1.1.

QUANTUM OPTICS
Hyperfine Interactions 37 (1987) 3-18 3

WAVE-PARTICLE DUALITY FOR SINGLE P H O T O N S

Alain ASPECT * and Philippe G R A N G I E R

lnstitut d'Optique Th~orique et AppliquJe, Bat. 503, UniversitJ Paris Sud, F91405 Orsay, France

By use of an atomic radiative cascade, we have been able to produce single photon states o f
the light. This light has been submitted to two different experiments. The first o n e s h o w s an
anticorrelation on both sides of a beam splitter (corpuscle-like behaviour). The second o n e
yields interferences (wave-like behaviour).

1. Introduction

Wave particle duality for the photon is considered a typical example of

quantum behaviour. As so, it is often introduced at the very be~nning of
elementary courses in quantum mechanics, where it is presented as a well
established fact. More generally, it is a widely spread belief that all the famous
historical "single photon interference experiments" are a definite experimental
evidence for the wave-particle duality of light.
We will first show that the situation is certainly not so clear. There is no doubt
that these experiments have shown interferences, even with very weak light,
evidencing the wave-like behaviour. But we will argue that there was no un-
questionnable evidence of a particle-like behaviour. More, in the light of the
modem discussions about non-classical effects in the statistical properties of light
[1,2], one can claim that it would have been impossible to observe any particle-like
behaviour. This claim is related to the type of source used in all these experi-
ments, producing chaotic light, for which it has now been recognized that there is
no non-classical effect (a non-classical effect is an effect that cannot be interpre-
ted by a model in which the light is treated by a classical field, represented by c.
numbers).
We will then develop the idea that light produced in "single photon states"
would exhibit a clear particle-like behaviour, anticorrelation on a beam-splitter.
A quantitative criterium will be produced, allowing to show that such a be-
haviour is incompatible with any classical model of the light.
We have been able to build a source of single photon pulses of light. This
source has allowed to observe the particle like behaviour, characteristic of single
photon states. We have then used the same source for an interference experiment,
that we thus think correct to call a single photon interference experiment [3].

* Now with: Coll6ge de France, Laboratoire de Spectroscopie Hertzienne de I'ENS, 24, rue
Lhomond, F75005 Paris, France.

9 J.C. Baltzer A.G. Scientific Publishing Company

4 A. Aspect, P. Grangier / Wave-particle duality

2. Feeble light interference experiments

Very early in the development of q u a n t u m mechanics, the question was raised

of knowing whether " a photon interferes with itself". M a n y experiments have
been carried out in this context (see table 1), after the first experiment by Taylor
(1909), where he made a photograph of a diffraction pattern with six m o n t h s of
exposure time. Although some conflicting results exist, almost all these experi-
ments have shown that the visibility of an interference pattern does not change
when the intensity of the light is strongly attenuated. The various authors have
thus concluded that " a photon interferes with itself" [4].
Their argument generally amounts to evaluating the energy flux q~ in the
interferometer, from which a p h o t o n flux r is calculated. They then show
that, in the average, there is less than one p h o t o n at a time in the interferometer.
Assuming the statistical independence between the emissions, the probability that
two photons are simultaneously present in the system is thus weaker than the
probability of presence of one photon. The conclusion is that the observed effect
is essentially due to single photons.

