PHYSICAL REVIEW A, VOLUME 63, 022106

Greenberger-Horne-Zeilinger nonlocality in phase space

P. van Loock and Samuel L. Braunstein
Quantum Optics and Information Group, School of Informatics, University of Wales, Bangor LL57 1UT, United Kingdom
共Received 9 June 2000; published 16 January 2001兲
We show that the continuous-variable analogues to the multipartite entangled Greenberger-Horne-Zeilinger
states of qubits violate Bell-type inequalities imposed by local realistic theories. Our results suggest that the
degree of nonlocality of these nonmaximally entangled continuous-variable states, represented by the maxi-
mum violation, grows with increasing number of parties. This growth does not appear to be exponentially large
as for the maximally entangled qubit states, but rather decreases for larger numbers of parties.

DOI: 10.1103/PhysRevA.63.022106 PACS number共s兲: 03.65.Ta, 03.67.⫺a, 42.50.Dv

Entanglement and nonlocality are the most outstanding
features of quantum mechanics. In the rapidly advancing
field of quantum communication and computation, entangled
states are the key ingredients: they enable quantum telepor-
tation 关1兴, quantum cryptography 关2兴, and many other poten-
tially useful schemes. Bell showed that nonlocality can be
revealed via the constraints that local realism imposes on the
statistics of two physically separated systems 关3兴. These con-
straints, expressed in terms of the Bell inequalities, can be
violated by quantum mechanics. Entanglement does not au-
tomatically imply nonlocality. The so-called Werner states
are mixed states that are inseparable, but do not violate any
Bell inequality 关4兴. Also pure entangled states can, if associ-
ated with a positive Wigner function, directly reveal a local
hidden-variable description 关3兴.
Towards possible applications in quantum communica-
tion, both theoretical and experimental investigations in-
creasingly focus on quantum states with a continuous spec-
trum defined in an infinite-dimensional Hilbert space. These
states can be relatively easily generated using squeezed light
and beam splitters, as for instance the entangled two-mode
squeezed vacuum state that has already proven its usefulness
for quantum teleportation 关5兴. The two-mode squeezed
vacuum state is an approximate version of the original
Einstein-Podolsky-Rosen 共EPR兲 state 关6兴 where the quadra-
ture amplitudes of the electromagnetic field play the roles of
position and momentum of a particle. Its Wigner function is
positive everywhere and hence it has a local hidden-variable
continuous-variable GHZ states 关13兴 and thereby to prove
their nonlocality. Since these states have a positive Wigner
function, we shall follow the convenient strategy of Ban-
aszek and Wodkiewicz 关8兴 who exploited the fact that the
Wigner function is connected to the quantum mean value of
the photon number parity operator. Relying on this connec-
tion, we will demonstrate N-party nonlocality using mean-
value inequalities 关12兴, and we do not follow the original
GHZ program utilizing a contradiction to local realism in a
single measurement.
Let us identify the ‘‘position’’ and ‘‘momentum’’ of a
particle with the quadrature amplitudes of a single electro-
magnetic mode 共the real and imaginary part of the mode’s
annihilation operator兲. In Ref. 关13兴, it has been shown that a
sequence of beam splitter operations,
B̂ N⫺1 N 共 ␲ /4兲 B̂ N⫺2 N⫺1 „cos⫺1 共 1/ 冑3 兲 …
⫻•••⫻B̂ 12„cos⫺1 共 1/冑N 兲 …,
applied to one momentum squeezed vacuum mode 1 and N
⫺1 position squeezed vacuum modes 2 through N, yields an
N-mode state with N-party entanglement between all modes.
Here, an ideal 共phase-free兲 beam splitter operation B̂ i j ( ␪ )
acts on a pair of modes i and j with annihilation operators â i
and â j like â i →â i cos ␪⫹â j sin ␪, and â j →â i sin ␪
⫺â j cos ␪. The Wigner function of the pure entangled
N-mode state is

冉冊 再 冋 冉兺 冊
description 关3兴. Thus, attempts to derive for this state viola- N N 2
2 2
tions of Bell inequalities based on homodyne measurements W 共 x,p兲 ⫽ exp ⫺e ⫺2r xi
of the quadratures failed 关7兴. However, whether nonlocality ␲ N i⫽1

册
is uncovered depends on the observables and the measure- N
1
ments considered in a specific Bell inequality and not only
on the quantum state itself. It was shown by Banaszek and
⫹
N 兺
i, j
共 p i⫺ p j 兲2

