REVIEWS OF MODERN PHYSICS, VOLUME 84, APRIL–JUNE 2012

Multiphoton entanglement and interferometry

Jian-Wei Pan,1,* Zeng-Bing Chen,1,† Chao-Yang Lu,1,‡ Harald Weinfurter,2,3,§
Anton Zeilinger,4,5,k and Marek Żukowski6,7,1,{
1
Hefei National Laboratory for Physical Sciences at Microscale and Department of Modern
Physics, University of Science and Technology of China, Hefei, Anhui 230026, China
2
Fakultät für Physik, Ludwig-Maximilians-Universität, D-80799 München, Germany
3
Max-Planck-Institut für Quantenoptik, D-85748 Garching, Germany
4
Vienna Center of Quantum Science and Technology (VCQ), Faculty of Physics,
Universität Wien, Boltzmanngasse 5, A-1090 Wien, Austria
5
Institut für Quantenoptik und Quanteninformation (IQOQI), Österreichische Akademie der
Wissenschaften, Boltzmanngasse 3, A-1090 Wien, Austria
6
Instytut Fizyki Teoretycznej i Astrofizyki, Uniwersytet Gdański, PL-80-952 Gdańsk, Poland
7
Institut für Experimentalphysik, Universität Wien, Boltzmanngasse 5, A-1090 Wien, Austria
(published 11 May 2012)

Multiphoton interference reveals strictly nonclassical phenomena. Its applications range from
fundamental tests of quantum mechanics to photonic quantum information processing, where a
significant fraction of key experiments achieved so far comes from multiphoton state manipulation.
The progress, both theoretical and experimental, of this rapidly advancing research is reviewed. The
emphasis is given to the creation of photonic entanglement of various forms, tests of the
completeness of quantum mechanics (in particular, violations of local realism), quantum informa-
tion protocols for quantum communication (e.g., quantum teleportation, entanglement purification,
and quantum repeater), and quantum computation with linear optics. The scope of the review is
limited to ‘‘few-photon’’ phenomena involving measurements of discrete observables.
DOI: 10.1103/RevModPhys.84.777 PACS numbers: 03.65.Ud, 03.67.
a, 42.50.
p

CONTENTS C. Two-photon interference due to indistinguishability

of photons 787
I. Introduction 778
D. Postselection of entanglement and quantum erasure 789
A. Quantum optics 778
E. Entangled-state analyzers 789
B. The essence of the quantum world: Entanglement 779
1. Bell-state analyzer 790
C. Sources of photonic entanglement 779
2. GHZ-state analyzer 790
D. Applications in quantum information 780
IV. Experimental realizations of photonic entanglement 791
E. Related reviews 780
A. Spontaneous parametric down-conversion 791
F. Our aims 780
B. Photonic entanglement in higher dimensions 793
II. Interference and quantum entanglement 780
1. Entangled qudits 793
A. Classical interference 781
2. Hyperentanglement 793
B. Quantum interference 781
C. Twin-beam multiphoton entanglement 794
1. Single-particle quantum interference 781
D. Multiphoton entanglement 795
2. Two-particle quantum interference 782
1. Entanglement construction 796
C. Quantum entanglement 782
2. New methods 797
D. Interferometry with entangled two-photon
3. First proposals 798
and multiphoton states 784
4. Experimental realizations 798
1. EPR interferometry 784
V. Falsification of a realistic world view 800
2. Greenberger Horne Zeilinger interferometry 784
A. Bell’s inequality 801
E. Interferometry with multiport beam splitters 785
B. GHZ theorem 803
III. Photonic qubits and linear optics 785
1. Impossibility of deriving realism via
A. Photonic qubits 785
perfect quantum correlations and locality 803
B. Simple linear-optical elements 786
2. GHZ theorem for laboratory measurement 804
3. Two-observer GHZ-like correlations 805
4. Hardy’s paradoxes 805
*[email protected] C. Refutation of a class of nonlocal realistic theories 806
†
[email protected] D. Noncontextual hidden-variable theories 807
‡
[email protected]
§ VI. Quantum communication 808
[email protected]
k A. Quantum dense coding 809
[email protected]
{
[email protected] B. Quantum teleportation 809

0034-6861= 2012=84(2)=777(62) 777 Ó 2012 American Physical Society

778 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

1. Theory: Qubit teleportation involving more photon interference and in intensity correlation mea-
an EPR channel and two bit transfer 809 surements: a plethora of classically impossible phenomena
2. Experimental quantum teleportation 811 occurs, most of them completely incomprehensible with any
3. Teleportation onto freely flying photons 811 classical concepts, neither particle nor wave. As always, in
4. Teleportation of a qubit carried by a photon the history of human scientific endeavor, harnessing of new
of the ancillary EPR pair 812 phenomena leads to new applications. The aim of this review
5. Teleportation with various physical systems 813 is to describe the recent theoretical and experimental advan-
6. More-involved teleportations 813 ces in multiphoton interference, entanglement, manipulation,
C. Entanglement swapping 814 and their applications in quantum communication and
1. Theory 814 computation.
2. First experimental demonstration 815
3. Other experiments on entanglement swapping 815 A. Quantum optics
D. Beating noisy environment 816
1. Entanglement distillation and concentration 817 An intensive research of the quantum properties of light
2. Entanglement purification 817 started around half a century ago. Its advances allow one to
E. Long-distance entanglement distribution 819 gain a coherent control of quantum optical systems, enabling
F. Quantum memory and quantum repeaters 820 true quantum engineering. As a result, quantum optical
1. Quantum repeater protocol 820 methods made possible to actually perform gedanken experi-
2. Quantum state transfer between ments concerning the foundations of quantum theory. This
matter and photons 822 control of quantum phenomena further allows one to search
VII. Photonic quantum computing 823 for novel information-processing protocols, which now prom-
A. Linear-optical two-qubit logic gates 823 ise new technologies based on quantum information science.
B. Cluster-state quantum computing 824 Soon after Einstein’s introduction of light quanta, Taylor
1. Constructing photonic cluster states 824 (1909) tried to find some new effects in a two-slit Young-type
C. Few-photon quantum computing experiments 825 experiment using extremely faint light, so faint that on aver-
D. Toward scalable optical quantum computing 826 age only one photon at a time was inside the apparatus. No
VIII. Concluding remarks 827 deviation from the classical interference was observed. Now,
Appendix A: The two-photon states produced by SPDC 828 with a fully developed theory of quantized light, we know that
experiments of this type cannot differentiate between the
classical explanation (based on the interference of electric
field waves) and the quantum explanation (based on the
I. INTRODUCTION interference of probability amplitudes for photons passing
through either of the two slits). The inherently ‘‘quantum
In his 1704 treatise, Opticks Newton claimed that light is nature of the electromagnetic field,’’ as we know now, is
composed of particles, and strongly opposed Huygens wave revealed directly in multiphoton experiments which were
picture. Later on with Young’s double-slit interference ex- not possible earlier.
periments, the wave picture seemed to be correct and suffi- Still, quantum interference of truly individual photons is
cient. This view was further strengthened by Maxwell’s certainly a fascinating phenomenon. The first precise experi-
electrodynamics. Yet, 201 years after Newton, during his ment aimed at exactly this was performed by Grangier,
annus mirabilis Einstein reintroduced lichtquanten (light Roger, and Aspect (1986). They used photon pairs emitted
particles) and in this way explained the photoelectric effect in atomic cascades, one of the photons was used as a trigger,
(Einstein, 1905).1 The ultimate consequences of Einstein’s and the other was fed into a Mach-Zehnder interferometer.
ideas, after fundamental works of Bohr, Heisenberg, and When detectors are placed in the two arms of the interfer-
Schroedinger, gave birth to quantum mechanics in 1925. ometer, besides background noise, no simultaneous detection
Quantum electrodynamics, the final theory of light, in which (i.e., coincidence) in both detectors was observed,2 i.e., the
photons are elementary excitations of the quantized electro- photon was found only in one of the two arms, a typical
magnetic field interacting with charges, was given by Dirac particlelike behavior. However, after overlapping the two
(1927), and its internal consistency was proved by Dyson, arms by the output beam splitter of the interferometer the
Feynman, Schwinger, and Tonomaga around 20 years later. usual (wavelike) interference pattern was observed. Recently,
According to these theories, photons, as all quantum Braig et al. (2003) demonstrated that, when also observing
particles, reveal both wavelike and particlelike properties, a interference depending on the phase difference between both
phenomenon known as wave-particle duality. The wave arms, the light from the interferometer output exhibited the
nature is revealed by interference, while the particle nature characteristic single-photon antibunching.
can be recognized in absorption and detection events, or Modern quantum optics was effectively born in 1956 when
more generally in the statistics of counts. The interference Hanbury Brown and Twiss (1956) introduced intensity
patterns involving single photons or, equally well, the light interferometry. It was the first serious attempt to study the
intensity does not reveal strictly nonclassical phenomena.
Some of the most counterintuitive effects begin with two or 2
As a matter of fact some coincidence was observed, however it
was below the case that was to be expected if the photons were
1
The current term photon was introduced by Lewis (1926). treated as classical wave packets

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 779

correlations between intensities recorded at two separated The EPR paradox and with it the entanglement of quantum
detectors. It motivated more sophisticated photon counting states remained a philosophical issue (Feynman, Leighton,
and correlation experiments. The quantum theory of optical and Sands, 1963), and seemed experimentally untestable for
coherence of Glauber (1963) gave theoretical clues to search almost 30 years. The breakthrough happened when Bell
for unambiguously quantum optical phenomena. Carmichael (1964) derived his remarkable inequalities which revealed
and Walls (1976a, 1976b) predicted photon antibunching in a that two-particle correlations for the two spin- 12 singlet
resonance fluorescence, which was observed experimentally disagree with any local realistic model. The pioneering
by Kimble, Dagenais, and Mandel (1977). The early experi- ‘‘Bell experiment’’ was done by Freedman and Clauser
ments used atomic beams as sources. Thus, atomic number (1972), followed by the famous ones by Aspect, Grangier,
(and emission statistics) fluctuations were unavoidable. Later, and Roger (1981, 1982) and Aspect, Dalibard, and Roger
Diedrich and Walther (1987) realized such experiments using (1982) and many others.4 The early experiments used
a single trapped ion and observed photon antibunching as polarization-entangled photon pairs from atomic cascades
well as sub-Poissonian statistics in the system. Squeezed (Clauser and Shimony, 1978). In late 1980s, parametric
states of light were experimentally generated by using a down-conversion was discovered as a convenient and robust
four-wave mixing in atomic sodium (Slusher et al., 1985) method to produce entangled photons (see Sec. V).
and in optical fibers (Shelby et al., 1986), or by using an A quarter of century after Bell’s paper, it turned out that the
optical parametric oscillator (Wu et al., 1986). Interested conflict of local realism with quantum mechanics is even
readers may find excellent reviews in Mandel and Wolf more striking for certain three or more particle entangled
(1995), Scully and Zubairy (1997), Walls and Milburn states. The Greenberger-Horne-Zeilinger (GHZ) theorem
(1994), and Lounis and Orrit (2005). (Greenberger, Horne, and Zeilinger, 1989; Greenberger
The study on photon statistics and photon-counting tech- et al., 1990; Mermin, 1990a) showed that the concept of
niques enables direct examination of some fundamental dis- EPR’s elements of reality is self-contradictory. That is, there
tinctions between quantum and classical concepts of light. are situations for which local realism and quantum mechanics
Parallel developments in neutron, atomic, and molecular make completely opposite predictions, even for perfectly
interferometry, as well as modern methods of cooling and correlated results, which were the starting point of the EPR
trapping ions, etc., allowed one to probe ever deeper the argumentation. The GHZ paper showed that three or more
properties of individual quantum systems and to realize particle interferometry is a rich untested area, full of exciting
many of the gedanken (thought) experiments testing the classically paradoxical phenomena. However, at that time no
foundations quantum physics. effective sources of three-photon or four-photon entangle-
ment were present. Thus, a new chapter in experimental
B. The essence of the quantum world: Entanglement multiphoton quantum optics was opened, but had to wait
for new ideas and experimental techniques.
Entanglement is a property of more than one quantum
system such that the state of one system cannot be seen C. Sources of photonic entanglement
independent of the other’s system. It forms the basis for the
most remarkable, purely quantum effects and is the main The standard source of entangled-photon pairs is nowadays
resource for the many applications of quantum information the nonlinear optical process of spontaneous parametric
processing. Initially, it was used by Einstein, Podolsky, and down-conversion (SPDC) ([for a pioneering observation,
Rosen (1935) to show that quantum mechanics is incomplete. see Burnham and Weinberg (1970)], the inverse of frequency
The trio, Einstein, Podolsky, and Rosen (EPR), argued that doubling and up-conversion. In SPDC, photons from a pump
the outcome of a measurement on any physical system is laser beam, within a nonlinear crystal, can spontaneously be
determined prior to and independent of the measurement converted into pairs that are momentum and frequency en-
(realism) and that the outcome cannot depend on any actions tangled, and in the so-called type II process can also be
in spacelike separated regions (Einstein’s locality).3 This polarization entangled (Kwiat et al., 1995). Today, SPDC
EPR criterion of, what is now called, ‘‘local realism’’ should sources of entangled-photon pairs of high quality and bright-
be fulfilled by every physical description of nature, and, ness can be routinely realized using various methods.
indeed, it looks quite reasonable to us, particularly as all However, since the GHZ paper, and even more after the
our classical world and experience fully adhere to it. They birth of quantum information, three or more photon entangle-
used the perfect correlations of entangled states (thus often ment was in demand. It turned out that using the primary
called EPR states) to define ‘‘elements of reality,’’ a notion two-photon entanglement, by a procedure which is called
which according to them was missing in quantum theory. entanglement swapping, one can entangle without any direct
Elements of reality are deterministic predictions for a mea- interaction particles which were independent of each other
surement result, which can be established without actually or, what is more important for us, construct entanglement of
performing the measurement, and without physically disturb- higher order (Żukowski, et al., 1993). Since photons basi-
ing the (sub)system to which they pertain. As elements of cally do not interact with each other, this method is of
reality in the studied case were argued to exist necessarily special importance for schemes aimed at creating multipho-
even for pairs of noncommuting observables, they claimed ton entanglements. Practical versions of this technique
they are contradicting the Heisenberg uncertainty relation. (Rarity, 1995; Zeilinger et al., 1997; Żukowski, Zeilinger,

3 4
A more detailed discussion of the EPR paradox is in Sec. V. See Aspect (1999); Tittel and Weihs (2001) for a recent survey.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

780 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

and Weinfurter, 1995) are thus the basis of all experiments resource overhead by three orders of magnitudes compared to
with three or more entangled photons, as well as of many the KLM scheme. Cluster states up to six entangled photons
realizations of quantum information protocols, up to have been realized (Kiesel, Schmid, Toth, et al., 2005;
measurement-based quantum computation. Walther, Resch, Rudolph et al., 2005; Lu et al., 2007),
and applied to demonstrate elementary quantum gates and
algorithms (Walther, Resch, Rudolph et al., 2005).
D. Applications in quantum information
The rapid growth of literature on photonic realizations of
Experimental quantum information processing was started quantum information processing still continues. One can
right after new experimental techniques allowed one to con- expect much more exciting new developments.
trol individual or compound quantum systems like atoms in
traps, pairs of entangled photons, etc., and to observe a new E. Related reviews
set of classically impossible phenomena. As always, new
controllable phenomena lead to new practical applications, The earlier stages of the research of photonic entanglement
especially in information transfer and processing. Quantum have been reviewed in Clauser and Shimony (1978) Mandel
information processing harnesses the superposition principle and Wolf (1995), and Mandel (1999). They contain a collec-
and nonclassical correlations of quantum mechanics and tion of descriptions of experiments demonstrating the very
employs them in communication and computation. In quan- nature of quantum mechanics. Because of a limit of space, we
tum cryptography [Bennett and Brassard (1984); Wiesner shall not discuss these experiments in this review. We start
(1983); Ekert (1991); for a review see Gisin et al. (2002)] our description more or less at the stage of developments at
complementary measurements on quantum systems are used which these earlier reviews ended. For detailed discussions
to establish a secret key shared by two partners, thus enabling on quantum entanglement, see Alber et al. (2001) and
for the first time a provably secure communication. Quantum Horodecki et al. (2008). Reviews on Bell’s theorem can be
teleportation (Bennett et al., 1993) enables a faithful transfer found in Laloë (2001), Werner and Wolf (2001a), and
of an unknown quantum state from one location to another, Genovese (2005). For an introduction to quantum information
using entangled states as a quantum channel. Quantum com- and quantum computation; see Bennett and DiVincenzo
puters promise to greatly increase the efficiency of solving (2000), Preskill (1998), Nielsen and Chuang (2000), and
problems such as factoring large integers, database search, Bouwmesster, Ekert, and Zeilinger (2001). Quantum cryp-
and simulation of some quantum phenomena. tography has been reviewed by Gisin et al. (2002). Linear-
Photons are the fastest information carrier, have a very optical quantum computing with photonic qubits has been
weak coupling to the environment, and are thus best suited for reviewed by Kok et al. (2007), O’Brien (2007), O’Brien,
quantum communication tasks. Thanks to this property, quan- Furasawa, and Vuckovic (2009), and Ralph and Pryde (2010).
tum key distribution with photons has now went long beyond Weinfurter (2000) and Zeilinger et al. (2005) gave a concise
the first few-meter laboratory demonstrations, to free-space review on experimental progress on photon interference and
or fiber-based distributions over 100 km [see, e.g., Rosenberg quantum information applications.
et al. (2007) and Ursin et al. (2007)], and is rapidly com-
mercialized into real life intercity cryptographic networks. To F. Our aims
ultimately extend the range of quantum communication to a
global scale, a collection of quantum toolkits still has to be In this review, we describe the progress in the last two
developed. A quantum repeater (Briegel et al., 1998) would decades or so, both theoretical and experimental, in multi-
allow in principle an efficient long-distance high-fidelity photon interferometry, and its applications ranging from
transmission of entanglement. Several ingredients of this fundamental tests of quantum mechanics to photonic quan-
scheme have been already demonstrated: entanglement swap- tum information processing. Emphasis will be put on creation
ping (Pan et al., 1998), purification (Pan, Gasparoni, Ursin and control of photonic entanglement with linear optics
et al., 2003), quantum memory (Yuan et al., 2008), etc. and its application in quantum communication and computa-
Despite the difficulty to ‘‘localize’’ photons, there has been tion. We limit the scope of our review to ‘‘few-photon’’
a considerable interest in linear-optical quantum computing. phenomena involving measurements of discrete observables,
This is motivated by the photon’s robustness against decoher- thus many fascinating experiments involving continuous
ence and the relative ease with which it can be manipulated variables will be not discussed here.
with high precision. Remarkably, by exploiting the nonline-
arity induced by measurement, Knill, Laflamme, and Milburn II. INTERFERENCE AND QUANTUM ENTANGLEMENT
(KLM) showed that scalable quantum computation is in
principle possible with linear optics, single-photon sources, Classical interference is a macroscopic expression of the
and detectors. A new and probably more practical approach is quantum one, i.e., the coherent or thermal states of the
the concept of a ‘‘one-way quantum computer’’ (Raussendorf electromagnetic fields can also be described with Maxwell’s
and Briegel, 2001; see Sec. VII.B.1). In this approach, one laws. The interference phenomena in the quantum realm are
starts with the so-called ‘‘cluster states’’ (Briegel and richer and more pronounced. We discuss here the basic
Raussendorf, 2001). The computation algorithm is then per- differences between the classical interference understood as
formed by applying a sequence of one-qubit measurements. interference of electromagnetic waves in space, and the
Optical quantum computing proposals (Browne and Rudolph, quantum one which is interference of various operationally
2005) based on the one-way model reduce the computational indistinguishable processes.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 781

A. Classical interference In addition to the phase fluctuations, one can also take into
account amplitude fluctuations. Nevertheless, the basic fea-
In classical physics interference results from the superpo- tures of the earlier formula must be retained. Amplitude
sition of waves. It may express itself in the form of intensity fluctuations tend to lower the visibility of the intensity corre-
variations or intensity correlations. lations patterns even further. Thus, the visibility of intensity
Consider two quasimonochromatic plane waves linearly correlations for fields with fluctuating phase differences is
polarized in the same direction, described by never full, maximally 50%. As we shall see, there is no bound
on visibility in the quantum case. For a broader treatment of
Ej ðr; tÞ ¼ Ej ei½kj r
!t

j ðtÞ þ c:c: (1) these matters, see Paul (1986) and Belinskii and Klyshko
(1993).
where Ej is the real amplitude of one of the fields, kj is the
wave vector, ! is the frequency of both waves, j ¼ 1, 2 the
index numbering the fields, and finally c.c. denotes the com- B. Quantum interference
plex conjugate of the previous expression. The intensity of
the superposed fields at a certain point in space is given by Quantum interference rests on the concept of superposition
of probability amplitudes of processes that contribute to the
Iðr; tÞ ¼ E21 þ E22 þ 2E1 E2 cos½
12 k  r

12
ðtÞ; given phenomenon.
(2)
1. Single-particle quantum interference
where
12 is the difference of the respective parameters for
the two fields, e.g.,
12
ðtÞ ¼
1 ðtÞ

2 ðtÞ. For
12
ðtÞ Single-particle interference looks almost identical to the
constant in time, or of values varying much less than , this classical one. We replace the fields (waves) by amplitudes
formula (after averaging over time) describes a Young-type Aðx; tÞ, which differ only by the fact that they must be
interference pattern. In the opposite case, of widely fluctuat- suitably normalized, if one wants to compute the probabil-
ing
12
ðtÞ no interference can be observed because the ities. Suppose that the (not normalized) amplitude to detect a
pattern is washed out. In the case of E1 ¼ E2 , one has photon at x is given by
maximal possible interference. This can be quantified in Ab1 ðx; tÞ ¼ ei½k1 ðx
b1 Þþx;b1 ðtÞ (7)
terms of the interferometric contrast, or visibility, V ¼
ðImax
Imin Þ=ðImax þ Imin Þ, which in the aforementioned if it originates from point b1 , and by
case equals 1. 0
The Hanbury-Brown and Twiss experiment introduced Ab2 ðx; tÞ ¼ ei½k1 ðx
b2 Þþx;b2 ðtÞ (8)
intensity correlation measurements to optics. Such correla-
tions between two points in space and two moments of time, if it originates from b2 . The quantum mechanical probability
for two classical fields, are described by an intensity corre- density that a particle is detected at x is given by
lation function Pðx; tÞ  jAb1 ðx; tÞ þ Ab2 ðx; tÞj2
Gð2Þ ðr1 ; t1 ; r2 ; t2 Þ ¼ hIðr1 ; t1 ÞIðr2 ; t2 Þiav : (3)  1 þ cos½
k1  x þ 0 þ x;b1 ðtÞ
x;b2 ðtÞ;
The average is taken over an ensemble, and for stationary (9)
fields this is equivalent to the temporal average. Even when
no intensity variations are observable (i.e., for averaged where
k1 ¼ k1
k01 , and 0 is an irrelevant constant
intensity constant in space), the intensity correlations can phase. Thus if the phase difference x;b1 ðtÞ
x;b2 ðtÞ is
reveal interference effects. Assume that the phases of the stable, one can have the Young-type interference patterns of
two fields fluctuate independently of one another. Then for up to 100% visibility. Such a stable phase difference can be
t1 ¼ t2 , the Gð2Þ function still exhibits a spatial modulation or observed with single photons in, e.g., a double-slit experi-
maximal visibility of 50% as exhibited by ment. Also, in the case of a classical wave description and
classical-like fields the observed intensity is proportional to
Gð2Þ ðr1 ; t; r2 ; tÞ ¼ ðI1 þ I2 Þ2 þ 2I1 I2 cos½ð
12 kÞðr1
r2 Þ; the probability density Pðx; tÞ, that is in this respect nothing
changes.
(4)
Nevertheless, the above description differs drastically from
where Ii ¼ E2i , i ¼ 1, 2. This formula can be easily reached the classical particle picture, in which one would expect that
by noting that the temporal average of a process originating with state A and with possible inter-
mediate stages B1 ; . . . BN , leading to an event C, would be
cos½ þ
12
ðtÞ cos½0 þ
12
ðtÞ; (5) described by
where
12
ðtÞ ¼
1 ðtÞ

2 ðtÞ is given by X
N
PðCjAÞ ¼ PðCjBj ÞPðBj jAÞ: (10)
cos cos0 hcos2
12
ðtÞiav þ sin sin0 hsin2
12
ðtÞiav j¼1

12 sinð þ 0 Þhsin2
12
ðtÞiav ; (6) In the quantum case PðCjAÞ ¼ jhCjAij2 , where
and due to the random nature of
12
ðtÞ only the first X
N
two terms survive because both hcos2
12
ðtÞiav and hCjAi ¼ hCjBj ihBj jAi; (11)
hsin2
12
ðtÞiav give 12 , whereas hsin2
12
ðtÞiav ¼ 0. j¼1

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

782 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

this means one sums over amplitudes not probabilities. For a As we shall see, the unbounded visibility is not the only
double slit, we have N ¼ 2, and hAjB1 i  hAjB2 i give the feature by which two-particle interference differs from the
amplitudes to reach the slits. Finally, one has hCjBj i ¼ Abj . classical one.
Please note that for classical particles the terms of the sum-
mation PðCjBj Þ etc., are real numbers, while in the quantum C. Quantum entanglement
case the amplitudes hCjBj i are complex numbers. Thanks to
that interference effects can be predicted. Entanglement according to Schrödinger (1935a, 1935b,
The difference between Eqs. (10) and (11) is in the as- 1935c, 1935d) contains ‘‘the essence of quantum mechanics.’’
sumption, inherent in Eq. (10), that the particle had to be in Consider a spin-0 particle which decays into two spin-1=2
one of the intermediate situations (states) Bi . In the quantum particles (Bohm, 1951). The quantum state is such that along
case any attempt to verify by measurement5 which of the any chosen direction, say the z axis, the spin of particle 1
situations actually took place puts one back to the classical when measured can be either up or down, which in turn, by
formula (10). The formula (11) leads to interference phe- angular momentum conservation, implies that for particle 2 it
nomena, and may be thought as a manifestation of a wave is, respectively, down or up. The state of the two spins is the
nature of quantum particles, whereas, if we make measure- rotationally invariant singlet
ments discriminating events Bi , we learn by which way
(welcher weg) the particles travel. The ‘‘which-way informa- 1
tion’’ is a clear signature of the particle nature. j c i12 ¼ pffiffiffi ðj "i1 j #i2
j #i1 j "i2 Þ; (15)
2

2. Two-particle quantum interference

where j "i1 (j #i1 ) describes the state of particle 1 with its spin
All this becomes much more puzzling once we consider a up (down) along the z direction. The minus sign is necessary
two-particle experiment. Already Einstein, Podolsky, and to get the rotational invariance. The state describes a coherent
Rosen (1935) pointed out some strange features of such a superposition of the two product states: there is no informa-
case. Schrödinger (1935a) noticed that these features are tion in the whole Universe on which of the two possibilities
associated with what he called ‘‘entangled states,’’ which will be detected at the measurement stage. None of those two
will be discussed in detail later. Consider now these as super- possibilities is the actual case. Actualization can happen only
positions of fully distinguishable products of single-particle via a measurement. This superposition, like any other, e.g., in
states, i.e., that there is a specific pair of local measurements the double-slit experiment, survives as long as no measure-
for which the results of the respective measurements on the ment actualizing one of those possibilities is performed, and
two subsystems are perfectly correlated (a result of a mea- any possible interaction of the particles with an environment
surement on one subsystem reveals the unique value of the does not destroy it. While none of the two possibilities
corresponding observable for the second subsystem). actually can be assigned without measurement, both of
Consider such a correlation: assume that if particle 1 is at them affect the predictions for all measurements. Another
b1 , then particle 2 is also at b1 , and whenever 1 is at b2 then property of the state (15) is that it does not make any
particle 2 is at b2 . Later on the particles are detected at two prediction about the result of spin measurement on only
different points, x1 and x2 . Then, according to the rules given one of the two particles: the result is random. The spin state
above of one of the particles is described by a reduced density
operator,6 which is a totally random state 12 Ik , where k ¼ 1,
Pðx1 ; x2 ; tÞ  jAb1 ðx1 ; tÞAb1 ðx2 ; tÞ þ Ab2 ðx1 ; tÞAb2 ðx2 ; tÞj2 2 is indexing the subsystems, and Ik ¼ j "ikk h" j þ j #ikk h# j is
 1 þ cosð
k1  x1 þ
k2  x2 þ
b1 ;b2 Þ; the unit operator for a given subsystem. All information
contained in the state in Eq. (15) defines only joint properties
(12) (Schrödinger, 1935d). The joint property can be put as fol-
lows: the two spins, if measured with respect to the same
with
direction, will be found opposite. As a matter of fact, this

b1 ;b2 ¼
x1 ;b ðtÞ þ
x2 ;b ðtÞ þ 0o ; (13)
6
Reduced density matrices (fully) describe states of subsystems.
where the amplitudes for the second particle are given by
Take two subsystems, 1 and 2. The average of any observable of the
formulas (7) and (8), with x2 replacing x1 and k2 replacing
composite system, say A^ 12 , is given by Tr12 ðA^ 12 %12 Þ, where %12 is
k1 , and
xi ;b ðtÞ ¼ xi ;b1 ðtÞ
xi ;b2 ðtÞ, with i ¼ 1, 2. Thus,
the density matrix of the full system 12. If the observable acts only
if the phase relation between the two amplitudes is stable, one on system 2, that is, it is of the form A^ 12 ¼ I^1  B^ 2 , where I^1 is the
can have absolutely noiseless interference with 100% visibil- unit operator for system 1, one has Tr12 ð%12 I^1  B^ 2 Þ ¼
ity, while there is no single-particle interference Tr2 ðB^ 2 Tr1 %12 Þ. This relation holds because, to calculate the trace,
Z one can always use a basis which consists of tensor products of basis
Pðx1 ; tÞ ¼ dx2 Pðx1 ; x2 ; tÞ ¼ const: (14) states of the two subsystems, jai i1 jbj i2 , with ranges of the indices i,
j defined by the dimensions of the subsystems (trace is basis
independent). Thus it is evident that the average is effectively
5
Or by securing the possibility of a postponed measurement defined by %2 ¼ Tr1 %12 , which is the density matrix describing
(which could be made by correlating our particle, while it is at all predictions concerning system 2 alone (disregarding its possible
the intermediate stage Bi , with another system, which could be correlations with system 1). This is the reduced density matrix of
measured later). system 2.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 783

property for any pair of complementary measurement settings Every pure state of two spins 1=2 can be put into the
fully defines the singlet state. following form:
Imagine that the two particles can be separated far apart,
one in Alice’s laboratory and the other one in Bob’s. As soon cosj "i1 j "i2 þ sinj #i1 j #i2 ;
as Alice measures the value of a spin projection along some where the states j "in and j #in , n ¼ 1, 2, are the eigenstates of
axis, new information is gained, and for her the state of Bob’s the zðnÞ  ðnÞ operator. The unit vectors zðnÞ are individually
particle is a well-defined pure one. This is independent of defined for each of the observers’ particles. They define the
the spatial separation between them. Thus a state like basis for the Schmidt decomposition for each of the
Eq. (15) is a perfect case to study (and to reveal) the EPR subsystems.
paradox.7 Basically, all earlier studies of entanglement con- The theory of entanglement of mixed states is more com-
centrated on entangled states of spins 1=2 or photonic polar- plicated. A state (pure or mixed) described by a density
izations; for a review, see Clauser and Shimony (1978). Much matrix AB of a composite quantum system consisting of
later, we saw an emergence of research on entanglement of two subsystems A and B is separable if and only if AB is a
three or more particles, practically together with the advent of convex combination of products densityPmatrices A and B
quantum information. of the two subsystems, namely, AB ¼  p A  B , where
Theoretical methods for entanglement analysis: The most P
p  0 and  p ¼ 1. Otherwise, AB is entangled (Werner,
important tool for the analysis of pure states of two subsys- 1989). For composite systems of more than two subsystems
tems is the so-called Schmidt decomposition. For any pure this definition can be generalized in a straightforward way.
state ji of a pair of subsystems, one described by a Hilbert Basic structural criteria, which decide whether a given
space of dimension N, the other by a space of dimension M, density operator represents an entangled state, were first
say N  M, one can find preferred bases, one for the first given by Peres (1996) and in the full form by Horodecki,
system, the other one for the second, such that the state is a Horodecki, and Horodecki (1996). The full set of separable
sum of bi-orthogonal terms, i.e., mixed states is a bounded convex set in a multidimensional
real space of Hermitian operators. Thus, any entangled state
X
N
ji ¼ ri jai i1 jbi i2 (16) is separated from the set of separable states by a hyperplane.
i¼1 The equation of such a hyperplane is defined by an element of
the space, namely, a Hermitian operator W, ^ which is called an
with nhxi jxj in ¼ ij , for x ¼ a, b and n ¼ 1, 2. The coef- ‘‘entanglement witness’’ (Horodecki, Horodecki, and
ficients ri are real and positive. The appropriate single sub- Horodecki, 1996; Lewenstein et al., 2000; Terhal, 2000;
system bases, here jai i1 and jbj i2 , depend upon the state. A Bruß et al., 2002; Bourennane, Eibl, Kurtsiefer, et al.,
proof8 of the Schmidt decomposition can be found in Peres 2004). Since the scalar product in the operator space is given
(2002). A generalization of the Schmidt decomposition to by TrðA^ y BÞ,
^ the equation of a hyperplane in the space is
more than two subsystems is not straightforward; see, e.g., given by TrðW%Þ ^ ¼ const. A Hermitian operator W^ is an
Carteret, Higuchi, and Sudbery (2000). It is easy to show that entanglement witness if for all separable states one has
the two reduced density matrices of Eq. (16) are endowed TrðW%^ sep Þ  0, whereas there exists an entangled state for
with the same spectrum. This does not hold for three or more which one has TrðW% ^ ent Þ < 0. Thus, via measurement of a
particle subsystems. suitably chosen witness operator, one can detect entanglement.
The original idea of Peres was the observation that pos-
7
Following Bohm (1951), one could apply the EPR reality itivity of a partial transposition (PPT) of a density matrix (i.e.,
criterion to the singlet state (15): ‘‘If, without in any way disturbing its transposition for just one subsystem) is a necessary con-
a system, we can predict with certainty (i.e., with probability equal dition for a state to be separable. This was extended by the
to unity) the value of a physical quantity, then there exists an Horodecki family to a fully general necessary and sufficient
element of physical reality corresponding to this physical quantity.’’ condition for separability, which is that a density matrix after
This would imply that to any possible spin measurement on any one any positive transformation (map) on one of the subsystems
of our particles we can assign such an element of physical reality on should remain a density operator.9 The spin offs of such
the basis of a corresponding measurement on the other particle. considerations are measures of entanglement via the negativ-
Whether or not we can assign an element of reality to a specific spin ity of the eigenvalues of a partial transpose of the density
component of one of the systems must be independent of which matrix, etc. Other methods that give a quantitative measure
measurement we actually perform on the other system and even of the degree of entanglement of bipartite entangled states
independent of whether we care to perform any measurement at all include the entanglement of formation (Bennett, DiVincenzo,
on that system. This approach was shown to be leading to a class of Smolin, and Wootters, 1996; Wootters, 1998a), concurrence
theories incompatible with quantum mechanics (Bell, 1964). The (Hill and Wootters, 1997), and tangle (Wootters, 1998b;
concept of elements of reality was shown to be strictly self- Coffman, Kundu, and Wootters, 2000). For more
contradictory via the GHZ theorem (see further on).
8
The crux of the proof is that the greatest of the coefficients is
given by Maxjai1 jbi2 jhjai1 jbi2 j and after finding it and the states 9
Partial transposition is a positive operation but is not ‘‘com-
that give the maximization, say ja1 i1 and jb1 i2 , one searches for the pletely positive,’’ while, e.g., the most general quantum evolution of
second greatest coefficient by performing maximization over the a subsystem is always represented by a ‘‘completely positive map,’’
linear subspace to which ja1 i1 and jb1 i2 do not belong. This is as such map leads from one density matrix to another one for the
recursively continued to get next coefficients and basis states, till the compound system even if the subsystem is entangled with another
procedure halts. one.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

784 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Rarity and Tapster, 1990), whose creation will be described in

Sec. IV. Before being combined at a 50:50 beam splitter (BS)
and then subject to single-photon detections, the two paths of
each photon acquire a relative phase shift.
By taking into account the phase shifts
1 and
2 and the
action of the two beam splitters, the probabilities of the
coincidence detections of two photons at the detector pairs
ðD1c=d ; D2c=d Þ read
FIG. 1. A two-photon interferometer with variable phase shifts
1
and
2 . Before being combined at the 50:50 beam splitter (BS) and p1c;2d ð
1 ;
2 Þ ¼ p1d;2c ð
1 ;
2 Þ
then subject to single-photon detections, the two paths of each
photon acquire a relative phase shift. For experimental realization of ¼ 14½1 þ cosð
1

2 Þ;
such a two-photon interferometer, see Fig. 11. Adapted from Horne, p1c;2c ð
1 ;
2 Þ ¼ p1d;2d ð;
2 Þ ¼ 14½1
cosð
1

2 Þ:
Shimony, and Zeilinger, 1989.
(18)

details on the entanglement theory, see Alber et al. (2001), Thus, by simultaneously monitoring the detectors on both
Mintert et al. (2005), Gühne and Toth (2009), and Horodecki sides of the interferometer, while varying the phase shifts
1
et al. (2008). and
2 , the interference fringes will be observed as shown
An interesting feature of the theory of entanglement of by the sinusoidal terms. In contrast, for any single detector
mixed states is that for two three-dimensional systems, or for the count rate shows no interference at all. For example,
more complicated ones, one can find states which are en- p1c ¼ p1c;2c ð
1 ;
2 Þ þ p1c;2d ð
1 ;
2 Þ ¼ 12 , independent of
tangled, but from which no maximally entangled state can be
1 and
2 .
distilled10 (Horodecki, Horodecki, and Horodecki, 1998).
Such states are called bound entangled. 2. Greenberger Horne Zeilinger interferometry
After many years of studying only two-particle entangle-
D. Interferometry with entangled two-photon ments, in 1989 a generalization of the EPR interferometry to
and multiphoton states three photons was proposed [Greenberger, Horne, and
Zeilinger (1989), later refereed to as GHZ]. The most ele-
Entanglement can manifest itself in strictly quantum inter-
mentary case is shown in Fig. 2. Though such a step from 2 to
ference phenomena, that is, these phenomena can neither be
3 seems to be small, it nevertheless leads to profound impli-
explained by a classical wave nor by a classical particle
cations, one of which is the GHZ theorem (Greenberger,
picture. To show this, below we present the basics of multi-
Horne, and Zeilinger, 1989; Greenberger et al., 1990;
particle interferometry.
Mermin, 1990a).
At the center of the interferometer is a source emitting
1. EPR interferometry
three photons in a so-called GHZ-entangled state
Recall that in a single-particle interferometer, as in 1
Young’s double-slit experiment, the interference pattern ap- jGHZi123 ¼ pffiffiffi ðjai1 ja0 i2 ja00 i3 þ jbi1 jb0 i2 jb00 i3 Þ: (19)
pears only if the particle’s two paths are indistinguishable. 2
However, for interferometers involving two or more particles, Here each photon has two different spatial modes, which are
dramatically new features arise. Figure 1 is a sketch of the for photon 1 jai and jbi. By taking into account the actions of
generalization of the concept of a Mach-Zehnder interfer- the relative phase shifts and the 50:50 beam splitters (for their
ometer to two-photon interferometry (Horne and Zeilinger,
1986; Żukowski and Pykacz, 1988; Horne, Shimony, and
Zeilinger, 1989; Horne, Shimony, and Zeilinger, 1990;
Rarity and Tapster, 1990; Greenberger, Horne, and
Zeilinger, 1993). We assume that a central source emits two
photons in an entangled state
1
j c i12 ¼ pffiffiffi ðjai1 ja0 i2 þ jbi1 jb0 i2 Þ: (17)
2
Here jai and jbi (ja0 i and jb0 i) are two different spatial modes
of photon 1 (photon 2). The entanglement of j c i12 is actually
called momentum entanglement (Horne and Zeilinger, 1986;

10
Distillation is the process in which two or more parties obtain
some amount of maximally entangled states out of a more numerous
set of copies of less entangled states by making only local opera-
tions and classical communication (Bennett, Bernstein, Popescu, FIG. 2. A three-photon interferometer with variable phase shifts
and Schumacher, 1996; Bennett, DiVincenzo, Smolin, and
1 ,
2 , and
3 . Adapted from Greenberger, Horne, and Zeilinger,
Wootters, 1996; Bennett et al., 1996). 1989.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 785

properties see Sec.II.E), one can deduce novel features of ment, see standard textbooks on quantum optics, e.g., Walls
three-particle interference (Greenberger et al., 1990). First, and Milburn (1994), Mandel and Wolf (1995), Scully and
the respective threefold coincidence detection probability for Zubairy (1997), and especially Bialynicki-Birula and
the output modes [ð1c; 2c; 3cÞ; ð1d; 2d; 3dÞ; etc.] reads Bialynicka-Birula (1975).
A single-photon pure state can be characterized by a
p1c;2c;3c ð
1 ;
2 ;
3 Þ ¼ 18½1 þ sinð
1 þ
2 þ
3 Þ; specific wave packet profile gk , i.e., by the quantum ampli-
p1d;2c;3c ð
1 ;
2 ;
3 Þ ¼ 18½1
sinð
1 þ
2 þ
3 Þ; etc: tudes for a given momentum k and polarization . According
(20) to the Born rule, jgk j2 gives the probability density of having
the single photonP with theR momentum ℏk and polarization .
The threefold coincidence rates given in Eq. (20) display Thus, one has ¼1;2 dkjgk j2 ¼ 1. The wave packet
sinusoidal oscillations depending on the sum
1 þ
2 þ
3 . profiles are vectors in a Hilbert space with a scalar product
Second, this three-particle interferometer does not exhibit given by
any two-particle fringes. For example, if only two-particle X Z
coincidences 2c-3c are detected, while the modes 1c and hgjhi ¼ dkg k hk : (21)
1d are ignored, the observed rate will be constant ¼1;2
p1c;2c;3c ð
1 ;
2 ;
3 Þ þ p1d;2c;3c ð
1 ;
2 ;
3 Þ ¼ 14 , and com-
One can introduce an arbitrary orthonormal basis glk ,
pletely independent of the phases.11 Finally, a similar argu-
where l are natural numbers and hgn jgm i ¼ nm . Two differ-
ment shows that, of course, no single-particle fringes can be
ent orthonormal bases, to be denoted, respectively, as primed
observed. Actually, an n-particle interferometer generalized
and
P1 unprimed, are related by a unitary operation: g0m ¼
along the above reasoning will only exhibit n-particle fringes, l
Uml g . The complex numbers Ulm satisfy
but no ðn
1Þ
; ðn
2Þ
; . . . , single-particle fringes Pl¼1
1 U U ¼ P1 U U ¼  . One can choose a spe-
(Greenberger et al., 1990). l¼1 ml kl l¼1 lm lk mk
cific basis of the wave packet profiles of the single photon,
The above two-photon and three-photon interferometry
say gl , and with each element of such a basis one associates a
was described using the photon’s path degrees of freedom.
quantum oscillatorlike construction to introduce number
However, similar interference effects can be observed using
states, namely, the Fock states. One introduces the vacuum
any of the photon’s degrees of freedom, e.g., polarization.
state ji
j0; 0; 0; . . .i, the state with no photons at all for
Moreover, the above argument should be understood as a
any modes. Next, for a chosen basis one associates a pair of
special case of a wider concept indicating that entangled
operators satisfying the usual relations for creation and anni-
massive particles (e.g., electrons and atoms) could also dis-
play multiparticle interference. hilation operators, namely, ½a^ l ; a^ yl  ¼ 1 and requires that for
all l a^ l ji ¼ 0. Using the standard oscillator algebra, one
pffiffiffiffiffiffi
constructs states like ða^ ynl = nl !Þji, which is a state of the
l
E. Interferometry with multiport beam splitters
electromagnetic field in which one has nl identical photons of
Novel interferometric effects can be obtained with N-port the same wave packet profile gl , and no other photons what-
beam splitters, which are devices which split light into more soever. This is denoted by j0; . . . ; 0; nl ; 0; 0; . . .i. Finally, one
than two output beams [for a general theory of such devices, assumes that ½a^ n ; a^ m  ¼ 0 and ½a^ n ; a^ ym  ¼ nm , that is, crea-
see Reck et al. (1994)]. Such devices can be utilized in tion and annihilation operators of photons with orthogonal
multiparticle interferometry (Zeilinger et al., 1993). With wave packet profiles always commute. A general (normal-
current technology, such experiments are becoming feasible. ized) basis state of the Fock space is therefore of the follow-
ing form:
Y1 ynl
III. PHOTONIC QUBITS AND LINEAR OPTICS a^ l
jn1 ; n2 ; n3 ; . . .i ¼ pffiffiffiffiffi ji: (22)
l¼1 n!
The possibility of performing quantum information-
processing tasks with photons is based on the fact that All vectors of the Fock space are linear combinations of the
quantum information can be encoded in quantum states of above basis states, which have a finite total number of
certain degrees of freedom (e.g., polarizations) of individual photons. It is easy to see that if one defines the creation
photons, and that individual photons can be manipulated operators with respect to an alternative basis of wave packet
either by simple optical elements (e.g., wave plates and profiles (here the primed ones), one has
interferometers) or by letting them interact with matter X
1
(trapped ions, atoms, etc.) at an atom-photon interface. a^ 0y
m ¼ Umn a^ yn : (23)
Here we show to what extent a photon can carry a qubit, n¼1
and the simplest elements that are used to manipulate it. To The vacuum state is invariant with respect to such trans-
this end, we begin with a formal definition of photons and formation, i.e., one still has a^ 0m ji ¼ 0. For more details,
their quantum states. see Bialynicki-Birula and Bialynicka-Birula (1975).
The formal theory of quantization of electromagnetic fields
was formulated by Dirac (1927). Here we only give its basic
A. Photonic qubits
mathematical devices for completeness; for detailed treat-
A quantum bit, or qubit is the most elementary unit of
11
This holds only for observables dependent on
i , like those quantum information. It is a generalization of the classical bit,
shown in the Fig. 2. which has two distinguishable states ‘‘0’’ and ‘‘1.’’ Similarly,

