DOI: 10.1126/science.1136303
, 966 (2007); 315Science
et al.Vincent Jacques,
Delayed-Choice Gedanken Experiment
Experimental Realization of Wheeler's
www.sciencemag.org (this information is current as of March 6, 2007 ):
The following resources related to this article are available online at
http://www.sciencemag.org/cgi/content/full/315/5814/966
version of this article at:
including high-resolution figures, can be found in the onlineUpdated information and services,
http://www.sciencemag.org/cgi/content/full/315/5814/966/DC1
can be found at: Supporting Online Material
http://www.sciencemag.org/cgi/content/full/315/5814/966#otherarticles
, 1 of which can be accessed for free: cites 7 articlesThis article
http://www.sciencemag.org/cgi/collection/physics
Physics
: subject collectionsThis article appears in the following
http://www.sciencemag.org/about/permissions.dtl
in whole or in part can be found at: this article
permission to reproduce of this article or about obtaining reprintsInformation about obtaining
registered trademark of AAAS.
c 2007 by the American Association for the Advancement of Science; all rights reserved. The title SCIENCE is a
CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.
Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
o
n
M
ar
ch
6
, 2
00
7
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m
A similar replacement of the active subset is not
apparent in the dentate gyrus, which suggests
that the change in the CA3 code is triggered by
direct projections from entorhinal grid cells to
the CA3 (32).
Pattern separation in the dentate gyrus is
thus different from separation processes in the
cerebellum (10, 11), where signals from the
brain stem spread out on a layer of granule cells
whose cell numbers exceed those of the input
layer by a factor of several million. The number
of granule cells in the dentate gyrus and pyram-
idal cells in the CA3 only marginally out-
numbers the projection neurons from layer II
of the entorhinal cortex [in the rat, 1,000,000,
300,000 and 200,000, respectively (15, 35, 36)],
which suggests that the same hippocampal
cells must participate in many representa-
tions even when the population activity is
sparse (13, 14). In such networks, orthogonal-
ization of coincidence patterns may be more
effective.
The decorrelated firing of the dentate cells
contrasts with the invariant discharge structure of
grid cells upstream in the medial entorhinal
cortex (30–32) (Fig. 4). The reduction in spatio-
temporal coincidence could be derived from the
lateral entorhinal cortex, but not by a straight-
forward relay mechanism, because cells in this
area do not exhibit reliable place modulation
(37). It is thus likely that many of the underlying
computations take place within the dentate gyrus
itself. The use of a dedicated neuronal popula-
tion for orthogonalization of small differences
in input to the CA fields enables the hippocam-
pal network to encode the full variety of expe-
rience in a more diversified manner than what
could be accomplished with attractor networks
alone.
References and Notes
1. B. L. McNaughton, L. Nadel, in Neuroscience and
Connectionist Theory, M. A. Gluck and D. E. Rumelhart,
Eds. (Lawrence Erlbaum, Hillsdale, NJ, 1989), pp. 1–63.
2. A. Treves, E. T. Rolls, Hippocampus 2, 189 (1992).
3. R. C. O'Reilly, J. L. McClelland, Hippocampus 4, 661
(1994).
4. P. E. Gilbert, R. P. Kesner, I. Lee, Hippocampus 11, 626
(2001).
5. D. Marr, Philos. Trans. R. Soc. Lond. B Biol. Sci. 262, 23
(1971).
6. M. E. Hasselmo, E. Schnell, E. Barkai, J. Neurosci. 15,
5249 (1995).
7. M. Tsodyks, Hippocampus 9, 481 (1999).
8. K. Nakazawa et al., Science 297, 211 (2002).
9. I. Lee, D. Yoganarasimha, G. Rao, J. J. Knierim, Nature
430, 456 (2004).
10. D. Marr, J. Physiol. 202, 437 (1969).
11. J. Albus, Math. Biosci. 10, 25 (1971).
12. P. Chadderton, T. W. Margrie, M. Hausser, Nature 428,
856 (2004).