冋 冉兺 冊 兺 册冎
Wodkiewicz 关8兴, that the two-mode squeezed vacuum state is N 2 N
nonlocal, as it violates a Clauser-Horne-Shimony-Holt 2 1
共CHSH兲 inequality 关9兴 when measurements of photon num- ⫺e ⫹2r pi ⫹ 共 x i ⫺x j 兲 2 , 共1兲
N i⫽1 N i, j
ber parity are considered.
The nonlocality of the multipartite entangled qubit where x⫽(x 1 ,x 2 , . . . ,x N ) and p⫽(p 1 ,p 2 , . . . ,p N ) are the
Greenberger-Horne-Zeilinger 共GHZ兲 states can in principle positions and momenta of the N modes and r is the squeezing
be manifest in a single measurement and need not be statis- parameter 共with equal squeezing in all initial modes兲. The
tical 关10兴 as the violation of a Bell inequality that relies on state W(x,p) is always positive, symmetric among the N
mean values. But Mermin and others 关11,12兴 also derived modes, and becomes peaked at x i ⫺x j ⫽0 (i, j
Bell-CHSH inequalities for N-particle systems. The aim of ⫽1,2, . . . ,N) and p 1 ⫹ p 2 ⫹•••⫹ p N ⫽0 for large squeezing
this paper is to apply those N-party inequalities to r. For N⫽2, it equals the well-known EPR-state Wigner

1050-2947/2001/63共2兲/022106共6兲/$15.00 63 022106-1 ©2001 The American Physical Society