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

786 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

we can have a qubit in two distinguishable, i.e., orthogonal arms of the interferometer. For more details, see Bennett
states j0i and j1i. However, in contradistinction to its classical (1992) and Gisin et al. (2002).
counterpart, a qubit can be prepared as, or transformed to any While in this review we are mainly concerned with the
superposition of these two states (normalization requires above three implementations of photonic qubits, one should
20 þ 21 ¼ 1) keep in mind that other implementations are possible.
Frequency qubits have first been implemented in quantum
jqubit i ¼ 0 j0i þ 1 j1i: (24) cryptography (Sun, Mazurenko, and Fainman, 1995;
Mazurenko, Giust, and Goedgebuer, 1997) and also more
Any isolated two-level system consisting of a pair of recently in entangled atom-photon systems (Madsen et al.,
orthogonal quantum states represents a qubit. Photons, mass- 2006).
less spin 1 particles, have only two eigenvalues of their spin Quantum d-level (high-dimensional) systems (‘‘qudit’’)
along the direction of their propagation (wave vector), ℏ. can also be realized using, e.g., orbital angular momentum
These two spin values correspond to right-handed and left- states of photons (Mair et al., 2001), or using simultaneously
handed circular polarization. Thus this property makes the two or more degrees of freedom listed above. For instance, for
photon an ideal candidate for a qubit. However, there are the latter case a polarized single photon in a coherent super-
other degrees of freedom of a photon that can be used to position of two spatial modes can be thought of as a quantum
encode qubit information. system in a four-dimensional Hilbert space (Boschi et al.,
Polarization qubits. The most commonly used photonic 1997; Michler, Weinfurter, and Żukowski, 2000; Chen et al.,
qubits are realized using polarization. In this case, arbitrary 2003; Kim, 2003).
qubit states can be 0 jHi þ 1 jVi, where H and V stand for Two-photon polarization-entangled states–The so-called
horizontal and vertical polarizations, respectively. The advan- Bell states14 form a basis in the four-dimensional two-qubit
tage of using polarization qubits stems from the fact that they Hilbert space. Bell states of photonic polarization qubits can
can easily be created and manipulated with high precision by be, for example,
simple linear-optical elements such as polarizing beam split-
ters (PBS), polarizers, and wave plates. Photon polarization 1
j c  i12 ¼ pffiffiffi ðjHi1 jVi2  jVi1 jHi2 Þ; (25)
states and spin states of a spin 1=2 particle are perfect qubits 2
given by nature, no human invention is required.
Spatial qubits–A single photon can also appear in two 1
j
 i12 ¼ pffiffiffi ðjHi1 jHi2  jVi1 jVi2 Þ: (26)
different spatial modes or paths a and b: the general state 2
reads 0 jai þ 1 jbi. This may occur if a single photon exits a
BS, with two output modes a and b. Any state of spatial As we shall see, such entangled states serve as a central
qubits can be prepared by using suitable phase shifters and physical resource in various quantum information protocols
BS. A disadvantage of using spatial qubits is that the coher- like quantum cryptography, quantum teleportation, entangle-
ence between jai and jbi is sensitive to the relative phase ment swapping, and in tests aimed at excluding hidden-
between the paths a and b, and this is difficult to control in variable models of quantum mechanics.
long-distance cases.
Time-bin qubits–For a more robust long-distance trans- B. Simple linear-optical elements
mission of quantum information, one may use time-bin qu-
bits. The computational basis12 consists of two states which In the photonic domain, quantum states of photons can be
are of the same spectral shape, but time shifted by much more easily, and with high precision, manipulated by simple pas-
than the coherence time.13 Time-bin qubits can be realized by sive linear-optical devices. These linear-optical elements in-
sending a single photon through an unbalanced Mach- clude BSs, PBSs, wave plates, and phase shifters. Classically,
Zehnder interferometer. Its wave packet is split by the first such devices conserve energy: The total input energy equals
BS, with transmission coefficient T ¼ j0 j2 and reflection the total output energy, and there is no energy transfer
coefficient R ¼ j1 j2 into two coherent parts. The transmit- between different frequencies. A passive linear-optical device
ted one propagates along the short arm, and the reflected one is described by a unitary transformation of annihilation
along the long arm. If the wave packet extension is shorter operators for the same frequency
than the arm length difference, the output from the ports of X
the second, 50:50 BS is two wave packets well separated in m ¼
a^ out Umn a^ in
n; (27)
m
time. If no photon is registered in, say output port I, in the
other output one has a single photon in a coherent superpo- where U is a unitary matrix, and the indices denote a basis of
sition of two time-bin states 0 jearlyi þ 1 jlatei. The phase orthogonal modes.
relation can be controlled with a phase shifter in one of the The BS is one of the most important optical elements. It
has two spatial input modes a and b and two output modes c
12
and d (Fig. 3). The theory of the lossless BS was developed
A basis of a qubit is called computational if one associates by Zeilinger (1981) and Fearn and Loudon (1987), for
logical 0 and 1 to its two orthogonal states
13
The coherence time is the time over which the relative phase of a
14
propagating wave remains stable. It can be approximately estimated Please note, that the term ‘‘Bell state’’ was earlier used with a
2
as   c
 , where  is the central wavelength of the source,
 is completely opposite meaning. E.g., Mann, Revzen and Schleich
the spectral width of the source, and c is the speed of light in (1992) define a Bell state as ‘‘a pure state which when split in any
vacuum. way cannot violate Bell’s inequality.’’

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 787

FIG. 3. The function of a BS. (a) The BS coherently transforms

two input spatial modes (a), (b) into two output spatial modes (c),
(d). (b) A Mach-Zehnder interferometer consists of two 50–50 BSs,
mirrors and phase shift (as a whole it forms a universal tunable
beam splitter).
FIG. 4. The operation of a polarizing beam splitter (PBS). (a) In
the usual configuration, the PBS transmits horizontal, and reflects
lossy BS; see Barnett et al. (1998). In a simplified theory of vertical, polarization. (b) Two photons, each entering via a different
BS, one assumes identical action for every relevant fre- input: If the two photons incident onto the PBS have identical
polarization, then they will always go out along different directions,
quency. The most commonly used BS is the symmetric
so there will be one photon in each of the two output modes. (c) If
50:50 BS characterized by the following transformation:
the two incident photons have opposite polarizations, they will
1 i ^ i 1 ^ always go out along the same direction, so there will be two photons
a^ ! pffiffiffi c^ þ pffiffiffi d; b^ ! pffiffiffi c^ þ pffiffiffi d: (28) in one of the two outputs and none in the other. In essence, a PBS
2 2 2 2
can thus be used as a polarization parity checker. From Pan et al.,
In such a case, an outgoing particle can be found with equal 1998; Pan, Simon, Brukner, and Zeilinger, 2001.
probability (50%) in either of the output modes c and d, no
matter through which single input beam it came. The factor i
in Eq. (28) is a consequence of unitarity. It describes a phase (1996) realized an all-fiber three-path Mach-Zehnder inter-
jump upon reflection (Zeilinger, 1981). ferometer, which is based on the idea of symmetric unbiased
Using two 50:50 BS and a phase shifter, one can build a multiport BS (Zeilinger et al., 1993).
Mach-Zehnder interferometer, which in fact can be used as a
universal BS of a variable transmitivity and reflectivity [see C. Two-photon interference due to indistinguishability
Fig. 1(b)]. The phase shifter can be some glass plate, a of photons
birefringent optical crystal, or a path-length tuner (e.g., an
optical trombone). Together with additional phase shifters in Quantum interference15 may also occur entirely due to
the input and output ports, the Mach-Zehnder interferometer indistinguishability of particles. We describe this phenome-
can perform an arbitrary SU(2) unitary transformation on a non in the case of photons, i.e., bosons. One can obtain all
qubit encoded in the two spatial modes [see, e.g., Englert, effects due to indistinguishability by a suitable symmetriza-
Kurtsiefer, and Weinfurter (2001)]. tion of the amplitudes for elementary processes, for which we
Another important component is the PBS (see Fig. 4). A do not know which particle ended up in which final state (for
standard PBS transmits the horizontal and reflects the vertical bosons amplitudes do not change sign when particles are
polarization. The transformations between the incoming interchanged). However, it is much more convenient to use
modes (a and b) and the outgoing modes (c and d) are as the formalism of bosonic creation and annihilation operators,
follows: as its algebra directly takes into account the symmetrization.
Here we present the most elementary optical effect due to the
a^ H ! c^ H and a^ V ! id^V ; indistinguishability of photons, the Hong-Ou-Mandel inter-
(29) ference (Hong, Ou, and Mandel, 1987) behind a 50:50 BS.
b^H ! d^H and b^V ! ic^ V :
As seen previously, upon reflection off a symmetric 50:50
For polarization qubits, any single-qubit operation can be BS a photon picks a phase shift i. We denote the input modes
accomplished by using a sequence of suitably oriented as a and b and output modes as c and d. If we have two
quarter-wave and half-wave plates. Simply, a half-wave plate spectrally identical photons (of the same polarization) each
retarding the 45 polarization acts on the H and V modes entering at exactly the same moment an opposite input port of
exactly as a certain 50:50 BS, etc. Now, the input modes are the BS, the initial state a^ y b^y ji is transformed into
two orthogonal polarizations, instead of two spatial input
modes of a Mach-Zehnder interferometer. 1 1 i
pffiffiffi ðc^ y þ id^y Þ pffiffiffi ðd^y þ ic^ y Þji ¼ ðc^ y2 þ d^y2 Þji:
With these simple optical elements, large optical networks 2 2 2
can be constructed, mapping an input state onto an output (30)
state via a linear transformation determined by the networks’
unitary transformation. As a possible generalization of the
usual interferometers with two ports, an N-port interferome- 15
See Mandel (1999) for a review on a series of pioneering
ter was proposed by Reck et al. (1994), which can realize any parametric down-conversion experiment revealing quantum effects
UðNÞ transformation for N optical spatial modes by using an in one-photon and two-photon interference. Our present review can
arrangement of BS, phase shifters, and mirrors. Weihs et al. be treated as a direct continuation of Mandel’s review.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

788 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

1 1
pffiffiffi ðc^ y1 þ id^y1 Þ pffiffiffi ðd^y2 þ ic^ y2 Þji: (31)
2 2
Since c^ 1
c^ 2 and d^1
d^2 , the terms that contribute to
the cases in which each photon exits by a different exit
port, namely 12 ðc^ y1 d^y2
d^y1 c^ y2 Þji, do not cancel with each
other.
We use to denote the degree of distinguishability be-
tween photon 1 and 2. The probability of finding a coinci-
dence count at exits c and d, which is given by the square
of the norm of 12 ðc^ y1 d^y2
d^y1 c^ y2 Þji, is ð1=2Þj j2 . Thus, if
j j ¼ 1 (the photons a^ and b^ are fully distinguishable), this
probability reads 1=2; if ¼ 0 (the photons are indistin-
guishable), it vanishes. Therefore, the Hong-Ou-Mandel ef-
fect depends critically on the distinguishability of photons.
The distinguishability was tuned in the original experiments
with the temporal delay between the two photons [Fig. 5(b)].
The original Hong-Ou-Mandel experiment used the two
photons of the same signal-idler pair from parametric down-
FIG. 5. (a) Individually incoming photons can be transmitted or conversion (see Sec. IV.A). Later experiments of this kind
reflected. However, since the two photons are identical, we cannot evolved into observations of a Hong-Ou-Mandel dip for
distinguish between the two cases when both are transmitted or both photons originating from two sources, which were progres-
reflected. The BS introduces a phase difference of  between the sively more and more independent of each other. This was
two amplitudes describing such possibilities. This leads to a de- motivated by both fundamental issues, such as whether
structive interference. Thus, no coincidence detection can be found. independent photons indeed interfere, and practical ones;
(b) Data from Hong, Ou, and Mandel (1987): Pairs of photons interference of photons emerging from different sources
impinging on a BS are produced by spontaneous parametric down- must be harnessed if one wants to build complicated schemes
conversion (see Sec. IV.A) and have the same polarization and
realizing quantum protocols, e.g., quantum repeaters
frequency distributions. The measured number of coincidence
(see Sec. VI.F) and distributed quantum computing (see
counts as a function of relative path displacement (temporal dis-
Sec. VII). Rarity, Tapster, and Loudon (1996) observed the
tinguishability) shows a ‘‘Hong-Ou-Mandel’’ dip, for equal optical
paths. From Hong, Ou, and Mandel, 1987. interference between independent photons, one of which was
a triggered single photon from a down-converted pair, and the
other one was in an attenuated beam17 derived from the
The two terms, which describe the cases in which each pumping laser light. Interference of two triggered single
photon exits by a different exit port cancels each other, photons created via parametric down-conversion by the
c^ d^
d^ c^ ¼ 0. This cancellation occurs if the two photons same pump pulse passing twice through a nonlinear crystal
are perfectly indistinguishable in terms of all of their other was achieved in the Innsbruck teleportation experiment
degrees of freedom such as frequency, time, or polarization. (Bouwmeester et al., 1997); for more details see Sec. VI.B.
The two photons exit the BS paired via only one (random) With a similar method of triggering, Keller, Rubin, and Shih
output port. This is a bunching effect due to the bosonic (1998) used photons generated by two mutually coherent but
character of photons. Thus, there are no coincidences be- time-separated pulses from the same mode-locked laser.
tween the output ports.16 Another (and more graphical) way Experiments aimed at observing the Hong-Ou-Mandel dip
to look at this is shown in Fig. 5(a). for fully independently emitted photons, like the one of
If the photons are at least partially distinguishable (in this Kaltenbaek et al. (2006), will be described in Sec. VI.C.
case we label the annihilation operators of the photons with The Hong-Ou-Mandel interference provides a powerful
different subscripts as a^ 1 and b^2 ), the initial state a^ y1 b^y2 ji tool to estimate the degree of indistinguishability of two
is transformed by a perfect 50:50 BS via the relations in separately emitted photons. For instance, two single photons
Eq. (28) into successively emitted from a single quantum dot or a single
trapped atom were overlapped on a BS and the Hong-Ou-
Mandel effect was observed (Santori et al., 2002; Legero
16
If the two incident photons are in an antisymmetric polarization- et al., 2004). Other examples include single photons from
entangled state j
i, the amplitudes for photons to exit via different independent trapped atoms and ions (Beugnon et al., 2006;
ports interfere constructively, as in this case their spatial wave Maunz et al., 2007), and from remote cold atomic ensembles
function has to be antisymmetric, too (Weinfurter, 1994). For proper (Yuan et al., 2007) or independent, tunable quantum dots
path length, one has a coincidence peak (instead of a dip) (Mattle (Patel et al., 2010). The interference of indistinguishable
et al., 1996). This observation is crucial for discriminating the j
i photons enables a process called entanglement swapping and
using a BS (see the ‘‘Bell-state analyzer’’ Sec. III.E). For a color (at
frequency !1 and !2 ) entangled state, one can further observe
17
‘‘spatial quantum beating,’’ where the two-photon detection exhibits The beam was in a pulsed weak coherent state: for each pulse
a modulation of the form cosð!1
!2 Þ [see, e.g., Ou and Mandel there was only a very small probability for it to contain a single
(1988a)]. photon.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 789

teleportation (Żukowski, et al., 1993, see Sec. VI.C), which

in turn opens up prospects distributing of entanglement be-
tween distant matter qubits such as ions (Moehring et al.,
2007) and atomic ensembles (Yuan et al., 2008).
A characteristic feature of the Hong-Ou-Mandel interfer-
ence is that it is sensitive to path length changes on the order
of the coherence length of the photons, while in a Mach-
Zehnder interferometer one has subwavelength sensitivity.
The photon’s coherence length can range from a few hundred FIG. 6. (a) Shih-Alley type polarization entanglement. One com-
micrometers, in pulsed parametric down-conversion, to a few bines two orthogonally polarized photons at a BS and registers the
meters for trapped ions. This makes the Hong-Ou-Mandel coincidence events behind it. (b) In a similar way, by superposing
interferometers very stable. two (H þ V) polarized photons at a PBS, one can observe an
entangled state jHic jHid þ jVic jVid in the postselected coincidence
counts. From Ou and Mandel, 1988b; Shih and Alley, 1988.
D. Postselection of entanglement and quantum erasure

We now discuss other tricks and states obtainable with a

single BS on which a pair of photons impinges. coincidence detection between the two outputs can originate
Note first that according to Eq. (30), in a Hong-Ou-Mandel only if either both photons are transmitted (resulting in a
interferometer, the electromagnetic field emerging from the jHHi case) or both are reflected (jVVi case). The two cases
BS associated with the pair of photons is in an entangled state are quantum mechanically indistinguishable, if the photons
½ðc^ y Þ2 þ ðd^y Þ2 ji ¼ j2c 0d i þ j0c 2d i, a so-called N00N are indistinguishable. Thus again, an entangled state j

icd
state (see Sec. IV.C), with N ¼ 2. This state has interesting can be generated using postselection. As in the case of the
interferometric properties which will be discussed later. Hong-Ou-Mandel interferometer, the entanglement quality is
With the simple device, one can also produce effects sensitive to optical path length changes of the order of the
characteristic for a maximally entangled polarization state. photons’ coherence length. Such interferometers play a cru-
Ou and Mandel (1988b) and Shih and Alley (1988) were the cial role in creation, manipulation, and projection of various
first to achieve this. In their experiments, the two photons multiphoton entangled states.
entering the BS, each by a different input port, were indis- When analyzing the above experiments, one could be
tinguishable in all degrees of freedom, except that they were mislead to suppose that the interference arises due to the
oppositely polarized [say, one horizontally polarized H and fact that the wave packets of the two photons overlap at the
the other vertically polarized V, see Fig. 6(a)]. After being BS. Indeed the Hong-Ou-Mandel dip is presented as a func-
superposed on the BS and leaving the two output ports c and tion of the temporal delay between the two photons, i.e.,
d, the photons emerged in a polarization-tagged two-photon effectively in terms of the overlap [Fig. 5(b)]. However, it
state is important to note that essentially the origin of this inter-
pffiffiffi ference is due to the indistinguishability of two-photon am-

þ þ
2 2j c icd þ iðj
icc þ j
idd Þ:
½
1 plitudes describing the various alternatives leading to a
coincidence count. To dispel a misconception that the pho-
A coincident detection at the two outputs c and d can occur tons must arrive at the BS at the same time for some type of
only due the entangled j c
icd component of the total state. classical local ‘‘agreement,’’ Pittman et al. (1996) performed
As in the case of the Hong-Ou-Mandel interferometer, the a ‘‘postponed compensation’’ two-photon Shih-Alley type
visibility of this polarization interference is reduced once the experiment. Interference is observed, even though the pho-
photons are partially distinguishable. For example, by vary- tons were arriving at the BS at different times. In the experi-
ing the relative optical paths before the photons reach the BS, ment, the optical path of one input light beam was much
the overlap of the photon wave packets behind the beam longer than of the other one with a difference exceeding the
splitter may be controlled, up to a total distinguishability. photon’s coherence length. However, after the interferometer,
The full effect occurs for perfect overlap. This kind of post- this delay was compensated (using a polarization-selective
selected entanglement was used to violate Bell’s inequality,18 unbalanced Mach-Zehnder interferometer) in such a way that
in the experiments of both Ou and Mandel (1988b) and Shih the firing times of the two detectors did not provide any
and Alley (1988). The effect was later demonstrated for two information whatsoever concerning which of the two-photon
indistinguishable photons from quasi-independent sources processes led to the coincidence detection. The experimental
(Pittman and Franson, 2003; Fattal et al., 2004). results confirmed that the quantum interference can indeed be
Figure 6(b) shows another setup for generating entangle- revived. A discussion of related ‘‘quantum erasure’’ experi-
ment, in a similar spirit, however using a PBS [such an effect ments can be found in Scully, Englert, and Walther (1991);
was used in a proposal for obtaining multiphoton entangle- Kwiat, Steinberg, and Chiao (1992); and Michler et al.
ment by Zeilinger et al. (1997); see Sec. IV.D]. As a PBS (1996).
customarily transmits H and reflects V polarization, a
E. Entangled-state analyzers
18
There was a controversy whether such a kind of experiment can
constitute a true Bell test. It was positively resolved in Popescu, The projection of two photons into a Bell state lies at the
Hardy, and Żukowski (1997). heart of many quantum information-processing protocols,

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

790 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

such as quantum dense coding (Bennett and Wiesner, 1992),

quantum teleportation (Bennett et al., 1993), and entangle-
ment swapping (Żukowski, et al., 1993). A deterministic
controlled-NOT (CNOT) gate19 would make such a the Bell-
state measurement possible. However, CNOT gates are diffi-
cult to realize with linear optics and single photons (see
Sec. VII). Nevertheless, by exploiting quantum interference
effects due to the bosonic nature of photons discussed above,
photonic Bell-state and GHZ-state analyzers can be realized
in a probabilistic way.

1. Bell-state analyzer
A linear-optical Bell-state analyzer was suggested by
Weinfurter (1994) and Braunstein and Mann (1995). It is
based on the two-photon interference effect at a 50:50 BS
and via a coincidence analysis can distinguish two of the four
Bell states. As shown in Fig. 7(a), the setup consists of a BS
followed by two-channel polarizers in each of its output
beams. As the polarization state j c
i is antisymmetric, it
results in a coincidence detection at the outputs of the BS FIG. 7. (a) A Bell-state analyzer using a BS. (b) A modified Bell-
(i.e., a coincidence at detectors DH1 and DV2 or at DH2 and state analyzer. The angle between the half-wave plate axis and the
DV1 ). In fact, the state j c
i can be encoded with any degree horizontal direction is 22.5. It corresponds to a 45 rotation of the
of freedom [e.g., color encoding, see Moehring et al. (2007)] polarization. (c) A GHZ-state analyzer. Adapted from Pan et al.,
and can be pinpointed by a two-channel coincidence behind a 1998.
BS. This property can be easily checked by reversing the
action of the Shih-Alley interferometer. The minus sign in pffiffiffi
j c
i leads to a cancellation of the photon bunching ampli- ð1= 2ÞðjHia jHib
jVia jVib Þ, we observe coincidence at
tudes, that is to ‘‘fermionic-like’’ behavior. For the remaining detectors DH1 and DV2 or DV1 and DH2 . The other two Bell
three states, both photons exit via the same output port of the states would lead to no coincidence at detectors in modes
BS. The state j c þ i can be distinguished from the other two 1 and 2.
by the fact that the emerging photons have different polar- Finally, we mention that by taking advantage of the prop-
izations.20 This results in a coincidence counts at detectors erties of hyperentanglement one can implement a complete
DH1 and DV1 or at DH2 and DV2 . The two states jþ i and Bell-state analysis (Kwiat and Weinfurter, 1998; Walborn,
j
i both lead to a two-photon event at a single detector, and Pádua, and Monken, 2003). Such a scheme was experimen-
thus cannot be distinguished. tally realized (Schuck et al., 2006) and was used to beat the
A modified version of a Bell-state analyzer, which can channel capacity limit for linear photonic superdense coding
directly be generalized to the N-particle case, was introduced [Barreiro, Wei, and Kwiat (2008), see Sec. VI.A for more
by Pan and Zeilinger (1998). Consider the arrangement of details on dense coding].
Fig. 7(b). Two spectrally identical photons enter the Bell-state
analyzer by modes a and b. Assume that they arrive at PBSab 2. GHZ-state analyzer
simultaneously, so that their wave functions overlap behind it.
The properties of a PBS depicted in Fig. 4, and a coincidence Bell-state measurement schemes can be generalized to the
detection in modes 1 and 2, allow one to distinguish j
þ i12 N-particle case. One can construct a GHZ-state analyzer (Pan
and j

i12 polarization
pffiffiffi Bell states. Specifically, for the and Zeilinger, 1998), with which one can identify two out of
incident state ð1= 2ÞðjHia jHib þ jVia jVib Þ we always ob- the 2N maximally entangled GHZ states.
serve a coincidence between either detectors DH1 and DH2 or In the case of three spectrally indistinguishable identical
DV1 and DV2 . On the other hand, if the incident state is photons, eight maximally entangled polarization GHZ states
could be given by
19 1
The quantum CNOT gate, a fundamental quantum circuit, is a j
0 i ¼ pffiffiffi ðjHijHijHi  jVijVijViÞ; (32)
two-qubit gate acting on a target qubit jit and a control qubit j ic . 2
It flips the target qubit (j0it ! j1it , j1it ! j0it ) if the control qubit
is in logic j1ic , and does nothing if the control p ffiffiffi is j0ic . Note
qubit 1
j
1 i ¼ pffiffiffi ðjVijHijHi  jHijVijViÞ; (33)
that if the control qubit is in a superposition ð1= 2Þðj0i þ j1iÞ, the 2
action of a CNOT gate produces a maximally entangled state of the
target and control qubits. 1
20
In addition to the fermionic-like antibunching for the state j c
i j
2 i ¼ pffiffiffi ðjHijVijHi  jVijHijViÞ; (34)
and ‘‘boson-like’’ bunching behavior of the state j c þ i, there are
2
also intermediate behaviors observed by tuning the phase between
the two Bell statej c
i, which could be used to simulate anyons
1
j
3 i ¼ pffiffiffi ðjHijHijVi  jVijVijHiÞ: (35)
(Michler et al., 1996). 2

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 791

The notation used is such that the index i in j i i designates a It governs the directional correlations of the emissions. It
GHZ state with the property that the polarization of photon i, also implies the emerging photon pairs have entangled fre-
in each of the terms of the superposition, is different from the quencies and linear momenta. There are two different types of
polarization of the other two. Consider now the setup of Fig. 7 the process: either signal and idler photons share the same
and suppose that three photons enter the GHZ analyzer by polarization (type I) or they have perpendicular polarizations
modes a, b, and c. The polarization beam splitters transmit H (type II).
and reflect V polarizations, thus a coincidence detection at the The quantum nature of SPDC was first studied by Klyshko
three outputs can only originate from the case that either all and Zel’dovich (Klyshko, 1967, 1988; Zel’dovich and
photons are transmitted (this corresponds to the input state Klyshko, 1969). With the works of Mollow (1973) and
jHijHijHi) or all reflected (jVijVijVi). The two cases are Hong and Mandel (1985) the theory reached its final form
fully indistinguishable if the photons perfectly overlap spa- (see the Appendix). The predicted strong quantum correla-
tially and temporally. Thus, two GHZ states, namely j 0 i¼ tions between the photon pairs created in SPDC were first
jHijHijHi  jVijVijVi, can be filtered out of the eight. One experimentally observed by Burnham and Weinberg (1970).
can further tell apart the states j i by placing a polarizer Quantum interference of (type I) SPDC photons was first used
after each PBS, setting it to distinguish the þ=
polarization to violate Bell’s inequality by Ou and Mandel (1988b) and
basis (by þ=
here we denote 45 linear polarizations). In Shih and Alley (1988). The process was shown to be a ready
such a case, the state jþ 0 i leads to coincidences þ þ þ, source of (path) entangled pairs by Horne, Shimony, and
þ

,
þ
, and

þ, while j
0 i causes totally
Zeilinger (1989). This was demonstrated independently by
different events: þ þ
, þ
þ,
þ þ, and


. The Rarity and Tapster (1990). Polarization entanglement in the
success probability of the GHZ analyzer is thus 1=4. type II process was discovered by Kwiat et al. (1995). For a
survey of SPDC, see Shih (2003).
We give a brief description of the SPDC process in the
IV. EXPERIMENTAL REALIZATIONS OF PHOTONIC Appendix and show below how to create photons entangled in
ENTANGLEMENT various degrees of freedom.
Polarization entanglement–Currently, the standard method
Sources of entangled photons play a central role in the
to produce polarization-entangled photons is the noncollinear
experimental study of foundations of quantum mechanics and
type II SPDC process (Kwiat et al., 1995). Its principle is
are an essential resource in optical quantum information
described in Fig. 8.
processing. The early Bell-test experiments used entangled
The state emerging through the two beams A and B is a
photons from atomic cascades; see Clauser and Shimony
superposition of jHijVi and jVijHi, namely,
(1978). Such a source has some drawbacks. The directions
of entangled-photon emissions are uncorrelated. This causes 1
very low collection efficiency. Moreover, the entanglement is pffiffiffi ðjHiA jViB þ ei jViA jHiB Þ; (36)
2
only perfect for photons that are emitted back to back, a
loophole that could allow a local hidden-variable model to where the relative phase  is due to the birefringence. Using
explain the experimental data (Santos, 1991). Meanwhile, it an additional birefringent phase shifter (or even slightly
was discovered that the process of spontaneous parametric rotating the down-conversion crystal itself), the value of 
down-conversion allows pairs of entangled photons to be can be set as desired, e.g., to 0 or . A net phase shift of 
collected in clearly specified directions, with reasonable may be obtained by a 90 rotation of a quarter-wave plate in
intensity and with very high purity. one of the paths. A half-wave plate in one path can be used to
Today, essentially all entangled-photon sources employ the change horizontal polarization to vertical and vice versa. One
second order optical nonlinearity leading to SPDC or more can thus produce any of the four Bell states in Eq. (26)
recently also the third order Kerr nonlinearity in four-wave (Mattle et al., 1996).
mixing in optical fibers. Such processes can be realized with The birefringence of the nonlinear crystal introduces com-
an increasing quality and brightness. For instance, Altepeter, plications. Since the ordinary and extraordinary photons have
Jeffrey, and Kwiat (2005) reported an entangled-photon pair different velocities and propagate along different directions,
source with an impressive count rate of over 1 106 per the resulting longitudinal and transverse walk-offs between
second and a fidelity of 97.7%. In this section, we shall focus the two terms in the state (36) are maximal for pair creation
on the creation of photonic entanglement of various forms. near the entrance face. This results in a relative time delay
T ¼ Lð1=uo
1=ue Þ (L is the crystal length, and uo and ue
A. Spontaneous parametric down-conversion are the ordinary and extraordinary group velocities, respec-
tively) and a relative lateral displacement d ¼ L tan ( is
If one shines strong laser light on a nonlinear crystal, the the angle between the ordinary and extraordinary beams
pump photons have some probability to split into correlated inside the crystal). If the coherence time c of the down-
pairs of lower energy. This is called spontaneous parametric converted light is shorter than T, the terms in Eq. (36)
down-conversion. The new photons, customarily called become, in principle, distinguishable, and no two-photon
‘‘signal’’ and ‘‘idler,’’ satisfy the following relations: for the polarization interference and thus no entanglement is observ-
wave vectors within the crystal one has k0  ki þ ks where able. Similarly, if d is larger than the coherence width, the
subscripts 0, s, and i denote, respectively, pump, signal, and terms can become partially labeled by their spatial location.
idler wave vectors, and the respective frequencies satisfy Fortunately, because the photons are produced coherently
!0  !i þ !s . This is usually called phase matching. along the entire length of the crystal, one can completely

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

792 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

FIG. 9. A Franson-type experiment allowing interference of time-

entangled photon pairs (under the condition of a strict coincidence).
The main feature of the scheme is a pair of two identical unbalanced
FIG. 8 (color). Type II parametric down-conversion. Inside a Mach-Zehnder interferometers. Adapted from Franson, 1989.
specially cut (birefringent) nonlinear crystal (e.g., BBO), a pump
photon can convert spontaneously into a pair of photons of lower
frequencies. The polarization of signals is orthogonal with respect to Temporal entanglement–If the crystal is pumped by a
the one of idlers. One can attempt to pick only photons of frequency continuous wave (CW) laser with a coherence time of
Lc ,
which is one-half of the frequency of the pumping field. In such a the time at which pair emission happens is undefined and
case, if the crystal is suitably oriented, H polarized photons are phase stable within
Lc . Now imagine that each of the
emitted into one cone (in the figure, the lower one), while V photons is sent to a different unbalanced Mach-Zehnder
polarized photons are emitted into the other cone, and the two interferometer. We assume that the interferometers have an
cones intersect. Due to the phase-matching pairs of photons pop up identical arm-length difference
L much longer than the
only along an intersection of a plane containing the pump beam with coherence length of the photons, but shorter than the coher-
the cones, emissions along the directions at which the two cones ence length of the pump laser (see Fig. 9). The coincidence
intersect have undefined polarizations. However, as one of the count rates at the outputs of both interferometers show a
photons is from the upper cone and the other one is from the lower sinusoidal interference pattern, which depends on the sum
one, and due to the indistinguishability of the two processes, we of the local phase shifts. If the coincidence gate is much
have no way to establish which is which (except from a direct shorter than
Lc , only the following coherent processes are
polarization measurement), photons in beams A and B are polar- selected and contribute to the interference:
ization entangled. This holds under condition that one erases all
possible distinguishing features of upper and lower cone photons 1
(which arise because one of them is an ordinary ray and the other pffiffiffi ðjlongi1 jlongi2 þ ei
jshorti1 jshorti2 Þ; (37)
2
one an extraordinary one) which can easily be done via compensa-
tion methods described in the text. where jshorti and jlongi denote the photon in short or long
arm of the given local interferometer. This is the principle of
Franson-type interferometry (Franson, 1989).21 The predicted
compensate for the longitudinal walk-off (Rubin et al.,
interference phenomena were observed by Brendel, Mohler,
1994). After the compensation, interference occurs pairwise
and Martienssen (1992) and Kwiat, Steinberg, and Chiao
between processes in which the photon pair is created at
(1993), many other experiments followed.
distances x from the middle of the crystal. The ideal
A pulsed-pump version of Franson interferometry uses so-
compensation is therefore to use two crystals, one in each
called time-bin entanglement (Brendel et al., 1999; Gisin
path, which are identical with the down-conversion crystal,
et al., 2002; Marcikic et al., 2002) (see Fig. 10). A short,
but only half as long. If the polarizations are rotated by 90
ultraviolet pump pulse is first sent through an unbalanced
(e.g., with a half-wave plate), the retardations of the o and e
Mach-Zehnder interferometer (the pump interferometer) cre-
components are exchanged and temporal indistinguishability
ating two mutually coherent pump pulses jshortip þ
is restored. The method also provides reasonable compensa-
jlongip , and next through a beta barium borate (BBO)
tion for the transverse walk-off effect.
An alternative method is using type I SPDC in two orthog- crystal. Then, if via SPDC a pump photon is converted into
onally oriented crystals. Two-photon states of tunable purity a pair, the latter is in the state
and degree of entanglement can be produced (White et al., jshorti1 jshorti2 þ jlongi1 jlongi2 : (38)
1999; Cinelli et al., 2004; Peters et al., 2004). Periodic
poling of nonlinear crystals modifies the phase-matching The two entangled photons can be separated and subjected to
conditions for SPDC and thus allows one to exploit the local measurements in unbalanced interferometers with
L
material’s nonlinear properties more efficiently compared to identical as in the pump interferometer [for more details, see
bulk crystals. Kim, Fiorentino, and Wong (2006) proposed Gisin et al. (2002)]. By varying the parameters of the beam
and implemented a polarization Sagnac interferometric con- splitters and phases in the pump interferometer, all two-qubit
figuration with bidirectional pumping of a type II phase-
matched periodically poled KTiOPO4 crystal. A pulsed 21
The original motivation for this type of interferometry was to
(Kuzucu and Wong, 2008), and narrow band (0.15 nm), have an alternative method of obtaining correlations that lead to
wavelength tunable entangled-photon source based on such violations of Bell inequalities. However, as shown by Aerts et al.
a configuration with a spectral brightness up to 273 000 pairs (1999), the postselection inherent in this type of experiment makes
ðs mW nmÞ
1 has been reported (Fedrizzi et al., 2007). In a this connection much more complicated: the original scheme in its
further development, via four-wave mixing in a photonic ideal form has an explicit local hidden-variable model. One has to
crystal fiber Fulconis et al. (2007) developed a bright, pulsed modify the experiment, and use nonstandard Bell inequalities to
source of photon pairs. make it a valid Bell test.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 793

FIG. 11. The Rarity-Tapster experiment with momentum entan-

glement from a type I SPDC. Using double apertures A two
correlated pairs of modes are selected from the emission spectrum
of a type I down-conversion source. On each of the two BS beams of
the same wavelength are superimposed. The detectors Da3 , Db3 ,
Da4 , and Db4 register two-photon coincidences between the a and b
outputs. From Rarity and Tapster, 1990.

