 14 nm in spectral width and spectrally separated by 15 nm. Because of the
short length of the pulses only a short interaction length is required to produce
spectral correlations [85]. The timing of the pulses is adjusted such that they
fully overlap at the end of the fibre. This locates the center of the collision.
Output spectra of colliding pulses (a) as well as of non-colliding pulses (b) are
displayed in Fig. 24.14 (right). The spectral shift of the respective signal and
probe spectra is clearly visible and amounts to approximately half the spectral
width of the single pulses (7 nm). In addition to this a modulation appears in
the spectrum due to interference with a dispersive background wave created by
the slight chirp of the input pulses.
After the pulses leave the fibre they are separated using a grating. In addition,
the probe pulse is spectrally filtered by a subsequent spatial slit. Note that the
photon number of the probe pulse is preserved in the interaction. The spectral
filtering then induces losses to the probe pulse and the transmitted photon
number is taken as the probe output. Correlations between the signal and
the probe outputs are detected as follows: The photocurrent of the probe is
attenuated by an amount a (variable gain 9 in Fig. 24.14) and added to the
signal photocurrent. The input-output relations can be written as

onpout ,,;, onout,un

p
+ Bonin 5 , (24.50)

where on~ut,un is the uncorrelated part of the photon-number fluctuations of

the probe. Assuming on~ut = on~n, the added mode photocurrent noise is
 Quantum solitons: towards experimental quantum communication 415

proportional to:

V(n~ut + an~ut) = V(n!n)(l + 20a + 02a2) + a2V(n~ut,un), (24.51)

where V(A) denotes a variance (A2) - (A)2 of an observable A. An anticorre-

lation (0 < 0) between signal and filtered probe photon number is anticipated
for frequency low-pass filtering of the probe pulse. In that case the combined
currents exhibit a noise reduction as a function of a for small a. Results are
shown in Fig. 24.15. The variances of the photocurrents are measured simulta-
neously with two RF spectrum analyzers at 20 MHz with a 300 kHz-resolution
bandwidth. A clear noise reduction below the signal noise is found in the

-
(l)
"0 6

'-
Q)
4
~
0
0.
2
Q)
en
·0
c 0

Q)
'-
-2
0.0 0.3
relative attenuation a
Figure 24.15 Reduction of the signal noise power below the signal shot noise level. A
conditional variance of 0.73 results.

added current noise levels, as expected for anticorrelations. Furthermore, the

noise reduces by 1.37 dB even below the shot noise level of the signal. This
corresponds to a conditional variance smaller than unity and indicates oper-
ation in the quantum domain [28]. The noise levels of the plus and minus
channels are fitted to a parabola according to Eq. 24.51 (solid lines). The con-
ditional variance of Vcond(nslnp) = 0.73 is obtained for the measurement in
Fig. 24.15, thus satisfying the quantum-state reduction criterion for QND mea-
surements. A strong negative correlation of C = -0.62 can be inferred [14].
The small amount of additional noise introduced to the signal is attributed to
the experimental inaccuracy and the influence of the stimulated Raman effect.
The second criterion for a QND measurement refers to the transfer coefficients
which have recently been measured to be in the QND region [14]. Thus the
measurement performed so far does fulfill the two QND criteria. Based on the
416 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

probe measurement the signal photon number fluctuations could be reduced

27% below the shot noise limit.
The amount of noise reduction in this experiment is limited for several
reasons. Firstly, the data were all obtained with a finite detection efficiency.
The linear losses were larger than 20% due to finite performance of lenses,
grating, beam splitters, mirrors and detectors. In a theoretical investigation it
was simulated, however, that the performance of such a system will ultimately
be limited by the internal noise structure of the solitons [14]. Compared to
the phase measuring QND scheme this scheme shows improved performance
due to robustness against phase noise. Although the QND scheme using
quadrature detection is predicted to yield better results, the frequency shift QND
measurement discussed above achieved the best performance of the soliton
QND measurements reported so far.
The soliton QND experiments demonstrate that a spectral filter in a non-
linear optical fibre transmission line can introduce crosstalk between informa-
tion channels, represented by different carrier frequencies. The technological
challenge is to utilize this effect for optical switching, optical tapping and ul-
trafast nonlinear optical gates. The function of a quantum optical tap of the
photon number was demonstrated [14, 26, 66] but other functions might be
implemented as well. These are the basic building blocks of any quantum
communication system.