Table 1
Feeble light interference experiments. All these experiments have been realized with attenuated light
from a usual source (atomic discharge).
Author Date Experiment Detector Photon Interferences
flux (s- 1)
Taylor (a) 1909 Diffraction Photography 10 6 Yes
Dempster et al. (b) 1927 (i) Grating Photography 10 2 Yes
(ii) Fabry Prrot Photography 105 Yes
Janossy et al. (c) 1957 Michelsoninterferometer Photomultiplier 105 Yes
Griffiths (d) 1963 Youngslits Image interferometer 2 x 103 Yes
Scarlet al. (e) 1968 Youngs l i t s Photomultiplier 2• 104 Yes
Donstov et al. (f) 1967 Fabry P~rot Image intensifier 10 3 No
Reynolds et al. (g) 1969 Fabry P&ot Image intensifier 102 Yes
Bozec et al. (h) 1969 Fabry Prrot Photography 102 Yes
Grishaev et al. (i) 1969 Jamininterferometer Imageintensifier 10 3 Yes
(a) G.I. Taylor, Prec. Cambridge Philos. Soc. 15 (1909) 114.
(b) A.J. Dempster and H.F. Batho, Phys. Rev. 30 (1927) 644.
(c) L. Janossy and Z. Naray, Acta Phys. Hungaria 7 (1967) 403.
(d) H.M. Griffiths, Princeton University Senior Thesis (1963).
(e) G.T. Reynoldset al., Advancesin electronics and electron physics 28 B (Academic Press, London,
1969).
(f) Y.P. Dontsov and A.I. Baz, Sov. Phys. JETP 25 (1967) 1.
(g) G.T. Reynolds, K. Spartalian and D.B. Scarl, Nuovo Cim. B 61 (1969) 355.
(h) P. Bozec, M. Cagnet and G. Roger, C.R. Acad. Sci. 269 (1969) 883.
(i) A. Grishaev et al., Sov. Phys. JETP 32 (1969) 16.
 A. Aspect, P. Grangier / Wave-particleduality 5

Ee 1//////////S le>

Eg I
Ig>

Fig. 1. Model of detector for the photoelectric effect. The atom has a ground state Ig), and a
continuum of ionized states. In state Ie), the free electron has a kinetic energy Ec - Eg - WT.

This argument is correct in its great context, as soon as the concept of photons
is accepted, i.e. if one admits that a light beam is constituted of quanta of energy
h v. But the question is still open to know whether it is necessary to introduce this
concept of photon. It is generally admitted that the discrete nature of the
detection process is an evidence of the concept of photon. More specifically,
Einstein's interpretation of the photoelectric effect is often considered a proof of
the necessity of describing the light as composed of photons. In fact, it is an
elementary exercice in q u a n t u m mechanics to show that all the characteristics of
the photoelectric effect can be obtained with a model where a quantized detector
interacts with a classical electromagnetic field [1,5].
Let us for instance consider a simple model where the detector is an atom with
a ground state l g ) and a continuum of ionized excited states l e), with a gap W T
(fig. 1). This atom interacts with a classical electromagnetic field via an electric
dipole interaction, re~resented by the Hamiltonian d'. D, where the electric field
is a number, and D is the atomic electric dipole operator. If d ~ is a field with
an amplitude ~'0 oscillating at the angular frequency ~, we can easily find the
transition rate from state I g ) to state l e), by use of Fermi's Golden Rule:

d ,~g --~ e =2h I(elblg)12"e2"p(e)8(E~-Eg-h~

dt (1)

Integration over the final states would yield the total ionization rate, but formula
(1) embodies all the features of the photoelectric effect. The existence of a
threshold is related to the fact that the density of excited states, p(e), is zero if
E e < Eg + W T. The probability of detection is proportional to the intensity ~2.
The final energy of the system is Ee = E g + ho~, i.e. the kinetic energy of the
electron is h t ~ - W T. Finally, the fact that the photodetection is constituted by
discrete events can be attributed to the quantization of the detector.
To conclude this discussion, we can say that there is no logical necessity to
introduce the concept of photon to describe these experiments of interference
with weak light. It is enough to use a model where the light is treated as a
6 A. Aspect, P. Grangier / Wave-particle duality

classical electromagnetic field (for which we are not surprised to have inter-
ference) and where the detector is quantized.
One may argue that we know n o w that there exist some effects - called
non-classical properties of light [1], [6,9] - that cannot be interpreted by such
semi-classical models. But these effects, observable on higher order intensity
correlation functions, can only be observed for light emitted in certain particular
states. A n d it is well k n o w n that the light emitted by usual sources - such as used
in the interference experiments under discussion - has absolutely no non-classi-
cal feature, even in weak intensity beams. In other words, it is i m p o r t a n t to
realize that a classical light (emitted by a usual source, or a laser) has no
non-classical features, even if strongly attenuated. In particular, it is definitely
different from fight emitted in single-photon states, as we discuss it now.