B. Photonic entanglement in higher dimensions

1. Entangled qudits

FIG. 10. Schematic of the pulsed time-entangled twin-photon Photonic entanglement in higher dimensions can in prin-
source. The interferometers are represented by curvy lines. This ciple be generated by SPDC processes in a form of general-
is to stress that practical realizations of this type of interferometers ization of path or temporal entanglement into more than two
are usually built out of optical fibers and couplers. From Brendel conjugate pairs of beams or time bins, respectively (Zeilinger
et al., 1999. et al., 1993), and analyzed with N-port beam splitters (Reck
et al., 1994). As shown by Żukowski, Zeilinger, and Horne
(1997) such configuration can lead to new types of EPR
entangled time-bin states can be generated. An advantage of correlations, and can be used for tests of local realism
the time-bin entanglement is that it is insensitive to polariza- [which are more discriminating than two qubit tests; see
tion fluctuations. Using reference laser pulses to actively lock Kaszlikowski et al. (2000)].
the phase, it can be robustly distributed over long distances in Another route is to use the photons’ orbital angular mo-
optical fibers. Note that because the pulses are only separated mentum. Orbital angular momentum eigenstates of photons
on the order of a few nanoseconds, and this is much shorter are states of the electromagnetic field with phase singular-
than the time scale of any phase drifts in the fiber, the drifts do ities. They can be utilized for observation of higher-
not affect the quality of entanglement. An experiment by dimensional entanglement (Mair et al., 2001; Vaziri,
Marcikic et al. in 2004 demonstrated the distribution of Weihs, and Zeilinger, 2002; Vaziri et al., 2003). This
time-bin entanglement over 50 km in optical fibers. approach has advantages in certain quantum communication
Path entanglement–Entanglement experiments involving protocols (Vaziri, Weihs, and Zeilinger, 2002; Molina-Terriza
path (momentum) entanglement were proposed by Horne et al., 2004; Gröblacher et al., 2006). High-dimensional
and Zeilinger (1986), and their feasible version by entangled qudits have also been created by transverse spatial
Żukowski and Pykacz (1988). Finally, Horne, Shimony, and correlations of two SPDC photons (Neves et al., 2005), or
Zeilinger (1989) proposed that SPDC is an ideal source in using transverse momentum and position entanglement of
case of such experiments. This was realized independently by photons emitted in SPDC, in a form called pixel entanglement
Rarity and Tapster (1990). Because of the phase-matching (O’Sullivan-Hale et al., 2005).
relation, idler and signal photons of given frequencies are
correlated in emission directions. One can use apertures, see
2. Hyperentanglement
Fig. 11, to select only two pairs of spatially conjugate modes
(directions). The photon pairs then emerge via the apertures As shown earlier, the SPDC photons are entangled in
such that they are either in the upper a mode (a1) and lower b energy and momentum, and, if suitably selected, can be
mode (b2) or in the lower a mode (a2) and upper b mode also entangled in polarization or path. If one selects
(b1). The resulting state is thus pairs which are entangled not only in polarization but
also in some other degree(s) of freedom, this specific
entanglement is called hyperentanglement (Kwiat, 1997).
1 Hyperentanglement may have interesting applications such as
ji ¼ pffiffiffi ðei
b ja1ijb2i þ ei
a ja2ijb1iÞ: (39)
2 Bell-state analysis (Kwiat and Weinfurter, 1998; Walborn,
Pádua, and Monken, 2003; Schuck et al., 2006), two-particle
GHZ-type tests of local realism (Michler, Weinfurter, and
The a modes enter a BS via opposite inputs, so do b modes. Żukowski, 2000; Chen et al., 2003), implementations of
Behind the BS upper and lower paths cannot be distinguished, single-photon two-qubit CNOT gate (Fiorentino and Wong,
leading to two-photon interference, which depends on the 2004), two-qubit swap gates (Fiorentino, Kim, and Wong,
difference of the relative phase shifts in a and b modes. 2005), and quantum cryptography (Chen, Z.-B., et al., 2006).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

794 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Polarization-path entanglement–Polarization-path entan-

glement can be generated by a double pass of a pump laser
through a BBO crystal (Chen et al., 2003). The pump passes
the crystal and is reflected to pass it again. While the polar-
ization state in each of the two possible emission processes is
given by the respective SPDCpffiffiffisetting, the path state of
the pairs is j c
ð
Þipath ¼ ð1= 2ÞðjuiA jdiB
ei
jdiA juiB Þ,
where the two orthonormal kets jdi and jui denote the two FIG. 12. (a) Experimental setup for the creation and analysis of
path states of photons. By properly adjusting the distance hyperentangled photons. (b) Spatial filtration (spa): hologram (holo)
between the mirror and the crystal such that
¼ 0, one gets and single-mode fiber (smf). (c) Energy time transformation (e–t):
thick quartz decoherer (dec) and liquid crystal (LC). (d) Polarization
ji ¼ j c
ipol  j c
ð0Þipath ; (40) filtration (poln): quarter-wave plate (qwp), half-wave plate (hwp),
and polarizer (pol). From Barreiro et al., 2005.
which is a two-photon state maximally entangled in both
polarization and path. It was independently realized by
Barbieri et al. (2005) using type I nonlinear crystals, and photon represents a qubit, and can be individually manipu-
by Yang et al. (2005) using a type II nonlinear crystal. lated and observed.
Polarization-time entanglement—A more robust distribu- The twin-beam multiphoton entanglement can be gener-
tion of hyperentanglement is possible with photon pairs ated via standard SPDC, but with a strong pump pulse.
which are entangled both in time (i.e., time-bin entanglement) Stimulated SPDC (Lamas-Linares, Howell, and
and in polarization (Genovese and Novero, 2002; Chen, Bouwmeester, 2001; Simon and Bouwmeester, 2003) can
Z.-B., et al., 2006; Schuck et al., 2006). To create such a be seen as an extension of interferometrically enhanced
polarization-time entanglement, we similarly combine the SPDC (Herzog et al., 1994) and may show an onset of a
creation of polarization entanglement with the method to laserlike emission of entangled photons, i.e., we can have an
obtain temporal entanglement. That is, either by using a short, (‘‘entanglement laser’’) in the sense that a (spontaneously
ultraviolet laser pulse sent first through an unbalanced Mach- created) photon pair in two polarization-entangled modes
Zehnder interferometer (the pump interferometer), to have stimulates, inside a nonlinear gain medium, emission of addi-
two pulses well separated in time, or by taking advantage of tional pairs.
the long coherence time of a CW-pump laser. Using a ‘‘time- A simplified Hamiltonian22 responsible for generation of
path transmitter’’ introduced by Chen, Z.-B., et al. (2006), polarization-entangled SPDC photons can be given as
one can realize a transformation between polarization-path H0 ¼ i
ðayH byV
ayV byH Þ þ H:c:. Horizontally (H) and verti-
and polarization-time hyperentanglement. cally (V) polarized photons occupy two spatial modes (a and
Entanglement in multiple degrees of freedom–In an experi- b);
is a coupling constant that depends on the nonlinearity
ment by Barreiro et al. (2005), besides the entanglement in of the crystal and the intensity of the pump pulse. After
polarization and in energy, photon pairs from a single non- the interaction time t the resulting photon state is given by
linear crystal were also entangled in orbital angular momen- j c i ¼ e
itH0 j0i (Kok and Braunstein, 2000; Lamas-Linares,
tum. The experimental setup is shown in Fig. 12. By pumping Howell, and Bouwmeester, 2001; Simon and Bouwmeester,
two BBO crystals with optical axes aligned in perpendicular 2003), and in the number state representation reads
planes, a two-photon ð2  2Þ  ð3  3Þ  ð2  2Þ-dimensional 1 pffiffiffiffiffiffiffiffiffiffiffiffi
1 X
hyperentangled state was produced, approximately described jc i ¼ n þ 1 tanhn j c

n i;
by cosh2  n¼0
Xn
ð
1Þm
ðjHHi þ jVViÞ  ðj1;
1iLG þ j0; 0iLG þ j
1; 1iLG Þ jc

ni¼ pffiffiffiffiffiffiffiffiffiffiffiffi jn
m; m; m; n
mi: (42)
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
polarization spatial m¼0 n þ 1

 ðjEEi þ jLLiÞ: (41) The ket jn
m; m; m; n
mi denotes a number state in the

|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
energy time respective modes aH , aV , bH , and bV , and  ¼
t is the
interaction parameter. To avoid multipair emission events,
Here  describes the orbital-angular-momentum spatial mode most SPDC experiments are restricted to   0:1. By going
balance, which is due the properties of the source (Torres, to higher values, bipartite entangled states constituting of
Alexandrescu, and Torner, 2003) and the selection via the large numbers of photons are generated. The state j c i is a
mode-matching conditions. superposition of the states j c

n i of n indistinguishable photon
pairs. Each j c
n i is an analog of a singlet state of two
spin-n=2 particles, thus j c i is invariant under joint rotations
C. Twin-beam multiphoton entanglement
of the polarization bases of both modes. The polarization of
It is also possible to produce entangled states involving each beam is completely undetermined, but the polarizations
large numbers of photons, approaching the macroscopic do-
main. Such entanglement is related to experiments on twin 22
It is interesting to note that this simplified Hamiltonian and the
beams (Smithey et al., 1992) and could be called a twin- state generated thereby [Eq. (42)] are closely related to squeezing
beam multiphoton entanglement. It is different from multi- and continuous-variable entanglement, see Braunstein and van
photon entanglement in which each spatially separated Loock (2005) for a review.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 795

FIG. 13 (color online). Super-resolving phase measurement with N00N states. (a) A schematic of the experimental setup of Nagata et al.
(2007): a Mach-Zehnder interferometer consisting of two 50:50 beam splitters. (b) Single-photon count rate in mode e as a function of phase
plate (PP) angle with single-photon input j10iab . (c) Two-photon count rate in modes e and f for input state j11iab . (d) Four-photon count rate
of three photons in mode e and one photon in mode f for the input state j11iab . The visibility of the fringes is 0.91(6), greater than the
threshold to beat the standard quantum limit. Adapted from Nagata et al., 2007.

of the two beams are always anticorrelated. The average single photons in each input of the Mach-Zehnder interfer-
photon-pair number is hni ¼ 2sinh2 . ometer (j11iab ), the state after the first BSpffiffiffiis, due to the
Out of such states one can extract, for example, the follow- Hong-Ou-Mandel effect, ðj20i ffiffiffi þ j02icd Þ= 2, which then
pcd
ing two-pair term of Eq. (42): evolves to ðj20i þ ei2’ j02iÞ= 2. After the second BS, the
1 probability of detecting two photons in the output modes e
jc

2 i ¼ pffiffiffi ðj2; 0; 0; 2i
j1; 1; 1; 1i þ j0; 2; 2; 0iÞ; (43) and f is P ¼ ð1
cos2’Þ=2 which shows a phase super
3 resolution [for the experimental data see Fig. 13(c)]. If two
which can be treated as a singlet state of two (compo- photons are fed into in each input of the interferometer
pffiffiffiffiffiffiffiffi
site) spin-1 systems [see Howell, Lamas-Linares, and (j22iab ), after the first BS we get 3=8ðj40icd þ j04icd Þ þ
Bouwmeester (2002) for a test of Bell’s inequality by en- ð1=2Þj22icd : a generalized multiphoton Hong-Ou-Mandel in-
tangled states of spin-1-like systems]. terference phenomenon. After the second BS, the probability
The theory of an entanglement laser was developed by of detecting three photons in one output e and one in f is P ¼
Simon and Bouwmeester (2003). The basic principle of a 1
ð3=8Þ cos4’ [see Fig. 13(d) for data]. The high-precision
stimulated entanglement creation was first experimentally optical phase measurements have many important applica-
demonstrated in the few-photon regime (Lamas-Linares, tions, e.g., overcoming the diffraction limit for classical light
Howell, and Bouwmeester, 2001). Later, twin-beam entan- (Boto et al., 2000; Kok et al., 2001).
glement of up to 12 photons (Eisenberg et al., 2004) was
experimentally observed.
D. Multiphoton entanglement
A special twin-beam entanglement is the so-called ‘‘high
N00N’’ type (Bouwmeester, 2004) state of two beams The original motivation to observe entanglement of more
(Dowling, 1998; Kok et al., 2001; Kok, Lee, and Dowling, than two particles, with measurements on the particles per-
2002)
formed at spatially separated stations, stems from the obser-
1 vation by GHZ that three-particle entanglement leads to a
jN00Ni ¼ jN; 0; 0; Ni ¼ pffiffiffi ðjNia j0ib þ j0ia jNib Þ: dramatic conflict between local realism and EPR’s ideas with
2
predictions of quantum mechanics (Greenberger, Horne, and
(44)
Zeilinger, 1989; Greenberger et al., 1990; Mermin, 1990a);
It was experimentally realized for N ¼ 3 (Mitchell, Lundeen, see Sec. V.B.1. However, in 1989 no ready sources of three or
and Steinberg, 2004) and N ¼ 4 (Walther et al., 2004) more particle entanglement were available. Yurke and Stoler
[for N ¼ 2, see, e.g., Rarity et al. (1990); Edamatsu, (1992a, 1992b) showed that in theory multiparticle entangle-
Shimizu, and Itoh (2002); and Sun et al. (2006)]. These ment effects should in principle be observable for particles
experiments demonstrated an interesting feature of N00N originating from independent sources. A general method for
states: the effective de Broglie wavelength of the multiphoton making such an interference observable, and also to swap
state is by 1=N shorter than the wavelength of the single entanglement, was given by Żukowski, et al. (1993); Rarity
photon (Jacobson et al., 1995). Nagata et al. (2007) not (1995); Żukowski, Zeilinger, and Weinfurter (1995); and
only measured the reduced de Broglie wavelength of four- Zeilinger et al. (1997). In the following, we first present
entangled photons, but also obtained a visibility that ex- the basic methods followed by numerous experiments in
ceeds the threshold to beat the standard quantum limit which multiphoton entanglement was observed. Once one is
(Fig. 13). See Fig. 13(a) for more details. If we put two able to entangle two photons that never interacted, one can

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

796 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

construct many types of entanglement (Zeilinger et al., and aðþÞ, and detection of the particles form B in dðþÞ
1997), which in turn can be utilized in many ways (Bose, and bðþÞ. Note that depending on the phase shifts the detec-
Vedral, and Knight, 1998). tion at, e.g., cðþÞ and dðþÞ, acts like a Bell-state measure-
ment,pffiffiffi projecting the two photons into the state
1. Entanglement construction ð1= 2Þðjc0 ijd0 i þ eið
c þ
d Þ jcijdiÞ. The other ptwo
ffiffiffi photons
are, due to this event, in the state ð1= 2Þðjb0 ija0 iþ
We have at hand only photon sources of two-particle
e
ið
c þ
d Þ jbijaiÞ. This process is called entanglement
entanglement. We show in detail an operational method to
swapping.
swap entanglement of two pairs of particles (Żukowski,
Enforcing source indistinguishability–Imagine now that
Zeilinger, and Weinfurter, 1995), which has been used in
the sources of entangled states are two crystals pumped by
many experiments [the pioneering one was the Innsbruck
independent, pulsed lasers operating synchronously. Assume
teleportation (Bouwmeester et al., 1997)]. The technique
that the time separation between two pulses is much larger
of essentially erasing which source information can be ap-
than all other time scales of the experiment, i.e., we study the
plied in many other configurations, e.g., in the case of a
radiation generated in each crystal by a single pulse. We omit
double pair emission from a single source, etc. It even works
retardation effects by assuming equal optical paths. We as-
for totally independent emissions (provided they are
sume that we pick the SPDC radiation with frequencies close
synchronized).
to 12 !op , where !op is the central frequency of the pump pulse.
Entangling two independent particles: the principle–
Suppose that the four SPDC photons are detected in almost
Figure 14 shows a configuration for obtaining interference
the same moment of time (up to a few nanosecond window),
effects for two pairs of particles originating from two inde-
one in each of the detectors aðþÞ, bðþÞ, cðþÞ, and dðþÞ. One
pendent sources. Assume that the sources of path-entangled
could determine that the photon detected at dðþÞ came from
states in Fig. 14, A and B, each spontaneously emits a pair of
crystal A (B) by noting the near simultaneity of the detection
particles in an entangled state (all particles are supposed to be
of photon dðþÞ and one of the photons at aðþÞ or bðþÞ [the
identical) at nearly the same
pffiffiffimoment of time, and the states of detection times of a true SPDC pair are extremely tightly
the pairs are jpAffiffiffii ¼ ð1= 2Þðjaijdi þ ja0 ijc0 iÞ for source A correlated, see the Appendix, Eq. (A18) and the discussion
and jB i ¼ ð1= 2Þðjb0 ijd0 i þ jbijciÞ for source B (the letters after it]. To ensure that the source of photons is unknowable,
represent beams taken by the particles in Fig. 14, and hejfi ¼ the (initially spectrally broadband) photons should be de-
ef ). The beams x and x0 , where x ¼ a, b, c, or d are tected behind a narrow-band filtering system (to be called
superposed at 50:50 BSs. Behind the BSs, we place detectors later simply a filter) whose inverse of the bandwidth (coher-
in the output ports xðÞ. In all unprimed beams, one can ence time) clearly exceeds the pulse duration  (e.g., by an
introduce a phase shift of
x . The detector stations differ in order of magnitude). Then, the temporal separation of true
their role: aðÞ and bðÞ observe radiation coming from one SPDC pairs, all created within , spreads over times much
source only, but this is not so for stations dðÞ and cðÞ. For longer than  and thereby prevents identifying the source of
instance, if a single particle is detected by dðþÞ, its origin the photon by comparison of the arrival times. Note that
may be, under suitable conditions, completely unknown. If it filtering is necessary only in modes c and d, while no filtering
cannot be determined, even in principle, which source pro- is required in front of the detectors aðÞ and bðÞ.23
duced the particle which activated the detectors, say dðþÞ and One can estimate the maximal visibility expected for the
cðþÞ, then four-particle interference effects may occur. interference process, using the results presented in the
Assume that detectors aðþÞ and bðþÞ also fired. Appendix, where the basic properties of the SPDC radiation
Simultaneous firings of the four detectors can exhibit inter- are derived. The amplitude of the four-photon detections at,
ference effects provided the two contributing processes are say, detectors aðþÞ, bðþÞ, cðþÞ, and dðþÞ at times ta , tb , tc ,
indistinguishable: detection of the particles from soure A in and td , is proportional to
dðþÞ and aðþÞ, and detection of the particles from B in cðþÞ
and bðþÞ; detection of the particles from source A in cðþÞ eið
a þ
b þ
c þ
d Þ Aad ðta ; td ÞAcb ðtc ; tb Þ
þ Ab0 d0 ðtb ; td ÞAa0 c0 ðta ; tc Þ; (45)

where
i , i ¼ a, b, c, and d, is the local phase shift in the

given beam. The probability amplitude Aðt; t0 Þ to detect one
photon of a SPDC pair at t and the other one at t0 , is the one
given by Eq. (A20). To get an overall probability of the
process, one has to integrate the square of the modulus of
the amplitude over the detection times. Since typical time
resolution of the detectors is of the order of nanoseconds,
which is much longer than the coherence times of typical

23
This method also precludes the possibility of inferring the source
of the photon from the frequency correlations. The frequency of the
FIG. 14 (color online). Four-particle interference effects for two photons reaching dðþÞ is better defined than the pumping pulse
pairs of particles originating from two independent sources. frequency, and it is the spread of the latter one that limits the
Adapted from Żukowski et al., 1993. frequency correlations of the idler-signal pair from one source.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 797

filters and the width of femtosecond (fs) pump pulses, the multiphoton states [see Barbieri et al. (2009) and Weinhold
integrations over time can be extended to infinity. The joint et al. (2008)].
probability to have counts in the four detectors behaves as Note that source indistinguishability in principle can also
 X  be achieved with an ultracoincidence technique, which does
1
Vð4Þ cos
x not require a pulsed pump, but an extremely good detection
x¼a;b;c;d time resolution
T much sharper that the coherence time of
the filtered SPDC radiation, and rejection of all events at cðþÞ
and the visibility Vð4Þ is given by and dðþÞ which are detected with time difference higher than,
R 4 say, 2
T [see Żukowski, et al. (1993)]. In such a case, the
d tjAad ðta ; td ÞAbc ðtb ; tc ÞAb0 d0 ðtb ; td ÞAa0 c0 ðta ; tc Þj
Vð4Þ ¼ R 4 ; pumping lasers may be CW ones.24
d tjAad ðta ; td ÞAbc ðtb ; tc Þj2
(46) 2. New methods
where d4 t ¼ dta dtb dtc dtd . Assume that the filter functions in The methods described above require a femtosecond
all beams are of identical Gaussian form: Ff ðtÞ ¼ pulsed laser pump. Unfortunately, a femtosecond pulse
e
2!p t jFf ðtÞj, whereas the pump beam is described by GðtÞ ¼
i o
pumped SPDC shows relatively poor quantum interference
e
i!p t jGðtÞj. Here !op is the central frequency of the pulse,
o
visibilities (Keller and Rubin, 1997). The following methods
and jFj and jGj functions are given by Fourier transforms of are used to increase the quantum interference visibility: (i) a
exp½
12 ð!
Þ2 =2 , where  ¼ 12 !op (for jFj) or !op (for thin nonlinear crystal (Sergienko et al., 1999), (ii) narrow-
jGj), and  is the respective spectral width. One gets band spectral filters in front of detectors, as shown above (Di
 1=2 Giuseppe et al., 1997; Grice and Walmsley, 1997; Grice
2p et al., 1998), and (iii) an interferometric technique (Branning
Vð4Þ ¼ ; (47)
2p þ 2F 2f =ð2p þ 2F þ 2f Þ et al., 1999, 2000) without spectral and amplitude postselec-
tion, which is making the spectral wave function of the two
where p is the spectral width of the pulse, f is the spectral photons much more symmetric.25 The first two methods
width of the filters in beams a, a0 , b, b0 , and the spectral width reduce the intensity of the entangled-photon pairs signifi-
of the filters in c and d is F . If one removes the filters cantly and cannot achieve perfect overlap of the two-photon
in beams a, a0 , b, and b0 , the formula simplifies to Vð4Þ ¼ amplitudes. For the theoretical and experimental details of the
½2p =ð2p þ 2F Þ1=2 ; see Żukowski, Zeilinger, and Weinfurter last method, see Kim, Kulik, and Shih (2001).
(1995). Namely, narrow filters in paths a, a0 and b, b0 are not A method gaining great importance is tuning the properties
necessary to obtain high visibility. The other filters, for of the down-conversion source and the pump such that one
detectors which observe radiation from both sources, should obtains frequency uncorrelated pairs of photons; see, e.g.,
be always sufficiently narrow. Grice, U’Ren, and Walmsley (2001); U’Ren, Banaszek, and
The influence of photon statistics–The visibility of the four- Walmsley (2003); Walton et al. (2003, 2004); Torres, Macià,
particle fringes in the setup of Fig. 14 can be impaired by the and Torner (2005); Garay-Palmett et al. (2007); Mosley
statistical properties of the emission process. The statistics of et al. (2008); and Halder et al. (2009) who demonstrated a
a single beam of a down converter is thermal-like. The state source of photon pairs based on four-wave mixing in photonic
of idler-signal pairs emerging via a pair of (perfectly phase crystal fibers [see also Fulconis et al. (2007)]. Engineering of
matched) pinholes is given by the phase-matching conditions in the fibers allowed creation
of photon pairs at 597 and 860 nm in an intrinsically factor-
X
1
able state of frequencies. Thus there were almost no spectral
j c i ¼ N
1 zm jm; sijm; ii; (48)
m¼0 correlations. The source is narrow band and bright. Two
separate sources of such a kind were used to generate a
where z is a number dependent on the strength of the pump, Hong-Ou-Mandel interference. The idlers were used to herald
jm; si (jm; ii) denotes an m-photon state in the signal (idler) the singles. The observed interference, conditioned on a joint
mode, and N is the normalization constant. It can be shown detection event of two idlers, had a raw visibility of 76.1%.
(Żukowski et p al.,
ffiffiffiffiffiffi1999) that the visibility is reduced below Since narrow-band filtering is unnecessary in case of such
50% if jzj2 > ð 17
3Þ=8  0:140. Thus, to have high visi-
bility the ratio of the probability of each pulse to produce a 24
single down-converted pair to the probability of not produc- See, e.g., an experiment reported by Halder et al. (2007) who
use single-photon detectors with a time resolution of 70 ps, which
ing anything must be less than 14%. This threshold is at quite
is much shorter than the coherence length ( 350 ps) of the tightly
high pump powers. Nevertheless, this puts a strong limitation
filtered photons in the experiment (see Sec. VI.C.3). Using an atom-
as how many particles can be entangled using such methods. cavity system, Legero et al. (2004) generated single photons with
Simply, creation of entanglement for many particles requires coherence time of 500 ns exceeding the time resolution of
more and more initial entangled pairs, thus one pumps employed photon detectors by 3 orders of magnitude, and observed
stronger. However, a strong pump leads to lower visibility quantum beat between photons of different frequencies with a near-
of quantum interference, which may prohibit to observe the unity visibility.
correlations due to the desired multiphoton entanglement 25
The method rests on two distinct processes for emission of a
(Laskowski et al., 2009). Recently, several experiments down-converted pair, which are coherently overlapped. The axes of
were performed to identify and quantify the experimental polarization are switched between the two processes. This gives a
imperfections that contribute error to the produced symmetrization of the spectral properties.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

798 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

sources, one can achieve a higher collection efficiency than in

the case of using passive filtering [see Kim, Kulik, and Shih
(2001) and Yao et al. (2012) where the collection efficiency
was reported to be about twice as high than in measurements
using 3 nm filters].
With mastering these phase-matching tune-up techniques,
one can expect that in future they may replace the ones based
on passive filtering, as a new method for entanglement swap-
ping and related processes.

3. First proposals
In the 1990’s, many proposals were made for observations
of multiphoton entanglement (Żukowski, et al., 1993; Rarity,
1995; Żukowski, Zeilinger, and Weinfurter, 1995; Zeilinger
et al., 1997; Pan and Zeilinger, 1998; see also Sec. III.E.2), or
FIG. 15 (color online). A three-photon polarization-entanglement
involving atoms (Cirac and Zoller, 1994; Haroche, 1995;
source. The photon sources A and B are pumped by short pulses.
Sleator and Weinfurter, 1995). The PBS1 transmits 45 polarization and reflects
45 polariza-
The generic idea of observing photonic GHZ entangle- tion, F is a narrow filter, and DT1 and DT2 are two single-photon
ment, later put into practice, was given by Zeilinger et al. detectors. A single-photon trigger event in one of the detectors
(1997); 26 see Fig. 15. Assume that sources A and B simul- signals that coincident detections in channels 1, 2, and 3 would re-
taneously emit a photon pair each. Pairs in beams x, yp(1, ffiffiffi 3, sult in GHZ correlations. The setup can also be used for observation
and 2, 4) are in identical polarization states ð1= 2Þ of four-photon GHZ interference. From Zeilinger et al., 1997.
ðjHx ; Hy i þ jVx ; Vy iÞ. The state of the four particles, after
the passage of 2 and 3 via PBS1, and provided the photons
are indistinguishable (which can be secured using the meth- Unfortunately, this is not the case in the actual SPDC
ods as described earlier), reads experiments. Because of the absence of perfect pair sources
and perfect single-photon detectors, in the experiments both
2ðjH1 ; H2 ; H3 ; H4 i þ jV1 ; V2 ; V3 ; V4 i þ jH1 ; H3 ; V3 ; V4 i
1
three-photon and four-photon entanglement (Bouwmeester,
þ jV1 ; V2 ; H2 ; H4 iÞ: (49) et al., 1999; Pan et al., 2001; Eibl et al., 2003; Zhao et al.,
2003; see also Sec. IV.D.4) is observed only under the
Only the superposition jH1 ; H2 ; H3 ; H4 i þ jV1 ; V2 ; V3 ; V4 i, condition that there is one and only one photon in each of
which is a GHZ state, leads to fourfold coincidence. the four outputs. As there are other detection events where
Therefore, fourfold coincidences can reveal four-particle two photons appear in the same output port, this condition
GHZ correlations. might raise doubts about whether such a source can be used
The scheme in Fig. 15 also allows one to generate uncondi- for a valid GHZ test of local realism (Sec. V.B.2). By further
tional three-particle GHZ states27 via a method based on the developing the ideas of Yurke and Stoler (1992a), Żukowski
notion of entangled entanglement (Krenn and Zeilinger, (2000) showed that the above procedure indeed permits a
1996). For example, one could analyze the polarization state valid GHZ test.
of photon 2 by passing it through a PBS selecting 45 and

45 polarizations. Then the polarization state of the remain-
pffiffiffi 4. Experimental realizations
ing three photons 1, 3, and 4 will be projected into ð1= 2Þ
ðjH1 ; H3 ; H4 i þ jV1 ; V3 ; V4 i, if and only if detector DT1 de- The first experiment involving techniques of forcing indis-
tects a single photon. A similar superposition, however with a tinguishability of photons from different parametric down-
minus sign, is obtained once detector DT2 detects a single conversion pairs was the teleportation experiment by
photon. The detection of photon 2 excludes the last two terms Bouwmeester et al. (1997). This, however, will be discussed
in Eq. (49), and projects the remaining three photons into a later in the context of quantum communication (Sec.VI). A
spatially separated and freely propagating GHZ state. GHZ-type entanglement among three spatially distributed
However, the scheme works only with photon-number dis- photons, using the above methods of entanglement construc-
criminating detectors, and if both EPR sources emit only a tion, was first observed by Bouwmeester et al. (1999). The
pair each without double pair (or more) emission events. main idea behind this experiment, as was put forward in
Zeilinger et al. (1997), is to transform two pairs of
26 polarization-entangled photons into three entangled photons.
Earlier proposals, Żukowski, et al. (1993) and Żukowski,
The fourth photon served the role of a trigger. Figure 16
Zeilinger, and Weinfurter (1995), are essentially showing an explicit
illustrates the experimental setup. Two pairs of polarization-
operational method to realize the ideas of Yurke and Stoler (1992a,
1992b). They involved techniques which required more complicated entangled photons are generated via a pulsed SPDC. The
optical setups and more sources, however the basic principles were probability per pulse to create a single pair in the desired
the same. modes was on the order of a few 10
4 with a correspondingly
27
The original proposal for the realization of three-photon GHZ smaller probability to create four photons and negligible for
states in Zeilinger et al. (1997) makes use of a slightly different three pair events.
pffiffiffiThe source was aligned to emit photon pairs
interferometric setup. in the state ð1= 2ÞðjHa ; Vb i
jVa ; Hb iÞ.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 799

detects a photon, there are only two possible polarization

combinations: jH1 H2 V3 i and jV1 V2 H3 i. Bouwmeester et al.
adjusted the two path lengths such that the two photons
arrived simultaneously at the last PBS. The photons were
spectrally filtered, with
 ¼ 4:6 nm, and monitored by
single-mode, fiber-coupled single-photon detectors. The fil-
tering process stretched the coherence time so that it was
substantially larger than the pump pulse duration ( 200 fs).
These processes erase the possibility of distinguishing the
photons from different pairs by their arrival time or spatial
modes (see previous sections). Thus the resulting counts
originate from a superposition between jH1 H2 V3 i and
jV1 V2 H3 i, that is, the three-photon GHZ state

1
pffiffiffi ðjHi1 jHi2 jVi3 þ jVi1 jVi2 jHi3 Þ: (50)
2

To test whether one indeed dealt with a the three-photon

GHZ state, Bouwmeester et al. performed a polarization
analysis (with polarizing filters in front of each detector,
not shown in Fig. 16). They first compared the counts of all
eight possible polarization combinations HHH; . . . ; VVV.
The results showed that the intensity ratio between the de-
sired events (HHV and VVH) and the six other undesired
ones was about 12:1. The dominance of the two terms is a
necessary but not yet sufficient condition for demonstrating a
GHZ entanglement, as there could, in principle, be just a
FIG. 16 (color online). Entanglement among three single photons. statistical mixture of the two states. Next, to test whether the
(a). Experimental setup for observing three-photon GHZ entangle- one indeed dealt with a coherent superposition of the two
ment. The UV-pulse incident on the BBO crystal produces four terms, Bouwmeester et al. performed measurements in the
photons, two in each mode. Conditioned on the registration of one ‘‘diagonal’’ basis þ=
¼ H  V. A result þ at detector D1
photon at the trigger detector T, the three photons registered at D1, should
pffiffiffi effectively project the state of the other photons to
D2, and D3 exhibit the desired GHZ correlations. (b). Experimental ð1= 2Þðjþi2 jþi3
j
i2 j
i3 Þ. The obtained data are consis-
data. The photon at D1 polarized at 45 and the photon at detector tent with this prediction, as can be seen from Fig. 16(b). This
D2 polarized at
45 . The two curves show the fourfold coinci- is so only within a short interval of delay between the
dences for a polarizer in front of detector D3 oriented at
45 and photons, while for path length differences larger than the
45 , respectively, as a function of the spatial delay in path a. (c). coherence length of the detected photons the coherence
Experimental data for the case of the polarizer in front of detector between the two GHZ-terms vanishes. Further control experi-
D1 set to 0. Adapted from Bouwmeester et al., 1999. ments were done by setting the polarizer in front of detector
D1 at 0 such that the GHZ state (50) is projected into a
separable state jVi2 jHi3 . In this case, the results in Fig. 16(c)
The experimental arrangement was such that the GHZ show no correlation.
entanglement could be observed under the condition that By extending the above technique, GHZ-type entangle-
both the trigger photon and the three entangled photons are ment among more particles was observed. A four-photon
detected. This is usually called a ‘‘postselection’’ (sometimes GHZ state was first observed by Pan et al. (2001) and its
also called ‘‘conditional detection’’ or ‘‘observation in coin- genuine entanglement was confirmed by Zhao et al. (2003).
cidence basis’’). We now go into more details on how this Later on, five-photon entanglement was demonstrated by
works. As initially there were four photons, provided that interfering a four-photon GHZ state (Pan et al., 2001) with
each of them is detected in a different output, one can infer a source of pseudosingle photons from an attenuated laser
that the two photons in mode a were horizontally and verti- light (Zhao et al., 2004). With further improvements in high-
cally polarized. The photon of polarization H was transmitted power pump source and photon collection efficiency, GHZ
through the PBS and detected by the trigger detector T, and entanglement among six photons (Lu et al., 2007) and eight
the other one, V polarized, must have been reflected by the photons (Yao et al., 2012) was also observed. The number of
PBS and consequently
pffiffiffi rotated by the half-wave plate (HWP) GHZ-like entangled qubits climbed up to ten by entangling
into state ð1= 2ÞðjHi þ jViÞ. Concerning the two photons in both the polarization and momentum degrees of freedom of
mode b, they must have been one of polarization H and the five photons (Gao, Lu, et al., 2010).
other of V, in order to match the polarizations in mode a. SPDC is quite versatile, as it enables one to observe a
After a random distribution by the BS, one photon can be number of other, genuinely multipartite entangled states.
detected by the detector D3 while the other one can be Emission of two photons into each of two modes gives
combined on a PBS with the photon from mode a. already a highly entangled state and is significantly different
Therefore, if each of the three detectors D1, D2, and D3 from simply emitting two pairs (Weinfurter and Żukowski,

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

800 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

2001). After beam splitting of these two modes into four correlations of all subsystems, such that there is no subsystem
modes and again conditioning on detection of one photon in which is just classically correlated with the other particles.
each of the four modes, the observed state can be written as a Detection of a genuine multiparticle entanglement with ap-
superposition of a four-photon GHZ state and a product of propriate witnesses usually requires an experimental effort
two EPR pairs. This state, first observed by Eibl et al. (2003), that increases only polynomially with the number of qubits. A
is the extension of the singlet state j
i to four photons toolbox for efficient witness operator construction has been
and is thus invariant under equal unitary transformations in created for some multiparticle states [such as GHZ, cluster,
the four output modes. This enables a decoherence-free and W state, see, e.g., Gühne et al. (2007)], and applied in a
communication of a qubit encoded in four-photon states number of multiphoton experiments [see, e.g., Bourennane,
(Bourennane, Eibl, Gaertner, et al., 2004). By splitting the Eibl, Kurtsiefer, et al. (2004); Kiesel, Schmid, Toth, et al.
two emission modes into three output modes the six-photon (2005); Lu et al. (2007), and Wieczorek, et al., 2009]. We
singlet was also observed recently by Radmark, Zukowski, refer the interested reader to Horodecki et al. (2008) and
and Bourennane (2009), with visibility of the six-photon Gühne and Toth (2009) for more details.
interference as high as 85%. The high visibility is possible
because the distinctive feature of this scheme is that it does
not involve interferometric overlaps; only beam splitting is V. FALSIFICATION OF A REALISTIC WORLD VIEW
necessary.
With a detailed analysis of the work of EPR and its
Another important multipartite entangled state, the sym-
extension by Bohm (1951), in a trailblazing paper Bell
metric Dicke state, can be obtained by using collinear type II
(1964) proved, that despite the hopes of Einstein et al., there
SPDC and splitting the four (six) photons into four (six)
is a deep conflict between quantum mechanics and local
output modes (Kiesel et al., 2007; Prevedel et al., 2009;
realistic theories. Not only a conceptual one, which was
Wieczorek, Krischek, Kiesel, Schmid, 2009; Wieczorek
stressed by EPR is their claims concerning incompleteness
et al., 2009). The high symmetry of this state makes it an
of quantum mechanics, but one which straightforwardly leads
ideal resource to synthesize a number of different multipartite
to drastically different predictions concerning two-particle
entangled states by projection as shown above for the GHZ
interference phenomena.
states (Wieczorek, Kiesel, Schmid, Weinfurter, 2009), but
Realism, the cornerstone of classical physics, is a view that
also for entanglement enhanced subshot-noise measurements.
for any physical system (also a subsystem of a compound
Finally, all the above states can also be observed by a single,
system), one can find a theoretical description (deterministic
tunable setup. There, a wave plate rotating the polarization in
or probabilistic) which involves all results of all possible
one of the two emission modes of SPDC, followed by a PBS
experiments that can be performed upon it no matter which
combining these two emission modes, serves to set any
experiment actually was performed. Evidently, this directly
superposition between a GHZ state and the product of two
contradicts the Bohr’s complementarity principle. A theory is
entangled pairs [Wieczorek et al. (2008); see also Lanyon
local if it assumes that information, and influences, cannot
and Langford (2009)].
travel faster than light. By assuming that local measurement
To characterize the created multiphoton entangled states,
events are determined by other events (i.e., causes) in their
various methods have been introduced. Quantum state tomog-
backward light cone, the term local realism could be replaced
raphy [for an introduction, see James et al. (2001)] uses
by local causality.
projective measurements on an ensemble of identically pre-
Bell’s famous theorem is of profound scientific and phil-
pared quantum states each probing the state from a different
osophical consequences. It also showed that the previously
‘‘perspective.’’ It is a tool to fully reconstruct the density
ignored class of entangled states is very important in experi-
matrix of a quantum system.28 Experimentally, this technique
mentally distinguishing between the classical and the quan-
was used for two-photon, three-photon, four-photon polariza-
tum. We first present the formal aspects of Bell’s theorem as
tion states, and also hyperentangled photon pairs (White
well as other forms of ‘‘no-go theorems’’ for hidden-variable
et al., 1999; Barreiro et al., 2005; Resch et al., 2005;
theories.29 Next, we present the most important photonic tests
Walther, Resch, Rudolph et al., 2005). A disadvantage
invalidating classes of such theories.
of such a method, however, is that the number of measure- An important line of research was opened with the discov-
ments grows exponentially with the number of photons, ery of the GHZ theorem, which pertains to three or more
thus the reconstruction of a n-photon state necessitates 4n particle systems, and reveals a contradiction between
n-fold coincidence measurements which is experimentally quantum mechanics and local realistic theories, even for
demanding.
Entanglement witnesses–The method of entanglement wit-
29
nesses (Horodecki, Horodecki, and Horodecki, 1996; Hidden variables are those hypothetical parameters that suppos-
Lewenstein et al., 2000; Terhal, 2000; Bruß et al., 2002) edly influence the results of individual measurement acts, but are
allows one to detect entanglement via measuring a suitable not present in the standard mathematical structure of quantum
observable. One speaks of genuine multipartite entanglement mechanics. This is why they are called hidden. If one introduces
(Tóth and Gühne, 2005) whenever the state involves quantum to the theory expressions containing algebraic functions of results
of, e.g., pairs of, noncommeasurable variables, this is tantamount to
the introduction of hidden variables (such operations are impossible
28
A similar technique, called quantum process tomography, has in quantum mechanics). Also, as causes of individual measurement
been used to fully characterize the quantum CNOT gates (O’Brien events do not appear in the quantum formalism, they are hidden
et al., 2004). variables as well.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 801

X X
definite predictions. This result was the initial motivation for Sðs1 ; s2 ÞðAj1 þ s1 Aj1 ÞðBj1 þ s2 Bj2 Þ ¼ 4;
experimental efforts to produce entanglement of more than s1 ¼1 s2 ¼1
two particles. With the advances in multiphoton entangled- (51)
state preparation, discussed in the previous section, a new
class of tests of the validity of local realistic theories became where Sðs1 ; s2 Þ is an arbitrary ‘‘sign’’ function of s1 and s2 ,
possible. Note that, as any classical information processing, that is, one always has jSðs1 ; s2 Þj ¼ 1. Imagine now that N
or communication, has a local realistic model, the theorems pairs of photons are emitted, pair by pair, and N is sufficiently
and experiments that reveal phenomena impossible to de- pffiffiffiffiffiffiffiffiffi
large, 1=N  1. The average value of the products of the
scribe by such formalism clearly indicate existence of a
local values for a joint test (the Bell correlation function),
different method of processing and transferring information.
during which, for all photon pairs, only one pair of observ-
That is why quantum information processing is able to per-
ables (say, A^ n and B^ m ) is chosen, is given by
form tasks impossible with the classical methods.
X j j
1 j¼N
A. Bell’s inequality EðAn ; Bm Þ ¼ An B m ; (52)
N j¼1
Consider pairs of photons simultaneously emitted in oppo-
site directions. They arrive at two very distant measuring where n ¼ 1, 2 and m ¼ 1, 2. Equation (51) after averaging
devices A and B operated by Alice and Bob, respectively. implies, together with Eq. (52), that for the four possible
Their apparatuses have a ‘‘knob,’’ which specifies which choices the following inequalities33 must be satisfied for local
dichotomic (i.e., two valued, yes no) observable they mea- realistic descriptions [see Weinfurter and Żukowski (2001)
sure.30 One can assign to the two possible results (eigen) and Werner and Wolf (2001b)]:
values 1 (for yes or no).31 Alice and Bob are at any time free
to choose the knob settings (also in a ‘‘delayed-choice’’ X

4 Sðs1 ; s2 Þ½EðA1 ; B1 Þ þ s2 EðA1 ; B2 Þ
mode, after an emission). s1 ¼1;s2 ¼1
Assuming realism allows one to introduce and compare
values of results of all possible experiments that can be þ s1 EðA2 ; B1 Þ þ s1 s2 EðA2 ; B2 Þ  4: (53)
performed on an individual system, without caring which
measurement would be actually done on the system. If one chooses a nonfactorizable function Sðs1 ; s2 Þ, say,
2 ð1þ s1 þ s2
s1 s2 Þ, the famous Clauser, Horne,
1
According to locality, random choices and the consecutive
observations made by Alice and Bob, which can be simulta- Shimony, and Holt (CHSH) (Clauser, et al., 1969) in-
neous in certain reference frames, cannot influence each equality is recovered34
other, and the choice made on one side cannot influence the
results on the other side, and vice versa.
To test local realism Alice chooses randomly, with equal SBell
jEðA1 ; B1 Þ þ EðA1 ; B2 Þ þ EðA2 ; B1 Þ
EðA2 ; B2 Þj
probability, to measure either observable A^ 1 or A^ 2 , and Bob  2: (54)
either B^ 1 or B^ 2 . Denote the hypothetical results that they may
get for the jth pair by Aj1 and Aj2 for Alice’s two possible
We now discuss one more assumption, sometimes provo-
choices,32 and Bj1 and Bj1 for Bob’s choices. The numerical
catively called of free will. For the actual experiment, we
values of these can be 1. The assumption of local realism assume that choices of actual observables are random and
allows one to treat Aj1 and Aj2 on equal footing as two independent from all other processes in the experiment. Note
numbers, one of them revealed in the experiment, the second that only in a part of the cases (around 1=4) the given pair of
one unknown, but still either 1. Thus, their sum and differ- observables [see Eq. (52)] would be measured. However, as N
ence always exist, and are algebraic expressions with two is large, the correlation function obtained on a randomly
unknowns. For the dichotomic values for all the possible preselected subensemble35 of pairs, due to the aforemen-
measurement results, one obtains either the combination tioned randomness and independence, cannot differ too
jAj1
Aj2 j ¼ 2 and jAj1 þ Aj2 j ¼ 0, or jAj1
Aj2 j ¼ 0 and much from the one that would pffiffiffiffi have been obtained for the
jAj1 þ Aj2 j ¼ 2, and similarly for Bob’s values. Thus, the full ensemble (say, by 2= N ), as we deal with two statis-
following trivial relation holds: tically independent processes. Therefore, the results of the

33 P
All such inequalities boil down to just a single one s1 ¼1
P
s2 ¼1 jEðA1 ;B1 Þ þ s2 EðA1 ;B2 Þ þ s1 EðA2 ;B1 Þ þ s1 s2 EðA2 ;B2 Þj  4.
30 34
E.g., for a device consisting of a polarizing beam splitter and two The simple algebra to reach this result rests on the fact that
P P
sj ¼1 sj ¼ 0, while sj ¼1 sj ¼ 2.
detectors behind its outputs, this knob specifies the orientation of the 2
35
polarizer, etc. Note that the subensemble is selected by the choice of observ-
31
We assume, that we have a perfect situation in which the ables made by Alice and Bob before the actual measurements. If the
detectors never fail to register a photon. choices are statistically independent of any other processes in the
32
Note that if the results were depending also on the settings a experiment, which is equivalent to Alice and Bob having free will,
Bob’s side, one would use a two index notation Aja;b , with a, b ¼ 1, expectation values of correlation functions conditioned on a par-
2, where a and b index Alice’s and Bob’s choices of the settings. As ticular choice of local settings do not differ from the unconditional
index b is missing, this is tantamount to the locality assumption. ones, like Eq. (52). For more details see (Gill et al., 2002).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