13. M. W. Jung, B. L. McNaughton, Hippocampus 3, 165
(1993).
14. M. K. Chawla et al., Hippocampus 15, 579 (2005).
15. D. G. Amaral, N. Ishizuka, B. Claiborne, Prog. Brain Res.
83, 1 (1990).
16. J. O’Keefe, J. Dostrovsky, Brain Res. 34, 171 (1971).
17. R. U. Muller, J. L. Kubie, J. B. Ranck Jr., J. Neurosci. 7,
1935 (1987).
18. R. U. Muller, J. L. Kubie, J. Neurosci. 7, 1951 (1987).
19. G. J. Quirk, R. U. Muller, J. L. Kubie, J. Neurosci. 10,
2008 (1990).
20. E. Bostock, R. U. Muller, J. L. Kubie, Hippocampus 1, 193
(1991).
21. E. J. Markus et al., J. Neurosci. 15, 7079 (1995).
22. C. Lever, T. Wills, F. Cacucci, N. Burgess, J. O'Keefe,
Nature 416, 90 (2002).
23. T. J. Wills, C. Lever, F. Cacucci, N. Burgess, J. O'Keefe,
Science 308, 873 (2005).
24. S. Leutgeb et al., Science 309, 619 (2005).
25. Materials and Methods and other supporting material are
available on Science Online.
26. J. K. Leutgeb et al., Neuron 48, 345 (2005).
27. D. A. Henze, L. Wittner, G. Buzsáki, Nat. Neurosci. 5, 790
(2002).
28. K. D. Harris, J. Csicsvari, H. Hirase, G. Dragoi, G. Buzsáki,
Nature 424, 552 (2003).
29. R. M. Hayman, S. Chakraborty, M. I. Anderson,
K. J. Jeffery, Eur. J. Neurosci. 18, 2825 (2003).
30. M. Fyhn, S. Molden, M. P. Witter, E. I. Moser,
M.-B. Moser, Science 305, 1258 (2004).
31. T. Hafting, M. Fyhn, S. Molden, M.-B. Moser, E. I. Moser,
Nature 436, 801 (2005).
32. M. H. Fyhn, T. F. Hafting, A. Treves, E. I. Moser,
M.-B. Moser, Nature, in press.
33. S. Leutgeb, J. K. Leutgeb, A. Treves, M.-B. Moser,
E. I. Moser, Science 305, 1295 (2004).
34. A. Vazdarjanova, J. F. Guzowski, J. Neurosci. 24, 6489
(2004).
35. P. R. Rapp, P. S. Deroche, Y. Mao, R. D. Burwell,
Cereb. Cortex 12, 1171 (2002).
36. B. D. Boss, G. M. Peterson, W. M. Cowan, Brain Res. 338,
144 (1985).
37. E. L. Hargreaves, G. Rao, I. Lee, J. J. Knierim, Science
308, 1792 (2005).
38. We thank A. Treves, C. A. Barnes, and M. R. Mehta for
discussion and A. M. Amundsgard, K. Haugen, K. Jenssen,
E. Sjulstad, R. Skjerpeng, and H. Waade for technical
assistance. This work was supported by a Centre of
Excellence grant from the Norwegian Research Council.
Supporting Online Material
www.sciencemag.org/cgi/content/full/315/5814/961/DC1
Materials and Methods
SOM Text
Figs. S1 to S12
Tables S1 and S2
References
2 October 2006; accepted 15 December 2006
10.1126/science.1135801
REPORTS
Experimental Realization of Wheeler’s
Delayed-Choice Gedanken Experiment
Vincent Jacques,1 E Wu,1,2 Frédéric Grosshans,1 François Treussart,1 Philippe Grangier,3
Alain Aspect,3 Jean-François Roch1*
Wave-particle duality is strikingly illustrated by Wheeler’s delayed-choice gedanken experiment,
where the configuration of a two-path interferometer is chosen after a single-photon pulse has
entered it: Either the interferometer is closed (that is, the two paths are recombined) and the
interference is observed, or the interferometer remains open and the path followed by the photon
is measured. We report an almost ideal realization of that gedanken experiment with single
photons allowing unambiguous which-way measurements. The choice between open and closed
configurations, made by a quantum random number generator, is relativistically separated from
the entry of the photon into the interferometer.