P. van LOOCK AND SAMUEL L. BRAUNSTEIN
function that approaches ␦ (x 1 ⫺x 2 ) ␦ (p 1 ⫹p 2 ) in the limit of
infinite squeezing. Any nonzero squeezing yields N-partite
entanglement and the position and momentum correlations
can be exploited for quantum teleportation 关14兴 between any
two of N parties with the assistance of the remaining N⫺2
parties 关13兴.
In order to prove the nonlocality exhibited by the state
W(x,p), we use the fact that the Wigner function is propor-
tional to the quantum expectation value of a displaced parity
operator 关15兴. We obtain the relation 关8兴
冉冊 冉冊
N N
2 2
W 共 ␣兲 ⫽ 具 ⌸̂ 共 ␣兲 典 ⫽
␲ ␲
where ␣⫽x⫹ip⫽( ␣ 1 , ␣ 2 , . . . , ␣ N ) and ⌸( ␣) is the quan-
tum expectation value of the operator
⌸̂ 共 ␣兲 ⫽ 丢 i⫽1
N
⌸̂ i 共 ␣ i 兲 ⫽ 丢 i⫽1
N
D̂ i 共 ␣ i 兲共 ⫺1 兲 n̂ i D̂ †i 共 ␣ i 兲 . 共3兲
The operators D̂ i ( ␣ i ) are phase-space displacement opera-
tors acting on mode i. Thus, ⌸̂( ␣) is a product of displaced
parity operators given by
⌸̂ i 共 ␣ i 兲 ⫽⌸̂ (⫹)
i 共 ␣ i 兲 ⫺⌸̂ i 共 ␣ i 兲 ,
(⫺)
with the projection operators
⬁ serve as the hidden-variable probability distribution. In this
i 共 ␣ i 兲 ⫽D̂ i 共 ␣ i 兲
⌸̂ (⫹) 兺
k⫽0
兩 2k 典具 2k 兩 D̂ †i 共 ␣ i 兲 ,
⬁
i 共 ␣ i 兲 ⫽D̂ i 共 ␣ i 兲
⌸̂ (⫺) 兺
k⫽0
兩 2k⫹1 典具 2k⫹1 兩 D̂ †i 共 ␣ i 兲 ,
corresponding to the measurement of an even 共parity ⫹1) or
an odd 共parity ⫺1) number of photons in mode i. This
means that each mode is now characterized by a dichotomic
variable similar to the single-particle spin or the single-
photon polarization. Different spin or polarizer orientations
are replaced by different displacements in phase space.
These different settings of a measurement with two possible
outcomes ⫾1 for each possible setting is exactly what we
need for the nonlocality test.
In the case of N-particle systems, such a nonlocality test is
possible using the N-particle generalization of the two-
particle Bell-CHSH inequality 关12兴. This inequality is based
on the following recursively defined linear combination of
joint measurement results
B N ⬅ 21 关 ␴ 共 a N 兲 ⫹ ␴ 共 a N⬘ 兲兴 B N⫺1
⫹ 12 关 ␴ 共 a N 兲 ⫺ ␴ 共 a N⬘ 兲兴 B N⫺1
⬘ ⫽⫾2,
where ␴ (a N )⫽⫾1 and ␴ (a N⬘ )⫽⫾1 describe two possible
outcomes for two possible measurement settings 共denoted by
a N and a N⬘ ) of measurements on the Nth particle. Provided
that B N⫺1 ⫽⫾2 and B N⫺1⬘ ⫽⫾2, Eq. 共7兲 is true for a single
PHYSICAL REVIEW A 63 022106
run of the measurements, where ␴ (a N ) becomes either ⫹1
or ⫺1 and so does ␴ (a N⬘ ). Thus, induction proves Eq. 共7兲 for
any N with
B 2 ⬅ 关 ␴ 共 a 1 兲 ⫹ ␴ 共 a 1⬘ 兲兴 ␴ 共 a 2 兲 ⫹ 关 ␴ 共 a 1 兲 ⫺ ␴ 共 a 1⬘ 兲兴 ␴ 共 a 2⬘ 兲 ⫽⫾2,
共8兲
which is trivially true 共the expressions B N⬘ are equivalent to
B N but with all the a i and a ⬘i swapped兲. Within the frame-
work of local realistic theories with the hidden variables ␭
⫽(␭ 1 ,␭ 2 , . . . ,␭ N ) and the normalized probability distribu-
tion P(␭), we obtain an inequality for the average value of
B N ⬅B N (␭),
⌸ 共 ␣兲 , 共2兲
冏冕 d␭ 1 d␭ 2 . . . d␭ N P 共 ␭兲 B N 共 ␭兲 ⭐2. 冏 共9兲
By the linearity of averaging, this is a sum of means of
products of the ␴ (a i ) and ␴ (a ⬘i ). For example, if N⫽2, we
obtain the CHSH inequality
兩 C 共 a 1 ,a 2 兲 ⫹C 共 a 1 ,a 2⬘ 兲 ⫹C 共 a 1⬘ ,a 2 兲 ⫺C 共 a 1⬘ ,a 2⬘ 兲 兩 ⭐2,
共10兲
with the correlation functions
共4兲 C 共 a 1 ,a 2 兲 ⫽ 冕 d␭ 1 d␭ 2 P 共 ␭ 1 ,␭ 2 兲 ␴ 共 a 1 ,␭ 1 兲 ␴ 共 a 2 ,␭ 2 兲 .
共11兲
Following Bell 关3兴, an always positive Wigner function can
共5兲 sense, the EPR-state Wigner function could prevent the
CHSH inequality being violated: W(x 1 ,p 1 ,x 2 ,p 2 )
⬅ P(␭ 1 ,␭ 2 ). The same applies to the general Wigner func-
tion in Eq. 共1兲: W(x,p)⬅ P(␭) could be used to construct
共6兲 correlation functions
C 共 a兲 ⫽ 冕 d␭ 1 d␭ 2 . . . d␭ N P 共 ␭兲
⫻ ␴ 共 a 1 ,␭ 1 兲 ␴ 共 a 2 ,␭ 2 兲 ••• ␴ 共 a N ,␭ N 兲 , 共12兲
where a⫽(a 1 ,a 2 , . . . ,a N ) . However, for parity measure-
ments on each mode with possible results ⫾1 and different
settings by different displacements, this would require un-
bounded ␦ functions for the local objective quantities
␴ (a i ,␭ i ) 关8兴, as in this case the relation
C 共 a兲 ⬅⌸ 共 ␣兲 ⫽ 共 ␲ /2兲 N W 共 ␣兲 共13兲
holds. This relation, which directly relates the correlation
function to the Wigner function, is indeed crucial for the
nonlocality proof of the continuous-variable states in Eq. 共1兲.
For the EPR state with N⫽2, we can now look at the com-
bination 关8兴
共7兲 B2 ⫽⌸ 共 0,0兲 ⫹⌸ 共 0,␤ 兲 ⫹⌸ 共 ␣ ,0兲 ⫺⌸ 共 ␣ , ␤ 兲 , 共14兲
which according to Eq. 共10兲 satisfies 兩 B2 兩 ⭐2 for local real-
istic theories. Here, we have chosen the displacement set-
tings ␣ 1 ⫽ ␣ 2 ⫽0 and ␣ 1⬘ ⫽ ␣ , ␣ ⬘2 ⫽ ␤ .
Let us write the states in Eq. 共1兲 as
022106-2
GREENBERGER-HORNE-ZEILINGER NONLOCALITY IN . . . PHYSICAL REVIEW A 63 022106

FIG. 1. Violations of the inequality 兩 BN 兩 ⭐2

imposed by local realistic theories with the en-
tangled two-mode EPR (N⫽2, as in Ref. 关8兴兲,
three-mode GHZ (N⫽3), four-mode GHZ (N
⫽4), and five-mode GHZ (N⫽5) states.