Acknowledgments
The authors are grateful to R. Loudon, D. Ostrowsky, R. Werner, B. Buchler,
S. Lorenz and T. C. Ralph for fruitful discussions and their support.

Notes
1. In principle deterministic perturbations of X. are allowed [29.54].
2. In fibres the absorption is typically as low as 0.2 dBlkm at >. = 1.5 j.tm.

3. For a discussion of the phase operator. see [61. 62].

4. GAWBS = Guided Acoustic Wave Brillouin Scattering [75]

References
[1] E. Schr6dinger, Naturwiss., 23, 807 (1935).
[2] A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935).
[3] D. Bohm, Quantum Theory. Prentice Hall, Englewood Cliffs, NJ, 1951.
[4] J. S. Bell, Physics 1,195 (1964).
[5] J. F. Clauser, M. A. Home, A. Shimony, and R. A. Holt, Rev. Prog. Phys.
23, 880 (1969); J. F. Clauser and A. Shimony, Rev. Prog. Phys. 41, 1881
 Quantum solitons: towards experimental quantum communication 417

(1978).
[6] A. Aspect, P. Grangier, and G. Roger, Phys. Rev. Lett. 47, 460 (1981);
A. Aspect, J. Dalibard, and G. Roger, Phys. Rev. Lett. 49, 1804 (1982).
[7] C. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and
W. K. Wootters, Phys. Rev. Lett. 70,1895 (1993); C. H. Benett, F. Bessette,
G. Brassard, L. Salvail, and J. Smolin, 1. Cryptology 5,3 (1992).
[8] For a review see W. Tittel, G. Ribordy, and N. Gisin, Physics World,41
(March 1998); N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, Rev. Mod.
Phys. 74, 145-195 (2002).
[9] D. Bouwmeester, A. Ekert, A. Zeilinger (Eds.), Physics of Quantum In-
formation. Springer, Berlin, 2000.
[10] Z. Y. Ou, S. F. Pereira, H. J. Kimble, K. C. Peng, Phys. Rev. Lett. 68,
3663 (1992).
[11] Ch. Silberhorn, P. K. Lam, O. Weill, F. Konig, N. Korolkova, and
G. Leuchs, Phys. Rev. Lett. 86,4267 (2001) and quant-phlOI03002.
[12] Y. Zhang, H. Wang, X. Li, J. Jing, C. Xie, and K. Peng, Phys. Rev. A 62,
023813 (2000).
[13] K. Bencheikh, J. A. Levenson, Ph. Grangier, O. Lopez, Phys. Rev. Lett.
75,3422 (1995).
[14] F. Konig, B. Buchler, T. Rechtenwald, G. Leuchs, and A. Sizmann,
"Soliton back-action evading measurement using spectral filtering", Phys.
Rev. A, submitted; F. Konig, T. Rechtenwald, M. A. Zielonka, R. Steidl,
G. Leuchs, and A. Sizmann, "Quantum-nondemolition measurement us-
ing spectral correlations between fibre-optical pulses." Conference on
Lasers and Electro-Optics/ Quantum Electronics and Laser Science Con-
ference CLEO/QELS'2000, San Fransisco, California, May 7-12, 2000,
Technical digest, QThI28, p. 206.
[15] F. Konig, M. A. Zielonka, and A. Sizmann, "Transient photon-number
correlations of interacting solitons", Phys. Rev. A 66, in print (2002);
A. Sizmann, F. Konig, M. A. Zielonka, R. Steidl, and T. Rechtenwald,
"Quantum correlations of colliding solitons", in Massive WDM and TDM
Soliton Transmission Systems (A ROSC Symposium). A. Hasegawa (Ed.),
Kluwer Academic Publishers, Dodrecht 2000, p. 289-298.
[16] A. Kuhn, M. Hennrich, T. Bondo, and G. Rempe, Appl. Phys., 69, 373
(1999); M. Hennrich, T. Legero, A. Kuhn, and G. Rempe, Phys. Rev. Lett.
85, 4872 (2000).
[17] c. K. Law and H. J. Kimble, 1. Mod. Opt. 44, 2067 (1997).
[18] B. Huttner and J. Brendel, "Photon-counting techniques for fiber mea-
surements", Lightwave, August 2000, pp. 112-120.
418 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[19] A. Karlsson, M. Bourennane, G. Ribordy, H. Zbinden, J. Bendel, J. Rarity,