3. Single photon behaviour: anticorrelation on a beam-splitter

All the observed non-classical properties of light are related to second order
(in intensity) coherence properties of light. But there had still been no test of the
conceptually very simple situation dealing with single-photon states of the light
impinging on a b e a m splitter. In this case, the q u a n t u m theory of light predicts a
perfect anticorrelation for detections on both sides of the b e a m splitter (a single
p h o t o n can only be detected once).. On the other hand, any description involving
classical waves would predict some a m o u n t of coincidences, since a wave is
divided on a b e a m splitter. It is thus possible to characterize experimentally
single p h o t o n behaviour, by the observation of an anticorrelation between the
detections on both sides of the beam splitter.
In order to find a more quantitative criterium, we consider a source emitting
light pulses, which impinge on a b e a m splitter (fig. 2). Synchronized with each
pulse, we have a gate which enables the photodetectors for a detection time w

PM~9/
m

Nr

Nc

TRG
IGER
F--"~ / Nt

t PMt
NI

Fig. 2. Study of detection correlations after a beam-splitter. The source s emits light pulses, which
impinge on a beam-splitter. A triggering system produces a gate of duration W, synchronized with the
light pulse. The detections are authorized only during the gates. A coincidence is counted when both
PM r and PM t register a detection during the same gate.
 A. Aspect, P. Grangier / Wave-particle duality 7

overlapping the arrival of the corresponding pulse. During that gate, one moni-
tors singles detections in the transmitted or reflected channels, and a coincidence
if both channels register a count during the same gate. Denoting N 1 the rate of
gates, N r and N t the rates of singles counts, and Nr the rate of coincidences, we
obtain the probabilities for a single count during a gate
Pt = Nt/N1 Pr = Nr/N1 (2a)
and the probability for a coincidence
Pc = UJU~ (2b)
For a classical-wave description of the light, the intensity I(t) is divided on the
beam splitter into a reflected and a transmitted part. The probabilities of
photodetection during the n-th gate are proportional to the average intensity
during this gate

i. = --I i i . + w i (t ) dt. (3)

W

Denoting by brackets an average over the ensemble of gates, we have

Pt = a t w ( i . ) Pr = arW(i.) (4a)
where ar and a t are overall detection efficiencies, including the splitting coeffi-
cient on the beam splitter.
Restraining this study to the case when the singles probabilities are small
compared to one, the coincidence probability then writes
Pc = ata~w 2 Q-2. ) . (4b)
The standard Cauchy-Schwartz inequality
(i2) >~ (i.)2 (5)
holds for the bracket average. Therefore, for any classical-wave description of the
experiment of fig. 2, we expect
(6a)
or equivalently

a>/1 with a = - - NcN~ (6b)

NrN,"
The intuitive meaning of this inequality is clear. For a classical wave divided
on the beam splitter, there is a minimum rate of coincidences, corresponding to
the "accidental coincidences".
We have thus obtained a criterium for empirically characterizing a single
particle behaviour of light pulses. The violation of inequality (6) will indicate that
8 A. Aspect, P. Grangier / Wave-particle duality

the light pulses should not be described as wave packets divided on a beam
splitter but rather as single photons that cannot be detected simultaneously on
both sides of the beam splitter.

4. Anticorrelation on a beam-splitter: experiments

We have built an experimental setup corresponding to the scheme of fig. 2, i.e.

allowing to measure singles and coincidences on the two sides of a beam-splitter,
during gates triggered by events synchroneous with the light pulses. This system
has been used to study light pulses emitted by a classical source (4a). Then, we
have studied pulses from a source designed to emit single-photons wave packets,
as explained in 4b.

4a. ATTENUATED CLASSICAL SOURCE

In order to confirm experimentally our arguments, and also to test the p h o t o n

counting system, we have first studied light from a pulsed photodiode. It
produced light pulses with a rise time of 1.5 ns and a fall time about 6 ns. The
gates, triggered by the electric pulses driving the photodiode, were 9 ns wide (as
in 4b) and overlapped almost completely the light pulses.
The source was attenuated to a level corresponding to one detection per 1000
emitted pulses. With a detector quantum efficiency about 10%, the average
energy per pulse can be estimated to about 0.01 photon. In the context of table 1,
this source would certainly have been considered a source of single photons.
Table 2 shows the results of the anticorrelation measurements. The quantity t~
(of inequality (6)) is consistently found equal to 1, i.e. no anticorrelation is
observed.
This experiment thus supports the claim that light emitted by an attenuated
classical source does not exhibit a single photon behaviour on a beam-splitter.
This has been found true, even with well separated light pulses, with an average
energy by pulse much less than one photon.