802 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

actually chosen measurements, under all the three assump- whether a detector registers a particle or not is statistically
tions, must satisfy a Bell inequality in the form of Eq. (54), up
pffiffiffiffi independent of all other processes in the experiment.39 Of
to minor statistical fluctuations of the order of 1= N . Note course, such an assumption is highly questionable. One can
that the Bell-type argument presented above avoids any ex- find many ad hoc local realistic models that violate it; see,
plicit introduction of hidden variables, other than the hypo- e.g., Cabello and Santos (1996). For example, for qubits one
thetical local realistic values. could assign three possible local outcomes: þ=
1 and no
Some quantum processes involving entangled states vio- count, or in the case of polarization experiments, one could
late the inequality.36 For example, the predictions for the expect the response of the detection systems might depend on
spin-1=2 singlet give correlations
pffiffiffi for which SBell can acquire the photon’s polarization, even without turning to hidden-
the maximal value 2 2 allowed by quantum mechanics, variable approaches. In case of some experiments, especially
known as the Cirel’son37 bound (Cirel’son, 1980; Landau, the early cascade ones, the polarization state of the photons
1987). In fact, predictions for any pure, nonfactorisable (i.e., depends on the direction of emission, etc.
not necessarily maximally entangled) two-system state lead The famous Aspect et al. experiments were the pioneering
to violations (Gisin and Peres, 1992). This is also the case for attempt to address the locality loophole. To close the locality
a wide range of mixed states (Horodecki, Horodecki, and loophole, one must freely and rapidly choose the directions of
Horodecki, 1995). local analyzers and register the particles in such a way that it
Experimental tests of Bell’s inequality: The initial experi- is impossible for any information about the setting and the
ments using photon pairs produced in an atomic cascade to detection to travel via any (possibly unknown) causal channel
test Bell’s inequalities [Freedman and Clauser (1972); to the other observer before he or she, in turn, chooses the
Aspect, Dalibard, and Roger (1982); Aspect, Grangier, and setting and finishes the measurement. Thus, the selection of
Roger (1982); see also Clauser and Shimony (1978) for more analyzer directions has to be completely unpredictable, which
experiments], falsified Bell’s inequalities, and thus chal- necessitates a (quantum) physical random number generator.
lenged the local realistic world view. However, this falsifica- A numerical pseudorandom number generator would not do:
tion was up to certain loopholes, which are due to its state at any time is predetermined. Furthermore, to achieve
experimental imperfections, and still allow one to build local a complete independence of both observers, one should avoid
realistic models for the measurement results obtained in the any common context, as would be a conventional use of
experiments. coincidence circuits. Individual events should be registered
The locality loophole is present in experiments which do on both sides completely independently, and compared only
not have independent, i.e., space-time separated measurement long after the measurements are finished. This requires inde-
settings, can be guaranteed only with random and fast switch- pendent and highly accurate time bases on both sides.
ing of the local measurement settings. In such a case, one of Aspect’s experiments were the first to use fast but periodic
the assumptions behind Bell’s inequalities, full certainty of switching of the local polarization analyzers. Although the
the independence of Alice’s results on Bob’s settings, or vice settings were quickly varying, they were derived from inde-
versa, is not enforced. The efficiency loophole emerges due to pendent function generators with certain correlation times
low collection and detection efficiency of the particles. For an and thus not fully random as it is assumed in the derivation
efficiency lower than about 83%, one can show that one of Bell inequalities. This independent randomness was ex-
cannot derive a (generalized) CHSH-type inequality that is perimentally realized by Weihs et al. (1998). The observers
violated by quantum predictions; see, for example, Garg and ‘‘Alice’’ and ‘‘Bob’’ were spatially separated by 400 m across
Mermin (1987) and Eberhard (1993) managed to lower this the Innsbruck University science campus. The individual
threshold to 67% by effectively, employing, the Clauser and measurements were finished within a time much shorter
Horne (1974) inequalities, albeit for nonmaximally entangled than 1:3 s, which is the distance of the two observation
states only.38 In the analysis of experiments with efficiency stations in light seconds. The actual orientation for local
loophole, many authors usually use the so-called fair sam- polarization analysis was determined independently by a
pling assumption, which states that one expects that the fact quantum physical random number generator (Jennewein
et al., 2000). The employed generator had a light-emitting
36
The CHSH inequality was the first experimentally testable Bell diode illuminating a BS whose two outputs were monitored
inequality. The original Bell (1964) inequality, since it assumes by a pair of photomultipliers. Upon receiving a pulse from
perfect correlations of the singlet state, cannot be tested experimen- one photomultiplier a 0 was produced, whereas the pulse
tally, as in such a case correlations are never perfect. A general- coming form the other one was giving a 1. This results in a
ization of the original inequality to the imperfect case leads to the set of binary random numbers (Fig. 17). The polarization-
CHSH inequality. Nevertheless, the original inequality clearly entangled photon pairs were distributed to the observers
reveals the conflict between local realism and quantum theory. through long optical fibers. A typical observed value of
37
In the meantime, the transliteration of this surname was changed SBell , the right side of inequality (54), was as high as 2:73 
to Tsirelson. 0:02. In 10 s 14 700 coincidence events were collected. This
38
All of these results are derived under the usual assumption that
efficiency of all detectors is independent of the locally measured
39
observable, and equal for all detectors. E.g., a local model explain- When discussing experiments which were not directly aimed at
ing correlations of a maximally entangled state in which detector closing the detection loophole, we shall give an analysis of the
efficiency is dependent on the measured observables was proposed experimental data which always would tacitly assume the fair
by Selleri and Zeilinger (1988). For a more recent discussion of the sampling assumption. However, we shall not repeat this statement
efficiency loophole, see, e.g., Vertesi, Pironio, and Brunner (2010). ad nauseam. This pertains also to the locality loophole.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 803

in ‘‘relativistic configurations’’ stress the oddness of quan-

tum correlations.40

B. GHZ theorem

1. Impossibility of deriving realism via perfect quantum

correlations and locality
If there are N > 2 maximally entangled quantum systems
(qubits), and if measurements on them are performed in N
mutually spatially separated regions by N independent ob-
servers, the correlations in such an experiment violate bounds
imposed by local realism much stronger than in the two-
particle case. More remarkably, instead of purely statistical
reasoning for deriving Bell’s inequality, one can fully follow
the spirit of the EPR paper, and first try to define ‘‘elements of
reality’’ based on specific perfect correlations of the en-
tangled state. In a further step, one then can show contra-
FIG. 17 (color). One of the two observation stations. A random dicting predictions between local realistic theories and
number generator is driving the electrooptic modulator. Silicon quantum mechanics with precisely those quantum predictions
avalanche photodiodes are used as detectors. A ‘‘time tag’’ is stored which were used to define the ‘‘elements of reality’’
for each detected photon together with the corresponding random (Greenberger, Horne, and Zeilinger, 1989).
number 0 or 1 and the code for the detector ‘‘þ’’ or ‘‘
’’ Take the GHZ experiment, Fig. 2. Assume that a click at
corresponding to the two outputs of the polarizer. From Weihs the local detector Dx1 , where x ¼ d, e, f is described as a
et al., 1998. result of value þ1, whereas clicks at Dx2 are ascribed
1.
According to the quantum probabilities (20), the average
values of the product of local results reads
corresponds to a violation of the local realistic threshold of 2 X
by 30 standard deviations. Still the experiment had a detec- Eð
a ;
b ;
c Þ ¼ ð
1Þiþjþkþ1 pdi ej fk ð
a ;
b ;
c Þ
i;j;k¼1;2
tion efficiency well below the required minimum allowing
one to avoid the fair sampling assumption. ¼ sinð
A þ
B þ
c Þ: (55)
Meanwhile, there were many ideas to close the detection
loophole; see, e.g., Eberhard (1993) and Kwiat et al. (1994). Here pdi ej fk ð
  Þ is the probability for a detection of one
It was first closed in an ion-trap experiment by Rowe et al. photon by detectors di , ej , and fk , given the phase settings
(2001) utilizing the nearly perfect detection efficiency of
a    . Therefore, if
a þ
b þ
c ¼ =2 þ k (where k
fluorescence detection of single ions. However, as the two is an integer), one has perfect correlations and perfect pre-
entangled ions were separated by approximately 3 m, the dictability of the common measurement result. For instance,
locality loophole was left widely open. A recent experiment for
a ¼ =2,
b ¼ 0, and
c ¼ 0, whatever may be the
by Matsukevich et al. (2008) involved two separate ion traps results of local measurements of the observables for, say, the
(1 m distance) and an entanglement-swapping procedure particles belonging to the nth triple represented by the GHZ
involving the photons emitted by the ions. The detection quantum state, they have to satisfy
loophole was again perfectly closed. This method gives
hope for a future experiment simultaneously closing both An ð=2ÞBn ð0ÞCn ð0Þ ¼ 1; (56)
loopholes (Simon and Irvine, 2003; Rosenfeld et al.,
2009). Violation of Bell’s inequality was also demonstrated where X n ð
Þ (X ¼ A, B, or C) is the value of a local
using Josephson phase qubits with deterministic entangled- measurement, by Alice, Bob, and Cecil, respectively, that
state preparation and single-shot readout (Ansmann et al., would have been obtained for the nth particle triple, if the
2009). setting of the measuring device is
. Locality assumption
Other aspects of entanglement have been demonstrated forces one to assume that X n ð
Þ depends solely on the local
distributing entanglement over long distances. For example, phase. Equation (56) indicates that we can predict with
over 10.5 km free space in Hefei by Peng et al. (2005), or
in an asymmetric arrangement, in the case of which one 40
As in a properly performed Bell experiment, the measurement
photon is sent over 144 km between the islands of
events have to be spatially separated, in the relativistic meaning of
La Palma and Teneriffe (Ursin et al., 2007). Bell experi- this word (the space-time interval between them has to be of the
ments were also performed using fiber-based entangled- spatial type), they are simultaneous in a certain reference frame.
photon sources, from which two photons were distributed However, if the detectors move with respect to each other and the
over a distance of more than 10 km apart; see, e.g., Tittel source, the detection events cease to be simultaneous in the rest
et al. (1998) and Zbinden et al. (2001). The later experi- reference frames of respective detectors, etc. One can have various
ment and the subsequent ones (Stefanov et al., 2002, 2003) temporal sequences. The experiments showed that even with mov-
were done with moving reference frames. These Bell tests ing detectors the expected quantum correlations occur.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

804 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

certainty the result of measuring the observable pertaining to an arbitrary number of parties.41 The noise resistance42 of
one of the particles (say, C) by choosing to measure suitable violations of such inequalities by N-qubit GHZ states is
observables for the other two. Hence, in a local realistic growing exponentially with N. This is an important fact,
model the values Xn ð
Þ for the angles specified in the equal- because usually one expects noise to increase with the num-
ity are EPR’s elements of reality. ber of photons involved in an interferometric experiment (due
Similarly, when measuring different settings, according to to the increasing complications and alignment problems).
Eq. (55), one would obtain Thus, nonclassicality of the GHZ correlations can be signifi-
cant even for many particles (Mermin, 1990b; Roy and Singh,
An ð0ÞBn ð0ÞCn ð=2Þ ¼ 1; (57) 1991; Klyshko, 1993; Żukowski, 1993; Gisin and Bechmann-
Pasquinucci, 1998).
An ð0ÞBn ð=2ÞCn ð0Þ ¼ 1: (58)
2. GHZ theorem for laboratory measurement
Now, in a local realistic model, from these results we The first laboratory test of the GHZ-type paradox was done
can deduce a further correlation by simply multiplying by Pan et al. (2000). The experiment used a three-photon
Eqs. (56)–(58). Since X n ð0Þ2 ¼ þ1, regardless of whether GHZ state
X n ð0Þ ¼ þ1 or
1, we obtain
1
j
i ¼ pffiffiffi ðjHi1 jHi2 jHi3 þ jVi1 jVi2 jVi3 Þ: (63)
An ð=2ÞBn ð=2ÞCn ð=2Þ ¼ 1: (59) 2
The data obtained in the form of the (necessarily imperfect)
This, however, is in a full contradiction with the quantum GHZ correlations were shown to violate local realism.
mechanical prediction obtained from Eq. (55) which reads The GHZ state (63) satisfies the following eigen equations:
x^ 1 y^ 2 y^ 3 j
i ¼
j
i; y^ 1 x^ 2 y^ 3 j
i ¼
j
i;
An ð=2ÞBn ð=2ÞCn ð=2Þ ¼
1: (60) (64)
y^ 1 y^ 2 x^ 3 j
i ¼
j
i; x^ 1 x^ 2 x^ 3 j
i ¼ j
i;

Thus, the EPR elements of reality program breaks down, where x^ denotes the observable discriminating between j45 i
because it leads to a 1 ¼
1 contradiction. Introduction of and j135 i polarizations, whereas y^ discriminates between
EPRs elements of reality leads to a prediction concerning one left and right circular polarizations. In both cases, the as-
of the perfect correlations, Eq. (59), which is opposite with cribed eigenvalues are 1 and
1, respectively. With these
respect to the quantum prediction. We have a ‘‘Bell theorem settings, one can get a GHZ paradox falsifying the elements
without inequalities’’ (Greenberger et al., 1990), which of reality of the form described in the previous section.
destroys any hopes to derive realism from locality and perfect Demonstration of the conflict between local realism and
correlations of the EPR type, as a necessary condition for quantum mechanics for GHZ entanglement consists of four
any reasoning to be logically valid is that it does not lead experiments. In the experiment by Pan et al. (2000), the
to a 1 ¼
1 contradiction. measured values for x^ 1 y^ 2 y^ 3 , y^ 1 x^ 2 y^ 3 , and y^ 1 y^ 2 x^ 3 followed the
Multiparticle Bell inequalities–In a laboratory one cannot values predicted by quantum physics in a fraction of 0:85 
expect perfect correlations, and even if they were possible, 0:04 of all cases. The fourth experiment, measuring x^ 1 x^ 2 x^ 3 ,
any necessarily finite sample would be endowed with a finite, was also performed, and yielded results as shown in Fig. 18.
nonzero uncertainty. Thus, any test of local realism based on The data again agree with quantum mechanics for the same
the GHZ correlations has to resort to some Bell-type inequal- fraction of events. The results are in a clear disagreement with
ities. Series of such inequalities were discovered by Mermin a prediction range that can be made with the data of the three
(1990b), Ardehali (1992), and Belinskii and Klyshko (1993). first measurements using a local realistic model.
To get the full set of such inequalities, for correlation func- The experimental results confirmed the quantum predic-
tions involving the product of the result of all parties, it is tions, within an experimental uncertainty. The obtained
enough to generalize Eq. (52) to the situation in question. For average visibility43 of ð71  4Þ% clearly surpasses the
example, extending Eq. (51) for three partners one has threshold of 50%, necessary for a violation of local realism
X in three-particle GHZ experiments (Mermin, 1990b; Roy and
Sðs1 ; s2 ; s3 ÞðAn1 þ s1 An2 ÞðBn1 þ s2 Bn2 Þ
s1 ;s2 ;s3 ¼1
41
As a matter of fact, this single inequality either implies all earlier
ðCn1 þ s3 Cn2 Þ ¼ 8: (61) derived tight inequalities, e.g., those of Mermin (1990b).
42
This leads to the following Bell inequality (Weinfurter and As indicated by the decreasing threshold visibility for mixed
states of the form ð1
VÞ 21N I^ þ VjGHZN ihGHZN j, which is suffi-
Żukowski, 2001; Werner and Wolf, 2001b):
cient to violate the inequalities.
X X 43
j s1 s2 s3 EðAk ; Bl ; Cm Þj  8; (62) The imperfect visibilities, which are typical in multiphoton
s1 ;s2 ;s3 ¼1 k;l;m¼1;2 experiments, are mainly caused by two reasons (see also
Sec. IV.D.1). First, higher-order emissions of entangled photons
which is the necessary and sufficient condition for correlation give rise to the undesired components in the multiphoton states.
functions EðAk ; Bl ; Cm Þ to have a local realistic model [for Second, the partial distinguishability of photons from different
proofs, see Werner and Wolf (2001b) and Żukowski and emissions or sources causes some degree of incoherence. There
Brukner (2002)]. The reasoning is trivially generalizable to may be also alignment problems.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 805

considerable debate (Cabello, 2001a; Cabello, 2001b;

Lvovsky, 2002; Cabello, 2003; Chen et al., 2003;
Marinatto, 2003), it was shown that an all-versus-nothing
violation of local realism can be shown for two-particle
four-dimensional entangled systems. In this new refutation
of local realism, one recovers EPR’s original situation of two-
party perfect correlations, but with much less complexity. This
becomes possible with a new approach for defining elements
of reality, which nevertheless strictly follows the EPR criteria.
A third protocol of the two-observer GHZ-like theorem
uses a two-photon hyperentanglement (Chen et al., 2003).
Because of the specific properties of the hyperentanglement,
nine variables for each party can be regarded as simultaneous
EPR elements of reality. The nine variables can be arranged
in three groups of three, and the three variables of each group
can be measured by one and the same apparatus. This elim-
inates the necessity of an argument based on noncontextuality
as it is not necessary to assume any of these variables to be
independent of local experimental context. Experimental
demonstrations of such a protocol were done by Cinelli
et al. (2005) and Yang et al. (2005)45 using a two-photon
hyperentangled state in Eq. (40) (for hyperentanglement see
Sec. IV.B.2).

4. Hardy’s paradoxes
Hardy’s thought experiment (Hardy, 1993) provides an
alternative way to demonstrate a violation of local realism
without inequalities for two spin-half particles, or equiva-
lently for polarizations of two photons. A crucial distinguish-
ing feature in Hardy’s thought experiment is that the two
particles are nonmaximally entangled. In such a case, in the
ideal situation, for a specific set of measurements quantum
FIG. 18. (a) Predictions of quantum mechanics and (b) of local mechanics predicts that approximately 9% of the pairs of
realism, and (c) observed results for the final x^ 1 x^ 2 x^ 3 measurement in photons must give measurement results in a form of coinci-
the first GHZ-type experiment. H 0 and V 0 denote here diagonal and
dence counts absolutely not allowed by local realism.46 The
antidiagonal linear polarizations. From Pan et al., 2000.
original proposal was experimentally demonstrated by
Torgerson et al. (1995) and White et al. (1999); as in the
Singh, 1991; Ryff, 1997; Żukowski and Kaszlikowski, GHZ-type experiments, the results of the experiments were
1997). fed into specific inequalities, specially derived to take into
Four-photon entanglement was later demonstrated by Pan account experimental imperfections (under the fair sampling
et al. (2001) and Eibl et al. (2003), and used for a corre- assumption, of course). Their violation indicates underlying
sponding multiphoton falsification of local realism [see also Hardy’s contradiction between local realism and quantum
Zhao et al. (2003)]. mechanics. Hardy’s scheme was later scaled up, so that
ð50
hÞ% of photon pairs lead to coincidence events pro-
3. Two-observer GHZ-like correlations hibited by local realism (where h is any small finite number).
The effect was demonstrated in an experiment by Boschi
Interestingly, the GHZ reasoning can be reduced to a two- et al. (1997).
party (thus two spacelike separated regions) case while its all- Another proposal suggested by Hardy (1992) was imple-
versus-nothing feature is still retained. One option is to mented using a pair of Mach-Zehnder interferometers that
encode three two-state quantum systems in distinct degrees couple via the bosonic photon bunching effect at a beam
of freedom of only two photons. Thereby, a GHZ-type argu- splitter (Irvine et al., 2005). The original idea was based
ment, now also necessarily involving noncontextuality,44 can on a double Mach-Zehnder interferometer that involved
be applied in this two-particle scheme (Żukowski, 1991; particles that annihilate each other (say, electron and
Michler, Weinfurter, and Żukowski, 2000). The second option
is to find suitable EPR elements of reality in the two-particle 45
case, and to show that they are internally inconsistent. Unfortunately, the experimental configuration in the former
Such an approach has been taken by Bernstein et al. experiment did not eliminate the necessity of the noncontextuality
(1993) for a (spinless) two-particle interferometer. Later, after assumption. Nevertheless, upon a permutation in the setup one
could get an arrangement avoiding this problem.
46
Note that in the GHZ scenario, one has situations in which 100%
44
See Sec. V.D for a discussion of this concept. coincidences are not allowed by local realism.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

806 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

positron). The right internal path of the electron interferome- Leggett (2003), and experimentally by Gröblacher et al.
ter is at some place partially overlapping with the left internal (2007). Subsequently, a reformulation enabled one to reduce
path of the positron interferometer. The individual interfer- the dependence on auxiliary assumptions as shown indepen-
ometers are tuned such that if only one of the particles is fed dently by Paterek et al. (2007) and by Branciard et al. (2007)
to its interferometer, it would always exit by the left exit port. [see also Romero et al. (2010)].
However, if both electron and positron are simultaneously fed We now discuss the description of the polarization of
into their interferometers, there is a nonzero probability photons within such theories. The following assumptions
amplitude of their annihilation. This is related to both parti- are made: (L1 ) realism, (L2 ) physical states are statistical
cles being in the overlapping arms of the interferometers mixtures of subensembles with definite polarizations, and
(within the story, in such a case the annihilation is assumed (L3 ) local expectation values for polarization measurements
to happen with probability 1). The two particles act on each taken for each subensemble obey Malus’ law.
other like (two) bombs of the Elitzur and Vaidman (1993) Importantly, locality is not assumed. Measurement out-
paradox.47 Thus, if one treats in a realistic way the presence, comes may depend on whatever the parameters in spacelike
or the absence, of the particles in the internal paths of the separated regions are. It can be shown that such theories can
interferometers, they can never be registered both in the right explain some important features of entangled states of two
exit ports of their interferometers, because this would mean particles: first, by assumption (L3 ), they do not allow infor-
for each of them that the other particle was traveling via the mation to be transmitted faster than at the speed of light;
overlapping arm. That is, both of them were there, thus second, they are capable to reproduce perfect anticorrela-
annihilation must occur. Still, quantum prediction gives a tions, a fundamental feature of the Bell singlet state; and
probability of 1=16 of both particles emerging via the right third, they provide a model for all Bell-type experiments in
exit ports. Of course, all that is a gedanken experiment per se. which the CHSH inequality was violated. Nevertheless, theo-
To formulate a feasible version of it, Irvine et al. resorted to ries based on assumptions ðL1 Þ–ðL3 Þ deliver predictions dif-
photons, and a pair of Mach-Zehnder interferometers with ferent from the quantum ones for certain other measurement
one of their internal mirrors replaced by a shared 50–50 beam outcomes.
splitter. Instead of annihilation, they relied on the Hong-Ou- We now discuss a general mathematical structure of such
Mandel bunching effect: if two indistinguishable photons models. We concentrate on the description of events at
enter the shared beam splitter, such that each enters by a Alice’s side; events at Bob’s side must follow a similar model.
different input port, they exit randomly via just one port (see Assumption (L1 ) allows an individual binary measurement
Sec. III.C). Thus, if two photons meet (instead of annihila- outcome A for any possible polarization measurement along
tion) one of the interferometers looses the light traveling any direction a (that is, whether a single photon is transmitted
through it, and no detection is possible in its exit ports. or absorbed by a polarizer set at a specific angle) to be a well-
Using a similar realistic reasoning as before, this leads to defined function of some set of hidden variables , and, by
the conclusion that no joint detection in the right exit ports of (L2 ), of a three-dimensional vector48 u. As locality is not
the two interferometers is possible. The quantum prediction is assumed, A can depend on some set of other possibly non-
different, and the experiment gave results agreeing with it. local parameters  and the remote setting of Bob, b. That is,
the measurement outcome A depends on these five variables
A ¼ Að; u; a; ; bÞ, and can take values 1 (two distinct
C. Refutation of a class of nonlocal realistic theories measurement outcomes). According to assumption (L3 ),
particles with the same u but with different s and s build
Violation of local realism implies that either locality or up subensembles of ‘‘definite polarizations’’ described by a
realism, or both, cannot provide a foundational basis for probability distribution u ð; Þ, and the local expectation
quantum theory (provided the freedom assumption of ran- 
value AðuÞ, obtained by averaging over R and , ful-
domness and independence of setting choices holds). In a fills Malus’ law, that is, AðuÞ  ¼ ddu ð; Þ
novel approach, Leggett (2003) discussed a broad class of Að; u; a; ; bÞ ¼ u  a. Finally, with assumption (L2 ), the
nonlocal hidden-variable theories, which, based on a very measured expectation value for a general physical state is
plausible type of realism, provide an explanation for all given by averaging R over a distribution FðuÞ of the subensem-
existing Bell-type experiments. Nevertheless, they are in bles, that is, hAi ¼ duFðuÞAðuÞ.  Of course, one introduces
conflict with quantum predictions as shown theoretically by a similar dependence for Bob’s measurement outcomes,
B ¼ Bð; v; b; 0 ; aÞ, now depending on Bob’s vector v.
47
Imagine again a Mach-Zehnder interferometer, tuned in such a The correlation function of measurement results for a
way that all photons emerge via its ‘‘left’’ exit. Just a single photon source emitting well-polarized photons is defined as the
enters it. In the meantime, somebody may put an ultrasensitive light average of the products of the individual measurement out-
detector into one of the paths, a bomb triggered by light in the comes:
Elizur-Vaidman anecdote. If the bomb is there, then we either have Z
an explosion, or the photon emerges with equal probability from ABðu; vÞ ¼ ddd0 u;v ð; ; 0 ÞAð; u; a; ; bÞ
both outputs of the interferometer (if the bomb does not register the
photon it must be propagating in the other internal path). In the case Bð; v; b; 0 ; aÞ: (65)
it exits via the ‘‘right’’ exit, we have detected the bomb, using light
(photon), without igniting it. This is often called ‘‘interaction-free
48
measurement.’’ If there is no bomb, nothing changes, and all We use here the Bloch sphere, or spin-1=2-like, parametrization
photons emerge via the left exit. of polarization states and measurement settings.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 807

For a general source producing mixtures of polarized pho-

tons, the observable correlations are averaged over a distri-
bution of the polarizations Fðu; vÞ, and the general correlation
function E is given by
Z
E ¼ hABi ¼ dudvFðu; vÞABðu; vÞ: (66)

It is an important trait of this model that there exist sub-

ensembles of definite polarizations (independent of measure-
ments) and that the predictions for the subensembles agree
with Malus’ law. It is clear that other classes of nonlocal
theories, possibly even fully compliant with all quantum
mechanical predictions, might exist that do not have this
property when reproducing entangled states. Such theories FIG. 19 (color online). The experimental violation of Leggett-type
may, for example, include additional communication or di- nonlocal realism in Gröblacher et al. (2007). The dashed line shows
mensions. A specific case deserving comment is Bohm’s the bound of inequality (68) for the particular class of nonlocal
theory (Bohm, 1951). There the nonlocal correlations are a realistic theories. The solid line is the quantum prediction reduced
consequence of the nonlocal quantum potential, which exerts to fit the experimentally achieved visibility. The experimental data
suitable torque on the particles leading to experimental re- were taken for various difference angles of local measurement settings
sults compliant with quantum mechanics (Dewdney, Holland, (parametrized as on the Poincaré sphere). From Gröblacher et al.,
and Kyprianidis, 1987). 2007.
Following Leggett (2003) one can use the following iden-
tity, which holds for any numbers A ¼ 1 and B ¼ 1:
SPDC entanglement source, and was well beyond the thresh-

1 þ jA þ Bj ¼ AB ¼ 1
jA
Bj: (67) old of 97.4% which is required for testing the Leggett-type
inequality. The observed SNLHV ¼ 3:8521  0:0227 violates
This, plus the above assumptions, implies a Leggett-type inequality (68) by 3.2 standard deviations (see the data shown
inequality [for details of the derivation see Gröblacher in Fig. 19). At the same time, measurements gave SBell ¼
et al. (2007)] of the following form: 2:178  0:0199, which violates the CHSH inequality by 9
standard deviations. This way, the Gröblacher et al. experi-
SNLHV
jE11 ð’Þ þ E23 ð0Þj þ jE22 ð’Þ þ E23 ð0Þj
  ment excluded a broad class of nonlocal hidden-variable
4
 ’



 theories.
4
 sin 
; (68)
 2 
D. Noncontextual hidden-variable theories
where Ekl ð’Þ is a uniform average of all correlation func-
tions, defined in the plane of ak and bl , with the same relative
Another class of theorems, which show the drastic differ-
angle ’. Inequalities avoiding the averaging were also
ence between the classical and the quantum, are the no-go
derived and tested (Branciard et al., 2007; Paterek et al.,
theorems for noncontextual hidden-variable (NCHV) inter-
2007). For the inequality to be applicable, the vectors a1 and
pretations of quantum mechanics [Specker (1960); Bell
b1 necessarily have to lie in a plane orthogonal to the one
(1966); Kochen and Specker (1967); Mermin, N.D. (1990);
defined by a2 and b2 . This contrasts with the CHSH inequal-
for a survey, see Mermin (1993)]. Such realistic theories
ity. Thus, if, as it is experimentally most easy, a~ 1 , a~ 2 , and b~1 assume that hidden variables fix the values of measurement
correspond to linear polarizations, then b~2 must correspond to results of all possible observables for the given system, and
an elliptical polarization. that such values are independent of the measurement context.
Quantum theory violates inequality (68). Consider quan- That is, they do not depend on which other observables are
tumppredictions
ffiffiffi for the polarization singlet state, j c
iAB ¼ measured together with them.49 It is interesting that already in
ð1= 2Þ½jHiA jViB
jViA jHiB . The quantum correlation the lowest dimension of 3 for which a degenerate observable
function for the local measurements defined by ak and bl
depends only on the relative angles between the vectors: 49
Ekl ¼
ak  bl ¼
cos’. Thus, the left-hand side of in- The measurement context is defined by a maximal observable.
An observable is maximal if it has a fully nondegenerate spectrum,
equality (68), for quantum predictions, reads j2ðcos’ þ 1Þj.
that is for a d dimensional system it is of the form M^ ¼
The maximal violation of inequality (68) is for ’max ¼ 18:8 : Pd
i¼1 i jbi ihbi j, where jbi i are the eigenstates and i the eigenval-
the bound given by inequality (68) equals 3.792 whereas the ues of the observable, which are such that i
j for all pairs of
quantum value is 3.893. the indices. A degenerate observable D^ does not satisfy this last
The Leggett-type inequality valid for nonlocal realistic requirement. That is, one has at least one pair of indices, say i ¼ 1,
theories of the discussed type was experimentally tested by P
2, such that 1 ¼ 2 . Thus, for our example D^ ¼ 1 i¼1;2 jbi i
Gröblacher et al. (2007). In the experiment a SPDC source of Pd
hbi j þ i¼3 i jbi ihbi j. The observable commutes with an infinite set
high-fidelity two-photon entangled states was used, with of different maximal observables. This is due to the fact that
P P
visibilities 99:0  1:2% in the H=V basis, 99:2  1:6% i¼1;2 jbi ihbi jÞ ¼
0 0
i¼1;2 jbi ihbi jÞ, where the primed eigenvectors
in the 45 basis, and 98:9  1:7% in the R=L basis. At are any pair of orthogonal normalized states in the two-dimensional
that time it was the highest reported visibility for a pulsed subspace spanned by the pair jb1 i and jb2 i.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

808 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

can exist (only such observables can be measurable in differ- into a contradiction with quantum formalism. The trick used
ent contexts) noncontextual hidden-variable models of quan- by Anselem et al. is to treat as subsystem 1 the polarization
tum mechanics are impossible (Kochen and Specker, 1967).50 degree of freedom of a photon, and as subsystem 2 the path
Bell’s theorem is a case of a no-go theorem for NCHV in degree of freedom, as it was the case in Żukowski (1991),
which noncontextuality is given ‘‘for free’’ by the locality Michler, Weinfurter, and Żukowski (2000), and Simon et al.
assumption. As locality forbids the result on Alice’s side to (2000). This allowed the construction of six elaborate inter-
depend on the actual observable chosen to be measured by ferometers equivalent to measurements of all the terms in the
Bob, etc., the required noncontextuality is enforced by the inequality (69). The observed value of the left-hand side of
relativistic causality. This is very appealing, because relativ- inequality (69) was for all 20 tested states close to 5.45, with
ity is generally assumed to be a principle setting theory for the highest measurement error at 0.0032. After averaging over
causal links. Noncontextuality, without the help from relativ- all states the standard deviation was just 0.0006, thus the
istic principles, seems to be a much stronger assumption, as it violation of the inequality was as high as by 655 standard
is difficult to argue why nature has to obey it.51 Nevertheless, deviations. The discrepancy between the ideal quantum value
both NCHV theories and local realistic ones can be reduced 6 was due to imperfections in the complicated interferometers
to the assumption of the existence of a joint probability (note that each one had eight exit ports), and the effective
distribution for noncommuting observables. Note that such observables were slightly deviating from the ideal ones con-
distributions are impossible in the quantum formalism. sidered in the theoretical reasoning of Cabello. Most recently,
We now present an example of a Kochen-Specker type a proposal by Klyachko et al. (2008) for a single qutrit
problem. Recently, Cabello (2009) showed that if nine ob- contextuality experiment involving only six different mea-
servables A, B, C, a, b, c, , , and have predefined surements was experimentally realized for photons prepared
noncontextual outcomes 1, they must satisfy the following in superposition of three modes (Lapkiewicz et al., 2011).
inequality:

S ¼ hABCi þ habci þ h i þ hAai VI. QUANTUM COMMUNICATION

þ hBb i
hCc i  4; (69)
Quantum communication ultimately aims at absolute se-
where hABCi denotes the ensemble average of the product of curity and faithful transfer of information, classical or quan-
the three outcomes of measuring the mutually compatible tum. Photons are the fastest information carrier, and due to
observables A, B, and C. For any four-dimensional quantum their very weak coupling to the environment are an obvious
system, one can find a set of observables for which the choice for quantum communication, especially for long dis-
prediction of quantum mechanics is S ¼ 6, irrespectively of tances. Hence, the ability of manipulating the quantum fea-
the actual state. This inequality was violated in an experiment tures (such as coherence and entanglement) of photons is a
by Amselem et al. (2009), where the chosen observables had precious resource.
the form In this section, we review several breakthroughs in the field
of quantum communication52: By exploiting entanglement
A ¼ ð1Þ
z ; B ¼ ð2Þ
z ; C ¼ ð1Þ ð2Þ
z  z ; one can efficiently encode classical messages (Bennett and
Wiesner, 1992; Sec. VI.A), transfer quantum information to a
a ¼ ð2Þ
x ; b ¼ ð1Þ
x ; c ¼ ð1Þ ð2Þ
x  x ;
remote location (Bennett et al., 1993; Sec. VI.B), entangle
 ¼ ð1Þ ð2Þ
z  x ; ¼ ð1Þ ð2Þ
x  z ; ¼ ð1Þ ð2Þ
y  y : two remote particles that have no common past (Żukowski,
et al., 1993; Sec. VI.C), and purify a large ensemble in a less
(70)
entangled states into a smaller ensemble with higher entan-
The above operators i (i ¼ x, y, z) are the usual Pauli glement (Bennett et al., 1996; Deutsch et al., 1996; Gisin,
operators, for two subsystems, 1 and 2. This set has the 1996; Horodecki, Horodecki, and Horodecki, 1996; Pan,
following properties (Peres, 2002): all operators have spec- Simon, Brukner, and Zeilinger, 2001; Sec. VI.D).
trum 1, all operators in each row commute, and so do all Needless to say, one of the ultimate dreams is long-
operators in each column. However, any two operators be- distance or even global (103 –104 km) quantum communi-
longing to different columns and rows do not commute. Thus, cation. As a combination of the ideas of entanglement
each operator belongs to two different contexts explicitly purification and swapping, the quantum repeater protocol
displayed in this array. Furthermore, each operator is the (Briegel et al., 1998, see Sec. VI.F) is an efficient method
product of the other ones in the column or in the row to for beating decoherence and photon losses in attempts to
which it belongs, with the sole exception that in the case of create long-distance high-quality entanglement.
the last column each operator is also such a product but times In Sec. VI.E, we discuss steps on the road toward satellite-

1. Thus, there is no way that these nine operators behave based quantum communication and its first step, i.e., free-
like real numbers upon multiplication. In other words, if one space distribution of entangled-photon pairs over a distance
ascribes to each of the operators whatever realistic values, of 10 km achieved in 2005 (Peng et al., 2005; Resch et al.,
either þ1 or
1, independent of the row or column, one runs 2005) and, more recently, over 144 km (Ursin et al., 2007;
Fedrizzi et al., 2009).
50
For dimension 2, a degenerate observable is just a constant.
51 52
However, please note that expectation values of all quantum We exclude quantum cryptography, which has been extensively
observables are noncontextual. reviewed by Gisin et al. (2002).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 809

A. Quantum dense coding

One can encode two bits of classical information with two

qubits in such a way that each qubit carries a single bit. To
this end, in the case of two qubits represented by polarization
states, one could use states HH, HV, VH, and VV. The idea
of quantum dense coding introduced by Bennett and Wiesner
(1992) is that, by manipulating only one of the two particles
in a Bell state, one can also encode two bits of information.
The procedure runs as follows:
Step 1: Sharing maximal entanglement. A maximally en-
tangled qubit pair (say, in the state j c þ iAB ) is shared by Alice
and Bob. They agree in advance that j c
iAB , j

iAB ,
j
þ iAB , and j c þ iAB represent the binary numbers 00, 01,
10, and 11, respectively.
Step 2: Coding of the message. According to the value Bob FIG. 20. Experimental setup for quantum dense coding. From
wants to transmit to Alice, he performs one out of four Mattle et al., 1996.
possible unitary transformations (identity operation I, ^ x ,
y , and z ) on his qubit B alone. The three nonidentity
operations transform, in an one-to-one way, the original state signal-to-noise ratio reduced the actual channel capacity to
j c þ iAB , respectively, into j
þ iAB , j

iAB , and j c
iAB . on average 1.13 bit for the cases of successful transfers.
Once this is done, Bob sends his qubit to Alice. Note that Nevertheless, the classical and Holevo values of 1 were
this possibility of transforming any of the four basis states to breached.
any other by only manipulating one of the two qubits holds Schuck et al. (2006) realized a complete linear-optical
only for the maximally entangled states. For product (and Bell-state analyzer which is able to distinguish all four Bell
states, provided the initial pair is entangled in two degrees of
classical) states it is always necessary to have control over
freedom (hyperentangled), in this case polarization entangle-
both qubits (bits) to encode two bits in four distinguishable
ment plus the intrinsic time-energy correlation of SPDC
states.
photons (Kwiat and Weinfurter, 1998). With this ability, the
Step 3: Decoding of the message. Upon reception, Alice
dense coding protocol was implemented for all four Bell
performs a Bell-state measurement, distinguishing between
states in the polarization degree of freedom, achieving an
the four code states and thus allowing her to read out both bits
overall channel capacity of 1.18(3) bits per photon. Later
of information. The quantum dense coding doubles the infor-
developments were using the observation that with hyper-
mation capacity of the transmission channel: what is actually
entanglement in at least two degrees of freedom four Bell
sent is only one qubit. This more efficient way of coding
states in one of the degrees of freedom can be distinguished
information at first glance seems to be at odds with Holevo
by local measurements (Walborn, Pádua, and Monken, 2003).
(1973), which states that maximally one bit can be encoded
Barreiro, Wei, and Kwiat (2008) exploited pairs of photons
on a single qubit. Entanglement, a property of pairs of qubits,
simultaneously entangled in spin and orbital angular momen-
allows one to circumvent this theorem and to encode infor-
tum and achieved a channel capacity of 1.630(6) bits, finally
mation entirely in the relative properties of the pair, i.e., in
beating the channel capacity limit of 1.58 bits of the conven-
their correlations.
tional linear-optics implementations.
The first experimental realization of quantum dense coding
was reported by Mattle et al. (1996). The experimental setup
is shown in Fig. 20. The preparation of the polarization- B. Quantum teleportation
entangled photon pairs, the single-qubit operations at Bob’s
station, and Alice’s Bell-state analyzer can all be done with The fascinating possibility of quantum teleportation was
the SPDC and linear-optical techniques presented in the discovered by Bennett et al. (1993). Quantum teleportation is
previous sections. In the Innsbruck experiment, each of the indeed not only a critical ingredient for many more quantum
two j c  iAB states could be distinguished, and they could be communication protocols and for quantum computation
distinguished from j
 iAB . However, with interferometric (Gottesman and Chuang, 1999; Knill, Laflamme, and
Bell-state analysis, there was no possibility to tell which of Milburn, 2001), its experimental realization allows new stud-
the states
 caused the given detection event. Thus, three ies of the fundamentals of quantum theory.
different messages could be operationally encoded by ma-
nipulating a single qubit only.53 Thus, an increase of channel 1. Theory: Qubit teleportation involving an EPR channel and
capacity to log2 3 ’ 1:58 bits was possible, the highest value two bit transfer
achievable using linear optics and classical communications;
The idea of quantum teleportation is illustrated in
see Lütkenhaus, Calsamiglia, and Suominen (1999) and
Fig. 21(a). Suppose particle 1 carries a qubit in the state
Vaidman and Yoran (1999). However, the observed
ji which Alice should teleport to Bob, that is to transfer
it to his particle. Consider pure states and we represent our
53
The states j c  iAB carried two different values, while j
 iAB the discussion in terms of qubits carried by polarization states
third. of photons. We assume that ji1 ¼ jHi1 þ jVi1 is the

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

810 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Still, according to Bennett et al. there is a way out. Suppose

that an ancillary pair of photons 2 and 3 is shared by Alice
and Bob, and that it is in the polarization-entangled state
j c
i23 . The entire system, comprising of the unknown state
of particle 1 and the entangled pair, is in a state ji1 j c
i23 .
By expanding the state of particles 1 and 2 (they are in the
hand of Alice) in the Bell basis, one gets

ji1 j c
i23 ¼ 12½j c
i12 ð
jHi3
jVi3 Þ þ j c þ i12

ð
jHi3 þ jVi3 Þ þ j

i12 ðjVi3
þ jHi3 Þ þ j
þ i12 ðjVi3
jHi3 Þ:
(71)
Now, if Alice performs a polarization Bell-state measurement
on her two particles then the four possible outcomes are
equally likely, regardless of the unknown state ji1 .
However, once particles 1 and 2 are projected into one of
the four entangled states, the polarization state of particle 3 is
instantaneously projected into one of the four pure states
appearing in Eq. (71). They can be rewritten in the following
form:

ji3 ;
^ z ji3 ; ^ x ji3 ; ^ y ji3 ; (72)
FIG. 21 (color). Experimental quantum teleportation. (a) Principle
of quantum teleportation. Alice has particle 1, whose qubit state she where the hatted symbols represent Pauli operators, which act
wants to teleport to Bob. Alice and Bob share an ancillary entangled as unitary transformations. Each of these possible resultant
pair of qubits carried by particles 2 and 3. Alice performs a Bell- states of Bob’s particle 3 is related in a one-to-one way with
state measurement jointly on particles 1 and 2. She sends a two-bit the original state ji1 , which Alice wanted to teleport. In the
classical information informing Bob about the result of her mea- case of the first (singlet) outcome, the state of polarization of
surement. Based on this, he performs one of four unitary trans-
particle 3 is the same as the initial state of particle 1 (except
formations (U) on his ancillary particle. The transformations are
for an irrelevant phase factor), so Bob does not need to do
such that particle 3 ends up in the state identical to the original state
anything to finish the transfer of the original state of polar-
of particle 1. (b) Setup of the Innsbruck teleportation experiment. A
pulse of ultraviolet light passing through a nonlinear crystal creates ization of particle 1. In the other three cases, Bob can apply
an ancillary pair of entangled photons 2 and 3 in a polarization state one of the unitary transformations of Eq. (72) to convert the
j c
i12 . The pulse is reflected, and during its second passage state of particle 3 into the original one of particle 1. However,
through the crystal another pair of photons can be created. One of all this is possible only if he additionally receives via a
them plays the role of particle 1. The other one serves as a trigger classical communication channel a two-bit information on
indicating that photon 1 is under way. Alice looks for coincidences the Bell-state measurement result obtained by Alice.54 After
after the BS at which photons 1 and 2 are superposed. Upon Bob’s unitary operation, the final state of polarization of
receiving a classical information indicating that Alice obtained a particle 3 becomes the new carrier of Alice’s unknown state,
coincidence count in detectors f1 and f2, which implies a collapse ji1 . The original state of Alice’s particle 1 is irrecoverably
into j c
i12 , Bob knows that his photon 3 is in the original erased by the Bell-state measurement, as the Bell-state mea-
polarization state of photon 1. He can check whether this is indeed surement does not reveal any information on the properties of
so by a polarization analysis with a PBS and the detectors d1 and d2 any of the particles prior to the measurement. This is why
behind it. From Bouwmeester et al., 1997. quantum teleportation circumvents the no-cloning theorem.
The transfer of quantum information from particle 1 to
original polarization state of particle 1. Alice does not particle 3 can happen over arbitrary distances. It is not
know the state of this qubit. Thus, the trivial idea that necessary for Alice to know where Bob is (although they
Alice performs certain measurements on ji1 by which she do need to share some reference frame information in order
would obtain all the information necessary for Bob to for the protocol to work). Furthermore, as quantum telepor-
reconstruct the state is ruled out: an experiment on a qubit tation is a linear operation applied to ji1 , it works for mixed
can give only one bit of information. This only suffices to states, or entangled states, equally well; the initial state ji1
determine which state can be ruled out, but is insufficient can be completely unknown not only to Alice but to anyone.
to reconstruct the actual state. To this end, we need A fascinating case is that ji1 could even be quantum me-
infinitely many measurements on identical copies of the chanically completely undefined at the time the Bell-state
state. The projection postulate makes it impossible to fully measurement takes place. This is the case when, as already
determine the state of a single quantum system, or, from remarked by Bennett et al. (1993), particle 1 itself is a
another point of view, the no-cloning principle (Wootters
54
and Zurek, 1982) excludes the possibility to create addi- Note that Alice and Bob before the teleportation must agree on
tional high-fidelity replicas of the original state. choice of the basis of her Bell-state measurement.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 811

member of an entangled pair, ultimately leading to entangle-

ment swapping (Żukowski, et al., 1993; Bose, Vedral, and
Knight, 1998). Quantum teleportation does not violate cau-
sality: a transfer of two bits of classical information is
absolutely necessary to conclude the process.
Generally speaking, the basic criteria to achieve a bona fide
qubit55 teleportation are (1) the experimental scheme without
any changes is capable of teleporting any pure or mixed qubit
state, this includes the possibility of entanglement swapping;
(2) a fidelity better than the classical one of 2=3, see Massar
(1995) can be achieved; (3) at least in principle, the scheme
should be extendable to long distances; and (4) the state to be
teleported should be of external nature, that is, it is carried by
a particle which plays no role in the preparation of the
quantum part of the teleportation channel (essentially the
EPR pair). This in principle allows one to teleport any
unknown qubit state delivered by some outside party.