and P. Tapster, "A single-photon counter for log-haul telecom", IEEE
Circuits & Devices, November 1999, p. 34.
[20] S. Lloyd and S. L. Braunstein, Phys. Rev. Lett. 82, 1784 (1999).
[21] U. Leonhardt, Measuring the quantum state of light. Cambridge Univer-
sity Press, 1997.
[22] A. Furusawa, J. L. SflSrensen, S. L. Braunstein, C. A. Fuchs, H. J. Kimble,
and E. S. Polzik, Science 283,706 (1998).
[23] G. Leuchs, T. C. Ralph, C. Silberhorn, and N. Korolkova, J. Mod. Opt.
46,1927 (1999); G. Leuchs and N. Korolkova, "Entangling fiber solitons:
Quantum noise engineering for interferometry and communication", Op-
tics & Photonic News, February 2002, p. 64-69.
[24] M. D. Reid, and P.D. Drummond, Phys. Rev. Lett. 60,2731 (1988); M. D.
Reid, Phys. Rev. A 40, 913 (1989); M. D. Reid, The Einstein-Podolsky-
Rosen Paradox and entaglement 1: Signatures of EPR correlations for
continuous variables, quant-ph/0112038 (2001).
[25] H. A. Haus, K. Watanabe, Y. Yamamoto, J. Opt. Soc. Am. B 6, 113 (1989).
[26] S.R. Friberg, S. Machida, and Y. Yamamoto, Phys. Rev. Lett. 69,3165
(1992); S. R. Friberg, S. Machida, M. J. Werner, A. Levanon, and T.
Mukai, Phys. Rev. Lett. 77, 3775 (1996).
[27] D. B. Horoshko and S. Ya. Kilin, Phys. Rev. A 61, 032304 (2000).
[28] M. J. Holland, M.J. Collett, D. F. Walls, and M. D. Levenson, Phys. Rev.
A 42,2995 (1990); J.-P. Poizat, J.-F. Roch, and P. Grangier, Ann. Phys.
Fr. 19,265 (1994).
[29] V. B. Braginsky, Y. I. Vorontsov and K. S. Thome, Science 209, 547
(1980).
[30] R. F. Werner, Phys. Rev. A 40, 4277 (1989).
[31] R. Simon, Phys. Rev. Lett. 84,2726 (2000).
[32] L.-M. Duan, G. Giedke, J. 1. Cirac, and P. Zoller, Phys. Rev. Lett. 84,
2722 (2000).
[33] D. F. Walls, G. J. Milburn, Quantum Optics. Springer, Berlin, 1995.
[34] R. F. Werner and M. M. Wolf, Phys. Rev. Lett. 86, 3658 (2001).
[35] G. Giedke, B. Kraus, M. Lewenstein, and J. I. Cirac, Phys. Rev. Lett.
87, 167904 (2001); G. Giedke, L.-M. Duan, J. I. Cirac, and P. Zoller,
Quant. Inf. Compo 1, 79 (2001); G. Giedke and J. 1. Cirac, "The char-
acterization of Gaussian operations and Distillation of Gaussian States",
quant-ph/0204085.
[36] S. Mancini, V. Giovannetti, D. Vitali, and P. Tombesi, Entangling macro-
scopic oscillators exploiting radiation pressure, quant-ph/0108044.
 Quantum solitons: towards experimental quantum communication 419

[37] R. Bruckmeier, H. Hansen, S. Schiller, and J. Mlynek, Phys. Rev. Lett.