再
⌸ 共 ␣兲 ⫽exp ⫺2 cosh 2r 兺
i⫽1
N

兩 ␣ i兩 2 B3 ⫽
3

兺 exp共 ⫺2J cosh 2r⫺

i⫽1
2
3 J sinh 2r cos 2 ␾ i 兲

⫹sinh 2r
2
N 冋 兺
N

i, j
共 ␣ i ␣ j ⫹ ␣ i* ␣ *j 兲 再
⫺exp ⫺6J cosh 2r⫺ 31 J sinh 2r

⫺ 兺
N

i⫽1
共 ␣ 2i ⫹ ␣ i* 2 兲 册冎 . 共15兲 ⫻ 兺
3

i⫽ j
关 cos 2 ␾ i ⫺4 cos共 ␾ i ⫹ ␾ j 兲兴 冎 . 共18兲

Apparently, because of the symmetry of the entangled three-

For N⫽2 and ␣ ⫽ ␤ ⫽i 冑J in terms of the real displace-
ment parameter J⭓0 关16兴, these states yield B2 ⫽1
⫹2 exp(⫺2J cosh 2r)⫺exp(⫺4Je ⫹2r ). In the limit of large
r(cosh 2r⬇e⫹2r/2) and small J, this B2 is maximized for
Je ⫹2r ⫽(ln 2)/3: B max
2 ⬇2.19 关8兴, which is a clear violation
of the inequality 兩 B2 兩 ⭐2. Smaller violations occur also for
smaller squeezing and bigger J. For any nonzero squeezing,
some violation takes place 共see Fig. 1兲.
Let us now examine the three-mode state and set N⫽3 in
Eq. 共15兲. According to the inequality of the correlation func-
tions derived from Eqs. 共7兲–共9兲 with N⫽3,
兩 C 共 a 1 ,a 2 ,a ⬘3 兲 ⫹C 共 a 1 ,a ⬘2 ,a 3 兲
⫹C 共 a ⬘1 ,a 2 ,a 3 兲 ⫺C 共 a 1⬘ ,a ⬘2 ,a ⬘3 兲 兩 ⭐2,
for the possible combination
B3 ⫽⌸ 共 0,0,␥ 兲 ⫹⌸ 共 0,␤ ,0兲 ⫹⌸ 共 ␣ ,0,0 兲 ⫺⌸ 共 ␣ , ␤ , ␥ 兲 ,
mode state, equal phases ␾ i should also be chosen in order to
maximize B3 . The best choice is ␾ 1 ⫽ ␾ 2 ⫽ ␾ 3 ⫽ ␲ /2, which
ensures that the positive terms in Eq. 共18兲 become maximal
and the contribution of the negative term minimal. Therefore
we again use equal settings ␣ ⫽ ␤ ⫽ ␥ ⫽i 冑J and obtain
B3 ⫽3 exp共 ⫺2J cosh 2r⫹2J sinh 2r/3兲 ⫺exp共 ⫺6Je ⫹2r 兲 .
共19兲
The violations of 兩 B3 兩 ⭐2 that occur with this result are simi-
lar to the violations 兩 B2 兩 ⭐2 obtained for the EPR state, but
the N⫽3 violations are even more significant than the N
⫽2 violations 共see Fig. 1兲. In the limit of large r 共and small
J), we may use cosh 2r⬇sinh 2r⬇e⫹2r/2 in Eq. 共19兲. Then
共16兲 B3 is maximized for Je ⫹2r ⫽3(ln 3)/16: B max 3 ⬇2.32. This
requires even smaller displacements J than in the N⫽2 case
for the same squeezing.
Let us now investigate the cases N⫽4 and N⫽5. From
Eqs. 共7兲–共9兲 with N⫽4, the following inequality for the cor-
relation functions can be derived:
共17兲 1
兩 C 共 a 1 ,a 2 ,a 3 ,a 4⬘ 兲 ⫹C 共 a 1 ,a 2 ,a 3⬘ ,a 4 兲 ⫹C 共 a 1 ,a 2⬘ ,a 3 ,a 4 兲
2

a contradiction to local realism does not occur only if 兩 B3 兩
⭐2. The corresponding settings here are ␣ 1 ⫽ ␣ 2 ⫽ ␣ 3 ⫽0
and ␣ ⬘1 ⫽ ␣ , ␣ 2⬘ ⫽ ␤ , and ␣ 3⬘ ⫽ ␥ . With the choice ␣
⫽ 冑Je i ␾ 1 , ␤ ⫽ 冑Je i ␾ 2 , and ␥ ⫽ 冑Je i ␾ 3 , we obtain
⫹C 共 a 1⬘ ,a 2 ,a 3 ,a 4 兲 ⫹C 共 a 1 ,a 2 ,a 3⬘ ,a 4⬘ 兲
⫹C 共 a 1 ,a 2⬘ ,a 3 ,a 4⬘ 兲 ⫹C 共 a 1⬘ ,a 2 ,a 3 ,a 4⬘ 兲
⫹C 共 a 1 ,a 2⬘ ,a 3⬘ ,a 4 兲 ⫹C 共 a 1⬘ ,a 2 ,a 3⬘ ,a 4 兲

022106-3
P. van LOOCK AND SAMUEL L. BRAUNSTEIN PHYSICAL REVIEW A 63 022106

⫹C 共 a 1⬘ ,a ⬘2 ,a 3 ,a 4 兲 ⫺C 共 a 1⬘ ,a ⬘2 ,a ⬘3 ,a 4 兲 tion is that in all four cases (N⫽2,3,4,5), violations occur for