2. Experimental quantum teleportation FIG. 22. The measured threefold coincidence rates at d2f1f2
( þ 45 ) and d1f1f2 (
45 ) for two cases of the state to be
Figure 21(b) is a schematic of the Innsbruck experimental
teleported being polarized under þ45 (a) and (b) or
45 (c)
setup of Bouwmeester et al. (1997).56 A pulse of ultraviolet
and (d), respectively. The coincidence rates are plotted as a function
laser passing back and forth through a BBO crystal (type II) of the delay (in m) between the arrival of photon 1 and 2 at Alice’s
creates two polarization-entangled EPR pairs. The pair used BS [see Fig. 21(b)]. The threefold coincidence rates are plotted after
as the ancillary one, labeled here as photons 2 and 3, is subtracting the spurious threefold background contribution. The
distributed to Alice and Bob. Photon 1 of the other pair passes data, together with similar results for other polarizations, constitute
a polarizer which prepares it in the initial state to be tele- a positive result of a test for teleportation of an arbitrary state. From
ported, and photon 4 is a trigger indicating that photon 1 is Bouwmeester et al., 1997.
under way. After photon 1 is given to Alice, she combines it
with her photon 2 and performs the Bell-state analysis.
To demonstrate that teleportation is allowed by Nature, it is polarization (Poincaré) sphere. The experimental results for
sufficient to identify one of the four Bell states. Bell-state teleportation of photon 1 polarized under þ45 (
45 ) is
measurement on photons 1 and 2 is done with the use of a BS. shown in the left (right) column of Fig. 22. Bouwmeester
As explained in Sec. III.E.1, if there is a coincidence detec- et al. demonstrated that quantum teleportation works for
tion between the two outputs of the beam splitter, then the orthogonal states jHi and jVi as well as for jHi þ jVi, jHi

photons are projected to the antisymmetric state j c
i12 . The jVi, and jHi þ ijVi. Thus, teleportation was tested for an
Bell-state analysis relies on the interference of two indepen- exhaustive set of mutually unbiased (in other words, fully
dently created photons. Therefore, one has to guarantee that complementary) bases of polarization (qubit) states. The
behind the BS the information which photon came from average fidelities measured for these states were 0.81(1),
which source is completely erased. This was done using the well above the 2=3 threshold.
methods described in Sec. IV.D.1 (Żukowski, Zeilinger, and
Weinfurter, 1995). In the experiment, the UV-pump pulse had 3. Teleportation onto freely flying photons
a duration of 200 fs. By using narrow bandwidth filters
Most applications of quantum teleportation include the
(
 ¼ 4 nm) in front of the detectors registering photons 1 subsequent manipulation of the teleported photon. Thus, a
and 2, a coherence time of about 500 fs could be obtained, freely propagating output state, which is teleported with
which was sufficiently longer than the pump pulse duration, high fidelity, is strongly desired. In the Innsbruck experi-
so that one could not infer anymore during which passage ment, however, owing to the probabilistic nature of SPDC,
through the crystal which of the two photons was created. there was also the chance to register a pf1f2 coincidence
This generated high visibility of the multiphoton interference. [Fig. 21(b)], seemingly indicating the preparation of a
Furthermore, single-mode fiber couplers acting as spatial single photon and the identification of a j c
i12 , which could
filters were used to guarantee good mode overlap of the occur unfortunately due to an unwanted event of two-pair
detected photons. emission during the second passage of the UV pulse, with no
To experimentally demonstrate that an arbitrary unknown emission in the first passage. In such a case, no photon
quantum state can be teleported, it is sufficient to show that propagates to Bob. Thus, in the experiment, a successful
the scheme works for all mutually orthogonal axes of the teleportation act had to be confirmed by a detection event at
Bob’s side. Because of this fact, Braunstein and Kimble
55
For a higher dimensional system this set of conditions has an (1998) classified the experiment as involving a ‘‘postselec-
extension. tion,’’ and implied that the fidelity of the process therefore
56
An operational blueprint for such an experiment is first put was not sufficient. Bouwmeester et al. (1998) pointed out
forward in Żukowski, et al. (1993). that the situation should be interpreted as reducing the

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

812 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

efficiency of teleportation rather than its quality. Possible teleported and one of the qubits of the EPR maximally
solutions (Braunstein and Kimble, 1998; Kok and entangled pair. This effectively boils down to an emulation
Braunstein, 2000) could include the discrimination of one- of the third particle (subsystem) in the process [for such an
photon and two-photon events at detector p [Fig. 21(b)], a emulation in the case of GHZ correlations see Żukowski
quantum nondemolition measurement of the photon number (1991) and the experiment Michler, Weinfurter, and
in mode 3. It should be noted that if one uses a single-photon Żukowski (2000)]. However, the fact that one of the particles
source or entangled pairs from a single quantum emitter, is emulated does not allow one to teleport a qubit state of an
e.g., from a quantum dot [cf. Akopian et al. (2006) and independently arriving particle, and it is difficult to imagine
Stevenson et al. (2006)], one in principle does not need to an entanglement swapping process which leaves as a result
worry about the double pair emission of SPDC at least up to two spatially separated qubits, previously independently
the antibunching quality of the source. However, due to the emitted, in a maximally entangled state. Thus, the compara-
lack of appropriate technology, such a scheme has not been tive straightforwardness of a Bell-state measurement has a
thus far realized. trade off: the process is not fully versatile, and in some
In the experiment of Pan, Gasparoni, Aspelmeyer, et al. respects does not mirror the original idea.
(2003) such problems were basically removed. The scheme In the protocol with the emulation, the quantum state to be
was such that a coincident registration of photons at the Bell teleported can prepared by performing a unitary operation on
measurement station was heralding that with a high proba- an additional degree of freedom of one of the EPR particles of
bility one has a propagating photon carrying the teleported the quantum channel. A protocol of this kind has been pro-
state; see Fig. 21(b). Such a process was called ‘‘teleportation posed by Popescu (1995) and was experimentally realized by
into freely propagating photons.’’ The basic idea of this Boschi et al. (1998) with a teleportation fidelity of 0.85(1). As
experimental method is that the entangled ancillary pair the protocol does not involve interference of photons from
was provided much more frequently than the photon to be two separate emissions, and as only one EPR pair is manipu-
teleported [a similar idea was also used in the teleportation lated, it avoids many difficulties. Just one SPDC source is
experiment using time-bin entanglement carried out by needed, and it works with just a one pair-emission process.
Marcikic et al. (2003)]. Thus, when a qubit which was to The main idea (see the experimental setup in Fig. 23) is to
be teleported arrived, the teleportation machinery was almost use the spatial and polarization degrees of freedom of pho-
always ready. Technically, the main idea was to reduce the tons. One emulates the particle which carries the to be tele-
number of unwanted f1-f2 coincidence counts. This was ported qubit with the use of an additional degree of freedom
accomplished by attenuating the beam 1 by a factor of , of an EPR particle sent to Alice. The first step is to produce
while leaving the intensity in modes 2–3 unchanged. With two photons entangled in their directions of propagation (this
such an arrangement, a threefold coincidence f1-f2-p occurs will serve as the EPR pair), i.e., entangled in momentum, but
with probability p2 for a successful teleportation (p is the each with a well-defined polarization. Thus, one starts with
probability of having a single pair creation during a SPDC
process). With a significantly lower probability ð pÞ2 one has 1
pffiffiffi ðja1 i1 ja2 i2 þ jb1 i1 jb2 i2 ÞjHi1 jVi2 : (73)
spurious coincidences without a photon at Bob’s side. Thus, 2
for a sufficiently low it is not necessary to actually detect
photon 3 to be certain that teleportation occurred. Photon 3 The area indicated as ‘‘EPR source’’ in Fig. 23 shows how
gives us a freely propagating beam of teleported qubits. this can be achieved.57 On the way to Alice, photon 1 is
To demonstrate a nonconditional teleportation, a series of intercepted by the preparer P who changes the polarization
neutral filters were inserted in mode 1, showing that the from H to an arbitrary quantum superposition
probability of a successful teleportation conditioned on an
f1-f2-p threefold coincidence increases with decreasing filter ji1 ¼ jHi1 þ jVi1 : (74)
transmission (e.g., the observed probability of success was
This is the quantum state that Alice will transmit to Bob. The
0:138  0:002 for ¼ 0:05). The average fidelity for the
preparer transforms the polarization in both paths a1 and b1 in
unconditional teleportation for three mutually unbiased bases
the same way. The total state ji of the two photons after his
was 0:80ð2Þ.
or her action is
4. Teleportation of a qubit carried by a photon of the ancillary 1
EPR pair ji ¼ pffiffiffi ðja1 i1 ja2 i2 þ jb1 i1 jb2 i2 Þji1 jVi2 ; (75)
2
It is well known that with standard optical devices (passive
linear optics plus detectors) one can measure any observable which is a formal analogue of the initial state in Eq. (71).
associated with a single photon. Thus, if the photon carries Next, Alice performs a Bell-state-like measurement on the
two qubits, any two-qubit measurement can be performed, two degrees of freedom of her (single) particle. The four
including a Bell-state measurement, involving states of two ‘‘Bell states’’ are represented by the following correlated
different photon ‘‘degrees of freedom,’’ e.g., polarization and polarization-path states of the photon:
path. Thus, as teleportation is from an algebraic point of view
a three-qubit operation, and as there is no easy solution for a 57
One first generates a two-photon polarization-entangled state
Bell-state measurement for two photons, each carrying a j c þ i12 . As a PBS transmits (deflects) H (V) photons, j c þ i12 is then
qubit, one can resort to the following. One can have a scheme transformed into momentum entanglement in Eq. (73), in which the
in which a single photon carries two qubits, the qubit to be photons with label 1 (2) are H (V) polarized.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 813

FIG. 24 (color). The principle of open-destination teleportation.

FIG. 23. Experimental setup for the ‘‘two qubit at one subsystem Adapted from Zhao et al., 2004.
scheme’’ of quantum teleportation. From Boschi et al., 1998.
The Innsbruck teleportation technique with its later im-
1 provements enables one to aim at a long-distance teleporta-
j c  i1 ¼ pffiffiffi ðja1 i1 jVi1  jb1 i1 jHi1 Þ; tion (Marcikic et al., 2003; Ursin et al., 2004) and toward
2 more complicated schemes (Zhao et al., 2004; Zhang, Q.
(76)
1 et al., 2006). There are other interesting developments.
j
  i1 ¼ pffiffiffi ðja1 i1 jVi1  jb1 i1 jHi1 Þ:
2 Marcikic et al. (2003) realized a teleportation of qubits at
telecommunication wavelengths over a fiber length of 2 km.
The measurement of photon 1 with respect to this basis can in
Adopting Boschi et al.’s protocol, Jin et al. (2010) emulated
principle be achieved with 100% success rate. (A photon
free-space quantum teleportation over 16 km.
detection by D1 , D2 , D3 , or D4 corresponds directly to a
projection onto one of the four Bell states; see Fig. 23.)
Since, in terms of the four single-photon Bell states, one 6. More-involved teleportations
has Open-destination quantum teleportation The so-called
1 þ
þ jb2 i2 ÞjHi2 þ j c
i1 ðja2 i2 open-destination teleportation of Karlsson and Bourennane
ji ¼ 2½j c i1 ð ja2 i2
(1998) is a protocol allowing to transfer a state to one of several
þ jb2 i2 ÞjHi2 þ j
 þ i1 ðja2 i2
jb2 i2 ÞjHi2 potential recipients. It can be decided who gets the state even
þ j

i1 ð ja2 i2
jb2 i2 ÞjHi2 ; (77) after the initial to be teleported state ji is wiped out in a Bell-
state measurement. Such a teleportation scheme was experi-
the final step of the protocol is that Alice informs Bob which mentally demonstrated for N ¼ 3 by Zhao et al. (2004).
detector clicked. With this information, Bob can reproduce Figure 24 shows the basic scheme. One uses a quantum
the initial polarization state by transforming the momentum channel in form of a N þ 1 qubit GHZ state, say for N ¼ 3
superposition of photon 2 [see Eq. (77)] into a corresponding and for polarization qubits:
polarization state, and applying suitable polarization trans-
1
formations (following the two bit classical information from ji2345 ¼ pffiffiffi ðjHi2 jHi3 jHi4 jHi5 þ jVi2 jVi3 jVi4 jVi5 Þ;
Alice). They represent the unitary corrections necessary to 2
put his photons into the polarization state that was set by the (78)
preparer at the other EPR photon. and requires, as always, a transfer of two bits of classical
information. The polarization state to be teleported ji1 ¼
5. Teleportation with various physical systems jHi1 þ jVi1 is first encoded into an N-particle coherent
superposition of these GHZ particles. By making a Bell-state
Each teleportation experiment done thus far has advan-
measurement on, say, photons 1 and 2, one projects the
tages and disadvantages [for a comparison between various
remaining photons into one of four states. For example.,
methods, see Bouwmeester, Pan, Weinfurter, and Zeilinger
whenever the result of the Bell-state measurement is a c þ
(2000)]. Quantum teleportation of continuous-variable states
state, one gets
(Furusawa et al., 1998; Braunstein and van Loock, 2005) has
the advantage that full Bell-state analysis is possible with 1
pffiffiffi ð jHi3 jHi4 jHi5 þ jVi3 jVi4 jVi5 Þ: (79)
linear optics (within the experimental bandwidth). Yet, it is 2
hard to extend to a long-distance case. The unavoidable
degradation of squeezed-states sets in during longer-distance The state ji can be read out at any of the three particles, by
transfers. This consequently leads to a rapid lowering of the performing a suitable projection measurement on N
1 of
quality of squeezed-state entanglement. Quantum teleporta- them, here on 2, and a unitary transformation dependent of
tion using nuclear magnetic resonance (Nielsen, Knill, and the received two bits of data, which is carried out on one
Laflamme, 1998) or trapped atoms (Barrett et al., 2004; of the GHZ particles.58 Assume that we want to transfer our
Chiaverini et al., 2004) has an obvious advantage in that
the input quantum state can be teleported with an efficiency 58
Either on the remaining one, upon which no measurement is
of 100%. Yet, it is difficult (if not impossible) to implement it done, or on any one of them before the N
1 measurements are
over long distances. done. Further on, we shall describe the first option, as it is simpler.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

814 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

state to particle 5. To this end upon the receipt of information

concerning the result of the Bell-state measurement (BSM),
the partner 5 must perform on his particle a x transforma-
tion, which interchanges polarizations H and V. The partners pffiffiffi
3 and 4 make measurements in the ji ¼ ðjHi  jViÞ= 2
basis. The recipient of the state is informed about the mea-
surement results.59 Once the recipient gets this additional
information, only if there was just one result associated
with a projection to j
i, he or she performs the sign flipping
z transformation. The state is recovered.
In contrast to the original teleportation scheme, after the
encoding operation the destination of teleportation is left
open until we perform a polarization measurement (‘‘decod-
ing’’) on two of the remaining three photons. This implies
that, even though photons 3, 4, and 5 are far apart, one can
still choose which particle should finally carry the teleported
FIG. 25 (color). Two-qubit quantum teleportation. Adapted from
state. No prior agreement on the final destination of the
Zhang, Q. et al., 2006.
teleportation is necessary.
In the Zhao et al. (2004) experiment, the required four-
photon GHZ entanglement was generated (conditional upon Zhang, et al. (2006) managed to obtain 105 photon pairs
joint detection) using the techniques of Secs. IV.D.3, and the per second from each EPR source. As a result, on average 10
pseudo-single photon state to be teleported was in the form of six-photon events per minute were registered. The two-qubit
polarization of an attenuated laser beam containing on aver- teleportation protocol was implemented for three different
age 0.05 photons per pulse. They detected fivefold coinci- initial states jXiA ¼ jHi1 jVi2 , jXiB ¼ ðjHi1 þ jVi
pffiffiffi1 ÞðjHi2

dence with a rate of 12 per hour and measured fidelities of ijVi2 Þ=2, and jXiC ¼ ðjHi1 jVi2
jVi1 jHi2 Þ= 2 ¼ j c
i.
teleportation from photon 1 to photon 5 and from photon 1 to The measured fidelity for jXiA , jXiB , and jXiC was 0.86(3),
photon 4. For þ=
linear and R=L circular polarization 0.75(2), and 0.65(3), respectively. All the measured fidelities
states, these were 0:80ð4Þ. were well beyond the state estimation limit of 0.40 for a two-
Quantum teleportation of composite two-qubit states– qubit system [for a derivation of the limit see Hayashi,
Zhang, Q. et al. (2006) demonstrated a teleportation of Hashimoto, and Horibe (2005)].
two-qubit states with a six-photon interferometer. Suppose
Alice wants to send an unknown state of a composite system
C. Entanglement swapping
consisting of qubits 1 and 2:
1. Theory
ji12 ¼ jHi1 jHi2 þ jHi1 jVi2 þ jVi1 jHi2 Entanglement swapping (Żukowski, et al., 1993) provides
a method of entangling two particles that never interacted or
þ jVi1 jVi2 (80)
even have no common past. It can also be interpreted as
teleportation of entanglement, i.e., teleportation of undefined
to a distant receiver Bob (Fig. 25). Before teleportation, Alice states of a particle entangled with another subsystem (Bennett
and Bob share two ancillary entangled-photon pairs (photon et al., 1993). We mention that one of the original motivations
pairs 3–5 and 4–6), which arepffiffiffiboth prepared in a Bell state, of entanglement swapping is the so-called ‘‘event-ready de-
tection’’ of the entangled particles, a concept suggested by
say, j
þ i ¼ ðjHHi þ jVViÞ= 2. Following the standard tel-
Bell (Clauser and Shimony, 1978; Bell, 1987). Entanglement
eportation protocol, Alice first teleports the state of photon 1
swapping, together with entanglement purification, is a key
to photon 5 by consuming the entangled pair 3–5. The result
element of the quantum repeater protocol (Briegel et al.,
of this step is ji52 . Similarly, Alice can also teleport the state
1998; Dür et al., 1999; see also Sec. VI.F) and opens a way to
of photon 2 to photon 6 by consuming the entangled pair 4–6.
efficiently distribute entanglement for massive particles
After a successful implementation of the two steps, the
(Bose, Vedral, and Knight, 1998).
original two-qubit state ji12 is teleported to qubits 5 and 6
Consider the arrangement of Fig. 26. We have two EPR
in ji56 .
sources. Assume that each source emits a pair of entangled
The teleportation of two-qubit states was realized by tele-
photons in a state, say, j c
i so that the total state of the four
porting two photonic qubits individually. Thus, neither the
photons is ji1234 ¼ j c
i12 j c
i34 . While pairs 1–2 and 3–4
two original qubits nor the teleported qubits have to be in the
are entangled, there is no entanglement of any of the photons
same place. Such a flexibility is desired in distributed quan-
1 or 2 with any of the photons 3 or 4.
tum information processing such as quantum telecomputation
Next, one performs a Bell-state measurement on photons 2
and quantum state sharing. The method can be easily gener-
and 3. According to the expansion
alized to teleport a state of an N-qubit composite system.
ji1234 ¼ 12ðj c þ i14 j c þ i23
j c
i14 j c
i23
59
The basis of measurement is earlier agreed by the partners of the
protocol.
j
þ i14 j
þ i23 þ j

i14 j

i23 Þ; (81)

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 815

FIG. 26. Principle of entanglement swapping. From Pan et al.,

1998.

this measurement always projects photons 1 and 4 also onto

a Bell state. For example, if the result of the Bell-state
measurement of photons 2 and 3 is j c
i, then the resulting FIG. 27. Experimental setup of entanglement swapping. The gen-
state for photons 1 and 4 is also j c
i. In all cases, photons 1 eration scheme for photon pairs 1–2 and 3–4, and the Bell-state
and 4 emerge entangled despite the fact that they never measurement identifying the j c
i23 state is identical as in the
interacted in the past. In Fig. 26, entangled particles are Innsbruck teleportation experiment. The Bell-state measurement
indicated by the same degree of darkness of the lines. Note on photons 2 and 3 projects the two remaining photons 1 and 4
that particles 1 and 4 become entangled after the Bell-state onto an entangled state. To analyze their entanglement, one looks at
measurement on particles 2 and 3. Without knowing which coincidences between detectors D1þ and D4, and between detectors
result of the BSM measurement occurred, however, the state D1
and D4, for different polarization angles . Note that, since the
of photons 1 and 4 would remain maximally mixed. detection of coincidences between detectors D1þ and D4 and
Given an ideal arrangement with sources that emit only a D1
and D4 are conditioned on the detection of the j c
i23 state
(which happens whenever one has a coincidence behind the beam
single pair of entangled photons each, the process of entan-
splitter), one in fact is looking for fourfold coincidences to signify
glement swapping also gives a means to generate event-ready
the event of entanglement swapping. Narrow bandwidth filters (F)
entanglement. Namely, as soon as Alice completes the Bell- are positioned in front of each detector to make photons from
state measurement on particles 2 and 3, we know that pho- different emissions indistinguishable. From Pan et al., 1998.
tons 1 and 4 are on their way, ready for detection in an
entangled state. In this way, one has the possibility to perform
an event-ready test of Bell’s inequality (Bell, 1987;
A test of Bell’s inequality involving swapping of entangle-
Żukowski, et al., 1993). For a further discussion on event-
ment has some appealing features, aside from being an
ready entanglement, see Sec. VII.B.1.
‘‘event-ready’’ one.60 In addition, it can be performed in a
delayed-choice mode, as suggested by Peres (2000) and
2. First experimental demonstration realized by Jennewein et al. (2001). In such an experiment,
The above scheme was realized using an SPDC source as one delays the instant of time at which the Bell-state mea-
shown in Fig. 27, by Pan et al. (1998). As in the Innsbruck surement is performed on photons 2 and 3. Thus, entangle-
teleportation experiment, only the antisymmetric Bell state ment between photons 1 and 4 in subensembles associated
j c
i23 was detectable in this Bell-state measurement. After with a specific result for 2 and 3 is revealed, a posteriori, after
such a detection event, signaled by a coincidence behind the they have already been measured and may no longer exist.
BS, according to the entanglement-swapping rules, photon 1 Recently, an experiment with active switching and spacelike
and 4 is projected into the same entangled state j c
i14 . This separation of the relevant decision was carried out by Ma
entanglement was tested by analyzing the polarization corre- et al. (2011).
lations between photons 1 and 4 conditioned on coincidences
between the detectors of the Bell-state analyzer. When vary- 3. Other experiments on entanglement swapping
ing the analysis angle  for the photons going to detector D4 ,
Recently, entanglement-swapping experiments with an in-
the coincidences with Dþ

1 and D1 should follow two com-
creased complexity involving pairs of entangled photons have
plementary sine curves in dependence on . The observed
been demonstrated: multistage entanglement swapping
sinusoidal behavior (the interference pattern) of the coinci-
(Goebel et al., 2008) and multiparticle entanglement swap-
dence rates had a visibility of 0.65(2), which clearly surpasses
ping (Lu, Yang, and Pan, 2009).
the 0.5 limit for a classical interference for coincidence
If one aims to build a quantum repeater (Briegel et al.,
measurements. A later experiment (Pan et al., 2001)
1998, see also Sec. VI.F), one has to achieve entanglement
achieved a visibility of 0:84, which is sufficient for violat-
swapping with synchronized entangled-photon sources
ing a Bell inequality (the threshold is 0.71) for photons 1 and
4. Further advancements were achieved by Jennewein et al. among all distributed segments. It thus requires stable inter-
(2001) by implementing a 2-state Bell analyzer. Using a fiber ference between two independently emitted photons.
coupler, an ideal spatial mode overlap was obtained. A Bell Kaltenbaek et al. (2006), Yang et al. (2006), and
inequality, for measurements on photons 1 and 4, was vio-
60
lated by the factor of 1.211(45), i.e., by 4.6 standard A successful BSM measurement defines the ensemble of photon
deviations. pairs, 1 and 4, which are subject to a Bell test.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

816 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Kaltenbaek et al. (2009) successfully realized the necessary ensembles. These experiments still suffer from low success
technique using synchronized fs lasers to solve the above probability and imperfect state fidelity. For instance, the
problem. Kaltenbaek et al. (2006)61 reported an active Moehring et al. (2007) experiment had a success probability
synchronization method: the two independent fs pulsed lasers of 3:6 10
9 and the fidelity of the states of the entangled
pumping the two separate SPDC sources were electronically ions was 0.63(3). The ion-ion entanglement fidelity was
synchronized to emit pulses at the same time. To enable improved to be 0.81 in a later experiment by Matsukevich
interference the two photons registered behind the BS cannot et al. (2008). Together with the high efficiency of the mea-
be distinguished in any way. To this end, the now standard surement of the quantum state of an ion, this high fidelity
methods suggested in Żukowski, Zeilinger, and Weinfurter allowed one to observe a Bell-inequality violation with an
(1995) and discussed in Sec. IV.D were used. efficiency high enough to close the detection loophole.
The entangled photons generated via the usual SPDC, as
used in the above experiments, have broadband linewidth D. Beating noisy environment
(usually on the order of several THz) determined by the
phase-matching condition. Thus, there the challenge was to So far, significant experimental progress has been achieved
achieve sufficiently sharp synchronization of the photons. in small-scale realizations of quantum information process-
Halder et al. (2007) took a different approach to achieve ing. However, interesting challenges arise in bringing quan-
entanglement swapping by a precise time measurement. The tum information processing to technologically useful scales.
photon detector used in the experiment was a niobium nitride This is primarily due to the unavoidable decoherence63
superconducting single-photon detector with a time resolu- caused by a coupling between the quantum system and the
tion of 74 ps.62 The photons were filtered using 10 pm- environment. In quantum communication, it is the noisy
bandwidth filters, which corresponds to a coherence time of quantum channel that degrades the quality of entanglement
350 ps well above the temporal resolution of the detectors. between particles the further they propagate. Yet, the imple-
Hence, ultracoincidence photon timing could be obtained, mentation of any of the quantum communication schemes (as
and pulsed sources could be replaced by continuous-wave reviewed above) over large distances requires that two distant
sources, which do not require any synchronization. parties share entangled pairs with high quality. Similarly,
The passive filtering used by Halder et al. is, however, during quantum computation the coherence of a quantum
extremely inefficient (the 10 pm-filter transmits <1% only of system also decreases exponentially with an increasing op-
all down-converted photons). Thus, a very bright narrow- eration time, consequently leading to failure in the quantum
band entangled-photon source is highly desirable. A recent computation. It is therefore necessary to overcome decoher-
experiment (Bao et al., 2008) realized such a source with a ence in any realistic large-scale realization of quantum infor-
linewidth of 9.6 MHz. Because of the long coherence time, mation processing.
synchronization for such sources is unnecessary, while coin- An important tool to overcome the noise in the quantum
cidence measurements with time resolution of several nano- communication channel is entanglement distillation, concen-
second with current commercial single-photon detectors will tration, and purification, proposed by Bennett, Bernstein,
be sufficient to see interference of photons originating from Popescu, and Schumacher (1996); Bennett et al. (1996);
independent sources. Bennett, DiVincenzo, Smolin, and Wootters (1996); and
Entanglement swapping provides a tool to entangle qubits Deutsch et al. (1996). A linear-optical implementation of
without direct interaction. An interesting application is that entanglement purification was suggested and experimentally
we can entangle distant, independent matter qubits through demonstrated by Pan, Simon, Brukner, and Zeilinger (2001)
photon-mediated entanglement swapping. Imagine we start and Pan, Gasparoni, Ursin et al. (2003). Quantum repeater
with two entangled atom-photon pairs (Blinov et al., 2004; (Briegel et al., 1998; Dür et al., 1999), based on entangle-
Volz et al., 2006). By implementing a Bell-state measure- ment purification and entanglement swapping, would provide
ment of the two photons, we can project the two atomic qubits an efficient way to generate highly entangled states between
into an maximally entangled state. Proof-of-principle experi- two distant locations. Remarkably, the quantum repeater
ments have been performed by Moehring et al. (2007) who protocol tolerates general errors on the percent level, which
entangled two trapped atomic ions separated 1 m apart using is reachable using entanglement purification based on linear
optics (Pan, Simon, Brukner, and Zeilinger, 2001; Pan,
entanglement swapping exploiting interference of photons
Gasparoni, Ursin et al., 2003). It has been shown that, despite
emitted by the ions, and by Yuan et al. (2008) in atomic
of local noise and/or errors on the percent level, the quantum
repeater can be used to establish a private and/or secure
61
The original purpose of the experiment was to demonstrate that quantum communication channel (Aschauer and Briegel,
independently emitted photons do interfere. Thus, the team used 2002). A study (Dür and Briegel, 2003) showed that entan-
two independently pumped parametric down-conversion crystals. glement purification can also be used to increase, by several
The only link between the two pumping lasers was via an electronic orders of magnitude, the quality of logical operations
pulse synchronization. A recording of a pair of idlers heralded that between two qubits. In essence, this implies that the threshold
two signals were on the way to the BS at which the Hong-Ou- for tolerable errors in quantum computation is within reach
Mandel coincidence dip was observed. The visibility was well using entanglement purification and linear optics.
surpassing the classical limit. Thus, nonclassical interference of
entirely independent photons was observed.
62 63
Conventional room-temperature silicon detectors have a time For general aspects on decoherence, we refer the reader to a
jitter of 500 ps. review by Zurek (2003).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 817

1. Entanglement distillation and concentration

Entanglement concentration aims to obtain with a nonzero
probability a higher entanglement from pure states with a
lower entanglement. There are two methods to achieve this.
The first is the so-called Procrustean method (Bennett,
Bernstein, Popescu, and Schumacher, 1996). It requires that
the photon pairs are all in a pure nonmaximally entangled
state, say, jinonmax ¼ jHijVi þ jVijHi, where  and
are two known amplitudes. In this case, the scheme only
involves local filtering operations (Gisin, 1996; Horodecki,
Horodecki, and Horodecki, 1996) on single pairs. Second, the
Schmidt decomposition scheme (Bennett, Bernstein,
Popescu, and Schumacher, 1996) works for photon pairs
that are all in a pure but unknown nonmaximally entangled
state jinonmax . In practice, this scheme is more difficult to
implement as it requires simultaneous collective measure- FIG. 28 (color). Scheme of entanglement purification by Bennett,
ments on many photons. Bernstein, Popescu, and Schumacher (1996); Pan, Simon, Brukner,
Kwiat et al. (2001) used the Procrustean method to and Zeilinger (2001). (a). Two poorly entangled (source and target)
demonstrate experimentally distillation of maximally en- pairs are initially shared by Alice and Bob. They both perform a
tangled states from nonmaximally entangled inputs. Using (local) CNOT operation on the two particles at their hands, measure
partial polarizers, they performed a filtering process to max- the particles belonging to the target pair in the 0=1 basis, and
imize the entanglement of pure polarization-entangled photon compare the measured results via classical communication. If the
pairs generated by SPDC. The method was also applied to results are the same, then the remaining pair will have a higher
initial states that were partially mixed [see also Wang et al. degree of entanglement than the original two pairs. In this case, they
keep the source pair. In the case of obtaining opposite results, they
(2006) for a later experiment]. After filtering, the distilled
discard it. By repeating the same procedure, always starting from
states show violations of a Bell’s inequality, while the initial
the pairs produced in the former purification step, it is possible to
states do not have this property. For some special two-qubit
distill pairs of arbitrarily high entanglement quality [for more
mixed states, Verstraete, Dehaene, and DeMoor (2001) con- details, see Bennett, Bernstein, Popescu, and Schumacher (1996)].
structed the optimal local filtering operations for distilling (b). An alternative and more feasible scheme which does not require
entanglement from the mixed state, with an experimental a CNOT operation but only polarizing beam splitters. The PBS
demonstration done by Wang et al. (2006). transmits horizontal (jHi), and reflects vertical (jVi), polarization.
The Schmidt decomposition scheme becomes practically By selecting only those events for which there is one, and only one,
feasible after the proposal of a linear-optical implementation photon in each output mode of the PBS, one can project two photons
of entanglement concentration (Yamamoto, Koashi, and input from different spatial modes into the subspace spanned by
Imoto, 2001; Zhao, Pan, and Zhan, 2001). Two independent jHijHi and jVijVi (for more details see text). Adapted from Pan,
experiments (Yamamoto et al., 2003; Zhao, Yang, Chen, Simon, Brukner, and Zeilinger, 2001.
Zhang, Pan, 2003) were reported for linear-optical entangle-
ment concentration.

2. Entanglement purification The linear-optical purification scheme [shown in Fig. 28(b)]

will be presented here using a specific example. Our initial
The underlying idea of entanglement purification is that, state is
by using local operations and classical communication only,
to extract from multiple copies of imperfect states (arbitrary ab ¼ Fj
þ iab h
þ j þ ð1
FÞj c
iab h c
j; (82)
mixed states) fewer copies of entangled state asymptotically
to near-unity fidelity. Schemes of entanglement purification where j c
iab is an unwanted admixture. The subscripts a and
ware introduced by Bennett, Bernstein, Popescu, and b indicate the particles at Alice’s and Bob’s locations,
Schumacher (1996) and Bennett, DiVincenzo, Smolin, and respectively.
Wootters, 1996; Deutsch et al. (1996), as illustrated in Alice and Bob share a large number of pairs described by
Fig. 28(a). However, a drawback of these theoretical schemes ab . They start by picking at random two such pairs. Each of
is that they require CNOT operations. In the context of long- them superimposes their photons on a PBS. An essential step
distance quantum communication, the probability of errors in the purification scheme is to select those cases for which
caused by the CNOT operation must be within a few percent, there is exactly one photon in each of the four spatial output
which, unfortunately, is somewhat beyond the current experi- modes. We shall refer to them as ‘‘four-mode cases.’’ This
mental techniques. A more feasible purification scheme was corresponds to a projection onto the subspace in which two
proposed by Pan, Simon, Brukner, and Zeilinger (2001). They photons at the same experimental location (Alice’s or Bob’s)
showed that purification does not have to entirely rely have equal polarization. This is similar to the bilateral CNOT
on CNOT operations. In some cases, a simple linear-optical operation of the original scheme. Note that the polarizations
element, a polarizing beam splitter, suffices [see at two different locations do not have to be the same. After
Fig. 28(b)]. performing the purification procedure (selection of four-mode

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

818 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

cases, measurements in modes a4 and b4 in the þ=
basis,

and local operations conditional on the measurement results),
Alice and Bob will finally create a new ensemble described
by the density operator

0ab ¼ F0 j
þ iab h
þ j þ ð1
F0 Þj c
iab h c
j; (83)

with a larger fidelity F0 ¼ F2 =½F2 þ ð1
FÞ2  (for F > 1=2)

of pairs in the desired state than before the purification.
Though it seems that only a rather special example, single
bit-flip error, has been considered, the same method actually
applies to the arbitrary mixed states ab , provided that they
contain a sufficiently large fraction F > 1=2 of photon pairs
in a maximally entangled state. This works as follows: one
can first purify away single bit-flip errors; phase errors can
then be easily transformed into bit-flip errors by a 45
polarization rotation and treated in a subsequent purification
step.
An experimental demonstration of the entanglement puri-
fication scheme has been reported by Pan, Gasparoni, Ursin
et al. (2003). The setup is shown in Fig. 29. For each run two
pairs of an initial, mixed state (82) were prepared with SPDC
and half-wave plates. Next, the two photons at Alice’s (Bob’s)
side in the mode a1–a2 (b1–b2) were interfered at a PBS. FIG. 29. Experimental setup for entanglement purification. UV
laser pulses pass through a BBO crystal twice to produce two
After the four photons’ passage through the two PBS, and
polarization-entangled photon pairs, i.e., pair 1 in a1–b1 and
under the condition that one detects one and only one photon
pair 2 in a2–b2. Four compensators (Comp.) are used to offset the
polarized along the  basis in each of the modes a4 and b4, birefringent effect caused by the BBO crystal during parametric
the two photons in the mode a3–b3, according to quantum down-conversion. The photons in modes b1 and b2 pass through a
mechanical calculations, have a higher fidelity to be in the half-wave plate (=2) to simulate a noise that reduces the entangle-
pure entangled state. There is, however, a complication in the ment quality. Next, the two pairs are sent via local PBSs. This
actual experiment. Owing to the probabilistic nature of results in entanglement purification. Adjusting the positions of the
SPDC, with a probability of the same order of magnitude, delay mirrors
1 and
2 tunes the optical paths in such a way that
two photon pairs can be emitted into a one mode pair. the photons at local measurement stations arrive at their PBS
Fortunately, as pointed out by Simon and Pan (2002), this simultaneously. Detections of exactly one photon in each of the
does not ruin the purification protocol. Simply, for the four outputs (a3, a4, b3, and b4) behind a 45 polarizer (Pol.) lead to
higher-order emissions causing a four-mode detection case a successful purification act. From Pan, Gasparoni, Ursin et al.,
the photons in a3-b3 are projected, due to interference, to the 2003.
entangled state. The scheme requires a fine stabilization of
the phases between the amplitudes of the four-mode
contributions. purification, SBell ¼ 2:29  0:13 was measured. This violates
In the first purification experiment, a mixed state as in the inequality by 2.2 standard deviations.
Eq. (82) with fidelity of F ¼ 0:75 was prepared. The If one estimates the accuracy of local operations at the
preparation was positively tested by measuring fractions used PBSs, one obtains theoretical values of fidelity better
both in the H=V and in the þ=
bases, as shown in than 98% for the process itself, or equivalently an error
Figs. 30(a) and 30(b), respectively. probability of at most 2%. Entanglement purification with
The measured results in Figs. 30(c) and 30(d) show a such a high quality is important not only for quantum
significant improvement of entanglement fidelity to the value communication, but also for quantum computation. With
of F0 ¼ 0:92  0:01 for the purified subensemble. In a second linear optics the accuracy of single-photon operations on
experiment, entanglement purification was performed for a polarization and spatial degrees of freedom can be ex-
mixed state with F ¼ 0:80. After purification, the observed tremely high (a typical accuracy of commercial products
entanglement fidelity for the subensemble in modes a3 and b3 is 10
3 ). These facts, together with the experimental
was about 0:94  0:01. For each initial mixed state the realization of high-fidelity teleportation (Pan, Gasparoni,
purification scheme worked. The two-photon interference Aspelmeyer, et al., 2003), imply that the threshold of
visibility of the original mixed states was 50% and 60%, tolerable error rates for quantum repeaters could be
respectively. In a single purification step it was increased to achieved.
84% and 88%. This is well above the threshold to violate a Although the efficiency of this entanglement purification
Bell inequality. In a subsequent development, a Bell experi- scheme (Pan, Simon, Brukner, and Zeilinger, 2001) is in
ment with purified states was performed (Walther, Resch, theory 1=4, the actual success probability in the experiment
Brukner et al., 2005). The states of initially poorly entangled (Pan, Gasparoni, Ursin, et al., 2003) was much lower as it
photons, below thresholds to violate a Bell-CHSH inequality, relied on the SPDC to probabilistically create two pairs of
were prepared by a controllable decoherence. After a entangled photons, thus only a small fraction of photons

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 819

quantum communication.64 This underlines the necessity

of developing quantum repeaters.
A promising way to realize long-distance quantum com-
munication is to exploit satellite-based free-space distribution
of single photons, or entangled-photon pairs (Aspelmeyer,
Jennewein, et al., 2003). In the scheme, the photonic quan-
tum states are sent from Earth’s surface and reflected from
one satellite to another, and finally sent back to the Earth.
Since the effective thickness of the atmosphere is on the order
of 5–10 km, while the outer space photon loss and decoher-
ence is negligible, with the help of satellites one can achieve
global free-space quantum communication, provided the
quantum states survive the passage through the aerosphere.
Along these lines, an important experimental progress has
been made in the free-space transmission of attenuated laser
pulses [over 23.4 km, see Kurtsiefer et al. (2002); over
144 km, see Schmitt-Manderbach et al. (2007)], of triggered
single photons from entangled-photon pairs [over 7.8 km, see
FIG. 30. Experimental results. (a), (b) Measured fractions, in the Resch et al. (2005); and over 144 km, see Ursin et al.
H=V and in the þ=
bases, for the original mixed state. (c), (2007)], of entangled-photon pairs [over 144 km, see Fedrizzi
(d) Measured fractions for the purified state in modes a3 and b3, et al. (2009)], and distribution of entangled-photon pairs to
in the same pair of bases. A comparison of the fractions in (a) and different locations [over 600 m, see Aspelmeyer et al. (2003);
(b), with the results shown in (c) and (d) indicates entanglement over 13 km, see Peng et al. (2005), see Table I for a
purification. From Pan, Gasparoni, Ursin et al., 2003. summary].
More recently, a 144-km free-space link was built between
two Canary Islands and used for transmission of one photon
of an entangled pair (Ursin et al., 2007), and later, both
actually went through the purification system. In addition to photons (Fedrizzi et al., 2009). The final photon states were
the low efficiency, another weakness of this experiment is found to preserve excellent, noise-limited fidelity, even
that, due to double pairs emission events in SPDC, a phase though they experienced extreme attenuation due to mainly
stabilization was required. This is unfeasible in long-distance turbulent atmospheric effects. The total channel loss of 64 dB
quantum communication. Considering possible applications corresponded to the estimated attenuation regime for a two-
of entanglement purification, these two problems can be photon satellite quantum communication scenario. The en-
solved by using entangled-photon pairs deterministically tanglement of the received two-photon states was confirmed
generated using quantum dots or other emitters. by violating the CHSH inequality by more than 5 standard
deviations. From a fundamental point of view, this means that
the photons are subject to virtually no decoherence during
E. Long-distance entanglement distribution their 0.5 ms long flight through air. For those aiming at a
world-wide quantum communication this is an encouraging
The ultimate goal of quantum communication is to work development. The photon-pair flight time of 0:5 ms
at long distances. A summary of the recent experimental
progress is listed in Table I [the distance for the distribu-
tion of entanglement in optical fibers was extended from
50 km (Marcikic et al., 2004) to the order of 100 km 64
In the context of quantum cryptography, recently revolutionary
(Honjo et al., 2007; Hübel et al., 2007; Takesue et al., progress has been achieved by introducing the idea of decoy states
2007)]. (Hwang, 2003; Lo, 2004, 2005; Wang, 2005). The decoy-state
For real life applications of fiber-based quantum commu- scheme, which is designed such that Alice randomly sends some
nication, one has to face several major limiting factors of her laser pulses with a lower average photon number can be used
including photon loss and photon detection noise (mainly to detect a photon-number-splitting attack, as Eve has no way to tell
dark counts). For quantum key distribution, the rate for dark which pulses are signal and which decoy. Thus, using classically
counts, etc., for a given photon detector is constant, while attenuated laser pulses, one can extend the secure quantum com-
munication distance from 30 km, as in the conventional scheme
the key rate decreases with increasing fiber length.
(Waks, Zeevi, and Yamamoto (2002), to  100 km with the decoy-
Therefore, the signal-to-noise ratio decreases exponentially
state protocol, and still gets higher key generation rates. Such a
with the length of the fiber. At a certain fiber length, the scheme was experimentally realized via optical fiber (Peng et al.
signal-to-noise ratio is so low that secure keys cannot be (2007); Rosenberg et al. (2007) and via free-space links (Schmitt-
generated. A further extension of the distance over which Manderbach et al., 2007). The decoy-state protocol allows the same
reliable quantum communication is possible requires detec- security level as in the case of true single-photon sources. Taking
tors of lower noise, fiber links of lower loss, quantum the advantage of ultra-low loss fibers and low-noise superconduct-
communication systems of faster working rates, etc. ing detectors Korneev et al. (2007) and Marsili et al. (2008)
In summary, the present-day technology puts a strong created a prototype of quantum key distribution working at a
limitation on the distance for practical fiber-based distance of 250 km.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