4b. SINGLE PHOTON PULSES

An excited atom emits a single photon, because of energy conservation. In

classical sources, many atoms are simultaneously in view of the detectors, and the
number of excited atoms fluctuates. As a consequence, the emitted light is
described by a density matrix reflecting these fluctuations, including the possibil-
ity that several photons are emitted simultaneously. For a Poisson fluctuation of
the number of emitting atoms, one can show that the statistical properties of the
light cannot be distinguished from the one of classical light.
 A. Aspect, P. Grangier / Wave-particle duality 9

Table 2
Anticorrelation experiment for light-pulses from an attenuated photodiode (0.01 photon/pulse). The
last column corresponds to the expected number of coincidences for a = l . All the measured
coincidences are compatible with a = 1; there is no evidence of anticorrelation.
Note that the singles rates are similar to the ones of table 3.

Trigger rates Singles rates Duration Measured Expected coincidences

coincidences for a = 1
N2r" N2t T
N 1 (s -1) N2r (s - t ) N2t (s - t ) T(s) NeT Nt

4 760 3.02 3.76 31200 82 74.5

8 880 5.58 7.28 31200 153 143
12130 7.90 10.2 25 200 157 167
20 400 14.1 20.0 25 200 341 349
35 750 26.4 33.1 12 800 329 313
50 800 44.3 48.6 18 800 840 798
67 600 69.6 72.5 12 800 925 955

In order to observe non-classical properties in fluorescence light, it is thus

necessary to isolate single atom emission. This was realized by Kimble et al. [8]
who had only one atom in their observation region when they demonstrated
antibunching. In our experiment, we have been able to isolate single atom
emission not in space but in time. Our source is composed of atoms that we
excite to the upper level of a two-photon radiative cascade (fig. 3) [10], emitting
two photons at different frequencies ~'1 and ~'2- The time intervals between the
detections of ~,1 and ~'2 are distributed according to an exponential law, corre-
sponding to the decay of the intermediate state with a life-time "rs = 4.7 ns. By
chosing the rate of excitation of the cascades much smaller than ('rs) -~, we have
cascades well separated in time. We can use the detection of 1,1 as a trigger for a
gate of duration o~ = 2 "rs, corresponding to the scheme of fig. 2. During a gate,

4p2 I S O ~
9D
5111nm -- ~51.3 rum
4s4p 1P 1
:4p l P 1
~)K
4~ nm

4s2 IS 0
Fig. 3. Radiative cascade of calcium used to produce the single photons wave packets. The detection
of photon Pl is used as trigger (el. fig. 2). The photon ~2 is sent onto the beam-splitter. The calcium
cascade used in this experiment was excited by two photon absorption from stabilized c.w. lasers, and
the cascade rate was held constant to a few percent.
10 A. Aspect, P. Grangier / Wave-particleduality

the probability for the detection of a photon p2 coming from the same atom that
emitted v1 is much bigger than the probability of detecting a photon v2 coming
from any other atom in the source. We are then in a situation close to an ideal
single-photon pulse [11], and we expect the corresponding anticorrelation be-
haviour on the beam splitter.
The expected values of the counting rates can be obtained by a straight-for-
ward quantum mechanical calculation. Denoting N the rate of excitation of the
cascades, and q , c t and c r the detection efficiencies of p h o t o n v 1 and v2
(including the collection solid angles, optics transmissions, and detector efficien-
cies) we obtain
N1= caN (7a)
N,= N#t[f(w) + Nw] (7b)
N r = Nl,r[f(w) + Nw] (7b')
Nc=Nleter[2f(w)Nw+(Nw)2]. (7c)
The quantity f(w), very close to 1 in this experiment, is the product of the factor
[ 1 - e x p ( - w / % ) ] (overlap between the gate and the exponential decay) by a
factor somewhat greater than I related to the angular correlation between v1 and
v2 [12].
The quantum mechanical prediction for a (eq. (6)) is thus
2f(w)Nw + ( Nw) 2
= (8)
[f(w) + Nw] 2
which is smaller than one, as expected. The anticorrelation effect will be stronger

Table 3
Anticorrelation experiment with single-photon pulses from the radiative cascade. The last column
corresponds to the expected number of coincidences for a = 1. The measured coincidences show a
clear anticorrelation effect.
These data can be compared to table 2.