79,43 (1997).
[38] S. Schmitt, J. Ficker, M. Wolff, F. Konig, A. Sizmann, and G. Leuchs,
Phys. Rev. Lett. 81,2446 (1998).
[39] D. Krylov, and K. Bergman, Optics Lett. 23,1390 (1998).
[40] M. J. Werner, Phys. Rev. Lett. 81,4232 (1998).
[41] N. Korolkova, G. Leuchs, S. Schmitt, C. Silberhorn, A. Sizmann, M.
Stratmann, O. Weill, and H. A. Bachor, Nonlinear Optics, 24, 223 (2000).
[42] M. Kitagawa, and Y. Yamamoto, Phys. Rev. A 34, 3974 (1986).
[43] A. Sizmann and G. Leuchs, in: E.Wolf (ed.), Progress in Optics XXXIV.
Elsevier Science Publishers B. V., Amsterdam, 1999, p. 373.
[44] L.-M. Duan, G. Giedke, J. 1. Cirac, and P. Zoller, Phys. Rev. Lett. 84,
4002 (2000).
[45] N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, Rev. Mod. Phys. 74,
145 (2002); N. Liitkenhaus, Phys. Rev. A 61, 052304 (2000); G. Brassard,
N. Liitkenhaus, T. Mor, and B. C. Sanders, Phys. Rev. Lett. 85, 1330
(2000).
[46] J. A. Levenson, 1. Abram, T. Rivera, P. Fayolle, J. C. Garreau, and P. Grang-
ier, Phys. Rev. Lett. 70,267 (1993); J. Ph. Poizat and P. Grangier, Phys.
Rev. Lett. 70, 271 (1993); E. Goobar, A. Karlsson, and G. Bjork, Phys.
Rev. Lett. 71,2002 (1993).
[47] Ph. Grangier, International Quantum Electronics Conference IQEC'2000,
Nice, September 10-15, 2000, Thtorial talk.
[48] M. Ozawa, Phys. Rev. Lett. 80,631 (1998).
[49] V. B. Braginsky and Y. 1. Vorontsov, Sov. Phys.-Usp. 17,644 (1974).
[50] G. J. Milburn and D. F. Walls, Phys. Rev. A 28, 2065 (1983).
[51] N. Imoto, H. A. Haus, and Y. Yamamoto, Phys. Rev. A 32,2287 (1985).
[52] B. Yurke, J. Opt. Soc. Am. B 2, 732 (1985).
[53] C. M. Caves, Phys. Rev. Lett. 45, 75 (1980).
[54] W. G. Unruh, Phys. Rev. D 19, 2888 (1979).
[55] G. P. Agrawal, Nonlinear Fiber Optics. Academic Press, Inc., San Diego,
California, 1995.
[56] Y. Lai and H. A. Haus, Phys. Rev. A 40,854 (1989).
[57] P. D. Drummond, R. M. Shelby, S. R. Friberg, and Y. Yamomoto, Nature
365,307 (1993).
[58] Ph. Grangier, J. A. Levenson, and J.-P. Poizat, Nature 396, 537 (1998).
[59] M. Kitagawa, N. Imoto, and Y. Yamamoto, Phys. Rev. A 35, 5270 (1987).
420 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

[60] H. A. Haus and Y. Lai, J. Opt. Soc. Am. B 7, 386 (1990).