820 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

TABLE I. Summary of some recent experimental advances on entanglement distribution over long
distances. For the details on different settings, see the text. S refers to the Bell-CHSH parameter. V
refers to the single-photon transmission and visibility of the interference.
Year Reference Distance S/Visibility Rate (Hz)
Via free-space
2003 Aspelmeyer, et al.(2003) 600 m S ¼ 2:41ð10Þ >15
2005 Resch, et al. (2005) 7.8 km S ¼ 2:27ð2Þ 84
2005 Peng, et al. (2005) 13 km S ¼ 2:45ð9Þ 150
2007 Ursin, et al.(2007) 144 km S ¼ 2:508ð37Þ 20–40
2009 Fedrizzi, et al. (2009) 144 km S ¼ 2:612ð114Þ 0.071
Via optical fiber
2004 Marcikic, et al. (2004) 50 km S ¼ 2:185ð12Þ 5
2007 Takesue, et al. (2007) 60 km V ¼ 75:8% 0.3
2007 Honjo, et al. (2007) 100 km V ¼ 81:6% 1.4

represents the longest lifetime of photonic Bell states re- can be stored and synchronized with the arrival of other
ported so far. photon pairs. This would, for instance, enable efficient gen-
eration of multiphoton states in a time which increase only
polynomially with number of involved qubits.
F. Quantum memory and quantum repeaters

We have shown that entanglement purification enables one 1. Quantum repeater protocol
to overcome the degradation of the quality of photon entan- In classical communication, the problem of exponential
glement. Still, a major drawback of schemes for communi- attenuation can be overcome by using repeaters at certain
cation between distant nodes is the exponential scaling of the points in the channel. They amplify the signal and restore it to
error probability with the length of the connecting channels. its original shape. In analogy to fault-tolerant quantum com-
The quantum repeater protocol (Briegel et al., 1998; Dür puting (Preskill, 1998; Nielsen and Chuang, 2000), the quan-
et al., 1999) provides a blueprint of a general framework to tum repeater proposal (Briegel et al., 1998; Dür et al., 1999)
remedy this problem by nesting entanglement purification is a cascaded entanglement purification protocol for commu-
and swapping steps. Once constructed, it would enable one nication systems.
to establish high-quality long-distance entanglement with The quantum repeater protocol comprises three elements:
resources increasing only polynomially with transmission (i) A method for creation of entanglement between particles
distance. at distant nodes, which uses auxiliary particles at intermediate
A quantum memory for single photons, with the ability of ‘‘connection points’’ and a nested purification protocol.
interconverting between stationary and flying qubits (see (ii) Entanglement purification, even with imperfect means.
Sec. VI.F.2), is a crucial element in the quantum repeater (iii) A protocol for which the time needed for entanglement
scheme. There are several candidates for localized qubits. For creation scales polynomially, whereas the required material
instance, one may use atomic internal states to store local resources per connection point grow only logarithmically
information. Mapping between the atomic and photonic qu- with the distance, as show in Fig. 31.
bits requires a strong coupling between atoms and photons Here we describe a scheme for the physical realizations of
via high-finesse cavities (Raimond, Brune, and Haroche, a quantum repeater which has been proposed by Duan, Lukin,
2001; Leibfried et al., 2003; Walther et al., 2006) or initial Cirac, and Zoller (Duan et al., 2001).66 The ensembles
atom-photon entanglement together with entanglement swap- suggested atomic ensembles as local memory qubits. They
ping. Below we focus on the atomic-ensemble based schemes have a collectively enhanced coupling to light, even without
[Duan et al. (2001); Duan, Cirac, and Zoller (2002); Chen the aid of high-finesse cavities. The scheme incorporates
et al. (2007); Jiang, Taylor, and Lukin (2007); Zhao et al. entanglement swapping, built-in entanglement purification,
(2007); see also Sangouard et al. (2009) for a review]. and quantum memory.
We emphasize that quantum memories have applications Figure 32 is a schematic of a setup for entangling two
not only in long-distance quantum communication, but they atomic ensembles (optically thick atomic cells of Na identical
also provide a route to a more efficient multiphoton entangle- atoms), L and R, which are spatially separated within the
ment (see Sec. IV.D) or linear-optics quantum computing (see channel attenuation length. A pair of metastable lower states
Sec. VII). So far, the majority of the reported multiphoton jgi and jsi can correspond to hyperfine or Zeeman sublevels
interferometry experiments face the problem of random of electronic ground states of alkali atoms. Long lifetimes for
arrivals of SPDC photon pairs. Thus, scalability of this relevant coherences in such systems have been observed
approach is questionable. Given a quantum repeater,
ideally with long storage time, high writing, and retrieval 66
efficiencies,65 the randomly generated SPDC photon pairs Other physical implementations include the quantum repeater
based on solid-state photon emitters (Childress et al., 2005;
Childress et al., 2006) and a hybrid quantum repeater using bright
65
Extensive efforts still need to be undertaken to make a quantum coherent light and electronic and nuclear spins (van Loock et al.,
memory usable for this purpose; see Sec. VI.F.2. 2006).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 821

vectors j0a i and j0p i are, respectively, the atomic and pho-
N
tonic vacuum states with j0a i
i jgii . For a large Na , the
collectively enhanced signal-to-noise ratio may strongly
boost the efficiency of the scheme.
This setup enables one to generate entanglement between
two distant ensembles, L and R, using the configuration
shown in Fig. 32. If two laser pulses excite both ensembles
simultaneously, the whole system is described by the state
jiL  jiR , where jiL and jiR are given by Eq. (84). The
subscripts L and R denote the respective cells [in Eq. (84) one
should add such subscripts to all mathematical objects]. The
forward-scattered Stokes signal from both ensembles is com-
FIG. 31 (color online). Quantum repeater scheme. (a) Creation of bined at the BS and a photodetector click in either D1 or D2
a sequence of entangled pairs. (b) Nested purification protocol
measures the combined radiation p from
ffiffiffi two samples, a^ yþ a^ þ or
which combines the methods of entanglement swapping and puri-
fication, assisted with repeated creation of auxiliary pairs. a^ y
a^
with a^  ¼ ða^ L  ei’ a^ R Þ= 2. The symbol ’ denotes
(c) Purification of entangled pairs stored in distant locations. an unknown difference of the phase shifts in the two channels.
Adapted from Briegel et al., 1998. Depending on which detector clicks, one applies a^ þ or a^
to
the whole state jiL  jiR . The resulting projected state of
the ensembles, L and R, is nearly maximally entangled. It
both in a room-temperature dilute atomic gas (Phillips et al., reads [we neglect Oðpc Þ terms]
2001) and in a sample of cold trapped atoms (Liu et al., pffiffiffi
j’ i ^y i’ ^ y
2001). LR ¼ ðSL  e SR Þ= 2j0a iL j0a iR : (85)
All the atoms are initially prepared in the ground states
jgii . A sample is illuminated by a short, off-resonant laser For each round, the probability for getting a click is given by
pulse that induces Raman transitions into states jsii . pc . Thus, we need to repeat the process about 1=pc times to
Particularly important is the forward-scattered Stokes light warrant a successful preparation of entanglement. The aver-
(the signal mode a).^ It is uniquely correlated with the exci- age preparation time is given by T0  t
=pc .
tation of
pffiffiffiffiffiffi P the symmetric collective atomic mode S^
The entanglement generation (as well as entanglement
ð1= Na Þ i jgiii hsj, where the summation is taken over all connection) in the DLCZ scheme is based on single-photon
the atoms. The light-atom interaction generates, after the interference at photodetectors,67 which requires a stable long-
interaction time t
, a two-mode (a^ and S) ^ squeezed state distance interferometric stability. The fluctuations of the
(Braunstein and van Loock, 2005), with the squeezing pa- relative phase ’ caused by the environment would wash
rameter rc proportional t
. If t
is very small, the two-mode out the coherence (i.e., entanglement) in Eq. (85). For in-
squeezed state can be written in the perturbative form stance, to maintain path-length phase stability at the level of
=10 ( is the wavelength) for single photons, typically of
pffiffiffiffiffiffi   1 m, generated from atomic ensembles (Eisaman
ji ¼ j0a ij0p i þ pc S^y a^ y j0a ij0p i þ Oðpc Þ; (84)
et al., 2005) requires a precise control of timing jitter at a
where pc ¼ tanh2 rc  1 is the small excitation probability subfemtosecond level, which is almost experimentally im-
and Oðpc Þ represents the terms with more excitations, whose possible (Holman et al., 2005). For more detailed analysis on
probabilities are equal or smaller than p2c . The Hilbert space the phase-stability problem of the DLCZ scheme, see Chen
et al. (2007).
The phase-stability problem can be overcome by interfer-
ing two photons, one coming from each remote ion or atom in
a cavity (Bose et al., 1999; Browne, Plenio, and Huelga,
2003; Feng et al., 2003; Simon and Irvine, 2003), which was
experimentally implemented by Maunz et al. (2007) and
Moehring et al. (2007). A robust implementation of a
quantum repeater using atomic ensembles was proposed by
Chen et al. (2007); Jiang, Taylor, and Lukin (2007); and
Zhao et al. (2007). With the help of two-photon interference
it eliminates the stringent requirement of long-distance phase
stabilization.
FIG. 32 (color). Schematic of a setup for generating entanglement Though the DLCZ scheme does not meet all the criteria for
between the two atomic ensembles L and R in the DLCZ scheme. long-distance quantum communication, it provides a prom-
The inset shows the relevant level structure of the atoms in the ising approach to a fully controllable single-photon source
ensemble with the ground state, jgi, the metastable state for storing based on atomic ensembles, which seems to be much easier
a qubit, jsi and the excited state, jei. The transition jgi ! jei is for experimental demonstrations. We now summarize the
coupled by a classical laser light, with the Rabi frequency . The
forward-scattered Stokes light comes from the transition jei ! jsi.
67
An off-resonant coupling with a large detuning
is assumed. Such a method was first proposed to entangle single atoms (Bose
Adapted from Duan et al., 2001. et al., 1999; Cabrillo et al., 1999).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

822 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

basic ideas behind it. The atomic ensemble generates a

correlated state in Eq. (84), which is an exact analog of the
SPDC radiation. By measuring the forward signal mode
with a single-photon detector, under the condition that the
detector clicks, the collective atomic mode is projected to a
single-excitation state. Such excitations can be stored for a
reasonably long time in metastable states (the so-called
ground-state manifold) of the atoms. On demand the single-
atomic excitation can be transferred to a single photon (still
within the storage time) with a method described in the next
section. This is with fully controllable properties: the emit-
ted single-photon pulse is directed forward; the emission
time is controllable by the repumping time; and the pulse
FIG. 33 (color). Quantum memory for photonic polarization qu-
shape is controllable by varying the time dependence of the
bits. Two identical ensembles are identically driven by a classical
Rabi frequency of the repumping pulse.
field, which is equally right-circularly and left-circularly polarized.
So far, significant advances have been achieved along these Classical and quantized light fields are fed into the first PBSR=L
lines. For a partial list,68 we mention the following: control- (‘‘rotated’’ PBS which reflects left-circular photons and transmits
lable generation, storage and retrieval of single photons with right-circular photons) and leave at two different outputs of the
tunable frequency, timing and bandwidth (Chou et al., 2004; second PBSR=L . Two half-wave plates (HWP), enabling the trans-
Eisaman et al., 2004; Chanelière et al., 2005; and Eisaman formation jRi $ jLi, are placed along the jLi output of the first
et al., 2005), a deterministic single-photon source using PBSR=L . As the atomic cell r (cell l) works as a quantum memory
measurement-based feedback protocol (Chen, S., et al., for single photons with right-circular (left-circular) polarization, via
2006; Laurat et al., 2006; Matsukevich et al., 2006b); the adiabatic transfer method, the whole setup is therefore a
conditional control of two atomic memories (Felinto et al., quantum memory of any single-photon polarization states. The inset
2006), entanglement of two atomic ensembles (Matsukevich shows the relevant level structure of the atoms, with the ground state
et al., 2006a) and its distribution between two quantum nodes jbi, the storage metastable state jci, and the excited state jei. From
located 3 meters apart (Chou et al., 2007), mapping photonic Chen et al., 2007.
entanglement into and out of an atomic-ensemble-based
quantum memory (Choi et al., 2008), optimal control of
light pulse storage and retrieval (Novikova et al., 2007), and quantum interfaces. Along this line, theoretical ideas on
the Hong-Ou-Mandel interference of photon pairs from two quantum light memory have been proposed (Fleischhauer
independent ensembles (Chanelière et al., 2007). A quantum and Lukin, 2000; Kozhekin, Mølmer, and Polzik, 2000;
repeater node following the robust protocol (Chen et al., Lukin, Yelin, and Fleischhauer, 2000; Duan et al., 2001;
2007; Zhao et al., 2007) was experimentally demonstrated by Duan, Cirac, and Zoller, 2002; Fleischhauer and Lukin, 2002;
Yuan et al. (2008). These experiments are currently limited Chen et al., 2007) and the relevant experimental advances
by the relatively short coherence time ( 20 s) of the (Kash et al., 1999; Liu et al., 2001; Schori et al., 2002;
memory qubits and the low conversion efficiency ( 15%) Phillips et al., 2001; Bajcsy, Zibrov, and Lukin, 2003; van der
between photonic and atomic states. See Kimble (2008) for a Wal et al., 2003; Julsgaard et al., 2004; Matsukevich and
more in-depth review on this topic. Kuzmich, 2004; Hétet et al., 2008) have been reported.69
The atomic-ensemble-based quantum memory consists of
2. Quantum state transfer between matter and photons a coherently driven atomic ensemble (N  1 atoms) of large
optical thickness with a level structure shown in the inset of
The technique of quantum state transfer between matter
Fig. 33. The jci-jei transition is coherently driven by a
and photons is indispensable for both long-distance quantum
classical field of Rabi frequency ðtÞ, and the jbi-jei tran-
communication and large-scale optical quantum computing
sition is coupled to a quantized single-mode (the multimode
(see Sec. VII). In such applications, the matter itself should be
case is similar) light field (described by an annihilation
endowed with a long storage time. This makes atoms strong
^ The coupling constant is denoted by g. Under
operator a).
candidates for localized photonic information carriers. The
the two-photon resonance (i.e., the two detunings for the two
early proposals (Cirac et al., 1997; van Enk, Cirac, and
transitions shown in the inset of Fig. 33 are both equal to
),
Zoller, 1997) along these lines use the strong coupling of
the classical driving field can induce transparency for the
photons and single atoms in high-finesse cavities.
quantized light field and a substantial group-velocity reduc-
The basic idea of quantum light memory is in transferring a
tion, and even the complete stopping of the light [for reviews,
photonic state to the excitations of atomic internal states. In
see Lukin and Imamoğlu (2001), Lukin (2003), and
such a way, it can be stored. After some controllable time, it
Fleischhauer, Imamoglu, and Marangos (2005)]. The
should be possible to transfer back the excitations to photons
Hamiltonian of the whole system (N atoms plus the quantized
restoring the original quantum state. The experimentally
light field), in a frame rotating at the optical frequency,
challenging technology at the interface of photons and single
atoms motivated search for alternative routes to matter-light
69
For comprehensive reviews, see Sangouard et al. (2009) and
68
For a comprehensive review, see Sangouard et al. (2009). Hammerer, Sorensen, and Polzik (2010).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 823

pffiffiffiffi
reads H ¼ ℏðtÞS^ec þ pffiffiffiffi PNN a^ Seb þ H:c:, where Sec ¼
ℏg ^ ^ topic has been reviewed earlier by Kok et al. (2007);
PN
i jeiii hcj, Seb ¼ ð1= N Þ i jeiii hbj, and H:c: denotes a
^ O’Brien (2007); O’Brien, Furasawa, and Vuckovic (2009);
Hermitian conjugate of the previous expression. This and Ralph and Pryde (2010). This section serves as a
Hamiltonian has its zero-energy eigenstates, the so-called supplement to these previous reviews. Thus, we skip
‘‘dark states.’’ When the atom number is much larger than some theoretical details and mainly focus on recent experi-
the photon number, the dark states represent elementary mental advances.
excitations of bosonic quasiparticles, i.e., the dark-state
polaritons. For more details on this concept, see A. Linear-optical two-qubit logic gates
Fleischhauer and Lukin (2000, 2002) and Lukin, Yelin,
and Fleischhauer (2000). Knill, Laflamme, and Milburn (2001) showed that the
By adiabaticallypffiffiffiffichanging ðtÞ between
pffiffiffiffi the two limiting success rate of the logic gates can be arbitrarily close to 1.
cases [ðtÞ  g N , or ðtÞ  g N ] one can coherently To this end, one can use ancilla photons and detectors. A
map dark-state polariton states onto either purely atomlike similar conclusion has been independently obtained by
states where the photons are stored or purely photonlike Koashi, Yamamoto, and Imoto (2001). Their solution was
states, which corresponds to the release of the stored photons. based on entangled ancilla photons. A novel aspect of this
In principle, a quantum memory based on the adiabatic trans- protocol is that, despite the lack of the photon-photon
fer method is reversible, preserves a pulse shape of the stored interaction, quantum measurements with photon-number
photons (Fleischhauer and Lukin, 2000, 2002), and may have resolving detectors can induce effective nonlinearity suffi-
an efficiency very close to unity. As there is no excited atomic cient for the realization of two-qubit gates. The original
state in the dark-state subspace, the storage time can be very KLM scheme was only very recently implemented in a
long. sophisticated setup up using polarization encoding and
The original quantum memory was proposed for storing a Sagnac interferometers for increased stability (Okamoto
coherent superposition of photon-number states. However, et al., 2010).
two atomic ensembles can be entangled by storage of two Further improvements reduced the complexity and im-
entangled light fields (Lukin, Yelin, and Fleischhauer, 2000). proved the efficiency of the original scheme by introducing
Recent experiments achieved single quantum excitation certain assumptions and restrictions, enabling a series of
memory times of 1 ms using cold atomic ensembles (Zhao, experiments and demonstrations. Hofmann and Takeuchi
B. et al., 2009), 6 ms using atomic rubidium confined in a (2002) and independently Ralph et al. (2001, 2002) proposed
one-dimensional optical lattice (Zhao, R. et al., 2009), and quantum gates working under the restriction of what is here
0.1 s using quantum memory confined in an optical lattice called a two-mode or four-mode case. That is, the successful
with laser compensation of the lattice light shifts (Radnaev operation of the gate can be verified if the two photons
et al., 2010). involved are detected in certain outputs (this is also called
‘‘detection in coincidence basis’’ or ‘‘conditioned detec-
VII. PHOTONIC QUANTUM COMPUTING tion’’). Essentially a single two-photon interference is
enough, together with a state-dependent filtering, to perform
As we have seen in the above section, the photon, thanks to probabilistic CNOT operations. The restriction does not allow
its high transmittance through air and glass fibers and its further operations on the two photons involved and thus limits
extremely long decoherence time, has arguably been the best the depth of calculations, however, the simplicity of the gate
candidate for quantum communication. However, things be- makes it a useful and reliable tool if no further joint opera-
come trickier when we come to the field of quantum compu- tions on the two photons are required. The original proposal
tation. The weak interaction between photons, which is of a used dual-rail encoding and was first implemented by
significant benefit in quantum communication, turns out to be O’Brien et al. (2003). With polarization encoding, the
a drawback where nontrivial two-qubit quantum gates are CNOT gate and conditional phase shift operations were
essential. For a long time it seemed obvious that linear-optical also demonstrated (Pittman, et al., 2003; Sanaka, et al.,
two-photon gates can be done only in a nondeterministic 2004). An even simpler setup becomes possible with polar-
fashion and thus quantum computing cannot be scalable. ization encoding (Kiesel, Schmid, Weber, et al., 2005;
However, in 2001, Knill, Laflamme, and Milburn (2001) Langford et al., 2005; Okamoto et al., 2005) which in
proved that scalable optical quantum computing is possible turn could be already applied, e.g., to observe cluster states
using only single-photon sources, linear-optical elements, for one-way quantum computing [see, Kiesel, Schmid, Toth,
and photon-number resolving detectors. The KLM scheme et al. (2005)]. Exploiting hyperentanglement, Lanyon, et al.
subsequently spurred new experiments demonstrating proba- (2008) demonstrated a three-qubit Toffoli gate. More
bilistic controlled two-photon gates. Despite the efforts of recently, Politi et al. (2008) reported high-fidelity silica-
KLM, the resource overhead required for optical quantum on-silicon integrated optical realizations of key quantum
computing is daunting. Several improvements of this proto- photonic circuits. Laing et al. (2010) reported a two-photon
col, particularly those based on cluster states or error encod- quantum interference visibility of 99.5(4)%, a CNOT gate
ing, have dramatically reduced this worrying resource [the obtained average fidelity of logical basis was 96.9(2)%]
overhead, and started to bridge the gap between the theoreti- and a path-entangled two-photon state (with fidelity of
cal scalability and practical implementations. >92%). Coffman, Kundu, and Wootters (2010) reported the
We witnessed considerable theoretical and experimental first probabilistic logic gates on integrated circuits for polar-
progress in optical quantum computing in these years. This ization qubits.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

824 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

case). Depending on the registered polarizations, up to a

specific unitary transformation, a nondestructive CNOT
gate operation is then performed on ji20 50 .

B. Cluster-state quantum computing

Another significant step is the discovery of ‘‘cluster-state

quantum computing’’ (Raussendorf and Briegel, 2001;
Raussendorf, Browne, and Briegel, 2003; Briegel et al.,
2009), which is based on the preparation of highly entangled
multiqubit states, the so-called ‘‘cluster states’’ (Briegel and
Raussendorf, 2001) and adaptive one-qubit measurements.
Besides its thought-provoking theoretical structure, this
model also brings a number of practical advantages for
physical realization of quantum computation. In scenarios
in which quantum gates can be performed directly in, at best,
a nondeterministic fashion, the one-way model is particularly
FIG. 34 (color). A quantum CNOT gate for photonic qubits. (a) A useful. Linear-optical cluster-state quantum computation is
nondestructive CNOT gate constituting of a PBS, half-wave plates the most prominent example. For existing, short surveys of
(HWP) and using an ancilla entangled photon pair j c
i34 (Pittman,
the topic, see O’Brien (2007) and Briegel et al. (2009).
Jacobs, and Franson, 2001). (b) Experimental results of two-photon
Cluster states can be created by a controllable Ising-type
CNOT gate. From Zhao et al., 2005.
interaction (Briegel and Raussendorf, 2001; Raussendorf and
Briegel, 2001; Raussendorf, Browne, and Briegel, 2003). It
In a separate development, Koashi, Yamamoto, and Imoto was recently shown that an efficient preparation of cluster
(2001) and Pittman, Jacobs, and Franson (2001) showed that states is possible with probabilistic two-qubit controlled
by using entangled pairs of photons as ancilla, the success of phase flip gates (Duan and Raussendorf, 2005; Chen, Q.,
the gate operation can be inferred by the detection of photons et al., 2006). Few-photon cluster states were created in
in ancilla outputs. This enables one to perform the gate several recent experiments (Kiesel, Schmid, Toth, et al.,
operation in a nondestructive manner. Such gates, assisted 2005; Walther, Resch, Rudolph et al., 2005; Zhang, A.-N.
by entangled or unentangled photon pairs, were reported et al., 2006; Lu et al., 2007; Tokunaga et al., 2008).
using four or five photons; see Gasparoni et al. (2004);
Zhao et al. (2005); Bao et al. (2007); Tokunaga et al. 1. Constructing photonic cluster states
(2008); and Gao, Yao, et al. (2010).
By combing the one-way model with linear-optical quan-
In this section, we present, for pedagogical purposes, the
tum computing, recent theoretical proposals require much
working principle and a proof-of-principle demonstration
less resources and effectively replace the original KLM
(Zhao et al., 2005) of a nondestructive CNOT gate for two
scheme [see Nielsen (2004); Browne and Rudolph (2005);
independent photons. Pittman, Jacobs, and Franson (2001)
Bodiya and Duan (2006); Chen, Zhao, and Pan (2008); and an
suggested that a successful operation of the gate can be
efficient parity-encoded optical quantum computing model
indicated by detection acts of measured ancilla photons.
by Gilchrist, Hayes, and Ralph (2007)]. Nielsen showed that
Such information is feed forwardable. This is an important
efficiency can be greatly enhanced by building photonic
feature for both the circuit model and the one-way model of
cluster states using easy nondeterministic gates. The resource
scalable optical quantum computing.
overhead (Bell states, operations, etc.) for a reliable entan-
Suppose that one aims to perform a CNOT gate on an
gling gate in Nielsen’s scheme is 103 , and thus about 2
arbitrary two-qubit state ji25 [Eq. (80)] using an ancilla
orders of magnitude less than the original KLM protocol.
entangled-photon pair in the Bell state j c
i34 . Note that
Furthermore, by introducing two linear-optical fusion opera-
PBS2 in Fig. 34(a) transmits jH þ Vi while it reflects
tions, Browne and Rudolph (2005) achieved a greater degree
jH
Vi polarization. The output state of the whole apparatus
of efficiency ( 102 ) and in a simpler scheme than the
is (Pittman, Jacobs, and Franson, 2001)
previous proposals. Matter qubits can also be constructed
1 into cluster states using linear optics and photon interference,
ji25 j c
i34 ! ½jV 0 i30 jVi40 ðCNOT20 50 ji20 50 Þ as proposed by Barrett and Kok (2005). Here we focus on a
4
þ jH 0 i30 jVi40 ðz^50 CNOT20 50 ji20 50 Þ linear-optical architecture for one-way quantum computing.

þ jH 0 i30 jHi40 ðx^ 50 z^50 CNOT20 50 ji20 50 Þ a. Linear-optical architecture

þ jV 0 i30 jHi40 ðx^ 50 CNOT20 50 ji20 50 Þ The Browne-Rudolph scheme has an important practical
pffiffiffi
3 advantage. It requires stable interferometry over only the
þ j . . .inot four-mode cases : (86) coherence length of the photons, i.e., the interferometer
2
path lengths differences only need to be kept constant to
Consider a detection of a pair of photons at the output modes within tens of microns, not fractions of a wavelength. Its
30 and 40 , only one photon at each output (the four-mode two basic operations are the type I and type II ‘‘two-qubit

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 825

operation fails (also with a probability of 50%) when zero

or two photons of either polarization are detected. The failure
outcomes have the effect of measuring both input qubits in
the z eigenbasis, which leaves the remaining qubits in a
cluster state of the same layout as before the measurement,
but now with all the bonds broken to the measured qubit.
Browne and Rudolph (2005) also showed that one can finally
prepare a square-lattice cluster of N qubits with a temporal
overhead scaling logarithmically with N, and with an opera-
FIG. 35 (color online). Nondeterministic photonic qubit fusion tional overhead (i.e., number of fusion operations) scaling as
gates. The type I fusion gate combines two input single photons N lnN (Chen, Zhao, and Pan, 2008). The described protocol
on a PBS and one of the outputs is measured in the H  V basis. is a linear-optical realization of the Duan and Raussendorf
The type II fusion gate combines two photons on a 45 rotated PBS (2005) proposal, and combines the advantages of the Browne-
(that is, both inputs and outputs are rotated using a HWP), and both Rudolph scheme, whose overall efficiency is thereby
outputs are detected. Adapted from Browne and Rudolph, 2005. demonstrated. A crucial element for a realistic realization is
quantum memory for polarization qubits, which was dis-
cussed in Sec. VI.F.2.
fusion’’ operations (Fig. 35). The physical mechanism of
type I fusion is again two-photon interference at the PBS, b. Event-ready entangler
which is used as a parity check (Cerf, Adami, and Kwiat,
1998; Pan et al., 1998; Pan, Simon, Brukner, and Zeilinger, In the above linear-optical architecture, two-photon en-
2001; Pittman, Jacobs, and Franson, 2001). As displayed in tangled pairs are the basic resources. In the case of SPDC,
Fig. 36, starting from a supply of two qubit polarization one usually does not know when a pair is emitted. Only a
cluster states 12 ðjHHi þ jVHi þ jHVi
jVViÞ (which are firing of photon detectors informs one that a spontaneous
emission act happened. However, most schemes of optical
equivalent to a Bell state and can be created via the methods
quantum computation, including the nondestructive CNOT
described in Sec. VII.B.1), the type I fusion operation allows
gates, scalable fusion of cluster states, require that the photon
one to efficiently generate arbitrarily long linear cluster
pairs should be created in an ‘‘event-ready’’ (or heralded)
states. If the type I fusion is applied to the end qubits of
way.
linear (i.e., one-dimensional) clusters of lengths n and m,
Zhang et al. (2008) proposed a way to generate one pair of
successful outcomes (with a probability of 50%) generate a
event-ready entangled photons from four single photons, with
linear cluster of length n þ m
1. The type I fusion
a method that requires only linear optics and photon-number-
discriminating detectors and performed an experimental
simulation using four photons from SPDC. Another scheme
of generating triggered photon pairs, which does not need true
single-photon sources but totally rests upon SPDC, was
proposed by Sliwa and Banaszek (2003). Following this
proposal, in two experiments, Barz et al. (2010) and
Wagenknecht et al. (2010), have demonstrated heralded
generation of photon states that are maximally entangled in
polarization.
Probably a more promising realization of a triggered
entangled-photon source will come from the biexciton (two
electron-hole pairs) radiative decay in a self-assembled quan-
tum dot. This was demonstrated by Stevenson et al. (2006).
A quantum dot can emit a single pair of entangled photons on
demand, with a probability close to 1. However, it has a very
low extraction efficiency. Recently, Dousse et al. (2010) used
a carefully fabricated cavity to increase the collection effi-
ciency, and created a source of polarization-entangled photon
pairs with a state fidelity of 0.67 and a rate of 0.12 per an
excitation pulse.

FIG. 36 (color online). Construction of two-dimensional cluster C. Few-photon quantum computing experiments
states using photon fusion (Browne and Rudolph, 2005; Duan and
Raussendorf, 2005). (a) Certain measurements on a cluster qubit In recent years, we also witnessed a number of proof-of-
will leave the remaining qubits in a new cluster state with a different principle demonstrations of quantum computing involving
layout. (b) The effect of type I and type II fusion operations on several photons and linear optics (experimental realizations
successful connection of two linear cluster states. (c) One method of of photonic CNOT gates are been discussed in Sec. VII.A).
efficient construction of two-dimensional cluster states [see also For example, Mohseni et al. (2003) and Tame et al. (2007)
Duan and Raussendorf (2005)]. demonstrated the two-qubit Deutsch-Josza algorithm in a

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

826 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

four photonic qubits by Lanyon et al. (2007) and Lu,

Browne, Yang, and Pan (2007). Aiming to solve the easiest
case of Shor’s algorithm (15 ¼ 3 5), these two experiments
designed a simplified linear-optical network to implement the
quantum circuits of the modular exponential execution. The
results yielded a high success probability (  0:99) of factor-
ing. It is notable that in these optical experiments during the
computation genuine multiparticle entanglement and multi-
path interference were observed, which did not appear in the
previous implementations using nuclear magnetic resonance
(Vandersypen et al., 2001). Optical implementations of
Shor’s algorithm later moved into an integrated photonic
chip devices (Politi, Matthews, and O’Brien, 2009).
Another interesting avenue of the ongoing multiphoton
experiments is to exploit the preliminary photonic quantum
computers for simulation of many-body physics, a powerful
application of quantum computers proposed in the early
1980s. Following proposals by Han, Raussendorf, and Duan
(2007) and Pachos (2007), four-qubit and six-qubit graph
states were created to mimic the ground state of the Kitaev
FIG. 37 (color online). Few-qubit cluster states and the quantum
spin-lattice model (Kitaev, 2003). Using the photonic graph
circuits they implement. For each three-qubit and four-qubit cluster,
its quantum state (jlin3 i, jlin4 i, j⊏4 i, j⊐4 i, or jh4 i) and the
states to simulate the creation and braiding of the anyons in
computation carried out in the one-way quantum computer model is the Kitaev model, a phase shift of  related to the anyon
shown. Adapted from Walther, Resch, Rudolph et al., 2005. braiding was observed, confirming the predictions for the
fractional statistics of Abelian 1=2 anyons (Lu et al., 2009;
Pachos et al., 2009). Recently, Lanyon et al. (2010) em-
circuit and a one-way model. The Grover’s search algorithm ployed a photonic quantum computer to solve a quantum
(Grover, 1997) has been realized by Kwiat et al. (2000) by chemistry problem: calculation of the energies of the hydro-
designing an optical circuit, and later by Walther, Resch, gen molecule. In their experiment, the iterative phase estima-
Rudolph et al. (2005); Chen, K. et al. (2007); Prevedel, tion quantum algorithm was performed in full. With the
Walther et al. (2007); and Vallone, Pomarico, and Martini assistance of a classical computer, it was possible to calculate
(2008), who used four-qubit cluster states. the complete energy spectrum, up to 20 bits of precision. Ma
In the experiment of Walther, Resch, Rudolph et al. (2005) et al. (2011) used polarization states of four photons to
via a SPDC a four-photon polarization-entangled cluster state simulate a frustrated Heisenberg spin system: a spin-1=2
was created (with a fidelity of F ¼ 0:63  0:02): tetramer.

jc i ¼ 12ðjHi1 jHi2 jHi3 jHi4 þ jHi1 jHi2 jVi3 jVi4

D. Toward scalable optical quantum computing
þ jVi1 jVi2 jHi3 jHi4
jVi1 jVi2 jVi3 jVi4 Þ: (87)
While in small-scale tests of optical quantum computing,
With local unitary operations jc i can be arranged to various we have witnessed a progress, serious problems exist in
cluster shapes jlin4 i, j⊏4 i, j⊐4 i, and jh4 i (see Fig. 37). scaling up this technique. Here, we briefly discuss the key
Using the cluster state jc i, a universal set of quantum logic challenges and the ongoing efforts in designing fault-tolerant
operations, single-qubit rotations, and nontrivial two-qubit architectures, fighting against experimental noise, and im-
gates were demonstrated. In addition, Walther et al. also proving single-photon sources and detectors.
implemented a two-qubit Grover’s quantum search algorithm Quantum computers will be very susceptible to noise,
using the box cluster jh4 i. The measured probability of the which rapidly destroys the fragile quantum information.
quantum computer to determine the correct outcome was Much effort has been devoted to the understanding of the
about 90%. A drawback of the experiment is that no feed scalability under realistic noise models. The threshold theo-
forward was used. This reduced the success rate of the rem has established that if the noise is below some value, an
computation by a factor of 2 for every measurement. An arbitrarily long quantum computation can be realized. Several
improved experiment by Prevedel, Walther et al. (2007) fault-tolerant architectures for optical quantum computing
incorporated active fast feed forward, so that the earlier have been proposed. Nielsen and Dawson (2005) showed
measurement outcomes could change the setting of a future that the thresholds are respectively <3 10
3 and <10
4
measurement in real time. Particularly, the computational for photon loss and depolarizing noise [see also Dawson,
step (i.e., the individual feed-forward cycle) could be oper- Haselgrove, and Nielsen (2006)]. In addition, there are pro-
ated in less than 150 ns using electrooptical modulators. posals of one-way quantum computing in decoherence-free
Shor’s quantum algorithm provides a way of factoring subspaces (Tame, Paternostro, and Kim, 2007; Jiang et al.,
large integers in polynomial time, a task for which no 2009) and topological cluster-state quantum computing
efficient classical method is known. Recently, a compiled (Raussendorf, Harrington, and Goya, 2006; Raussendorf,
version of Shor’s algorithm has been demonstrated using Harrington, and Goya, 2007). The latter proposal achieved

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 827

a high error threshold of 0:75 10
3 , the highest known for a the promise of a ready integration, and much experimental
local architecture. For photon loss alone, Ralph, Hayes, and effort has recently been devoted to improving the single-
Gilchrist (2005), Varnava, Browne, and Rudolph (2006), and photon quality, collection efficiency, and interference of
Varnava, Browne, and Rudolph (2008) have designed photons from remote independent quantum dots. Other con-
loss-tolerant quantum computer schemes within the circuit trollable single-photon sources can be devised using trapped
model and one-way model; the latter scheme can tolerate an single atoms in high-finesse optical cavities, which are spec-
overall optical loss, including source inefficiency and collec- trally narrow and have a well-defined spatial mode. However,
tion loss, up to an impressive 33%. Gong et al. (2010) there are problems associated with low out-coupling effi-
proposed a new scheme where the efficiency threshold for ciency. Single photons emitted by atomic ensembles, as we
loss tolerance requires the product of source and detector discussed in Sec. VI.F, are another promising source. This type
efficiencies to be >50%. Despite the progress, one should of single-photon sources naturally enjoys a very narrow
note that when the losses are high, the resource requirements ( MHz) line width and good indistinguishability. However,
become impractically high. Moreover, unfortunately, these in addition to the low photon extraction efficiency, much work
loss-tolerant codes tend to amplify the depolarizing errors needs to be done to improve the retrieval efficiency (converting
(e.g., bit flips and phase flips); the trade off has been atomic collective excitations into photons) up to unity.
discussed and new schemes have been designed which To meet the stringent demands of scalable optical quantum
tolerate both errors in Rohde, Ralph, and Munro (2007). computing, one faces yet another challenge: new single-
Some basic quantum error correction codes have been photon detectors that have near-unity efficiencies, high repe-
tested in optical experiments. O’Brien et al. (2005) demon- tition rates, low dark count rates, and the ability to resolve the
strated a two-qubit code for correction of a Z-measurement photon number. Currently, mostly used room-temperature
error. With a continuous-variable encoding, Aoki et al. silicon single-photon detectors can be operated at 10 MHz
(2009) realized a nine-qubit Shor’s code, which is able to with a peak efficiency of 65%, a dark count rate of about
correct an arbitrary single-qubit error. Decoherence-free sub- 100 Hz, and a timing jitter of typically 500 ps; work is in
spaces, a type of passive error-preventing codes, have been progress to improve these parameters [see, e.g., Kardynal,
experimentally realized using two photons by Kwiat et al. Yuan, and Shields (2008)]. Significant progress (Rosenberg
(2000) and four photons by Bourennane, Eibl, Gaertner, et al. et al., 2005; Divochiy et al., 2008; Lita, Miller, and Nam,
(2004). Furthermore, the decoherence-free subspace ap- 2008) has been made on superconducting detectors capable of
proach was applied in an optical demonstration of the resolving photon number, with an ultralow dark count rate
Deutsch-Jozsa algorithm (Mohseni et al., 2003), and for (0.15 Hz at the wavelength of 1:3 m), and high efficiency up
reliable measurement-based one-way information transfer to 95%. See Hadfield (2009) for more details.
(Prevedel, Tame et al., 2007). To fight against the qubit- Finally, it is worth mentioning that chip-scale waveguide
loss error, an especially serious problem for photonic qubits, quantum circuits have been created recently, and used to
Lu et al. (2008) demonstrated a four-qubit Grassl erasure demonstrate high-visibility Hong-Ou-Mandel interference,
correction code [for the theoretical proposals see Grassl, CNOT gates, a realization of an elementary Shor’s algorithm
Beth, and Pellizzari (1997) and Ralph, Hayes, and Gilchrist (Politi et al., 2008; Politi, Matthews, and O’Brien, 2009), and
(2005)], and a tree-shaped graph state (Varnava, Browne, and quantum walk (Peruzzo, 2010). Current silica waveguide
Rudolph, 2006). The tested method is applicable both in the circuits have dimensions of about 1 cm per logic gate. This
quantum circuit model and in the one-way model. is a step toward integrated optics architecture for improved
Despite the progress, the fault-tolerant thresholds are still performance, miniaturization, and scalability. An open chal-
well beyond what is achievable with today’s technology. lenge is to integrate such devices with single-photon sources
Optical quantum computing makes critical use of sources of and detectors.
on-demand single photons which are indistinguishable and can
be collected efficiently. The majority of experimental demon- VIII. CONCLUDING REMARKS
strations so far have relied on the SPDC photons, which suffers
from undesired higher-order photon emissions [Weinhold We reviewed the principles and experimental techniques
et al. (2008), this has been considered as the major source of for manipulation of multiphoton entangled states, which have
error for most experiments reviewed here, see Sec. IV.D.1], enabled a series of pioneering experiments in the field of
large bandwidth, and the probabilistic manner of photon pair quantum information. A number of important applications
emission. New generations of single-photon sources have been have been highlighted: Laboratory tests demonstrating the
developed: they are based on solid-state devices, atoms, mole- contradiction between quantum mechanics and local realism
cules, ions, etc. [see Lounis and Orrit (2005) for a recent performed with entangled photons, that is, the Bell and
review]. These new single-photon emitters include single Greenberger-Horne-Zeilinger experiments (Weihs et al.,
quantum dots (Michler et al., 2000; Pelton et al., 2002; 1998; Pan, Simon, Brukner, and Zeilinger, 2001). Quantum
Santori et al., 2002; Bennett et al., 2005; Englund et al., teleportation, the transfer and reconstruction of quantum
2007; Shields, 2007; Strauf et al., 2007), nitrogen-vacancy states over arbitrary distances, which became an experimental
color centers (Brouri et al., 2000; Kurtsiefer et al., 2000), reality with four-photon interferometry (Bouwmeester et al.,
neutral atoms (Kuhn, Hennrich, and Rempe, 2002; McKeever 1997; Marcikic et al., 2003). A variation of teleportation,
et al., 2004; Darquié, M. J. et al., 2005), ions (Keller et al., entanglement swapping (Pan et al., 1998), which together
2004), and molecules (Brunel et al., 1999; Lounis and with entanglement purification (Pan, Gasparoni, Ursin et al.,
Moerner, 2000). Solid-state sources of single photons hold 2003) and quantum memory [see, e.g., Yuan et al. (2008)] are