Trigger rates Singles rates Duration Measured Expected coincidences

coincidences for a = 1

N 1 (S -1) N2r (s -1) N2t (s -1) T (s) NcT N2r"

N1N2t T
4720 2.45 3.23 1200 6 25.5
8 870 4.55 5.75 17 200 9 50.8
12100 6.21 8.44 14800 23 64.1
20400 12.6 17.0 19 200 86 204
36 500 31.0 40.6 13 200 273 456
50 300 47.6 61.9 8 400 314 492
67100 71.5 95.8 3 600 291 367
 A. Aspect, P. Grangier / Wave-particle duality 11

O(

f
0 .s 1 x==N
Fig. 4. Anficorrelation parameter a as a function of wN (number of cascades excited during a gate).
The indicated error is + 1 standard deviation. The full line is the theoretical prediction of quantum
mechanics. The inequality a < 1 characterizes the quantum domain.

( a small compared to 1) if Nw can be chosen much smaller than f(w). This

condition corresponds actually to the intuitive requirement that N is smaller than
(%)-1 (w is of the order of ~'s)-
The counting electronics, including the gating system, was a critical part in
this experiment. The gate w was actually realized by time-to-amplitude con-
verters followed by threshold circuits. These single-channel analyzers are fed by
shaped pulses from PM1 (detecting vz) on the START input, and from PM r or
PM t on the STOP input. This allows to adjust the gates with an accuracy of 0.1
ns. A third time-to-amplitude converter measures the delays between the various
detections, and allows to build the various time delay spectra, useful for the
control of the system.
Table 3 shows the measured counting rates, for different values of the
excitation rate of the cascade. The corresponding values of a have been plotted
on fig. 4, a s a function of Nw. As expected, the violation of inequality (6)
increases when Nw decreases, but the signal decreases simultaneously, and it
becomes necessary to accumulate the data for long periods of time to achieve a
reasonable statistical accuracy. A maximum violation of more than 13 standard
deviations has been obtained for a counting time of 5 hours. The value of a is
0.18 + 0.06, corresponding to a total number of coincidences of 9, instead of a
minimum value of 50 expected for a classical model of the light.

4c. CONCLUSION

In these experiments, we have thus confirmed that light pulses emitted by a

classical source do not exhibit a single-photon behaviour on a beam splitter, even
when the average energy by pulse is much less than one photon.
12 A. Aspect, P. Grangier / Wave-particle duality

PMz2

SINGLE-PHOTON

INPUT
./~.d~/
/

Fig. 5. Mach-Zehnder interferometer. The photomultipliers PMzl and PMz2 are gated as in the
previous experiments, so that the interferences axe due to single photon wave packets. The path
difference is controlled by moving the mirrors.

We have also demonstrated a source that produces single photon wave-packets,

with a synchronized triggering signal. On a beam splitter, these single photon
pulses exhibit a very clear anticorrelation, that is to say a behaviour characteristic
of single photons, as predicted by quantum mechanics.
With such a source, it is thus possible to revisit the question of single photon
interferences.

5. Single-photon interference

Starting with the same source ar~.dbeam-splitter we have built a Mach-Zehnder

interferometer (fig. 5). The detectors in the two outputs Z1 and Z2 are gated as in
the previous experiments, so that we measure the probabilities that a single
photon is transmitted to output Z1 or Z2. According to quantum mechanics,
these probabilities Pzl and Pz2 are modulated with a visibility unity when the
path difference 8 is varied.
The interferometer has been designed to accept the large optical spread of the
beam from the source (diameter 40 mm, divergence 25 mrad), without alterating
the visibility of the fringes. The interference is thus observed in the focal planes
of lenses in the outputs Z1 and Z2, and 8 is varied around zero. The planeities of
the beam splitters and of the mirrors are close to >,/50 on the whole beam
diameter. Their orientations are controlled by mechanical stages at the same
precision. A translation system, piezzo driven, allows to maintain the orientation
of the mirrors while changing the path difference.
The interferometer has first been checked using light from the cascade source,
but without the gating system (in order to have a greater signal). We define the
fringe visibility as
NMaX_ NMin
V = NM ~ + NMin (9)
 A. Aspect, P. Grangier / Wave-particle duality 13