[61] P. Carruthers and M. Nieto, Phys. Rev. Lett. 14,387 (1965).
[62] W. Vogel and D.-G. Welsch, Lectures on Quantum Optics. Akademie
Verlag, Berlin, 1994.
[63] V. E. Zakharov and A. B. Shabat, Zh. Eksp. Teor. Fiz. 61, 118 (1971),
[Sov. Phys. JETP 34,62 (1972)].
[64] M. D. Levenson, R. M. Shelby, M. Reid, and D. F. Walls, Y. Yamamoto,
Phys. Rev. Lett. 57,2473 (1986).
[65] H. A. Bachor, M. S. Levenson, D. F. Walls, S. H. Perlmutter, and
R. M. Shelby, Phys. Rev. A 38,180 (1988).
[66] S. R. Friberg, T. Mukai, and S. Machida, Phys. Rev. Lett. 84,59 (2000).
[67] M. D. Levenson and R. M. Shelby, J. Mod. Opt. 34,775 (1987).
[68] J.-F. Roch, K. Vigneron, P. Grelu, A. Sinatra, J.-P. Poizat, and Ph. Grangier,
Phys. Rev. Lett. 78,634 (1997).
[69] B. Buchler, P. K. Lam, H. A. Bachor, U. Anderson and T. C. Ralph, Phys.
Rev. A 65, 011803(R).
[70] N. Imoto and S. Saito, Phys. Rev. A 39, 675 (1989).
[71] Ph. Grangier, J. M. Courty, and S. Reynaud, Opt. Commun. 89,99 (1992).
[72] P. D. Drummond, J. Breslin, and R. M. Shelby, Phys. Rev. Lett. 73,2837
(1994).
[73] J.M. Courty, S. SpaIter, F. Konig, A. Sizmann, and G. Leuchs, Phys. Rev.
A 58, 1501 (1998).
[74] Y. Sakai, R. J. Hawkins, and S. R. Friberg, Opt. Lett. 15,239 (1990).
[75] R. M. Shelby, M. D. Levenson, and P. W. Bayer, Phys. Rev. Lett. 54, 939
(1985); R. M. Shelby, M. D. Levenson, and P. W. Bayer, Phys. Rev. B 31,
5244 (1985).
[76] P. D. Townsend and A. J. Poustie, R. M. Shelby, Opt. Lett. 20,37 (1995).
[77] S. S. Yu and Y. Lai, "Quantum non-demolition measurements of the pho-
ton number with a fiber ring: enhancement of the correlation through
multiple soliton collisions", in International Quantum Electronics Con-
ference Technical digest, OSA Technical Digest Series, Sidney, Australia,
1996, V. 20, p. 246.
[78] S. Spalter, P. van Loock, A. Sizmann, and G. Leuchs, Appl. Phys. B 64,
213 (1997).
[79] S. J. Carter, P. D. Drummond, M. D. Reid, and R. M. Shelby, Phys. Rev.
Lett. 58, 1841 (1987).
[80] M. Rosenbluh and R. M. Shelby, Phys. Rev. Lett. 66, 153 (1991).
 Quantum solitons: towards experimental quantum communication 421

[81] S. R. Friberg, S. Machida, M. J. Werner, A. Levenson, and T. Mukai,

Phys. Rev. Lett. 77, 3775 (1996).
[82] S. Spillter, N. Korolkova, F. Konig, A. Sizmann, and G. Leuchs, Phys.
Rev. Lett. 81,786 (1998).
[83] M. Werner and S. Friberg, Phys. Rev. Lett. 79,4143 (1997).
[84] E. Schmidt, L. Knoll, D.-G. Welsch, M. Zielonka, F. Konig, and A.
Sizmann, Phys. Rev. Lett. 85, 3801 (2000).
[85] S. Spillter, M. Burk, U. StroBner, A. Sizmann, and G. Leuchs, Optics
Express 2,77 (1998).
Index

Abrams-Lloyd, 42 Decoherence, 40
Algorithm Dense coding, 64, 95
Abrams-Lloyd, 42 Deutsch-lozsa, 7, 32
Deutsch-lozsa, 7, 32 Displacement-covariant, 280
Grover, 8, 33 Distillability, 174
Shor, 7, 41 Down conversion, 79
Anticlone, 284 Eavesdropper, 290
BB84 protocol, 332 Efficient simulation, 47
Beam splitter, 27, 112, 117, 256, 285-286 Entanglement, 60, 250
Bell inequality, 5 bipartite, 112
Bell-state, 63, 78, 86 bound, 174, 180,211,213,218
Bit error rate, 298, 302 distillation, 173,207,323
Bit, 4 entropy, 194
Cauchy-Schwarz inequality, 123 fidelity,73
Characteristic function, 175 measure, 61, 194
Check pairs, 327 mixed,115
CHSH inequality, 133 multipartite, 111-112, 114, 125
Circuit partial, 141
quantum, 116 purification, 65, 173, 193, 323
Classical teieportation, 89 quantification, 194
Clifford swapping, 64
algebra, 49 EPR,59,68,255,382,390,396,400,404
group, 49-50 beams, 81,97
Cloning, 277 pair,62,78
fidelity, 283, 286 state, 69, 281
CNOT, 48, 50, 116,281 Error
Coherent correction, 19
amplitude, 71 probability, 298, 305
state, 71, 300 Excess noise, 279
Commutation relation, 106 Exponential speedup, 43
Completely-positive, 280 Factorize, 7
Computational complexity, 7 Feed-forward, III
Concatenation of c\oners, 286 Fidelity, 68, 79, 89
Conditional variance, 382 cloning, 283, 286
Conjugate observables, 278 Fourier, 117
Correlation matrix, 138, 175 Gate
Covariance maxtix, 175 Pauli,48
Covariant, 280 phase, 48, 51
CSS code, 330 SUM, 50, 52
Damping channel, 348 Gaussian, 43, 70, 219