⫺C 共 a 1⬘ ,a ⬘2 ,a 3 ,a ⬘4 兲 ⫺C 共 a ⬘1 ,a 2 ,a ⬘3 ,a 4⬘ 兲
⫺C 共 a 1 ,a 2⬘ ,a 3⬘ ,a ⬘4 兲 ⫺C 共 a 1 ,a 2 ,a 3 ,a 4 兲
⫺C 共 a 1⬘ ,a ⬘2 ,a ⬘3 ,a 4⬘ 兲 兩 ⭐2.
It is symmetric among all four parties as any inequality de-
rived from Eqs. 共7兲–共9兲 is symmetric among all parties. For
the settings ␣ 1 ⫽ ␣ 2 ⫽ ␣ 3 ⫽ ␣ 4 ⫽0 and ␣ 1⬘ ⫽ ␣ , ␣ 2⬘ ⫽ ␤ , ␣ 3⬘
⫽ ␥ , and ␣ 4⬘ ⫽ ␦ , complying with local realism implies 兩 B4 兩
⭐2, where
B4 ⫽ 12 关 ⌸ 共 0,0,0,␦ 兲 ⫹⌸ 共 0,0,␥ ,0兲 ⫹⌸ 共 0,␤ ,0,0 兲 ⫹⌸ 共 ␣ ,0,0,0兲
⫹⌸ 共 0,0,␥ , ␦ 兲 ⫹⌸ 共 0,␤ ,0,␦ 兲 ⫹⌸ 共 ␣ ,0,0,␦ 兲
⫹⌸ 共 0,␤ , ␥ ,0兲 ⫹⌸ 共 ␣ ,0,␥ ,0兲 ⫹⌸ 共 ␣ , ␤ ,0,0 兲
⫺⌸ 共 ␣ , ␤ , ␥ ,0兲 ⫺⌸ 共 ␣ , ␤ ,0,␦ 兲 ⫺⌸ 共 ␣ ,0,␥ , ␦ 兲
⫺⌸ 共 0,␤ , ␥ , ␦ 兲 ⫺⌸ 共 0,0,0,0兲 ⫺⌸ 共 ␣ , ␤ , ␥ , ␦ 兲兴 .
Similarly, for N⫽5 one finds
B5 ⫽ 12 关 ⌸ 共 0,0,0,␦ , ⑀ 兲 ⫹⌸ 共 0,0,␥ ,0,⑀ 兲 ⫹⌸ 共 0,␤ ,0,0,⑀ 兲
⫹⌸ 共 ␣ ,0,0,0,⑀ 兲 ⫹⌸ 共 0,0,␥ , ␦ ,0兲 ⫹⌸ 共 0,␤ ,0,␦ ,0兲
⫹⌸ 共 ␣ ,0,0,␦ ,0兲 ⫹⌸ 共 0,␤ , ␥ ,0,0 兲 ⫹⌸ 共 ␣ ,0,␥ ,0,0 兲
any nonzero squeezing. This requires the presence of
N-partite entanglement for any nonzero squeezing, which is
consistent with the results in Ref. 关13兴. Moreover, we see
that not only for large squeezing but also for modest finite
共20兲 squeezing, the significance of the violations 共at optimal dis-
placements J) grows with increasing N.
In the following, we will examine the general case of N
parties. How does the maximum violation of the Bell-type
inequalities derived with the continuous-variable GHZ states
in general evolve with increasing number of parties, in par-
ticular, compared to the exponential growth for the qubit
GHZ states 关11,12兴? At least for N⭐5, the maximum viola-
tion grows, and this growth does not appear to be exponen-
tially large, but rather seems to decrease. This conjecture has
not been proven, since we did not consider all possible set-
tings 共all possible combinations of ␣ i and ␣ i⬘ ). However,
there are strong hints that our choice of ␣ i ⫽0 and ␣ ⬘i
⫽i 冑J is near optimal. In particular, that the nonlocality is
always revealed for arbitrarily small squeezing 共any nonzero
共21兲
squeezing兲 lets our choice appear more appropriate than
other possible combinations. Having now much confidence
in the choice of settings that we used for small numbers of
parties, we will use the same settings for larger numbers of
parties.
Considering odd numbers of parties N, we find the follow-
ing expression for BN ,

⫹⌸ 共 ␣ , ␤ ,0,0,0兲 ⫺⌸ 共 ␣ , ␤ , ␥ , ␦ ,0兲 ⫺⌸ 共 ␣ , ␤ , ␥ ,0,⑀ 兲 (N⫺1)/2

⫺⌸ 共 ␣ , ␤ ,0,␦ , ⑀ 兲 ⫺⌸ 共 ␣ ,0,␥ , ␦ , ⑀ 兲 ⫺⌸ 共 0,␤ , ␥ , ␦ , ⑀ 兲 if N⫽3⫹8M : BN ⫽2 (3⫺N)/2 兺

k⫽0
共 ⫺1 兲 k

⫺⌸ 共 0,0,0,0,0 兲兴 , 共22兲

which has to statisfy 兩 B5 兩 ⭐2 and contains the same settings

as for N⫽4, but in addition ␣ 5 ⫽0 and ␣ ⬘5 ⫽ ⑀ .
⫻ 冉 N
2k⫹1 冊
⌸ 共 ␣ 1⬘ , ␣ ⬘2 , . . . , ␣ 2k⫹1
⬘ ,