Z1 (a) Z2
. 9 ~ .,~..
..." -. .: -..
20 9 :" "'~ .. ' 28 .. ,- . . . . . ..".. u
7 ": : ' . . ~-~" : -.. . ='.. f~ (.'- ..-..
- 9 ~" " ~. a .'. ": .
i:: 9~ ,./" :"..~..."
' " : ~ r"'" : :-" " .;.
)~ " u .-- ia.~ Is--": I

lee 1t~1}

Z1 Z2
(b) l- ,;
,
99 .
~.
:9
~: 9 .. ~.- ~..;
200
f 9 o.
,~
.~
.
I,
. ~
9
~ 9
,Q 9
9

/ :-/, ~-;, :~', L~,

100

CHRNNEL NUMBER

Fig. 6. N u m b e r of counts in outputs Z1 a n d Z2 as a function of the p a t h difference 8 (one channel

corresponds to h / 5 0 ) . (a) 1 s counting time per channel; (b) 15 s counting time per channel
(compilation of 15 elementary sweeps). This experiment corresponds to a n anticorrelation p a r a m e t e r
a = 0.18.

where N M~x and N Mi~ are the m a x i m u m and m i n i m u m counting rates on output
Z1 when 8 is varied (dark counts of the photomultipliers are subtracted for this
calculation). This fringe visibility (without gating) was found equal to 98.7% +
0.5%, the uncertainty corresponding to variations in the alignment from day to
day.
For a single-photon interference experiment, the gating system is used, and
is varied around ~ = 0 in 256 steps of k / 5 0 each, with a counting time of 1 s at
each step. All the data are stored in a computer, that also controls the path
difference. Various sweeps can be compiled to improve the signal to noise ratio.
A single sweep and the compiled results are shown on fig. 6. This experiment was
performed with the source running in the regime corresponding to an anticorrela-
tion parameter a = 0.18, measured in the previous experiment (section 4b). The
visibility of the fringes is clearly very close to 1.
The exact measurement of the visibility of the fringes is a delicate question.
One can use the definition (9), but this procedure uses only a small fraction of
the information stored, since only the m a x i m u m and m i n i m u m counting rates are
used. Another method, using all the data, consists in searching the linear
regression between the gated rates and the non-gated rates (that were also
monitored and stored). Standard statistical methods yield the coefficients of this
regression, which allows to express the visibility of the gated fringes as a function
of the visibility of the non-gated fringes for which the uncertainty is negligible.
Both procedures have given consistent results, but the second one yields
results with a better accuracy. These results are presented on fig. 7, where it is
14 A. Aspect, P. Grangier / Wave-particle duality

>-'100~'~ ~I g
~- 98.7
z 0 o 0 0 0 0
m

95 i I i I I I

0 5 1 ton

Fig. 7. Visibility of the fringes in the single-photon regime, as a function of wN. A correction (less
than 0.3%) has been made for the dark counts of the photomultipliers.

clear that the visibility appears constant when the anticorrelation increases (the
error bars increase because of the diminuation of the signal). We see for instance
that for WN= 0.06, the visibility is found equal to 0.98 + 0.01. This regime
corresponds to an anticorrelation parameter a = 0.11, i.e. a clear single-photon
situation.