Young’s double-slit experiment, realizedwith particles sent one at a time throughan interferometer, is at the heart of
quantum mechanics (1). The striking feature is
that the phenomenon of interference, interpreted
as a wave following two paths simultaneously, is
incompatible with our common-sense represen-
tation of a particle following one route or the
other but not both. Several single-photon inter-
ference experiments (2–6) have confirmed the
wave-particle duality of the light field. To un-
derstand their meaning, consider the single-
photon interference experiment sketched in
Fig. 1. In the closed interferometer configuration,
a single-photon pulse is split by a first beam-
splitter BSinput of a Mach-Zehnder interferometer
and travels through it until a second beamsplitter
BSoutput recombines the two interfering arms.
When the phase shift F between the two arms is
varied, interference appears as a modulation of
the detection probabilities at output ports 1 and 2,
respectively, as cos2 F and sin2 F. This result is
the one expected for a wave, and as Wheeler
pointed out, “[this] is evidence … that each ar-
1Laboratoire de Photonique Quantique et Moléculaire,
Ecole Normale Supérieure de Cachan, UMR CNRS 8537,
94235 Cachan, France. 2Key Laboratory of Optical and
Magnetic Resonance Spectroscopy, East China Normal
University, 200062 Shanghai, China. 3Laboratoire Charles
Fabry de l’Institut d’Optique, Campus Polytechnique, UMR
CNRS 8501, 91127 Palaiseau, France.
*To whom correspondence should be addressed. E-mail:
roch@physique.ens-cachan.fr
16 FEBRUARY 2007 VOL 315 SCIENCE www.sciencemag.org966
o
n
M
ar
ch
6
, 2
00
7
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m
riving light quantum has arrived by both routes”
(7). If BSoutput is removed (the open configu-
ration), each detector D1 or D2 on the output
ports is then associated with a given path of the
interferometer, and, provided one uses true single-
photon light pulses, “[either] one counter goes off,
or the other. Thus the photon has traveled only
one route” (7). Such an experiment supports
Bohr’s statement that the behavior of a quantum
system is determined by the type of measurement
performed on it (8). Moreover, it is clear that for
the two complementary measurements consid-
ered here, the corresponding experimental set-
tings are mutually exclusive; that is, BSoutput
cannot be simultaneously present and absent.
In experiments where the choice between
the two settings is made long in advance, one
could reconcile Bohr’s complementarity with
Einstein’s local conception of the physical
reality. Indeed, when the photon enters the inter-
ferometer, it could have received some “hidden
information” on the chosen experimental con-
figuration and could then adjust its behavior
accordingly (9). To rule out that too-naïve
interpretation of quantum mechanical comple-
mentarity, Wheeler proposed the “delayed-
choice” gedanken experiment in which the
choice of which property will be observed is
made after the photon has passed BSinput: “Thus
one decides the photon shall have come by one
route or by both routes after it has already done
its travel” (7).
Since Wheeler’s proposal, several delayed-
choice experiments have been reported (10–15).
However, none of them fully followed the
original scheme, which required the use of the
single-particle quantum state as well as rel-
ativistic space-like separation between the choice
of interferometer configuration and the entry of
the particle into the interferometer. We report the
realization of such a delayed-choice experiment
in a scheme close to the ideal original proposal
(Fig. 1). The choice to insert or remove BSoutput
is randomly decided through the use of a quan-
tum random number generator (QRNG). The
QRNG is located close to BSoutput and is far
enough from the input so that no information
about the choice can reach the photon before it
passes through BSinput.