We can now use the entangled states in Eq. 共15兲 with N

⫽4 and N⫽5 and apply the inequalities to them. For the ␣ 2k⫹2 , ␣ 2k⫹3 , . . . , ␣ N ), 共24兲
same reason as for N⫽3 共symmetry among all modes in the
states and in the inequalities兲, the choice ␣ ⫽ ␤ ⫽ ␥ ⫽ ␦ ⫽ ⑀
⫽i 冑J appears to be optimal 共maximizes positive terms and where the first 2k⫹1 arguments of ⌸ are ␣ 1⬘ ⫽ ␣ 2⬘ ⫽•••
minimizes negative contributions兲. ⬘ ⫽i 冑J, and the remaining ones are ␣ 2k⫹2 ⫽ ␣ 2k⫹3
⫽ ␣ 2k⫹1
With this choice, we obtain ⫽•••⫽ ␣ N ⫽0, and M ⫽0,1,2,3, . . . . Because of the sym-
metry of the states ⌸( ␣) in Eq. 共15兲, all possible permuta-
B4 ⫽2 exp共 ⫺2J cosh 2r⫹J sinh 2r 兲 ⫺2 exp共 ⫺6J cosh 2r tions of the (2k⫹1) ␣ ⬘i ’s with ␣ i⬘ ⫽i 冑J and the 关 N⫺(2k
⫹1) 兴 ␣ i ’s with ␣ i ⫽0 can be described by the same func-
⫺3J sinh 2r 兲 ⫹3 exp共 ⫺4J cosh 2r 兲
tion ⌸( ␣ 1⬘ , ␣ 2⬘ , . . . , ␣ 2k⫹1
⬘ , ␣ 2k⫹2 , ␣ 2k⫹3 , . . . , ␣ N ).
⫺ 12 exp共 ⫺8Je ⫹2r 兲 ⫺ 21 , Similarly, with the same settings ␣ ⬘i ⫽i 冑J and ␣ i ⫽0, and
again by exploiting symmetry, we obtain
B5 ⫽5 exp共 ⫺4J cosh 2r⫹4J sinh 2r/5兲
(N⫺1)/2
⫺ 52 exp共 ⫺8J cosh 2r⫺24J sinh 2r/5兲 ⫺ 21 . 共23兲
for N⫽5⫹8M : BN ⫽2 (3⫺N)/2
兺 共 ⫺1 兲 k⫹1
As shown in Fig. 1, the maximum violation of 兩 BN 兩 ⭐2 共for k⫽0

our particular choice of settings兲 grows with increasing num-

ber of parties N. The asymptotic analysis 共large r and small
J) yields for N⫽5: B max5 ⬇2.48 with Je
⫹2r
⫽5(ln 2)/24. At
a certain amount of large squeezing, smaller displacements J
⫻ 冉 冊
N
2k
⌸ 共 ␣ 1⬘ , ␣ 2⬘ , . . . , ␣ ⬘2k , ␣ 2k⫹1 ,

than for N⭐4 共at the same squeezing兲 are needed to ap-
proach this maximum violation. Another important observa- ␣ 2k⫹2 , . . . , ␣ N ), 共25兲

022106-4
GREENBERGER-HORNE-ZEILINGER NONLOCALITY IN . . . PHYSICAL REVIEW A 63 022106

FIG. 2. Maximum violations of the inequality

兩 BN 兩 ⭐2 imposed by local realistic theories in the
limit of large squeezing. BN is plotted as a func-
tion of A⬅Je ⫹2r for different N.

(N⫺1)/2
⫹8M and 兩 BN (r⫽0) 兩 ⬍2 if J⬎0. In the limit N→⬁, we
for N⫽7⫹8M : BN ⫽2 (3⫺N)/2
兺
k⫽0
共 ⫺1 兲 k⫹1 obtain BN (r⫽0)→0 for any J⬎0. Similar expressions as in
Eq. 共30兲 can be found for BN (r⫽0) in the other cases of odd