6. Conclusion - Wave particle duality for a single photon

We have presented two complementary experiments, bearing on the fight

pulses emitted by our source. The same gating procedure, synchronized in the
same way, has been used in both experiments. In the first experiment, we have
observed a clear anticorrelation on both sides of the b e a m splitter. We have
concluded that the source emits single-photon wave packets, as predicted by our
analysis. In the second experiment, we have observed interference with a visibility
close to 1, in the single photon regime. We think that this last experiment is a
genuine single photon interference experiment.
These two experiments illustrate in a direct way the notion of wave particle
duality for a single photon. If we want to use classical pictures (or concepts) to
interpret these experiments, we must use a particle picture for the first one ( " a
photon is not sprit on a beam-splitter"). On the contrary, the second experiment
can only be understood in the framework of a wave theory (" the electromagnetic
field is coherently sprit on the first b e a m splitter, and recombined on the second,
and this recombination depends on the path difference"). The logical conflict
between these two classical pictures is at the very basis of famous difficulties of
interpretation of q u a n t u m mechanics.
Not pretending to solve these difficulties, we can anyway m a k e some remarks
which are the application to this particular situation of general statements
 A. Aspect, P. Grangier / Wave-particle duality 15

appearing in the old discussions about the interpretation of wave-particle duality.

It is true that the two classical pictures (wave or particle) corresponding to the
two experiments are mutually incompatible, but the two corresponding experi-
ments are also mutually exclusive, in the sense that they could not be performed
simultaneously. One has to choose between the measurements described on fig. 2
or on fig. 5. The impossibility of performing simultaneously experiments corre-
sponding to complementary descriptions was an often repeated argument of
Bohr, who concluded that Quantum Mechanics is a coherent and complete
theory. It may also be worth noting that the problem of incompatible descrip-
tions arises only if we insist on using classical concepts such as waves or particles.
But if we stand to the quantum mechanical description, there is a unique
description of the light, by the same state vector (or density-matrix) for experi-
ments of fig. 2 or fig. 5. It is the observed quantity which changes, according to
the measurement performed, but not the description of the light.

ZI .01 s / c a n a l Z2

Zl , I s/canal Z2

Zl 1 s/canal Z2
/
D O

28 9 9 . . .. ~. ~ . J,
~ " 7 ~, ".~ ,. ." ~ [:" " , "" " :-
"': "'" "," "" "'~ "" ~'" "" 9 ": ,~ ". "..:" ." ,-" V '
9i ": .'; ~ , ' ~ ~" '~ "~" ~'~'-r':~'<'~'":'~":':'
8

Zl 10 s / c a n a l Z2
' - "; I' :',......*.
~.
288 . .; . .". . d: .~..~ ~t ,~
-
.. .., :" .~ ": ." 9 ." '. . . . .
9 * "~ *.t' ;* *. t * 9 .. ~; ~. .* . . . ; . ~
9 ; . , , . t 9 9 . " 9 . . . 9 9 . . ,,, . , ; .. , . .

a ": t* .t .. ". ." .:" ". ". ., 9 ". 9 ."

V v ~ :," ",; V '," :v. . .v. . . ~
. . .- v ,," "'"2~,

Fig. 8. Apparition of the fringes " p h o t o n by photon". T h e figure shows the n u m b e r of counts in the
outputs Z1 a n d Z2 as a function of the p a t h difference ~ (one channel corresponds to ~ / 2 5 ) . The
counting time per channel is varied from 0.01 s to 10 s. The regime of the source corresponds to a n
anticorrelation parameter a = 0.2.
16 A. Aspect, P. Grangier / Wave-particle duality

To finish, we cannot resist to present the apparition of the interferences

"photon by photon", since this is a classics of this topics. Figure 8 shows how the
interference pattern is built when the data are accumulated. In spite of the
excitement we experienced when observing this apparition of the fringes, we do
not think that it is the most relevant result for wave particle duality, which is
really illustrated by the combination of experiments of fig. 2 and 5.

References

[1] R. Loudon, Rep. Progr. Phys. 43 (1980) 913.