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

828 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

the essential components of quantum repeaters necessary for dominating part is H^ a
f 
^ e  Eðx; tÞ, i.e., scalar product of
quantum networking and long-distance quantum communi- the dipole moment
of the atoms or molecules with the local
cation. Proof-of-principle demonstrations of linear-optics electric field. Since the electric polarization pðx; tÞ of a
quantum computing [see, e.g., O’Brien et al. (2003); medium is given by the mean dipole moment of the atoms
Walther, Resch, Rudolph et al., (2005); Lu et al. (2008); or molecules per unit volume, the principal term of the field-
and Lanyon et al. (2010)] and super-resolving phase mea- crystal
R interaction Hamiltonian Hint is proportional to
V pðx; tÞ  Eðx; tÞd x, where V is the volume of the crystal.
surements [see, e.g., Nagata et al. (2007)] with multiphoton 3

devices. One can assume that Eðx; tÞ interacts with pðx; tÞ only in the
The ultimate goals are long-distance quantum communi- point x, thus the ith component of polarization is in the most
cation and scalable optical quantum computing. However, general case given by
many technological challenges remain. Parametric down-
conversion (Kwiat et al., 1995; White et al., 1999) has X
3
pi ðx; tÞ ¼ ð1Þ
ij ðxÞEj ðx; tÞ
been serving as the main workhorse for the multiphoton j¼1
experiments reviewed here; up to eight entangled photons
have been observed (Yao et al., 2012). However, due to its X
3
þ ð2Þ
ijk ðxÞEj ðx; tÞEk ðx; tÞ þ    ; (A1)
intrinsic limitations, there is a bottleneck with regard to the j;k¼1
attainable brightness and fidelity of multiphoton states based
on it. This calls for the development of a next generation of where ð1Þ ð2Þ
ij and ijk are the (macroscopic) polarizability
more reliable and scalable single-photon sources (Lounis and tensors. For any crystal with centrosymmetric structure the
Orrit, 2005). Other challenges include efficient coupling and quadratic term of the polarizability vanishes. Thus, as we
detection of single photons and quantum memories for pho- shall see, the SPDC effect exists only for birefringent media
tons with long storage times and a high retrieval efficiency, having a nonzero value of ð2Þ . If one assumes that ð2Þ
ijk ðxÞ has
etc. Continuing effort is devoted in this direction and encour- the same value for all points within the crystal, one gets
aging results have been obtained. For instance, single photons
Z
and entangled photons have been generated from self- Hint  pðx;tÞ  Eðx;tÞd3 x
assembled quantum dots embedded in a microcavity, with V
extraction efficiency up to 80% (Dousse et al., 2010). In the Z Z
¼ plin ðx;tÞ  Eðx;tÞd3 x þ pnl ðx;tÞ  Eðx;tÞd3 x;
case of a long-lived quantum memory based on atomic V V
ensembles, storage times were reported to reach up to 8 ms
(Zhao, R. et al., 2009). Employment of cavities in this case where plin (pnl ) is the linear (nonlinear) term of polarization.
may lead to additional potential improvements. Armed with The nonlinear (NL) part of the Hamiltonian is
these new techniques, the control of multiphoton states will Z X
reach a higher level. In any event, we expect that the tech- H NL  ð2Þ
ijk Ei ðx; tÞEj ðx; tÞEk ðx; tÞd x:
3 (A2)
V ijk
niques reviewed in this article will be in the forthcoming
future in the mainstream of further progress of experimental The quantized electric field can be expressed (in the inter-
multi-photon interferometry. action picture) as
2 Z
X i
ACKNOWLEDGMENTS Eðx;tÞ¼ d3 k pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðk;Þaðk;Þe
^ iðkx
!tÞ þH:c:
¼1 2!ð2Þ3
We would like to thank Y.-J. Wei, Z.-S. Yuan, B. Zhao, X.-
¼EðþÞ ðx;tÞþEð
Þ ðx;tÞ; (A3)
H. Bao, M. Zhang, and N.-L. Liu for valuable comments and
helps. This work was supported by the National Natural where Eð
Þ ðx; tÞ ¼ ½EðþÞ ðx; tÞy , and the summation is over
Science Foundation of China, the National Fundamental two orthogonal linear polarizations, H.c. denotes the
Research Program (under Grant No. 2011CB921300) and Hermitian conjugate of the previous term, and ðk; ^ Þ is a
the Chinese Academy of Sciences. M. Ż. was supported by unit vector defining the linear polarization. The symbol
Professorial Subsidy of FNP and the MNiI Grant No 1 P03B aðk; Þ denotes the annihilation operator of a monochromatic
04927, MNiSW Grant No. N202 208538, and EU programs photon with wave vector k and polarization ðk; ^ Þ. The
QAP and Q-ESSENCE. He acknowledges Austrian-Polish principal commutation rules for such creation and annihi-
and German-Polish exchange programmes, and finally CAS lation operators are given by70 ½aðk; Þ; ay ðk0 ; 0 Þ ¼
visting professorship. H. W. acknowledges the support from ;0 ð3Þ ðk
k0 Þ, ½ay ðk; Þ; ay ðk0 ; 0 Þ ¼ 0, and
DFG (MAP) and BMBF (QPENS and QuORep) and the EU 0 0
½aðk; Þ; aðk ;  Þ ¼ 0.
programs QAP and Q-ESSENCE. A. Z. acknowledges the The relevant terms in the Hamiltonian–One can neglect the
supports from EPSRC, QIPIRC, FWF, and EC under the depletion of the laser field and assume that the total field is
Integrated Project Qubit Applications. ELaser ðx; tÞ þ Eðx; tÞ, where ELaser is a classical field. The
quantum field E describes the emitted photons. The down-
conversion takes place, thanks to only the terms in Eq. (A2) of
APPENDIX A: THE TWO-PHOTON STATES PRODUCED the form
BY SPDC

70
Crystal-field interaction–In the interaction Hamiltonian of These new operators are linked with the ones discussed earlier
P R
the electromagnetic field with an atom or a molecule, the (see Sec. III) by the relation al ¼  d3 kglk aðk; Þ.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 829

Z X XZ Z
ð2Þ Laser ð
Þ ð
Þ 3
ijk Ei Ej Ek d x: (A4) d3 k d3 k0 Fo ðk0 Þ
ðk0
k
k0 Þð! þ !0
V ijk ;0

!0 Þay ðk; Þay ðk0 ; 0 Þji; (A8)

Simply, only Eð
Þ contains the creation operators, and thus
acting on the vacuum state ji can give rise to a two-photon and the frequencies of the emissions satisfy the relation72
state. Thus, we forget about all other terms and analyze only !0 ¼ ! þ !0 .
H NL in the form of Eq. (A4) plus its Hermitian conjugate. Directions of emissions–Since ! ¼ jkjc=nð!; Þ, where
We now describe the laser field as a monochromatic plane c=nð!; Þ ¼ cð!; Þ is the speed of light in the given me-
wave zE^ 0 ðeiðk0 x
!0 t

Þ þ c:c:Þ, where E0 is the field ampli- dium, which depends on frequency and polarization, the
71
tude. Then, from Eq. (A4), one gets for HNL condition for frequencies becomes jk0 jcð!0 Þ ’ jkjcð!; Þ þ
Z X jk0 jcð!0 ; 0 Þ. This together with k0 ’ ks þ ki fixes the pos-
 d3 x fð2Þ
3jk E0 ½e
iðk0 x
!0 t

Þ
þ c:c: sible emission directions, frequencies, and polarizations.73
V jk Time correlations–The probability of a detection of a
XZ photon, of, e.g., the horizontal polarization H, at a detector
d3 kfð!Þ^ j ðk; Þay ðk; Þe
iðkx
!tÞ situated at point x and at time t, is proportional to
Tr½%ðtÞEð
Þ ðþÞ

XZ H ðx; tÞEH ðx; tÞ, where  is the coefficient which
0
d3 k0 fð!0 Þ^ k ðk0 ; Þay ðk0 ; Þe
iðkx
! tÞ0 þ H:c:; characterizes the quantum efficiency of the detection process,
0 % is the density operator, and EH is the horizontal component
(A5) of the field in the detector. For the above relation to be true,
we also assume that only photons of a specified direction of
with fð!Þ a factor dependent on !. Its specific structure is the wave vector enter via the aperture of the detector. For a
irrelevant here. Extracting only those elements of the above pure state, this reduces to pðx; t; HÞ ’ h c jEð
Þ ðþÞ
H EH j c i. The
expressions which contain x and t, one sees that their overall probability of a joint detection of two photons, of polarization
contribution to H NL is given Rby
ðk0
k
H, at the locations x1 and x2 , and at the moments of time t and
0 0
k0 Þe
ið!o
!
! Þt , where
 ð. . .Þ ¼ V d3 xeiðk0
k
k Þx . t0 , is proportional to
0
The terms with the time dependent factors eið!o þ!þ! Þt aver-
pðx1 ; t; x2 ; t0 Þ
age out in any time integration (see below), and thus we can
drop them. If we assume that our crystal is a cube L L L,  h c jEð
Þ ð
Þ 0 ðþÞ 0 ðþÞ
H ðx1 ; tÞEH ðx2 ; t ÞEH ðx2 ; t ÞEH ðx1 ; tÞj c i:
then for L ! 1,
 approaches ðk0
k
k0 Þ. Thus,
(A9)
emission of the photon pairs is possible only for the directions
for which the condition k0  k þ k0 is met. Finally, one has If the detectors are very far away from each other, and from
the crystal, then the photon field reaching them can be treated
XZ Z 0 as free evolving. We put into Eq. (A9) the photon state
H NL  d3 k d3 k0
ðk0
k
k0 ÞAeff
;0 e

ið!0
!
! Þt

;0
Eq. (A8). Let t ¼ t1 and t0 ¼ t2 , and j c i ¼ j c ðt ¼ 1Þi,
then Eq. (A9) can be written down as
ay ðk;Þay ðk0 ;0 Þ þ H:c:; (A6)
pðx1 ; tjx2 ; t0 Þ ’ h c jEð
Þ ð
Þ 0
H ðx1 ; tÞEH ðx2 ; t Þ
P ð2Þ
;0 ¼
where Aeff ^ j ðk; Þ^ k ðk0 ; 0 Þ
j;k E0 3jk  is the effective EðþÞ ðþÞ 0
H ðx1 ; tÞEH ðx2 ; t Þj c i: (A10)
strength of the laser-crystal coupling. Henceforth, we shall
;0 by Fo ðk0 Þ.
replace the symbol Aeff To simplify the description, we replace the annihilation
The state of photons emitted in the SPDC process–The and creation operators, which were used above, with new
pump-crystal coupling is weak. The evolution of the state operators ai ð!Þ and their conjugates, which describe ‘‘uni-
jD ðtÞi (in the interaction picture) is given by directional’’ excitations of the photon field [i.e., we assume
iℏðd=dtÞjD ðtÞi ¼ H NL ðtÞjD ðtÞi. In the first order in the that the detectors see only the photons of a specified direction
perturbation expansion of propagation, a good assumption if the detectors are far
from the crystal, and the apertures are narrow, see Fearn and
1 Z t NL 0 Loudon (1987)]. The index i defines the direction (fixed) of
jD ðtÞi ’ jD ðt0 Þi þ H ðt ÞjD ðt0 Þidt0 : (A7)
iℏ t0 the wave vector. The new operators satisfy commutation
relations, which are a modification of those given above to
Put t0 ¼
1, and take the vacuum state (no photons) ji as the current specific case ½ai ð!Þ; ayj ð!0 Þ ¼ ij ð!
!0 Þ,
the initial state jð0Þi. Only in the term with the integral can ½ai ð!Þ; aj ð!0 Þ ¼ 0. If we choose just two propagation
one find creation of pairs of photons.
R For t ! 1 it contains an
0 it0 ð!þ!0
!0 Þ
integral of the following form: þ1
1 dt e , which is 72
2ð! þ !0
!0 Þ. Thus, the two-photon component of the One should add here a note that in reality this relation is not
state, at t ¼ 1, is effectively given by absolutely sharp. The molecular polarization was treated here
phenomenologically. Still, once a more refined model is used, the
relationship is sharp enough, so that the deviations from perfect
71
Since an arbitrary electromagnetic field is a superposition of the equality are beyond the resolution of the present measuring setups.
plane waves, starting with this trivial case it is very easy to get the 73
If one has ! ’ !0 , then we have a frequency degenerate SPDC,
general description. and if k^ ’ k^0 , then we have a colinear one.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

830 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

directions that fulfill the phase-matching conditions, then The probability of detection of two photons at the moments
effectively one can put t and t0 reads
Z
EðþÞ ðx ; tÞ ¼ d!e
i!t fi ð!Þai ð!Þ (A11) pðx1 ; tjx2 ; t0 Þ
H i

 
2
Z
 0 Þ 2
2½ð!
!Þ2 =2  

  d!e
i!ðt
t C e 

2 0 2
e
ð =2Þðt
t Þ :
with i ¼ 1, 2, and where f1 and f2 are the frequency response 

c


functions of the filter-detector system. (A18)
We assume that the maxima of the functions agree with the
frequencies given by the phase-matching conditions. As  ! 1, Eq. (A18) approaches ðt
t0 Þ. We have a
P
Introducing a unit operator I^ ¼ 1 i¼0 jb i ihb i j, where jbi i is perfect time correlation. For a realistic case of final band-
a basis states, into Eq. (A10), we obtain widths, the degree of time correlation of the detection of the
SPDC photons depends entirely on the frequency response of
pðx1 ; tjx2 ; t0 Þ ’ h c jEð
Þ ð
Þ 0 ^ ðþÞ
H ðx1 ; tÞEH ðx2 ; t ÞIEH ðx1 ; tÞ the detectors (plus interference filters, if any, in front of
EðþÞ 0
H ðx2 ; t Þj c i: (A12) them).
The output state of pulsed pumped SPDC–Since the pump
Since EðþÞ
H contains only the annihilation operators, they pulse is a superposition of monochromatic waves, the output
transform the two-photon state ji into the vacuum state. state for this case is an integral of the monochromatic
Thus, Eq. (A12) can be put as (Mollow, 1973) case over the momentum profile of the pulse: j c pulse i ¼
R 3
d k0 j c ðFo ðk0 ÞÞi, where j c ðFo ðk0 ÞÞi is the state for the
pðx1 ; tjx2 ; t0 Þ ’ h c jEð
Þ 0ð
Þ ðþÞ 0ðþÞ
H EH jihjEH EH j c i; monochromatic case with wave vector k0 and field amplitude
(A13) Fo ðk0 Þ. Since the frequency of the pulse and the wave vector
are not strictly defined, if the pulse is too short the SPDC
where the primed expressions pertain to the moment of time t0
photons are less tightly correlated directionally.
and the position x2 . Thus, we have pðx1 ; tjx2 ; t0 Þ ’
The two-photon state produced by a SPDC process is given
jA12 ðt; t0 Þj2 , where A12 ðt; t0 Þ ¼ hjEðþÞ 0ðþÞ 0
H ðx1 ; tÞEH ðx2 ; t Þj c i. by (in the unidirectional approximation)
With the use of the new creation operators, the state ji
can be approximated by Z Z Z
ji ¼ d!0 Fo ð!0 Þ d!1 d!2 ð!0
!1
!2 Þ
Z Z
ji þ d!1 d!2 Fo ð!
!1
!2 Þay1 ð!1 Þay2 ð!2 Þji:
ay1 ð!1 Þay2 ð!2 Þji; (A19)
(A14) where we have replaced the effective pump amplitude by the
Therefore, one gets the following formula for the detection spectral decomposition of the laser pulse Fo ð!0 Þ.
amplitude: Two-photon detection amplitude: the pulsed-pump case–If
we have a pulsed pump, we have to integrate the amplitude
A12 ðt;t0 Þ (A16) over the frequency content of the pump [just like it is in
Z 0 0
Z the case of the state (A19)]:
¼ hj d!0 e
i! t f2 ð!0 Þa2 ð!0 Þ d!e
i!t f1 ð!Þa1 ð!Þ
Z 0
Z Z Aðt; t0 Þ ¼ d!o Fo ð!o Þe
!o t
d!1 d!2 Fo ð!0
!1
!2 Þay2 ð!2 Þay1 ð!1 Þji:
Z 0
d!e
i!ðt
t Þ f2 ð!o
!Þf1 ð!Þ
(A15)
Z
Since the creation and annihilation operators for different ¼ dtp Fo ðtp Þf1 ðt
tp Þf2 ðt0
tp Þ; (A20)
modes commute, and since one can use
0 y 0
hjai ð! Þaj ð!Þji ¼ ij ð!
!Þ, we get where Fo ðtÞ is the Fourier transformation (time profile) of
Z Fo ð!Þ. Namely, the time correlation of the detections is
0 0
A12 ðt; t0 Þ ¼ Fo e
i!0 t d!e
i!ðt
t Þ f2 ð!0
!Þf1 ð!Þ; defined by the resolution of the respective filters, while the
events happen at times dictated by the pulse. This is clearly
(A16) visible in the case of no filters and broad band radiation.
and we have The (unphysical) limiting case is reached by replacing fs
by ðt
tp Þ and ðt0
tp Þ. This gives FðtÞðt0
tÞ.
pðx1 ; tjx2 ; t0 Þ  jA12 ðt; t0 Þj2

 

Z
 
 2
’
 d!e
i!ðt
t0 Þ
f ð!
!Þf ð!Þ

 ; REFERENCES
 2 0 1 
(A17) Aerts, S., P. Kwiat, J.-A. Larsson, and M. Zukowski, 1999, Phys.
Rev. Lett. 83, 2872.
i.e., the probability depends on the difference of the detection Akopian, N., N. Lindner, E. Poem, Y. Berlatzky, J. Avron, D.
times. Gershoni, B. Gerardot, and P. Petroff, 2006, Phys. Rev. Lett. 96,
For instance, assume that f1 ¼ f2 ¼ f, and that they are 130501.
Gaussian, fð!Þ ¼ Ce
ð!c
!Þ = , with the central frequency
2 2
Alber, G., T. Beth, M. Horodecki, P. Horodecki, R. Horodecki, M.
!c ¼ !0 =2. Then we have f1 ð!Þ ¼ f2 ð!0
!Þ ¼ fð!Þ. Rotteler, H. Weinfurter, R. Werner, and A. Zeilinger, 2001,

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 831

Quantum Information: An Introduction to Basic Theoretical Bennett, C., G. Brassard, S. Popescu, B. Schumacher, J. Smolin, and
Concepts and Experiments (Springer-Verlag, Berlin). W. Wootters, 1996, Phys. Rev. Lett. 76, 722.
Altepeter, J., E. Jeffrey, and P. Kwiat, 2005, Opt. Express 13, 8951. Bennett, C., and D. DiVincenzo, 2000, Nature (London) 404, 247.
Amselem, E., M. Radmark, M. Bourennane, and A. Cabello, 2009, Bennett, C., D. DiVincenzo, J. Smolin, and W. Wootters, 1996,
Phys. Rev. Lett. 103, 160405. Phys. Rev. A 54, 3824.
Ansmann, M., H. Wang, Radoslaw, C. Bialczak, M. Hofheinz, E. Bennett, C., and S. Wiesner, 1992, Phys. Rev. Lett. 69, 2881.
Lucero, M. Neeley, A. D. O’Connell, D. Sank, M. Weides, J. Bennett, C. H., 1992, Phys. Rev. Lett., 68, 3121.
Wenner, A. N. Cleland, and J. M. Martinis, 2009, Nature (London) Bernstein, H., D. M. Greenberger, M. A. Horne, and A. Zeilinger,
461, 504. 1993, Phys. Rev. A 47, 78.
Aoki, T., G. Takahashi, T. Kajiya, J. Yoshikawa, S. L. Braunstein, P. Beugnon, J., M. P. A. Jones, J. Dingjan, B. Darquie, G. Messin, A.
van Loock, and A. Furusawa, 2009, Nature Phys. 5, 541. Browaeys, and P. Grangier, 2006, Nature (London) 440, 779.
Ardehali, M., 1992, Phys. Rev. A 46, 5375. Bialynicki-Birula, I., and Z. Bialynicka-Birula, 1975, Quantum
Aschauer, H., and H. Briegel, 2002, Phys. Rev. Lett. 88, 047902. Electrodynamics (Pergamon, New York).
Aspect, A., 1999, Nature (London) 398, 189. Blinov, B., D. Moehring, L.-M. Duan, and C. Monroe, 2004, Nature
Aspect, A., J. Dalibard, and G. Roger, 1982, Phys. Rev. Lett. 49, (London) 428, 153.
1804. Bodiya, T., and L.-M. Duan, 2006, Phys. Rev. Lett. 97, 143601.
Aspect, A., P. Grangier, and G. Roger, 1981, Phys. Rev. Lett. 47, 460. Bohm, D., 1951, Quantum Theory (Prentice-Hall, Englewood Cliffs,
Aspect, A., P. Grangier, and G. Roger, 1982, Phys. Rev. Lett. 49, 91. NJ).
Aspelmeyer, M., H. R. Böhm, T. Gyatso, T. Jennewein, R. Boschi, D., S. Branca, F. D. Martini, and L. Hardy, 1997, Phys. Rev.
Kaltenbaek, M. Lindenthal, G. Molina-Terriza, A. Poppe, K. Lett. 79, 2755.
Resch, M. Taraba, R. Ursin, P. Walther, and A. Zeilinger, 2003, Boschi, D., S. Branca, F. D. Martini, L. Hardy, and S. Popescu,
Science 301, 621. 1998, Phys. Rev. Lett. 80, 1121.
Aspelmeyer, M., H. R. Bohm, T. Gyatso, T. Jennewein, R. Bose, S., P. Knight, M. Plenio, and V. Vedral, 1999, Phys. Rev. Lett.
Kaltenbaek, M. Lindenthal, G. Molina-Terriza, A. Poppe, K. 83, 5158.
Resch, M. Taraba, R. Ursin, P. Walther, and A. Zeilinger, 2003, Bose, S., V. Vedral, and P. Knight, 1998, Phys. Rev. A 57, 822.
Science 301, 621. Boto, A., P. Kok, D. Abrams, S. Braunstein, C. Williams, and J.
Aspelmeyer, M., T. Jennewein, M. Pfenningbauer, W. Leeb, and A. Dowling, 2000, Phys. Rev. Lett. 85, 2733.
Zeilinger, 2003, IEEE J. Sel. Top. Quantum Electron. 9, 1541. Bourennane, M., M. Eibl, S. Gaertner, C. Kurtsiefer, A. Cabello, and
Bajcsy, M., A. Zibrov, and M. Lukin, 2003, Nature (London) 426, H. Weinfurter, 2004, Phys. Rev. Lett. 92, 107901.
638. Bourennane, M., M. Eibl, C. Kurtsiefer, S. Gaertner, H. Weinfurter,
Bao, X.-H., T.-Y. Chen, Q. Zhang, J. Yang, H. Zhang, T. Yang, and O. Guhne, P. Hyllus, D. Bru, M. Lewenstein, and A. Sanpera,
J.-W. Pan, 2007, Phys. Rev. Lett. 98, 170502. 2004, Phys. Rev. Lett. 92, 087902.
Bao, X.-H., Y. Qian, J. Yang, H. Zhang, Z.-B. Chen, T. Yang, and Bouwmeester, D., A. Ekert, and A. Zeilinger, 2001, Eds., The
J.-W. Pan, 2008, Phys. Rev. Lett. 101, 190501. Physics of Quantum Information, (Springer-Verlag, Berlin).
Barbieri, M., C. Cinelli, P. Mataloni, and F. D. Martini, 2005, Phys. Bouwmeester, D., 2004, Nature (London) 429, 139.
Rev. A 72, 052110. Bouwmeester, D., J.-W. Pan, M. Daniell, H. Weinfurter, and A.
Barbieri, M., T. J. Weinhold, B. Lanyon, A. Gilchrist, K. Resch, M. Zeilinger, 1999, Phys. Rev. Lett. 82, 1345.
Almeida, and A. White, 2009, J. Mod. Opt. 56, 209. Bouwmeester, D., J.-W. Pan, M. Daniell, H. Weinfurter, M.
Barnett, S., J. Jeffers, A. Gatti, and R. Loudon, 1998, Phys. Rev. A Zukowski, and A. Zeilinger, 1998, Nature (London) 394, 841.
57, 2134. Bouwmeester, D., J.-W. Pan, K. Mattle, M. Eibl, H. Weinfurter, and
Barreiro, J., N. Langford, N. Peters, and P. Kwiat, 2005, Phys. Rev. A. Zeilinger, 1997, Nature (London) 390, 575.
Lett. 95, 260501. Bouwmeester, D., J.-W. Pan, H. Weinfurter, and A. Zeilinger, 2000,
Barreiro, J., T.-C. Wei, and P. Kwiat, 2008, Nature Phys. 4, 282. J. Mod. Opt. 47, 279.
Barrett, M., J. Chiaverini, T. Schaetz, J. Britton, W. M. Itano, J. D. Braig, C., P. Zarda, C. Kurtsiefer, and H. Weinfurter, 2003, Appl.
Jost, E. Knill, C. Langer, D. Leibfried, R. Ozeri, and D. J. Phys. B 76, 113.
Wineland, 2004, Nature (London) 429, 737. Branciard, C., A. Ling, N. Gisin, C. Kurtsiefer, A. Lamas-Linares,
Barrett, S., and P. Kok, 2005, Phys. Rev. A 71, 060310(R). and V. Scarani, 2007, Phys. Rev. Lett. 99, 210407.
Barz, S., G. Cronenberg, A. Zeilinger, and P. Walther, 2010, Nature Branning, D., W. Grice, R. Erdmann, and I. Walmsley, 1999, Phys.
Photon. 4, 553. Rev. Lett. 83, 955.
Belinskii, A., and D. Klyshko, 1993, Phys. Usp. 36, 653. Branning, D., W. Grice, R. Erdmann, and I. Walmsley, 2000, Phys.
Bell, J., 1964, Physics 1, 195, http://www.drchinese.com/David/ Rev. A 62, 013814.
Bell_Compact.pdf. Braunstein, S., and H. Kimble, 1998, Nature (London) 394, 840.
Bell, J., 1987, Speakable and Unspeakable in Quantum Mechanics Braunstein, S., and A. Mann, 1995, Phys. Rev. A 51, R1727.
(Cambridge University Press, Cambridge). Braunstein, S., and P. van Loock, 2005, Rev. Mod. Phys.
Bell, J. S., 1966, Rev. Mod. Phys. 38, 447. 77, 513.
Bennett, A., D. Unitt, P. Atkinson, D. Ritchie, and A. Shields, 2005, Brendel, J., N. Gisin, W. Tittel, and H. Zbinden, 1999, Phys. Rev.
Opt. Express 13, 50. Lett. 82, 2594.
Bennett, C., H. Bernstein, S. Popescu, and B. Schumacher, 1996, Brendel, J., E. Mohler, and W. Martienssen, 1992, Europhys. Lett.
Phys. Rev. A 53, 2046. 20, 575.
Bennett, C., and G. Brassard, 1984, Proceedings of the IEEE Briegel, H., and R. Raussendorf, 2001, Phys. Rev. Lett. 86, 910.
International Conference on Computers and Systems and Singal Briegel, H.-J., D. E. Browne, W. Dur, R. Raussendorf, and M. V. den
Proceeding and Bangalore and India (IEEE and New York), 175. Nest, 2009, Nature Phys. 5, 19.
Bennett, C., G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. Briegel, H.-J., W. Dür, J. Cirac, and P. Zoller, 1998, Phys. Rev. Lett.
Wootters, 1993, Phys. Rev. Lett. 70, 1895. 81, 5932.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

832 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Brouri, R., A. Beveratos, J.-P. Poizat, and P. Grangier, 2000, Opt. Coffman, V., J. Kundu, and W. K. Wootters, 2000, Phys. Rev. A 61,
Lett. 25, 1294. 052306.
Browne, D., M. Plenio, and S. Huelga, 2003, Phys. Rev. Lett. 91, B. Darquié, M. J., J. Dingjan, J. Beugnon, S. Bergamini, Y.
067901. Sortais, G. Messin, A. Browaeys, and P. Grangier, 2005,
Browne, D., and T. Rudolph, 2005, Phys. Rev. Lett. 95, 010501. Science 309, 454.
Brunel, C., B. Lounis, P. Tamarat, and M. Orrit, 1999, Phys. Rev. Dawson, C., H. Haselgrove, and M. Nielsen, 2006, Phys. Rev. A 73,
Lett. 83, 2722. 052306.
Bruß, D., J. Cirac, P. Horodecki, F. Hulpke, B. Kraus, M. Deutsch, D., A. Ekert, R. Jozsa, C. Macchiavello, S. Popescu, and
Lewenstein, and A. Sanpera, 2002, J. Mod. Opt. 49, 1399. A. Sanpera, 1996, Phys. Rev. Lett. 77, 2818.
Burnham, D., and D. Weinberg, 1970, Phys. Rev. Lett. 25, 84. Dewdney, C., P. Holland, and A. Kyprianidis, 1987, J. Phys. A 20,
Cabello, A., 2001a, Phys. Rev. Lett. 86, 1911. 4717.
Cabello, A., 2001b, Phys. Rev. Lett. 87, 010403. Diedrich, F., and H. Walther, 1987, Phys. Rev. Lett. 58, 203.
Cabello, A., 2003, Phys. Rev. Lett. 90, 258902. Di Giuseppe, G., L. Haiberger, F. De Martini, and A. V. Sergienko,
Cabello, A., 2009, Phys. Rev. Lett. 102, 040401. 1997, Phys. Rev. A 56, R21.
Cabello, A., and E. Santos, 1996, Phys. Lett. A 214, 316. Dirac, P., 1927, Proc. R. Soc. A 114, 243.
Cabrillo, C., J. Cirac, P. Garca-Fernandez, and P. Zoller, 1999, Phys. Divochiy, A., F. Marsili, D. Bitauld, A. Gaggero, R. Leoni, F.
Rev. A 59, 1025. Mattioli, A. Korneev, V. Seleznev, N. Kaurova, O. Minaeva, G.
Carmichael, H., and D. Walls, 1976a, J. Phys. B 9, L43. Gol’tsman, K. G. Lagoudakis, M. Benkhaoul, F. Lévy, and A.
Carmichael, H., and D. Walls, 1976b, J. Phys. B 9, 1199. Fiore, 2008, Nature Photon. 2, 302.
Carteret, H., A. Higuchi, and A. Sudbery, 2000, J. Math. Phys. Dousse, A., J. Suffczynski, A. Beveratos, O. Krebs, A. Lematre, I.
(N.Y.) 41, 7932. Sagnes, J. Bloch, P. Voisin, and P. Senellart, 2010, Nature
Cerf, N., C. Adami, and P. Kwiat, 1998, Phys. Rev. A 57, R1477. (London) 466, 217.
Chanelière, T., D. Matsukevich, S. Jenkins, S.-Y. Lan, T. Kennedy, Dowling, J., 1998, Phys. Rev. A 57, 4736.
and A. Kuzmich, 2005, Nature (London) 438, 833. Duan, L.-M., J. Cirac, and P. Zoller, 2002, Phys. Rev. A 66, 023818.
Chanelière, T., D. N. Matsukevich, S. D. Jenkins, S.-Y. Lan, R. Duan, L.-M., M. Lukin, J. Cirac, and P. Zoller, 2001, Nature
Zhao, T. A. B. Kennedy, and A. Kuzmich, 2007, Phys. Rev. (London) 414, 413.
Lett. 98, 113602. Duan, L.-M., and R. Raussendorf, 2005, Phys. Rev. Lett. 95,
Chen, K., C.-M. Li, Q. Zhang, Y.-A. Chen, A. Goebel, S. Chen, A. 080503.
Mair, and J.-W. Pan, 2007, Phys. Rev. Lett. 99, 120503. Dür, W., and H.-J. Briegel, 2003, Phys. Rev. Lett. 90, 067901.
Chen, Q., J. Cheng, K.-L. Wang, and J. Du, 2006, Phys. Rev. A 73, Dür, W., H.-J. Briegel, J. Cirac, and P. Zoller, 1999, Phys. Rev. A 59,
012303. 169.
Chen, S., Y.-A. Chen, T. Strassel, Z.-S. Yuan, B. Zhao, J. Eberhard, P., 1993, Phys. Rev. A 47, R747.
Schmiedmayer, and J.-W. Pan, 2006, Phys. Rev. Lett. 97, Edamatsu, K., R. Shimizu, and T. Itoh, 2002, Phys. Rev. Lett. 89,
173004. 213601.
Chen, Z.-B., J.-W. Pan, Y.-D. Zhang, C. Brukner, and A. Zeilinger, Eibl, M., S. Gaertner, M. Bourennane, C. Kurtsiefer, M. Zukowski,
2003, Phys. Rev. Lett. 90, 160408. and H. Weinfurter, 2003, Phys. Rev. Lett. 90, 200403.
Chen, Z.-B., Q. Zhang, X.-H. Bao, J. Schmiedmayer, and J.-W. Pan, Einstein, A., 1905, Ann. Phys. (Leipzig) 322, 132.
2006, Phys. Rev. A 73, 050302. Einstein, A., B. Podolsky, and N. Rosen, 1935, Phys. Rev. 47, 777.
Chen, Z.-B., B. Zhao, Y.-A. Chen, J. Schmiedmayer, and J.-W. Pan, Eisaman, M., F. M. A. André, M. Fleischhauer, A. Zibrov, and M.
2007, Phys. Rev. A 76, 022329. Lukin, 2005, Nature (London) 438, 837.
Chen, Z.-B., B. Zhao, and J.-W. Pan, 2008 (unpublished). Eisaman, M., L. Childress, F. M. A. André, A. Zibrov, and M. Lukin,
Chiaverini, J., et al., 2004, Nature (London) 432, 602. 2004, Phys. Rev. Lett. 93, 233602.
Childress, L., J. Taylor, A. Srensen, and M. Lukin, 2005, Phys. Rev. Eisenberg, H., G. Khoury, G. Durkin, C. Simon, and D.
A 72, 052330. Bouwmeester, 2004, Phys. Rev. Lett. 93, 193901.
Childress, L., J. Taylor, A. Srensen, and M. Lukin, 2006, Phys. Rev. Ekert, A., 1991, Phys. Rev. Lett. 67, 661.
Lett. 96, 070504. Elitzur, A., and L. Vaidman, 1993, Found. Phys. 23, 987.
Choi, K., H. Deng, J. Laurat, and H. Kimble, 2008, Nature (London) Englert, B.-G., C. Kurtsiefer, and H. Weinfurter, 2001, Phys. Rev. A
452, 67. 63, 032303.
Chou, C., J. Laurat, H. Deng, K. Choi, H. de Riedmatten, D. Felinto, Englund, D., A. Faraon, B. Zhang, Y. Yamamoto, and J. Vučković,
and H. Kimble, 2007, Science 316, 1316. 2007, Opt. Express 15, 5550.
Chou, C., S. Polyakov, A. Kuzmich, and H. Kimble, 2004, Phys. Fattal, D., K. Inoue, J. Vuckovic, C. Santori, G. S. Solomon, and Y.
Rev. Lett. 92, 213601. Yamamoto, 2004, Phys. Rev. Lett. 92, 037903.
Cinelli, C., M. Barbieri, R. Perris, P. Mataloni, and F. D. Martini, Fearn, H., and R. Loudon, 1987, Opt. Commun. 64, 485.
2005, Phys. Rev. Lett. 95, 240405. Fedrizzi, A., T. Herbst, A. Poppe, T. Jennewein, and A. Zeilinger,
Cinelli, C., G. D. Nepi, F. D. Martini, M. Barbieri, and P. Mataloni, 2007, Opt. Express 15, 15377.
2004, Phys. Rev. A 70, 022321. Fedrizzi, A., R. Ursin, T. Herbst, M. Nespoli, R. Prevedel, T.
Cirac, J., and P. Zoller, 1994, Phys. Rev. A 50, R2799. Scheidl, F. Tiefenbacher, T. Jennewein, and A. Zeilinger, 2009,
Cirac, J., P. Zoller, H. Kimble, and H. Mabuchi, 1997, Phys. Rev. Nature Phys. 5, 389.
Lett. 78, 3221. Felinto, D., C. Chou, J. Laurat, E. Schomburg, H. de Riedmatten,
Cirel’son, B., 1980, Lett. Math. Phys. 4, 93. and H. Kimble, 2006, Nature Phys. 2, 844.
Clauser, J., and M. Horne, 1974, Phys. Rev. D 10, 526. Feng, X.-L., Z.-M. Zhang, X.-D. Li, S.-Q. Gong, and Z.-Z. Xu,
Clauser, J., and A. Shimony, 1978, Rep. Prog. Phys. 41, 1881. 2003, Phys. Rev. Lett. 90, 217902.
Clauser, J. F., M. A. Horne, A. Shimony, and R. A. Holt, 1969, Phys. Feynman, R., R. Leighton, and M. Sands, 1963, The Feynman
Rev. Lett. 23, 880. Lectures on Physics III (Addison-Wesley, Reading, MA).

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 833

Fiorentino, M., T. Kim, and F. Wong, 2005, Phys. Rev. A 72, Gühne, O., C. Lu, W. Gao, and J. Pan, 2007, Phys. Rev. A 76,
012318. 030305.
Fiorentino, M., and F. N. C. Wong, 2004, Phys. Rev. Lett. 93, Gühne, O., and G. Toth, 2009, Phys. Rep. 474, 1.
070502. Hadfield, R. H., 2009, Nature Photon. 3, 696.
Fleischhauer, M., A. Imamoglu, and J. Marangos, 2005, Rev. Mod. Halder, M., A. Beveratos, N. Gisin, V. Scarani, C. Simon, and H.
Phys. 77, 633. Zbinden, 2007, Nature Phys. 3, 692.
Fleischhauer, M., and M. Lukin, 2000, Phys. Rev. Lett. 84, 5094. Halder, M., J. Fulconis, B. Cemlyn, A. Clark, C. Xiong, W. J.
Fleischhauer, M., and M. Lukin, 2002, Phys. Rev. A 65, 022314. Wadsworth, and J. G. Rarity, 2009, Opt. Express 17, 4670.
Franson, J., 1989, Phys. Rev. Lett. 62, 2205. Hammerer, K., A. Sorensen, and E. Polzik, 2010, Rev. Mod. Phys.
Freedman, S., and J. Clauser, 1972, Phys. Rev. Lett. 28, 938. 82, 1041.
Fulconis, J., O. Alibart, J. L. O’Brien, W. J. Wadsworth, and J. G. Han, Y.-J., R. Raussendorf, and L.-M. Duan, 2007, Phys. Rev. Lett.
Rarity, 2007, Phys. Rev. Lett. 99, 120501. 98, 150404.
Furusawa, A., J. Srensen, S. Braunstein, C. Fuchs, H. Kimble, and Hanbury Brown, R. and R. Q. Twiss, 1956, Nature (London) 177,
E. Polzik, 1998, Science 282, 706. 27.
Gao, W.-B., C.-Y. Lu, X.-C. Yao, P. Xu, O. Guhne, Y.-A. Hardy, L., 1992, Phys. Rev. Lett. 68, 2981.
Chen, C.-Z. Peng, Z.-B. Chen, and J.-W. Pan, 2010, Nature Hardy, L., 1993, Phys. Rev. Lett. 71, 1665.
Phys. 6, 331. Haroche, S., 1995, Ann. N.Y. Acad. Sci. 755, 73.
Gao, W.-B., X.-C. Yao, P. Xu, O. Guhne, C.-Y. Lu, C.-Z. Peng, Z.-B. Hayashi, A., T. Hashimoto, and M. Horibe, 2005, Phys. Rev. A 72,
Chen, and J.-W. Pan, 2010, Phys. Rev. Lett. 104, 020501. 032325.
Garay-Palmett, K., H. J. McGuinness, O. Cohen, J. S. Lundeen, R. Herzog, T. J., J. G. Rarity, H. Weinfurter, and A. Zeilinger, 1994,
Rangel-Rojo, A. B. U’ren, M. G. Raymer, C. J. McKinstrie, S. Phys. Rev. Lett. 72, 629.
Radic, and I. A. Walmsley, 2007, Opt. Express 15, 14870. Hétet, G., B. Buchler, O. Glöckl, M. Hsu, A. Akulshin, H.-A.
Garg, A., and N. Mermin, 1987, Phys. Rev. D 35, 3831. Bachor, and P. Lam, 2008, Opt. Express 16, 7369.
Gasparoni, S., J.-W. Pan, P. Walther, T. Rudolph, and A. Zeilinger, Hill, S. and W. K. Wootters, 1997, Phys. Rev. Lett. 78, 5022.
2004, Phys. Rev. Lett. 93, 020504. Hofmann, H., and S. Takeuchi, 2002, Phys. Rev. A 66, 024308.
Genovese, M., 2005, Phys. Rep. 413, 319. Holevo, A., 1973, Probl. Inf. Transm. 9, 177, http://www
Genovese, M., and C. Novero, 2002, Eur. Phys. J. D 21, 109. .mathnet.ru/php/archive.phtml?wshow=paper&jrnid;=ppi&paperid;=
Gilchrist, A., A. Hayes, and T. Ralph, 2007, Phys. Rev. A 75, 903&option_lang=eng.
052328. Holman, K., D. D. Hudson, J. Ye, and D. Jones, 2005, Opt. Lett. 30,
Gill, R., G. Weihs, A. Zeilinger, and M. Zukowski, 2002, Proc. Natl. 1225.
Acad. Sci. U.S.A. 99, 14632. Hong, C., and L. Mandel, 1985, Phys. Rev. A 31, 2409.
Gisin, N., 1996, Phys. Lett. A 210, 151. Hong, C., Z. Ou, and L. Mandel, 1987, Phys. Rev. Lett. 59,
Gisin, N., and H. Bechmann-Pasquinucci, 1998, Phys. Lett. A 246, 2044.
1. Honjo, T., H. Takesue, H. Kamada, Y. Nishida, O. Tadanaga, M.
Gisin, N., and A. Peres, 1992, Phys. Lett. A 162, 15. Asobe, and K. Inoue, 2007, Opt. Express 15, 13957.
Gisin, N., G. Ribordy, W. Tittel, and H. Zbinden, 2002, Rev. Mod. Horne, M., A. Shimony, and A. Zeilinger, 1989, Phys. Rev. Lett. 62,
Phys. 74, 145. 2209.
Glauber, R., 1963, Phys. Rev. 130, 2529. Horne, M., A. Shimony, and A. Zeilinger, 1990, Nature (London)
Goebel, A., C. Wagenknecht, Q. Zhang, Y.-A. Chen, K. Chen, J. 347, 429.
Schmiedmayer, and J.-W. Pan, 2008, Phys. Rev. Lett. 101, Horne, M., and A. Zeilinger, 1986, Ann. N.Y. Acad. Sci. 480, 469.
080403. Horne, M. A., and A. Zeilinger, 1985, Symposium on the
Gong, Y.-X., X.-B. Zou, T. C. Ralph, S.-N. Zhu, and G.-C. Guo, Foundations of Modern Physics and Joensuu and P. Lahti and
2010, Phys. Rev. A 81, 052303. P. Mittelstaedt (World Scientific, Singapore).
Gottesman, D., and I. Chuang, 1999, Nature (London) 402, 390. Horodecki, R., P. Horodecki, and M. Horodecki, 1995, Phys. Lett. A
Grangier, P., G. Roger, and A. Aspect, 1986, Europhys. Lett. 1, 173. 200, 340.
Grassl, M., Beth, Th., and Pellizzari, T., 1997, Phys. Rev. A 56, 33. Horodecki, R., P. Horodecki, and M. Horodecki, 1996, Phys. Lett. A
Greenberger, D., M. Horne, A. Shimony, and A. Zeilinger, 1990, 223, 1.
Am. J. Phys. 58, 1131. Horodecki, R., P. Horodecki, and M. Horodecki, 1998, Phys. Rev.
Greenberger, D., M. Horne, and A. Zeilinger, 1989, Bell’s Theorem Lett. 80, 5239.
and Quantum Theory and Conceptions of the Universe (Kluwer Horodecki, R., P. Horodecki, M. Horodecki, and K. Horodecki,
Academic, Dordrecht). 2009, Rev. Mod. Phys. 81, 865.
Greenberger, D., M. Horne, and A. Zeilinger, 1993, Phys. Today 46 Howell, J., A. Lamas-Linares, and D. Bouwmeester, 2002, Phys.
(8), 22. Rev. Lett. 88, 030401.
Grice, W., R. Erdmann, I. Walmsley, and D. Branning, 1998, Phys. Hübel, H., M. Vanner, T. Lederer, B. Blauensteiner, T. Lorunser, A.
Rev. A 57, R2289. Poppe, and A. Zeilinger, 2007, Opt. Express 15, 7853.
Grice, W., A. U’Ren, and I. Walmsley, 2001, Phys. Rev. A 64, Hwang, W., 2003, Phys. Rev. Lett. 91, 057901.
063815. Irvine, M. W. T., J. F. Hodelin, C. Simon, and D. Bouwmeester,
Grice, W., and I. Walmsley, 1997, Phys. Rev. A 56, 1627. 2005, Phys. Rev. Lett. 95, 030401.
Gröblacher, S., T. Jennewein, A. Vaziri, G. Weihs, and A. Zeilinger, Jacobson, J., G. Björk, I. Chuang, and Y. Yamamoto, 1995, Phys.
2006, New J. Phys. 8, 75. Rev. Lett. 74, 4835.
Gröblacher, S., T. Paterek, R. Kaltenbaek, Č. Brukner, M. James, D. F. V., P. G. Kwiat, W. J. Munro, and A. G. White, 2001,
Żukowski, M. Aspelmeyer, and A. Zeilinger, 2007, Nature Phys. Rev. A 64, 052312.
(London) 446, 871. Jennewein, T., U. Achleitner, C. Simon, G. Weihs, H. Weinfurter,
Grover, L., 1997, Phys. Rev. Lett. 79, 325. and A. Zeilinger, 2000, Rev. Sci. Instrum. 71, 1675.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