⫻ 冉 N
2k⫹1 冊
⌸ 共 ␣ 1⬘ , ␣ ⬘2 , . . . , ␣ 2k⫹1
⬘ ,
N, N⫽5⫹8M , N⫽7⫹8M , and N⫽9⫹8M , and in fact, no
violations occur. The inequality 兩 BN 兩 ⭐2 imposed by local
realistic theories always remains satisfied for zero squeezing.
␣ 2k⫹2 , ␣ 2k⫹3 , . . . , ␣ N ), On the other hand, inferring from the results for N⭐5
共26兲 parties, the maximum violations of 兩 BN 兩 ⭐2 occur for large
squeezing. Let us again consider the limit of large squeezing
(N⫺1)/2 ( cosh 2r⬇ sinh 2r⬇e⫹2r/2) and define A⬅Je ⫹2r . Now we
for N⫽9⫹8M : BN ⫽2 (3⫺N)/2 兺
k⫽0
共 ⫺1 兲 k can write Eq. 共28兲 and Eq. 共29兲 as

⌸ 共 ␣ 1⬘ , ␣ 2⬘ , . . . , ␣ 2k
⬘ , ␣ 2k⫹1 , ␣ 2k⫹2 , . . . , ␣ N 兲
⫻
N
2k冉 冊
⌸ 共 ␣ 1⬘ , ␣ 2⬘ , . . . , ␣ 2k
⬘ , ␣ 2k⫹1 , ⫽exp关 ⫺2A共 2k 兲 2 /N 兴 , 共31兲

␣ 2k⫹2 , . . . , ␣ N ). 共27兲 ⌸ 共 ␣ 1⬘ , ␣ 2⬘ , . . . , ␣ 2k⫹1

⬘ , ␣ 2k⫹2 , ␣ 2k⫹3 , . . . , ␣ N 兲

The functions concerned in these formulas are explicitly ⫽exp关 ⫺2A共 2k⫹1 兲 2 /N 兴 . 共32兲
given by 关see Eq. 共15兲兴 Figure 2 shows the maxima of the violations of 兩 BN 兩 ⭐2 共for
⌸ 共 ␣ 1⬘ , ␣ 2⬘ , . . . , ␣ 2k
⬘ , ␣ 2k⫹1 , ␣ 2k⫹2 , . . . , ␣ N 兲 our particular choice of settings兲, calculated with Eqs. 共24兲–
共27兲 and the asymptotic results from Eqs. 共31兲–共32兲 for large

再
⫽exp ⫺2J cosh 2r 共 2k 兲 ⫹2J sinh 2r
squeezing. The maximum violation grows from B max
for N⫽5 to B 85 max
5 ⬇2.48
⬇2.8 for N⫽85. Within this range, a maxi-

冋 册冎
mum violation near 2.8 is already attained with N⫽45 par-
共 2k 兲 2 ties and there is only a very small increase from N⫽45 to
⫻ 2k⫺2 , 共28兲
N N⫽85. On the other hand, between N⫽5 and N⫽9, the
maximum violation goes up from 2.48 to about 2.6, which is
⌸ 共 ␣ ⬘1 , ␣ 2⬘ , . . . , ␣ 2k⫹1
⬘ , ␣ 2k⫹2 , ␣ 2k⫹3 , . . . , ␣ N 兲 still significantly less than the increase between N⫽2

再 2 ⬇2.19) and N⫽5. This confirms our conjecture based

(B max
⫽exp ⫺2J cosh 2r 共 2k⫹1 兲 ⫹2J sinh 2r on the results for N⭐5: apparently, the maximum violation
indeed grows with increasing number of parties, but this

冋 册冎
共 2k⫹1 兲 2
⫻ 2k⫹1⫺2
N maximum emerging rather than a significant further increase
Let us first consider the case of zero squeezing, r⫽0. The
sum from Eq. 共24兲 becomes in this case
BN 共 r⫽0 兲 ⫽2 (3⫺N)/2共 1⫹e ⫺4J兲 N/2 sin关 N arctan共 e ⫺2J兲兴 .
共30兲
As expected, without squeezing, no violations of the Bell-
type inequalities are obtained for the unentangled, separable
N-mode states: we find BN (r⫽0)⫽2 if J⫽0 for any N⫽3
growth seems to continuously decrease for larger numbers of
. 共29兲 parties. In fact, from N⫽45 to N⫽85, we see a second local
of the absolute maximum violation.
In Fig. 3, calculated with Eqs. 共24兲–共27兲 and Eqs. 共28兲–
共29兲, violations of 兩 BN 兩 ⭐2 are compared between different
numbers of parties at certain amounts of squeezing of the
corresponding GHZ states. As stated earlier, the violations
grow with N also for modest finite squeezing, but this in-
crease is smaller than the increase of the maximum viola-
tions and becomes unrecognizable for small squeezing. An
illustrating example is that a violation comparable to the