Our single-photon source, previously devel-
oped for quantum key distribution (16, 17), is
based on the pulsed, optically excited photo-
luminescence of a single nitrogen-vacancy (N-V)
color center in a diamond nanocrystal (18). At the
single-emitter level, these photoluminescent cen-
ters, which can be individually addressed with
the use of confocal microscopy (19), have shown
unsurpassed efficiency and photostability at room
temperature (20, 21). In addition, it is possible to
obtain single photons with a well-defined polar-
ization (16, 22).
The delayed-choice scheme is implemented
as follows. Linearly polarized single photons are
sent by a polarization beamsplitter BSinput
through an interferometer (length 48 m) with
two spatially separated paths associated with
orthogonal S and P polarizations (Fig. 2). The
movable output beamsplitter BSoutput consists of
the combination of a half-wave plate, a polariza-
tion beamsplitter BS′, an electro-optical modula-
tor (EOM) with its optical axis oriented at 22.5°
from input polarizations, and a Wollaston prism.
The two beams of the interferometer, which are
spatially separated and orthogonally polarized,
are first overlapped by BS′ but can still be
unambiguously identified by their polarization.
Then, the choice between the two interferometer
configurations, closed or open, is realized with
the EOM, which can be switched between two
different configurations within 40 ns by means of
a homebuilt fast driver (16): Either no voltage is
applied to the EOM, or its half-wave voltageVp is
applied to it. In the first case, the situation
corresponds to the removal of BSoutput and the
two paths remain uncombined (open configu-
ration). Because the original S and P polar-
izations of the two paths are oriented along prism
polarization eigenstates, each “click” of one
detector D1 or D2 placed on the output ports is
associated with a specific path (path 1 or path 2,
respectively). When the Vp voltage is applied, the
EOM is equivalent to a half-wave plate that
rotates the input polarizations by an angle of 45°.
The prism then recombines the two rotated
polarizations that have traveled along different
optical paths, and interference appears on the two
output ports. We then have the closed interfer-
ometer configuration (22).
To ensure the relativistic space-like separation
between the choice of the interferometer config-
uration and the passage of the photon at BSinput,
we configured the EOM switching process to be
randomly decided in real time by the QRNG
located close to the output of the interferometer
(48 m from BSinput). The random number is gen-
erated by sampling the amplified shot noise of a
white-light beam. Shot noise is an intrinsic
quantum random process, and its value at a
given time cannot be predicted (23). The timing
of the experiment ensures the required rel-
ativistic space-like separation (22). Then, no
information about the interferometer configu-
ration choice can reach the photon before it
enters the interferometer.
The single-photon behavior was first tested
using the two output detectors feeding single and
Fig. 1. Wheeler’s delayed-choice gedanken
experiment proposal. The choice to intro-
duce or remove beamsplitter BSoutput (closed
or open configuration) is made only after
the passage of the photon at BSinput , so
that the photon entering the interfer-
ometer “cannot know” which of the two
complementary experiments (path differ-
ence versus which-way) will be performed
at the output.
Path 1
Path 2
detectors
BSinput
Single-photon
pulse
mirror
mirror
D2
BSoutput
D2
D1
Fig. 2. Experimental realization of
Wheeler’s gedanken experiment. Single
photons emitted by a single N-V color
center are sent through a 48-m polar-
ization interferometer, equivalent to a
time of flight of about 160 ns. A binary
random number 0 or 1, generated by
the QRNG, drives the EOM voltage
between V = 0 and V = Vp within
40 ns, after an electronic delay of
80 ns. Two synchronized signals from
the clock are used to trigger the single-
photon emission and the QNRG. In the
laboratory frame of reference, the
random choice between the open and
the closed configuration is made simul-
taneously with the entry of the photon
into the interferometer. Taking advantage of the fact that the QNRG is located at the output of the interferometer, such timing ensures that the photon enters the
future light cone of the random choice when it is at about the middle of the interferometer, long after passing BSinput.