834 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Jennewein, T., G. Weihs, J.-W. Pan, and A. Zeilinger, 2001, Phys. Kozhekin, A., K. Mølmer, and E. Polzik, 2000, Phys. Rev. A 62,
Rev. Lett. 88, 017903. 033809.
Jiang, L., A. Rey, O. Romero-Isart, J. Garcı́a-Ripoll, A. Sanpera, Krenn, G., and A. Zeilinger, 1996, Phys. Rev. A 54, 1793.
and M. Lukin, 2009, Phys. Rev. A 79, 022309. Kuhn, A., M. Hennrich, and G. Rempe, 2002, Phys. Rev. Lett. 89,
Jiang, L., J. Taylor, and M. Lukin, 2007, Phys. Rev. A 76, 012301. 067901.
Jin, X.-M., J.-G. Ren, B. Yang, Z.-H. Yi, F. Zhou, X.-F. Xu, S.-K. Kurtsiefer, C., S. Mayer, P. Zarda, and H. Weinfurter, 2000, Phys.
Wang, D. Yang, Y.-F. Hu, S. Jiang, T. Yang, H. Yin, K. Chen, C.- Rev. Lett. 85, 290.
Z. Peng, and J.-W. Pan, 2010, Nature Photon. 4, 376. Kurtsiefer, C., P. Zarda, M. Halder, H. Weinfurter, P. Gorman, P.
Julsgaard, B., J. Sherson, J. Cirac, J. Fiurášek, and E. Polzik, 2004, Tapster, and J. Rarity, 2002, Nature (London) 419, 450.
Nature (London) 432, 482. Kuzucu, O., and F. N. C. Wong, 2008, Phys. Rev. A 77, 032314.
Kaltenbaek, R., B. Blauensteiner, M. Żukowski, M. Aspelmeyer, Kwiat, P., 1997, J. Mod. Opt. 44, 2173.
and A. Zeilinger, 2006, Phys. Rev. Lett. 96, 240502. Kwiat, P., S. Barraza-Lopez, A. Stefanov, and N. Gisin, 2001,
Kaltenbaek, R., R. Prevedel, M. Aspelmeyer, and A. Zeilinger, Nature (London) 409, 1014.
2009, Phys. Rev. A 79, 040302(R). Kwiat, P., A. Berglund, J. Altepeter, and A. White, 2000, Science
Kardynal, B. E., Z. L. Yuan, and A. J. Shields, 2008, Nature Photon. 290, 498.
2, 425. Kwiat, P., P. Eberhard, A. Steinberg, and R. Chiao, 1994, Phys. Rev.
Karlsson, A., and M. Bourennane, 1998, Phys. Rev. A 58, 4394. A 49, 3209.
Kash, M., V. Sautenkov, A. Zibrov, L. Hollberg, G. Welch, M. Kwiat, P., K. Mattle, H. Weinfurter, A. Zeilinger, A. Sergienko, and
Lukin, Y. Rostovtsev, E. Fry, and M. Scully, 1999, Phys. Rev. Lett. Y. Shih, 1995, Phys. Rev. Lett. 75, 4337.
82, 5229. Kwiat, P., A. Steinberg, and R. Chiao, 1993, Phys. Rev. A 47, R2472.
Kaszlikowski, D., P. Gnaciński, M. Żukowski, W. Miklaszewski, Kwiat, P., and H. Weinfurter, 1998, Phys. Rev. A 58, R2623.
and A. Zeilinger, 2000, Phys. Rev. Lett. 85, 4418. Kwiat, P. G., A. M. Steinberg, and R. Y. Chiao, 1992, Phys. Rev. A
Keller, M., B. Lange, K. Hayasaka, W. Lange, and H. Walther, 2004, 45, 7729.
Nature (London) 431, 1075. Laing, A., A. Peruzzo, A. Politi, M. Verde, M. Halder, T. Ralph,
Keller, T., and M. Rubin, 1997, Phys. Rev. A 56, 1534. M. Thompson, and J. O’Brien, 2010, Appl. Phys. Lett. 97,
Keller, T., M. Rubin, and Y. Shih, 1998, Phys. Lett. A 244, 507. 211109.
Kiesel, N., C. Schmid, G. Toth, E. Solano, and H. Weinfurter, 2007, Laloë, F., 2001, Am. J. Phys. 69, 655.
Phys. Rev. Lett. 98, 063604. Lamas-Linares, A., J. Howell, and D. Bouwmeester, 2001, Nature
Kiesel, N., C. Schmid, U. Weber, G. Tóth, O. Gühne, R. Ursin, and (London) 412, 887.
H. Weinfurter, 2005, Phys. Rev. Lett. 95, 210502. Landau, L., 1987, Phys. Lett. A 120, 54.
Kiesel, N., C. Schmid, U. Weber, R. Ursin, and H. Weinfurter, 2005, Langford, N., T. Weinhold, R. Prevedel, K. Resch, A. Gilchrist, J.
Phys. Rev. Lett. 95, 210505. O’Brien, G. Pryde, and A. White, 2005, Phys. Rev. Lett. 95,
Kim, T., M. Fiorentino, and F. N. C. Wong, 2006, Phys. Rev. A 73, 210504.
012316. Lanyon, B. P., M. Barbieri, M. P. Almeida, T. Jennewein, T. C.
Kim, Y.-H., 2003, Phys. Rev. A 67, 040301(R). Ralph, K. J. Resch, G. J. Pryde, J. L. O’Brien, A. Gilchrist, and
Kim, Y.-H., S. Kulik, and Y. Shih, 2001, Phys. Rev. Lett. 86, 1370. A. G. White, 2009 Nature Phys. 5, 134.
Kimble, H., 2008, Nature (London) 453, 1023. Lanyon, B. P., and N. K. Langford, 2009, New J. Phys. 11,
Kimble, H., M. Dagenais, and L. Mandel, 1977, Phys. Rev. Lett. 39, 013008.
691. Lanyon, B. P., T. J. Weinhold, N. K. Langford, M. Barbieri, D. F. V.
Kimble, H., M. Dagenais, and L. Mandel, 1978, Phys. Rev. A 18, James, A. Gilchrist, and A. G. White, 2007, Phys. Rev. Lett. 99,
201. 250505.
Kitaev, A., 2003, Ann. Phys. (N.Y.) 303, 2. Lanyon, B. P., J. D. Whitfield, G. G. Gillett, M. E. Goggin, M. P.
Klyachko, A. A., M. A. Can, S. Binicioglu, and A. S. Shumovsky, Almeida, I. Kassal, J. D. Biamonte, M. Mohseni, B. J. Powell, M.
2008, Phys. Rev. Lett. 101, 20403. Barbieri, A. Aspuru-Guzik, and A. G. White, 2010, Nature Chem.
Klyshko, D., 1967, JETP Lett. 6, 23. 2, 106.
Klyshko, D., 1988, Photons and Nonlinear Optics (Gordon and Lapkiewicz, R., L. Peizhe, S. Christoph, N. Langford, S. Ramelow,
Breach, New York). M. Wiesniak, and A. Zeilinger, 2011, Nature (London) 474, 490.
Klyshko, D., 1993, Phys. Lett. A 172, 399. Laskowski, W., M. Wieśniak, M. Zukowski, M. Bourennane, and H.
Knill, E., R. Laflamme, and G. Milburn, 2001, Nature (London) Weinfurter, 2009, J. Phys. B 42, 114004.
409, 46. Laurat, J., H. de Riedmatten, D. Felinto, C.-W. Chou, E.
Koashi, M., T. Yamamoto, and N. Imoto, 2001, Phys. Rev. A 63, Schomburg, and H. Kimble, 2006, Opt. Express 14, 6912.
030301. Legero, T., T. Wilk, M. Hennrich, G. Rempe, and A. Kuhn, 2004,
Kochen, S., and E. Specker, 1967, J. Math. Mech. 17, 59. Phys. Rev. Lett. 93, 070503.
Kok, P., A. Boto, D. Abrams, C. Williams, S. L. Braunstein, and J. Leggett, A., 2003, Found. Phys. 33, 1469.
Dowling, 2001, Phys. Rev. A 63, 063407. Leibfried, D., R. Blatt, C. Monroe, and D. Wineland, 2003, Rev.
Kok, P., and S. Braunstein, 2000, Phys. Rev. A 61, 042304. Mod. Phys. 75, 281.
Kok, P., H. Lee, and J. Dowling, 2002, Phys. Rev. A 65, 052104. Lewenstein, M., B. Kraus, J. Cirac, and P. Horodecki, 2000, Phys.
Kok, P., W. Munro, K. Nemoto, T. Ralph, J. Dowling, and G. Rev. A 62, 052310.
Milburn, 2007, Rev. Mod. Phys. 79, 135. Lewis, G., 1926, Nature (London) 118, 874.
Korneev, A., Y. Vachtomin, O. Minaeva, A. Divochiy, K. Smirnov, Lita, A., A. Miller, and S. Nam, 2008, Opt. Express 16, 3032.
O. Okunev, G. Gol’tsman, C. Zinoni, N. Chauvin, L. Balet, F. Liu, C., Z. Dutton, C. Behroozi, and L. Hau, 2001, Nature (London)
Marsili, D. Bitauld, B. Alloing, L. Lianhe, A. Fiore, L. Lunghi, A. 409, 490.
Gerardino, M. Halder, C. Jorel, C., and H. Zbinden, 2007, IEEE J. Lo, H.-K., 2004, Proceedings of IEEE ISIT (IEEE, New York),
Quantum Electron. 13, 944. p. 137.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 835

Lo, H.-K., X. Ma, and K. Chen, 2005, Phys. Rev. Lett. 94, 230504. Michler, M., H. Weinfurter, and M. Żukowski, 2000, Phys. Rev.
Lounis, B., and W. Moerner, 2000, Nature (London) 407, 491. Lett. 84, 5457.
Lounis, B., and M. Orrit, 2005, Rep. Prog. Phys. 68, 1129. Michler, P., A. Kiraz, C. Becher, W. Schoenfeld, P. M. Petroff, L.
Lu, C.-Y., D. E. Browne, T. Yang, and J.-W. Pan, 2007, Phys. Rev. Zhang, E. Hu, and A. Imamoḡlu, 2000, Science 290,
Lett. 99, 250504. 2282.
Lu, C.-Y., W.-B. Gao, O. Gühne, X.-Q. Zhou, Z.-B. Chen, and Mintert, F., A. Carvalho, M. Kuś, and A. Buchleitner, 2005, Phys.
J.-W. Pan, 2009, Phys. Rev. Lett. 102, 030502. Rep. 415, 207.
Lu, C.-Y., W.-B. Gao, J. Zhang, X.-Q. Zhou, T. Yang, and J.-W. Pan, Mitchell, M., J. Lundeen, and A. Steinberg, 2004, Nature (London)
2008, Proc. Natl. Acad. Sci. U.S.A. 105, 11050. 429, 161.
Lu, C.-Y., T. Yang, and J.-W. Pan, 2009, Phys. Rev. Lett. 103, Moehring, D., P. Maunz, S. Olmschenk, K. Younge, D. N.
020501. Matsukevich, L.-M. Duan, and C. Monroe, 2007, Nature
Lu, C.-Y., X.-Q. Zhou, O. Gühne, W.-B. Gao, J. Zhang, Z.-S. Yuan, (London) 449, 68.
A. Goebel, T. Yang, and J.-W. Pan, 2007, Nature Phys. 3, 91. Mohseni, M., J. Lundeen, K. Resch, and A. Steinberg, 2003, Phys.
Lukin, M., 2003, Rev. Mod. Phys. 75, 457. Rev. Lett. 91, 187903.
Lukin, M., and A. Imamoğlu, 2001, Nature (London) 413, 273. Molina-Terriza, G., A. Vaziri, J.Řeháček, Z. Hradil, and A.
Lukin, M., S. Yelin, and M. Fleischhauer, 2000, Phys. Rev. Lett. 84, Zeilinger, 2004, Phys. Rev. Lett. 92, 167903.
4232. Mollow, B., 1973, Phys. Rev. A 8, 2684.
Lütkenhaus, N., J. Calsamiglia, and K.-A. Suominen, 1999, Phys. Mosley, P., J. Lundeen, B. Smith, P. Wasylczyk, A. B. U’Ren, C.
Rev. A 59, 3295. Silberhorn, and I. Walmsley, 2008, Phys. Rev. Lett. 100,
Lvovsky, A., 2002, Phys. Rev. Lett. 88, 098901. 133601.
Ma, X.-S., B. Dakic, W. Naylor, A. Zeilinger, and P. Walther, 2011, Nagata, T., R. Okamoto, J. O‘Brien, K. Sasaki, and S. Takeuchi,
Nature Phys. 7, 399. 2007, Science 316, 726.
Madsen, M., D. Moehring, P. Maunz, R. Kohn, Jr., L.-M. Duan, and Neves, L., G. Lima, J. G. A. Gómez, C. H. Monken, C. Saavedra,
C. Monroe, 2006, Phys. Rev. Lett. 97, 040505. and S. Pádua, 2005, Phys. Rev. Lett. 94, 100501.
Mair, A., A. Vaziri, G. Weihs, and A. Zeilinger, 2001, Nature Nielsen, M., 2004, Phys. Rev. Lett. 93, 040503.
(London) 412, 313. Nielsen, M., and I. Chuang, 2000, Quantum Computation and
Mandel, L., 1999, Rev. Mod. Phys. 71, S274. Quantum Information (Cambridge University Press, Cambridge,
Mandel, L., and E. Wolf, 1995, Optical Coherence and Quantum England).
Optics (Cambridge University Press, Cambridge). Nielsen, M., and C. Dawson, 2005, Phys. Rev. A 71, 042323.
Mann, A., M. Revzen, and W. Schleich, 1992, Phys. Rev. A 46, Nielsen, M., E. Knill, and R. Laflamme, 1998, Nature (London)
5363. 396, 52.
Marcikic, I., H. de Riedmatten, W. Tittel, V. Scarani, H. Zbinden, Novikova, I., A. Gorshkov, D. Phillips, A. S. rensen, M. Lukin, and
and N. Gisin, 2002, Phys. Rev. A 66, 062308. R. Walsworth, 2007, Phys. Rev. Lett. 98, 243602.
Marcikic, I., H. de Riedmatten, W. Tittel, H. Zbinden, and M. Legré, O’Brien, J., 2007, Science 318, 1567.
and N. Gisin, 2004, Phys. Rev. Lett. 93, 180502. O’Brien, J., G. Pryde, A. White, T. Ralph, and D. Branning, 2003,
Marcikic, I., H. de Riedmatten, W. Tittel, H. Zbinden, and N. Gisin, Nature (London) 426, 264.
2003, Nature (London) 421, 509. O’Brien, J. L., A. Furasawa, and J. Vuckovic, 2009, Nature Photon.
Marinatto, L., 2003, Phys. Rev. Lett. 90, 258901. 3, 687.
Marsili, F., D. Bitauld, A. Fiore, A. Gaggero, F. Mattioli, R. Leoni, O’Brien, J. L., G. J. Pryde, A. Gilchrist, D. F. V. James, N. K.
M. Benkahoul, and F. Lévy, 2008, Opt. Express 16, 3191. Langford, T. C. Ralph, and A. G. White, 2004, Phys. Rev. Lett.
Massar, S., and S. Popescu, 1995, Phys. Rev. Lett. 74, 1259. 93, 080502.
Matsukevich, D., T. Chanelière, S. D. Jenkins, S.-Y. Lan, T. O’Brien, J. L., G. J. Pryde, A. G. White, and T. C. Ralph, 2005, Phys.
Kennedy, and A. Kuzmich, 2006a, Phys. Rev. Lett. 96, Rev. A 71, 060303.
030405. Okamoto, R., H. Hofmann, S. Takeuchi, and K. Sasaki, 2005, Phys.
Matsukevich, D., T. Chanelière, S. D. Jenkins, S. Y. Lan, T. Rev. Lett. 95, 210506.
Kennedy, and A. Kuzmich, 2006b, Phys. Rev. Lett. 97, Okamoto, R., J. O’Brien, H. Hofmann, and S. Takeuchi, 2010,
013601. arXiv:1006.4743.
Matsukevich, D., and A. Kuzmich, 2004, Science 306, 663. O’Sullivan-Hale, M., I. Khan, R. Boyd, and J. Howell, 2005, Phys.
Matsukevich, D., P. Maunz, D. Moehring, S. Olmschenk, and C. Rev. Lett. 94, 220501.
Monroe, 2008, Phys. Rev. Lett. 100, 150404. Ou, Z., and L. Mandel, 1988a, Phys. Rev. Lett. 61, 54.
Mattle, K., H. Weinfurter, P. Kwiat, and A. Zeilinger, 1996, Phys. Ou, Z., and L. Mandel, 1988b, Phys. Rev. Lett. 61, 50.
Rev. Lett. 76, 4656. Pachos, J., 2007, Ann. Phys. (N.Y.) 322, 1254.
Maunz, P., D. Moehring, S. Olmschenk, K. Younge, D. N. Pachos, J., W. Wieczorek, C. Schmid, N. Kiesel, R. Pohlner, and H.
Matsukevich, and C. Monroe, 2007, Nature Phys. 3, 538. Weinfurter, 2009, New J. Phys. 11, 083010.
Mazurenko, Y., R. Giust, and J. Goedgebuer, 1997, Opt. Commun. Pan, J.-W., D. Bouwmeester, M. Daniell, H. Weinfurter, and A.
133, 87. Zeilinger, 2000, Nature (London) 403, 515.
McKeever, J., A. Boca, A. Boozer, R. Miller, J. Buck, A. Kuzmich, Pan, J.-W., D. Bouwmeester, H. Weinfurter, and A. Zeilinger, 1998,
and H. Kimble, 2004, Science 303, 1992. Phys. Rev. Lett. 80, 3891.
Mermin, N., 1990a, Phys. Today 43(6), 9. Pan, J.-W., M. Daniell, S. Gasparoni, G. Weihs, and A. Zeilinger,
Mermin, N., 1990b, Phys. Rev. Lett. 65, 1838. 2001, Phys. Rev. Lett. 86, 4435.
Mermin, N., 1993, Rev. Mod. Phys. 65, 803. Pan, J.-W., S. Gasparoni, M. Aspelmeyer, T. Jennewein, and A.
Mermin, N. D., 1990, Phys. Rev. Lett. 65, 3373. Zeilinger, 2003, Nature (London) 421, 721.
Michler, M., K. Mattle, H. Weinfurter, and A. Zeilinger, 1996, Phys. Pan, J.-W., S. Gasparoni, R. Ursin, G. Weihs, and A. Zeilinger,
Rev. A 53, R1209. 2003, Nature (London) 423, 417.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

836 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

Pan, J.-W., C. Simon, C. Brukner, and A. Zeilinger, 2001, Nature Rarity, J., P. Tapster, E. Jakeman, T. Larchuk, R. A. Campos, M.
(London) 410, 1067. Teich, and B. Saleh, 1990, Phys. Rev. Lett. 65, 1348.
Pan, J.-W., and A. Zeilinger, 1998, Phys. Rev. A 57, 2208. Rarity, J., P. Tapster, and R. Loudon, 1996, Quantum Interferometry,
Patel, R., A. Bennett, I. Farrer, C. Nicoll, D. Ritchie, and A. Shields, edited by F. De Martini, G. Denardo, and Y. Shih (VCH,
2010, Nature Photon. 4, 632. Weinheim, Germany).
Paterek, T., A. Fedrizzi, S. Gröblacher, T. Jennewein, M. Żukowski, Raussendorf, R., and H. Briegel, 2001, Phys. Rev. Lett. 86, 5188.
M. Aspelmeyer, and A. Zeilinger, 2007, Phys. Rev. Lett. 99, Raussendorf, R., D. Browne, and H. Briegel, 2003, Phys. Rev. A 68,
210406. 022312.
Paul, H., 1986, Rev. Mod. Phys. 58, 209. Raussendorf, R., J. Harrington, and K. Goya, 2006, Ann. Phys.
Pelton, M., C. Santori, B. Z. J. Vučković, G. Solomon, J. Plant, and (N.Y.) 321, 2242.
Y. Yamamoto, 2002, Phys. Rev. Lett. 89, 233602. Raussendorf, R., J. Harrington, and K. Goya, 2007, New J. Phys. 9,
Peng, C.-Z., T. Yang, X.-H. Bao, J. Zhang, X.-M. Jin, F.-Y. Feng, B. 199.
Yang, J. Yang, J. Yin, Q. Zhang, N. Li, B.-L. Tian, and J.-W. Pan, Reck, M., A. Zeilinger, H. J. Bernstein, and P. Bertani, 1994, Phys.
2005, Phys. Rev. Lett. 94, 150501. Rev. Lett. 73, 58.
Peng, C.-Z., J. Zhang, D. Yang, W.-B. Gao, H.-X. Ma, H. Yin, H.-P. Resch, K., M. Lindenthal, B. Blauensteiner, H. Böhm, A. Fedrizzi,
Zeng, T. Yang, X.-B. Wang, and J.-W. Pan, 2007, Phys. Rev. Lett. C. Kurtsiefer, A. Poppe, T. Schmitt-Manderbach, M. Taraba, R.
98, 010505. Ursin, P. Walther, H. Weier, H. Weinfurter, and A. Zeilinger, 2005,
Peres, A., 1996, Phys. Rev. Lett. 77, 1413. Opt. Express 13, 202.
Peres, A., 2000, J. Mod. Opt. 47, 531. Rohde, P., T. Ralph, and W. Munro, 2007, Phys. Rev. A 75, 010302
Peres, A., 2002, Quantum Theory: Concepts and Methods (Kluwer (R).
Academic Publishers, New York). Romero, J., J. Leach, B. Jack, S. Barnett, M. Padgett, and S. Franke-
Peruzzo, A., 2010, Science 329, 1500. Arnold, 2010, New J. Phys. 12, 123007.
Peters, N., J. Altepeter, D. Branning, E. J., and T. C. Wei, and P. Rosenberg, D., J. Harrington, P. Rice, P. Hiskett, C. Peterson, R.
Kwiat, 2004, Phys. Rev. Lett. 92, 133601. Hughes, A. Lita, S. Nam, and J. Nordholt, 2007, Phys. Rev. Lett.
Phillips, D., A. Fleischhauer, A. Mair, R. Walsworth, and M. Lukin, 98, 010503.
2001, Phys. Rev. Lett. 86, 783. Rosenberg, D., A. Lita, A. Miller, and S. Nam, 2005, Phys. Rev. A
Pittman, T., and J. Franson, 2003, Phys. Rev. Lett. 90, 240401. 71, 061803.
Pittman, T., B. Jacobs, and J. Franson, 2001, Phys. Rev. A 64, Rosenfeld, W., M. Weber, J. Volz, F. Henkel, M. Krug, A. Cabello,
062311. M. Zukowski, and H. Weinfurter, 2009, Adv. Sci. Lett. 2,
Pittman, T. B., M. J. Fitch, B. C. Jacobs, and J. D. Franson, 2003, 469.
Phys. Rev. A 68, 032316. Rowe, M., D. Kielpinski, V. Meyer, C. Sackett, W. M. Itano, C.
Pittman, T. B., D. V. Strekalov, A. Migdall, M. H. Rubin, A. V. Monroe, and D. J. Wineland, 2001, Nature (London)
Sergienko, and Y. H. Shih, 1996, Phys. Rev. Lett. 77, 1917. 409, 791.
Politi, A., M. Cryan, J. Rarity, S. Yu, and J. O’Brien, 2008, Science Roy, S., and V. Singh, 1991, Phys. Rev. Lett. 67, 2761.
320, 646. Rubin, M., D. Klyshko, Y. Shih, and A. Sergienko, 1994, Phys. Rev.
Politi, A., J. C. F. Matthews, and J. L. O’Brien, 2009, Science 325, A 50, 5122.
1221. Ryff, L., 1997, Am. J. Phys. 65, 1197.
Popescu, S., 1995, arXiv:quant-ph/9501020. Sanaka, K., T. Jennewein, J.-W. Pan, K. Resch, and A. Zeilinger,
Popescu, S., L. Hardy, and M. Żukowski, 1997, Phys. Rev. A 56, 2004, Phys. Rev. Lett. 92, 017902.
R4353. Sangouard, N., C. Simon, H. de Riedmatten, and N. Gisin, 2009,
Preskill, J., 1998, http://www.theory.caltech.edu/people/preskill/ arXiv:0906.2699.
ph229/. Sansoni, L., F. Sciarrino, G. Vallone, P. Mataloni, A. Crespi, R.
Prevedel, R., G. Cronenberg, M. Tame, M. Paternostro, P. Walther, Ramponi, and R. Osellame, 2010, Phys. Rev. Lett. 105, 200503.
M. Kim, and A. Zeilinger, 2009, Phys. Rev. Lett. 103, 020503. Santori, C., D. Fattal, G. S. J. Vučković, and Y. Yamamoto, 2002,
Prevedel, R., M. Tame, A. Stefanov, M. Paternostro, M. Kim, and A. Nature (London) 419, 594.
Zeilinger, 2007, Phys. Rev. Lett. 99, 250503. Santos, E., 1991, Phys. Rev. Lett. 66, 1388.
Prevedel, R., P. Walther, F. Tiefenbacher, P. Böhi, R. Kaltenbaek, T. Santos, E., 1992, Phys. Rev. Lett. 68, 2702.
Jennewein, and A. Zeilinger, 2007, Nature (London) Schmitt-Manderbach, T., H. Weier, M. Fürst, R. Ursin, F.
445, 65. Tiefenbacher, T. Scheidl, J. Perdigues, Z. Sodnik, C. Kurtsiefer,
Radmark, M., M. Zukowski, and M. Bourennane, 2009, Phys. Rev. J. G. Rarity, A. Zeilinger, and H. Weinfurter, 2007, Phys. Rev.
Lett. 103, 150501. Lett. 98, 010504.
Radnaev, A., Y. O. Dudin, R. Zhao, H. H. Jen, S. D. Jenkins, A. Schori, C., B. Julsgaard, J. S. rensen, and E. Polzik, 2002, Phys. Rev.
Kuzmich, and T. A. B. Kennedy, 2010, Nature Phys. 6, 894. Lett. 89, 057903.
Raimond, J., M. Brune, and S. Haroche, 2001, Rev. Mod. Phys. 73, Schrödinger, E., 1935a, Naturwissenschaften 23, 807.
565. Schrödinger, E., 1935b, Naturwissenschaften 23, 823.
Ralph, T., A. Hayes, and A. Gilchrist, 2005, Phys. Rev. Lett. 95, Schrödinger, E., 1935c, Naturwissenschaften 23, 844.
100501. Schrödinger, E., 1935d, Math. Proc. Cambridge Philos. Soc. 31,
Ralph, T., N. Langford, T. Bell, and A. White, 2002, Phys. Rev. A 555.
65, 062324. Schuck, C., G. Huber, C. Kurtsiefer, and H. Weinfurter, 2006, Phys.
Ralph, T., and G. J. Pryde, 2010, Prog. Opt. 54, 209. Rev. Lett. 96, 190501.
Ralph, T., A. White, W. Munro, and G. Milburn, 2001, Phys. Rev. A Scully, M., and M. Zubairy, 1997, Quantum Optics (Cambridge
65, 012314. University Press, Cambridge, England).
Rarity, J., 1995, Ann. N.Y. Acad. Sci. 755, 624. Scully, M. O., B.-G. Englert, and H. Walther, 1991, Nature (London)
Rarity, J., and P. Tapster, 1990, Phys. Rev. Lett. 64, 2495. 351, 111.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry 837

Selleri, F., and A. Zeilinger, 1988, Found. Phys. 18, 1141. Vandersypen, L., M. Steffen, G. Breyta, C. Yannoni, M. Sherwood,
Sergienko, A., M. Atatüre, Z. Walton, G. Jaeger, B. Saleh, and M. and I. Chuang, 2001, Nature (London) 414, 883.
Teich, 1999, Phys. Rev. A 60, R2622. van der Wal, C., M. Eisaman, A. Andre, R. Walsworth, D. Phillips,
Shelby, R., M. Levenson, S. Perlmutter, R. DeVoe, and D. Walls, A. Zibrov, and M. Lukin, 2003, Science 301, 196.
1986, Phys. Rev. Lett. 57, 691. van Enk, S., J. Cirac, and P. Zoller, 1997, Phys. Rev. Lett. 78,
Shields, A., 2007, Nature Photon. 1, 215. 4293.
Shih, Y., 2003, Rep. Prog. Phys. 66, 1009. van Loock, P., T. Ladd, K. Sanaka, F. Yamaguchi, K. Nemoto,
Shih, Y., and C. Alley, 1988, Phys. Rev. Lett. 61, 2921. W. Munro, and Y. Yamamoto, 2006, Phys. Rev. Lett. 96,
Simon, C., and D. Bouwmeester, 2003, Phys. Rev. Lett. 91, 053601. 240501.
Simon, C., and W. Irvine, 2003, Phys. Rev. Lett. 91, 110405. Varnava, M., D. Browne, and T. Rudolph, 2006, Phys. Rev. Lett. 97,
Simon, C., and J.-W. Pan, 2002, Phys. Rev. Lett. 89, 257901. 120501.
Simon, C., M. Żukowski, H. Weinfurter, and A. Zeilinger, 2000, Varnava, M., D. Browne, and T. Rudolph, 2008, Phys. Rev. Lett.
Phys. Rev. Lett. 85, 1783. 100, 060502.
Sleator, T., and H. Weinfurter, 1995, Phys. Rev. Lett. 74, 4087. Vaziri, A., J.-W. Pan, T. Jennewein, G. Weihs, and A. Zeilinger,
Sliwa, C., and K. Banaszek, 2003, Phys. Rev. A 67, 030101. 2003, Phys. Rev. Lett. 91, 227902.
Slusher, R., L. Hollberg, B. Yurke, J. Mertz, and J. F. Valley, 1985, Vaziri, A., G. Weihs, and A. Zeilinger, 2002, Phys. Rev. Lett. 89,
Phys. Rev. Lett. 55, 2409. 240401.
Smithey, D., M. Beck, M. Belsley, and M. Raymer, 1992, Phys. Rev. Verstraete, F., J. Dehaene, and B. DeMoor, 2001, Phys. Rev. A 64,
Lett. 69, 2650. 010101(R).
Specker, E. P., 1960, Dialectica 14, 239. Vertesi, T., S. Pironio, and N. Brunner, 2010, Phys. Rev. Lett. 104,
Stefanov, A., H. Zbinden, N. Gisin, and A. Suarez, 2002, Phys. Rev. 060401.
Lett. 88, 120404. Volz, J., M. Weber, D. Schlenk, W. Rosenfeld, J. Vrana, K. Saucke,
Stefanov, A., H. Zbinden, N. Gisin, and A. Suarez, 2003, Phys. Rev. C. Kurtsiefer, and H. Weinfurter, 2006, Phys. Rev. Lett. 96,
A 67, 042115. 030404.
Stevenson, R., R. Young, P. Atkinson, K. Cooper, D. Ritchie, and A. Wagenknecht, C., C.-M. Li, A. Reingruber, X.-H. Bao, A. Goebel,
Shields, 2006, Nature (London) 439, 179. Y.-A. Chen, Q. Zhang, K. Chen, and J.-W. Pan, 2010, Nature
Strauf, S., N. Stoltz, M. Rakher, L. Coldren, P. M. Petroff, and D. Photon. 4, 549.
Bouwmeester, 2007, Nature Photon. 1, 704. Waks, E., A. Zeevi, and Y. Yamamoto, 2002, Phys. Rev. A 65,
Sun, F. W., B. H. Liu, Y. F. Huang, Z. Y. Ou, and G. C. Guo, 2006, 052310.
Phys. Rev. A 74, 033812. Walborn, S., S. Pádua, and C. Monken, 2003, Phys. Rev. A 68,
Sun, P., Y. Mazurenko, and Y. Fainman, 1995, Opt. Lett. 20, 042313.
1062. Walls, D., and G. Milburn, 1994, Quantum Optics (Springer-Verlag,
Takesue, H., S. Nam, Q. Zhang, R. Hadfield, T. Honjo, K. Tamaki, Berlin).
and Y. Yamamoto, 2007, Nature Photon. 1, 343. Walther, H., B. Varcoe, B.-G. Englert, and T. Becker, 2006, Rep.
Tame, M. S., M. Paternostro, and M. S. Kim, 2007, New J. Phys. 9, Prog. Phys. 69, 1325.
201. Walther, P., J.-W. Pan, M. Aspelmeyer, R. Ursin, S. Gasparoni, and
Tame, M., R. Prevedel, M. Paternostro, P. Bohi, M. Kim, and A. A. Zeilinger, 2004, Nature (London) 429, 158.
Zeilinger, 2007, Phys. Rev. Lett. 98, 140501. Walther, P., K. Resch, C. Brukner, A. Steinberg, J.-W. Pan, and A.
Taylor, G., 1909, Proc. Camb. Phil. Soc. Math. Phys. Sci. 15, 114. Zeilinger, 2005, Phys. Rev. Lett. 94, 040504.
Terhal, B., 2000, Phys. Lett. A 271, 319. Walther, P., K. Resch, T. Rudolph, E. Schenck, H. Weinfurter, V.
Tittel, W., J. Brendel, H. Zbinden, and N. Gisin, 1998, Phys. Rev. Vedral, M. Aspelmeyer, and A. Zeilinger, 2005, Nature (London)
Lett. 81, 3563. 434, 169.
Tittel, W., and G. Weihs, 2001, Quantum Inf. Comput. 1, 3. Walton, Z., M. Booth, A. Sergienko, B. Saleh, and M. Teich, 2003,
Tokunaga, Y., S. Kuwashiro, T. Yamamoto, M. Koashi, and N. Phys. Rev. A 67, 053810.
Imoto, 2008, Phys. Rev. Lett. 100, 210501. Walton, Z., A. Sergienko, B. Saleh, and M. Teich, 2004, Phys. Rev.
Torgerson, J., D. Branning, C. Monken, and L. Mandel, 1995, Phys. A 70, 052317.
Lett. A 204, 323. Wang, X.-B., 2005, Phys. Rev. Lett. 94, 230503.
Torres, J., A. Alexandrescu, and L. Torner, 2003, Phys. Rev. A 68, Wang, Z.-W., X.-F. Zhou, Y.-F. Huang, Y.-S. Zhang, X.-F. Ren, and
050301. G.-C. Guo, 2006, Phys. Rev. Lett. 96, 220505.
Torres, J., S. C. F. Macià, and L. Torner, 2005, Opt. Lett. 30, 314. Weihs, G., T. Jennewein, C. Simon, H. Weinfurter, and A. Zeilinger,
Tóth, G., and O. Gühne, 2005, Phys. Rev. Lett. 94, 060501. 1998, Phys. Rev. Lett. 81, 5039.
U’Ren, A., K. Banaszek, and I. Walmsley, 2003, Quantum Inf. Weihs, G., M. Reck, H. Weinfurter, and A. Zeilinger, 1996, Opt.
Comput. 3, 480. Lett. 21, 302.
Ursin, R., T. Jennewein, M. Aspelmeyer, R. Kaltenbaek, M. Weinfurter, H., 1994, Europhys. Lett. 25, 559.
Lindenthal, P. Walther, and A. Zeilinger, 2004, Nature (London) Weinfurter, H., 2000, Adv. At. Mol. Opt. Phys. 42, 489.
430, 849. Weinfurter, H., and M. Żukowski, 2001, Phys. Rev. A 64,
Ursin, R., F. Tiefenbacher, T. Schmitt-Manderbach, H. Weier, T. 010102.
Scheidl, M. Lindenthal, B. Blauensteiner, T. Jennewein, J. Weinhold, T., A. Gilchrist, K. Resch, A. Doherty, J. O’Brien, G.
Perdigues, P. Trojek, B. Ömer, M. Fürst, M. Meyenburg, J. Pryde, and A. White, 2008, arXiv:0808.0794.
Rarity, Z. Sodnik, C. Barbieri, H. Weinfurter, and A. Zeilinger, Werner, R., 1989, Phys. Rev. A 40, 4277.
2007, Nature Phys. 3, 481. Werner, R., and M. M. Wolf, 2001a, Quantum Inf. Comput. 1(3), 1.
Vaidman, L., and N. Yoran, 1999, Phys. Rev. A 59, 116. Werner, R., and M. M. Wolf, 2001b, Phys. Rev. A 64, 032112.
Vallone, G., E. Pomarico, and F. D. Martini, 2008, Phys. Rev. A 78, White, A., D. James, P. Eberhard, and P. Kwiat, 1999, Phys. Rev.
042335. Lett. 83, 3103.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012

838 Jian-Wei Pan et al.: Multiphoton entanglement and interferometry

White, A. G., D. James, W. J. Munro, and P. G. Kwiat, 2001, Phys. Zel’dovich, Y., and D. Klyshko, 1969, JETP Lett. 9, 40.
Rev. A 65, 012301. Zhang, A.-N., C.-Y. Lu, X.-Q. Zhou, Y.-A. Chen, Z. Z., and T. Yang,
Wieczorek, W., N. Kiesel, C. Schmid, and H. Weinfurter, 2009, and J.-W. Pan, 2006, Phys. Rev. A 73, 022330.
Phys. Rev. A 79, 022311. Zhang, Q., X.-H. Bao, C.-Y. Lu, X.-Q. Zhou, T. Yang, T. Rudolph,
Wieczorek, W., R. Krischek, N. Kiesel, P. Michelberger, G. Tóth, and J.-W. Pan, 2008, Phys. Rev. A 77, 062316.
and H. Weinfurter, 2009, Phys. Rev. Lett. 103, 020504. Zhang, Q., A. Goebel, C. Wagenknecht, Y.-A. Chen, B. Zhao, T.
Wieczorek, W., C. Schmid, N. Kiesel, R. Pohlner, O. Guhne, and H. Yang, A. Mair, J. Schmiedmayer, and J.-W. Pan, 2006, Nature
Weinfurter, 2008, Phys. Rev. Lett. 101, 010503. Phys. 2, 678.
Wiesner, S., 1983, SIGACT News 15, 78. Zhao, B., Y.-A. Chen, X.-H. Bao, T. Strassel, C.-S. Chuu, X.-M. Jin,
Wootters, W. K., 1998a, Phys. Rev. Lett. 80, 2245. J. Schmiedmayer, Z.-S. Yuan, S. Chen, and J.-W. Pan, 2009,
Wootters, W. K., 1998b, Phil. Trans. R. Soc. A 356, 1717. Nature Phys. 5, 95.
Wootters, W., and W. Zurek, 1982, Nature (London) 299, 802. Zhao, B., Z.-B. Chen, Y.-A. Chen, J. Schmiedmayer, and J.-W. Pan,
Wu, L.-A., H. Kimble, J. Hall, and H. Wu, 1986, Phys. Rev. Lett. 57, 2007, Phys. Rev. Lett. 98, 240502.
2520. Zhao, R., Y. Dudin, S. Jenkins, C. Campbell, D. Matsukevich, T.
Yamamoto, T., M. Koashi, and N. Imoto, 2001, Phys. Rev. A 64, Kennedy, and A. Kuzmich, 2009, Nature Phys. 5, 100.
012304. Zhao, Z., Y.-A. Chen, A.-N. Zhang, T. Yang, H. Briegel, and J.-W.
Yamamoto, T., M. Koashi, Ş. K. Özdemir, and N. Imoto, 2003, Pan, 2004, Nature (London) 430, 54.
Nature (London) 421, 343. Zhao, Z., J.-W. Pan, and M. Zhan, 2001, Phys. Rev. A 64,
Yang, T., Q. Zhang, T.-Y. Chen, S. Lu, J. Yin, J.-W. Pan, Z.-Y. Wei, 014301.
J.-R. Tian, and J. Zhang, 2006, Phys. Rev. Lett. 96, 110501. Zhao, Z., T. Yang, Y.-A. Chen, A.-N. Zhang, M.Żukowski, and
Yang, T., Q. Zhang, J. Zhang, J. Yin, Z. Zhao, M. Żukowski, Z.-B. J.-W. Pan, 2003, Phys. Rev. Lett. 91, 180401.
Chen, and J.-W. Pan, 2005, Phys. Rev. Lett. 95, 240406. Zhao, Z., T. Yang, Y.-A. Chen, A.-N. Zhang, and J.-W. Pan, 2003,
Yao, X.-C., T.-X. Wang, P. Xu, H. Lu, G.-S. Pan, X.-H. Bao, C.-Z. Phys. Rev. Lett. 90, 207901.
Peng, C.-Y. Lu, Y.-A. Chen, and J.-W. Pan, 2012, Nature Photon. Zhao, Z., A.-N. Zhang, Y.-A. Chen, H. Z. J.-F. Du, T. Yang, and
6, 225. J.-W. Pan, 2005, Phys. Rev. Lett. 94, 030501.
Yuan, Z.-S., Y.-A. Chen, S. Chen, B. Zhao, M. Koch, T. Strassel, Y. Żukowski, M., 1991, Phys. Lett. A 157, 198.
Zhao, G.-J. Zhu, J. Schmiedmayer, and J.-W. Pan, 2007, Phys. Żukowski, M., 1993, Phys. Lett. A 177, 290.
Rev. Lett. 98, 180503. Żukowski, M., 2000, Phys. Rev. A 61, 022109.
Yuan, Z.-S., Y.-A. Chen, B. Zhao, S. Chen, J. Schmiedmayer, and Żukowski, M., and Č. Brukner, 2002, Phys. Rev. Lett. 88,
J.-W. Pan, 2008, Nature (London) 454, 1098. 210401.
Yurke, B., and D. Stoler, 1992a, Phys. Rev. Lett. 68, 1251. Żukowski, M., and D. Kaszlikowski, 1997, Phys. Rev. A 56,
Yurke, B., and D. Stoler, 1992b, Phys. Rev. A 46, 2229. R1682.
Zbinden, H., J. Brendel, N. Gisin, and W. Tittel, 2001, Phys. Rev. A Żukowski, M., and J. Pykacz, 1988, Phys. Lett. A 127, 1.
63, 022111. Żukowski, M., A. Zeilinger, and M. A. Horne, 1997, Phys. Rev. A
Zeilinger, A., 1981, Am. J. Phys. 49, 882. 55, 2564.
Zeilinger, A., H. J. Bernstein, D. M. Greenberger, M. A. Horne, and Żukowski, M., A. Zeilinger, M. A. Horne, and A. K. Ekert, 1993,
M. Żukowski, 1993, Quantum Control and Measurement, edited Phys. Rev. Lett. 71, 4287.
by H. Ezawa and Y. Murayama (North Holland, Amsterdam) p. 9. Żukowski, M., A. Zeilinger, M. Horne, and H. Weinfurter, 1999, Int.
Zeilinger, A., M. Horne, H. Weinfurter, and M. Żukowski, 1997, J. Theor. Phys. 38, 501.
Phys. Rev. Lett. 78, 3031. Żukowski, M., A. Zeilinger, and H. Weinfurter, 1995, Ann. N.Y.
Zeilinger, A., G. Weihs, T. Jennewein, and M. Aspelmeyer, 2005, Acad. Sci. 755, 91.
Nature (London) 433, 230. Zurek, W., 2003, Rev. Mod. Phys. 75, 715.

Rev. Mod. Phys., Vol. 84, No. 2, April–June 2012