022106-5
P. van LOOCK AND SAMUEL L. BRAUNSTEIN
maximum violation with the two-mode EPR state for large
2 ⬇2.19) can be attained with a five-mode
squeezing (B max
GHZ state built from five modestly squeezed states 共about
2.6 dB each兲.
We conclude with a summary and an assessment of our
results. We have considered pure multipartite entangled
states described by continuous quantum variables and shown
that they violate Bell-type inequalities imposed by local re-
alism. An experimental nonlocality test based on these states
and on our scheme is possible, but it would require detectors
capable of resolving the number of absorbed photons 关17兴.
Nevertheless, the N-mode states, which we have unambig-
uously proven to exhibit nonlocality, can be relatively easily
generated in practice, as opposed to the discrete-variable
GHZ states on which all current multiparty nonlocality
proofs rely. Furthermore, entangled N-mode states similar to
those considered here can even be produced using only one
single-mode squeezed vacuum state and linear optics instead
of N squeezed states 关13兴. Since it has been shown already
that the entangled two-mode state created this way is nonlo-
cal with respect to parity measurements 关18兴, one can apply
our analysis to the corresponding N-mode states and expect
that they too are nonlocal.
关1兴 C. H. Bennett et al., Phys. Rev. Lett. 70, 1895 共1993兲.
关2兴 A. K. Ekert, Phys. Rev. Lett. 67, 661 共1991兲.
关3兴 J. S. Bell, Physics 共Long Island City, N.Y.兲 1, 195 共1964兲.
关4兴 R. F. Werner, Phys. Rev. A 40, 4277 共1989兲.
关5兴 A. Furusawa et al., Science 282, 706 共1998兲.
关6兴 A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777
共1935兲.
关7兴 Z. Y. Ou, S. F. Pereira, H. J. Kimble, and K. C. Peng, Phys.
Rev. Lett. 68, 3663 共1992兲; B. Yurke and D. Stoler, ibid. 79,
4941 共1997兲; A. Gilchrist, P. Deuar, and M. D. Reid, Phys.
Rev. Lett. 80, 3169 共1998兲; W. J. Munro, Phys. Rev. A 59,
4197 共1999兲.
关8兴 K. Banaszek and K. Wodkiewicz, Phys. Rev. A 58, 4345
共1998兲.
关9兴 J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt, Phys.
Rev. Lett. 23, 880 共1969兲.
关10兴 D. M. Greenberger, M. A. Horne, A. Shimony, and A.
022106-6
PHYSICAL REVIEW A 63 022106
FIG. 3. Violations of the inequality 兩 BN 兩 ⭐2
imposed by local realistic theories for different N
at certain amounts of squeezing of the N-mode
GHZ states: r⫽0.1 (⬇0.9 dB), r⫽0.3
(⬇2.6 dB), r⫽0.8 (⬇6.9 dB), and r⫽1.5
(⬇13 dB). BN is plotted as a function of J. Note
that the axes of the displacement parameter J
vary in scale. The larger N becomes, the smaller
become the displacements required.
The degree of nonlocality of the continuous-variable GHZ
states, if represented by the maximum violation of the corre-
sponding Bell-type inequalities, seems to grow with increas-
ing number of parties. This growth, however, continuously
decreases for larger numbers of parties. Thus, the evolution
of the continuous-variable states’ nonlocality with increasing
number of parties and the corresponding evolution of nonlo-
cality for the qubit GHZ states are qualitatively equal but
quantitatively different 共with an exponential increase for the
qubits兲. The reason for this may be that the latter always
relies on maximally entangled states, whereas the former de-
pends on nonmaximally entangled states as long as the
squeezing remains finite. In fact, an observation of the non-
locality of the continuous-variable states requires small but
nonzero displacements J⬀e ⫺2r , which is not achievable
when the singular maximally entangled states for infinite
squeezing are considered.
This research was funded by a DAAD Doktorandensti-
pendium 共HSP III兲 and by the EPSRC Grant No. GR/
L91344. P.v.L. thanks T. C. Ralph, W. J. Munro, and A. K.
Pati for useful comments on the present paper and general
discussions concerning nonlocality.
Zeilinger, Am. J. Phys. 58, 1131 共1990兲.
关11兴 N. D. Mermin, Phys. Rev. Lett. 65, 1838 共1990兲.
关12兴 D. N. Klyshko, Phys. Lett. A 172, 399 共1993兲; N. Gisin and H.
Bechmann-Pasquinucci, ibid. 246, 1 共1998兲.
关13兴 P. van Loock and S. L. Braunstein, Phys. Rev. Lett. 84, 3482
共2000兲.
关14兴 S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 80, 869
共1998兲.
关15兴 A. Royer, Phys. Rev. A 15, 449 共1977兲; H. Moya-Cessa and P.
L. Knight, ibid. 48, 2479 共1993兲.
关16兴 This choice of two equal settings leads to the same result as the
choice by Banaszek and Wodkiewicz 共Ref. 关8兴兲 who used op-
posite signs: ␣ ⫽ 冑J and ␤ ⫽⫺ 冑J.
关17兴 K. Banaszek and K. Wodkiewicz, Phys. Rev. Lett. 82, 2009
共1999兲.
关18兴 J. W. Wu, Phys. Rev. A 61, 022111 共2000兲.