EOM
WPBS'
N-V color
center
S
P /2
N
N
N
48 m
trigger
pulses
c
2
1
BSinput
QRNG
VEOM = Vπ
VEOM = 0
-40
0
40
N
oi
se
(m
V)
1086420
Time (µs)
1
R
an
do
m
BSoutput
Path 2
Path 1
D1D1
D2D2
single
photon
4.2 MHz
CLOCK
www.sciencemag.org SCIENCE VOL 315 16 FEBRUARY 2007 967
REPORTS
o
n
M
ar
ch
6
, 2
00
7
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m
coincidence counters with BSoutput removed
(open configuration). We used an approach
similar to the one described in (2) and (6).
Consider a run corresponding toNT trigger pulses
applied to the emitter, with N1 counts detected in
path 1 of the interferometer by D1, N2 counts
detected in path 2 by D2, and NC detected co-
incidences corresponding to joint photodetec-
tions on D1 and D2 (Fig. 2). Any description in
which light is treated as a classical wave, such as
the semiclassical theory with quantized photo-
detectors (24), predicts that these numbers of
counts should obey the inequality
a ¼ NC � NT
N1 � N2 ≥ 1 (1)
Violation of this inequality thus gives a quanti-
tative criterion that characterizes nonclassical
behavior. For a single-photon wavepacket, quan-
tum optics predicts perfect anticorrelation (i.e.,
a = 0) in agreement with the intuitive image that
a single particle cannot be detected simulta-
neously in the two paths of the interferometer
(2). We measured a = 0.12 ± 0.01, hence we are
indeed close to the pure single-photon regime.
The nonideal value of the a parameter is due to
residual background photoluminescence of the
diamond sample and to the two-phonon Raman
scattering line, which both produce uncorrelated
photons with Poissonian statistics (6).
With single-photon pulses in the open
configuration, we expected each detector D1
and D2 to be unambiguously associated with a
given path of the interferometer. To test this
point, we evaluated the “which-way” information
parameter I = (N1 − N2)/(N1 + N2) (25–28) by
blocking one path (e.g., path 2) and measur-
ing the counting rates at D1 and D2. Avalue of
I higher than 0.99 was measured, limited by
detector dark counts and residual imperfections
of the optical components. The same value was
obtained when the other path was blocked (e.g.,
path 1). In the open configuration, we thus have
an almost ideal which-way measurement.
The delayed-choice experiment itself is
performed with the EOM randomly switched
for each photon sent into the interferometer,
corresponding to a random choice between the
open and closed configurations. The phase shift
F between the two interferometer arms is varied
by tilting the second polarization beamsplitter
BS′with a piezoelectric actuator (PZT). For each
photon, we recorded the chosen configuration,
the detection events, and the PZT position. All
raw data were saved in real time and were pro-
cessed only after a run was completed. For each
PZT position, detection events on D1 and D2
corresponding to each configuration were sorted
(Fig. 3). In the closed configuration, we observed
interference with 0.94 visibility. We attribute
the departure from unity to an imperfect overlap
of the two interfering beams. In the open con-
figuration, interference totally disappears, as
evidenced by the absence of modulation in
the two output ports when the phase shift F
was varied. We checked that in the delayed-
choice configuration, parameters a and I kept
the same values as measured in the prelimi-
nary tests presented above.
Our realization of Wheeler’s delayed-choice
gedanken experiment demonstrates that the
behavior of the photon in the interferometer
depends on the choice of the observable that is
measured, even when that choice is made at a
position and a time such that it is separated from
the entrance of the photon into the interferometer
by a space-like interval. In Wheeler’s words, as
no signal traveling at a velocity less than that of
light can connect these two events, “we have a
strange inversion of the normal order of time.
We, now, by moving the mirror in or out have an
unavoidable effect on what we have a right to say
about the already past hi