[2] R. Loudon, The Quantum Theory of Light, 2nd ed. (Clarendon, Oxford, 1983).
[3] P. Grangier, G. Roger and A. Aspect, Europhysics Lett. 1 (1986) 173;
P. Grangier, th6se de doctorat d'&at, Universit6 ParAs XI, Orsay (1986).
[4] P.A.M. Dirac, The Principles of Quantum Mechanics (Oxford University Press, 1958).
[5] W.E. Lamb and M.O. Scully, in: Polarisation, Mati~re et Rayonnement, ed. Soci&6 Fran~aise de
Physique (Presses Universitaires de France, Paris, 1969).
[6] D.C. Burnham and D.L. Weinberg, Phys. Rev. Lett. 25 (1970) 84;
S. Friberg, C.K. Hong and L. Mandel, Phys. Rev. Lett. 54 (1985) 2011.
[7] J.F. Clauser, Phys. Rev. D9 (1974) 853.
[8] H.J. Kimble, M. Dagenais and L. Mandel, Phys. Rev. Lett. 39 (1977) 691;
J.D. Cresser, J. Hager, G. Leuchs, M. Rateike and H. Walther, in: Topics in Current Physics, Vol
27 (Springer Verlag, Berlin, 1982).
[9] R. Short and L. Mandel, Phys. Rev. Lett. 51 (1983) 384;
M.C. Teich and B.E.A. Saleh, J.O.S.A. B2 (1985) 275. All the recent observations of squeezing
also constitute an evidence for "non-classical states of light".
[10] A. Aspect, P. Grangier and G. Roger, Phys. Rev. Lett. 47 (1981) 460.
[11] Since our experiment, single photon pulses have been produced by a different technique, which
consists in producing pairs of photons emitted in parametric splitting, and using one of these
photons for the gating system. A theoretical advantage of these techniques is a better time and
angular correlation between the two photons of a pair. See C.K. Hong and L. Mandel, Phys.
Rev. Lett. 56 (1986) 58;
J.G. Walker and E. Jakeman, Optica Acta 32 (1985) 1303.
[12] E.S. Fry, Phys. Rev. A8 (1973) 1219.

Discussion

Stig Stenholm: You can have not only particle or wave properties, but a situation where there is
some information about either one simultaneously?
Alain Aspect: It is correct to say that we have observed only two extreme situations. One can
in principle detect the passage or non-passage of a photon without absorbing
this photon (this has been proposed, but never realized). By use of such devices
in the arms of the interferometer, one could know in which arm the photon
passes. A careful analysis of such devices show that the passage of the photon
can only be known at a certain level of confidence. On the other hand, this
measurement provokes a perturbation of the phase of the fight beam, which is
 A. Aspect, P. Grangier / Wave-particle duality 17

higher when the detection of passage is more certain. Eventually, it is thus

possible to observe fringes but with a diminished contrast - a n d to know
simultaneously in which arm the p h o t o n passes - but not with certainty. Such
an experiment would thus reveal simultaneously a certain a m o u n t of particle-like
behaviour and a certain a m o u n t of wave-like behaviour. For properly defined
complementary observables associated t o these measurements, one can find
Heisenberg's inequalities.
G. Raoindra Kumar:The explanation that I have heard for the interference of light from two
i n d e p e n d e n t lasers is the following: The p h o t o n is a state of the radiation field.
It does not matter whether it comes from one laser or the other. Since the
property of interference is inherent to a photon, we can observe it. W h a t is your
c o m m e n t on this?
Alain Aspect: Let us try to point out the difficulty. It is certainly not difficult to write a
calculation - in the framework of q u a n t u m theory of light - showing that
interferences (more precisely a beat pattern) should be observed between two
single-line stabilized lasers.
It is also not difficult to understand this interference if one describes the light
beams as classical monochromatic electromagnetic waves. The problem arises if
one wants to consider the laser beams as streams of photons. Then, the
explanation requires to introduce amplitude of probabilities, and to show that
the observation results from an addition of amplitudes. It is not straight
forward. A standard c o m m e n t is also that since the two beams have equal
wavelengths, it is impossible to know from which laser the p h o t o n comes, and
there is no problem in observing interferences. This line of reasoning is also
somewhat intricate.
T h e conclusion is that the problem arises only when one looks for pictures.
Then, we find that one picture is perfectly clear, while the other is quite hard to
deal with. The lesson is that we should use the best adapted picture for the
situation u n d e r investigation. In the case of a laser beam, correctly described by
a coherent state, it is not a surprise that the best picture is a classical
electromagnetic wave. As a matter of fact a coherent state is the q u a n t u m state
which is the closest to a classical wave.