423
424 QUANTUM INFORMATION WITH CONTINUOUS VARIABLES

additive-noise, 289 Partial transpose, 113-114, 180,218

doner, 278, 280, 283 Pauli
distribution, 288 gate, 48
noise, 283 group,49
operation, 207 operator, 323
raw key,289 Phase gate, 48,51
state, 105, 114, 141, 174,384 Phase-insensitive amplifier, 285-286
Generalized uncertainty principle, 297, 299 Photon absorption, 345
GHZ, 105, 114-115 Position, 116
operator, 106 Privacy amplification, 305
paradox, 105 QND,26, 182, 187,244-245,248-250,381,405
state, 108 Quadrature, Ill, 116,382
Gottesman-Knill,47 Quantum
Group entanglement, 5
Clifford, 49-50 circuit, 116
Heisenberg-Weyl,49 cloning, 277
Pauli,49 communication, III
Grover, 8, 33 computation, 4
Hadamard, 20,48, 116 fault-tolerant, 15
Hamiltonian, 10 universal, 10
Kerr, 12 cryptography, 288, 295
quadratic, 11 duty, 71, 89
Harmonic oscillator, 38 error correction, 15,21
Heisenberg-Weyl group, 49 floating point, 10
Homodyne,99,120 key distribution, 288, 317
detection, III noise, 84, 296
measurement, 179 parallelism, 6
Hybrid, 37, 40 repeater, 174
Inequality teleportation, 312
Bell,5 uncertainty, 78, 388, 396, 400
Cauchy-Schwarz, 123 Quasifree transfonnation, 176
CHSH,133 Qubit, 5, 32, 38
Infonnation exclusion principle, 291 Quduty, 71-72
Inseparability, 121,177-178,384,387,402 Qunat, 13,32,38
multipartite, 112, 116 Reconciliation protocol, 290
Joint measurement, 278 Reduction criterion, 174
Kerr, 121 Rotation-covariant, 283
effect, 14,399,407 Sagnac interferometer, 397
Hamiltonian, 12 Schmidt
nonlinearity, 13, 187,381,396 decomposition, 61, 112, 114
Key pairs, 327 number, 61, 220, 224
Kraus operators, 62 Separability
Linear optics, 112, 120 full,I23
Local-hidden variable, 105 partial, 123
Local-realistic theories, 113 Separable state, 384
LOCC, 60, 65, 211 . Shannon theory, 289
Loss, 307 Shor, 7, 41
Majorization, 114 Signal transfer coefficient, 302, 309
Momentum, 116 Signal-to-noise ratio, 289
Multipartite inseparability, 112, 116 Soliton, 406
Mutual information, 100,304 collision, 407
Nat, 13 Spin squeezing, 244
Nona-splitter, 27 Spin states
Nonlocality, 105, 131, 137 coherent, 234
OPO, 81, 381 EPR,235
Optical fibre, 381, 408 squeezed, 234
Parity operator, 132 SPM, 410
 INDEX 425

Squeezed Transform
light, 112, 308 discrete Fourier, 287
sUrte, 42, 112, 180,236,284,319,388,413 Fourier, 20, 282
vacuum, 13,81 Hadamard, 20
Stabilizer, 48 Transpose
generators, 320 partial, 113-114, 180,218
Standard form, 176 Tritter,27
State swapping, 260 Two-mode cat sUrte, 202
SUM gate, 50, 52 Vacuum fluctuations, 82, 283
Symplectic, 176 Virial theorem, 26
Syndrome, 25-26 Von Neumann entropy, 62, 96, 113, 116
Telecloning, 130 W state, lIS
Teleportation, 63,67, 77,255,258 Wave-packet, 22
classical, 89 Weyloperator, 175
Theorem Wigner
Gottesman-Knill, 47 function, 68, 73, 99, 128, 384, 396
no-cloning, 278 representation, 128
virial,26 XOR, 21, 33, 243-244
Transfer coefficient, 409 XPM coupling, 407