Quantum effects in one-photon and two-photon interference

L. Mandel
Department of Physics and Astronomy, University of Rochester, Rochester,
New York 14627

After introducing some basic definitions, the article describes several optical interference experiments
in which quantum effects appear. An analysis of these experiments leads to some new and improved
measurement techniques and to a better understanding of the quantum state. [S0034-6861(99)02102-9]

I. INTRODUCTION We now turn to interference effects that are defined in

terms of an ensemble average. Let us start by distin-
Although interference is intrinsically a classical wave guishing between second-order or one-photon, and
phenomenon, the superposition principle which under- fourth-order or two-photon interference experiments. In
lies all interference is also at the heart of quantum me- the simplest and most familiar type of experiment, one
chanics. Feynman has referred to interference as really photodetector, say D 1 , is used repeatedly to measure
‘‘the only mystery’’ of quantum mechanics. Further- the probability P 1 (x 1 ) of detecting a photon in some
more, in some interference experiments we encounter short time interval as a function of position x 1 [see Fig.
1(a)]. Interference is characterized by the (often, but not
the idea of quantum entanglement, which has also been
necessarily, periodic) dependence of P 1 (x 1 ) on the op-
described as really the only quantum mystery. Clearly
tical path lengths S A D 1 and S B D 1 or on the correspond-
interference confronts us with some quite basic ques- ing phase shifts f A1 and f B1 . Because P 1 (x 1 ) depends
tions of interpretation. Despite its long history, going on the second power of the optical field and on the de-
back to Thomas Young at the beginning of the 19th cen- tection of one photon at a time, we refer to this as
tury, optical interference still challenges our understand- second-order, or one-photon, interference. Sometimes
ing, and the last word on the subject probably has not two photodetectors D1 and D2 located at x 1 and x 2 are
yet been written. With the development of experimental used in coincidence repeatedly to measure the joint
techniques for fast and sensitive measurements of light, probability P 2 (x 1 ,x 2 ) of detecting one photon at x 1 and
it has become possible to carry out many of the Gedan- one at x 2 within a short time [see Fig. 1(b)]. Because
ken experiments whose interpretation was widely de- P 2 (x 1 ,x 2 ) depends on the fourth power of the field, we
bated in the 1920s and 1930s in the course of the devel- refer to this as fourth-order, or two-photon, interfer-
ence. For the purpose of this article, a photon is any
opment of quantum mechanics. Although this article
eigenstate of the total number operator belonging to the
focuses entirely on experiments with light, interference eigenvalue 1. That means that a photon can be in the
has also been observed with many kinds of material par- form of an infinite plane wave or a strongly localized
ticles like electrons, neutrons, and atoms. We particu- wave packet. Because most photodetectors function by
larly draw the reader’s attention to the beautiful experi- photon absorption, the appropriate dynamical variable
ments with neutron beams by Rauch and co-workers
and others (see, for example, Badurek et al., 1988).
Quantum optical interference effects are key topics of a
recent book (Greenstein and Zajonc, 1997), an extended
rather thorough review (Buzek and Knight, 1995) and
an article in Physics Today (Greenberger et al., 1993).
The essential feature of any optical interference ex-
periment is that the light from several (not necessarily
primary) sources like S A and S B (see Fig. 1) is allowed
to come together and mix, and the resulting light inten-
sity is measured at various positions. We characterize
interference by the dependence of the resulting light in-
tensities on the optical path length or phase shift, but we
need to make a distinction between the measurement of
a single realization of the optical field and the average
over an ensemble of realizations or over a long time. A
single realization may exhibit interference, whereas an
ensemble average may not. We shall refer to the former
as transient interference, because a single realization
usually exists only for a short time. Transient interfer-
ence effects have been observed in several optical ex- FIG. 1. Principle of photon interference: (a) one-photon or
periments in the 1950s and 1960s. (Forrester et al., 1955; second-order interference; (b) two-photon or fourth-order in-
Magyar and Mandel, 1963; Pfleegor and Mandel, 1967, terference. S A and S B are sources. D 1 and D 2 are photodetec-
1968). tors.

S274 Reviews of Modern Physics, Vol. 71, No. 2, Centenary 1999 0034-6861/99/71(2)/274(9)/$16.80 ©1999 The American Physical Society
 L. Mandel: Quantum effects in photon interference S275

for describing the measurement is the photon annihila- interference is the physical manifestation of the intrinsic
tion operator. If we make a Fourier decomposition of indistinguishability of the sources or of the photon
the total-field operator Ê(x) at the detector into its paths. If the different possible photon paths from source
positive- and negative-frequency parts Ê (1) (x) and to detector are indistinguishable, then we have to add
the corresponding probability amplitudes before squar-
Ê (2) (x), then these play the roles of photon annihila- ing to obtain the probability. This results in interference
tion and creation operators in configuration space. Let terms as in Eq. (3). On the other hand, if there is some
Ê (1) (x 1 ), Ê (1) (x 2 ) be the positive-frequency parts of way, even in principle, of distinguishing between the
the optical field at the two detectors. Then P 1 (x 1 ) and possible photon paths, then the corresponding probabili-
P 2 (x 1 ,x 2 ) are given by the expectations in normal or- ties have to be added and there is no interference.
der: Let us see how this argument works when each source
consists of a single two-level atom. When the atom is in
P 1 ~ x 1 ! 5 a 1 ^ Ê ~ 2 ! ~ x 1 ! Ê ~ 1 ! ~ x 1 ! & , (1)
the fully excited state (an energy eigenstate), its energy
P 2 ~ x 1 ,x 2 ! 5 a 1 a 2 ^ Ê ~ 2 ! ~ x 1 ! Ê ~ 2 ! ~ x 2 ! Ê ~ 1 ! ~ x 2 ! Ê ~ 1 ! ~ x 1 ! & , can be measured, in principle, without disturbing the
atom. Suppose that both sources are initially in the fully
(2)
excited state and that the energy of each atom is mea-
where a 1 , a 2 are constants characteristic of the detec- sured immediately after the detection of a photon by
tors and the measurement times. D 1 . If source A is found to be in the ground state
whereas source B is found to be still excited, then, obvi-
II. SECOND-ORDER INTERFERENCE ously, S A can be identified as the source of the photon
detected by D 1 . Therefore there is no second-order in-
Let us decompose Ê (1) (x 1 ) and Ê (1) (x 2 ) into two terference in this case, and this conclusion holds regard-
normal modes A and B, such that â A ,â B are the annihi- less of whether the energy measurement is actually car-
lation operators for the fields produced by the two ried out. In this case, the optical field is in a one-photon
sources S A and S B , respectively. Then we may put Fock state u 1 & A u 0 & B for which ^ â †A â B & 50. On the other
Ê (1) (x 1 )5f A e i f A1 â A 1f B e i f B1 â B , where f A , f B are hand, if the atoms are in a superposition of upper and
complex parameters, and similarly for Ê (1) (x 2 ). From lower states initially, then the atomic energy has no well-
Eq. (1) we then find defined initial value and it cannot be measured without
disturbing the atom. The source of the detected photon
P 1 ~ x 1 ! 5 a 1 @ u f A u 2 ^ n̂ A & 1 u f B u 2 ^ n̂ B & therefore cannot be identified by measuring the atomic
i ~ f B1 2 f A1 ! † energy, or in any other way, and, as a result, second-
1f *
Af Be ^ â A â B & 1c.c.# . (3)
order interference is observed. This argument can be
If second-order interference is characterized by the de- made more quantitative in that the degree of second-
pendence of P 1 (x 1 ) on the optical path lengths or on the order coherence u g (1,1)AB u in Eq. (4) can be shown to equal
phase difference f B1 2 f A1 , then clearly the condition the degree of path indistinguishability (Mandel, 1991).
for the system to exhibit second-order interference is It should be clear from the foregoing that in these
that ^ â †A â B & Þ0. This is usually achieved most easily if experiments one photon does not interfere with another
the fields from the two sources S A and S B are at least one; only the two probability amplitudes of the same
partly correlated or mutually coherent. We define the photon interfere with each other. This has been con-
degree of second-order mutual coherence by the nor- firmed more explicitly in interference experiments with
malized correlation ratio a single photon (Grangier et al., 1986) and in experi-
1,1! ments with two independent laser beams, in which inter-
u g ~AB u [ u ^ â †A â B & u / ~ ^ â †A â A &^ â †B â B & ! 1/2, (4)
ference was observed even when the light was so weak
so that, by definition, u g (1,1) AB u lies between 0 and 1. But that one photon passed through the interferometer and
such correlation is not necessary for interference. Even was absorbed by the detector long before the next pho-
with two independent sources it is apparent from Eq. (3) ton came along (Pfleegor and Mandel, 1967, 1968).
that interference can occur if ^ â A & Þ0Þ ^ â B & . An ex-
ample would be the two-mode coherent state III. FOURTH-ORDER INTERFERENCE
u v A & A u v B & B , with complex eigenvalues v A , v B , for
which ^ â †A â B & 5 v *
A v B , which is nonzero because of the We now turn to the situation illustrated in Fig. 1(b), in
definite complex amplitude of the field in a coherent which two photodetectors are used in coincidence to
state. The field of a single-mode laser is often well ap- measure the joint probability P 2 (x 1 ,x 2 ) of detecting one
proximated by a coherent state for a short time. On the photon at x 1 and one at x 2 . Fourth-order interference
other hand the corresponding expectations vanish for a occurs when P 2 (x 1 ,x 2 ) depends on the phase differ-
field in a Fock (photon number) state u n A & A u n B & B , for ences f A1 2 f B2 , and this happens when the different
which ^ â †A â B & 50. Therefore there is no second-order in- paths of the photon pair from the sources to the detec-
terference in this case. Needless to say, this situation has tors are indistinguishable. Then we again have to add
no obvious counterpart in classical optics. the corresponding (this time two-photon) probability
In order to understand why interference effects occur amplitudes before squaring to obtain the probability.
in some cases and not in others, we need to recall that From Eq. (2) one can show that

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

S276 L. Mandel: Quantum effects in photon interference

P 2 ~ x 1 ,x 2 ! 5 a 1 a 2 $ u f A u 4 ^ :n̂ 2A : & 1 u f B u 4 ^ :n̂ 2B : & 12 u f A u 2 u f B u 2 ^ n̂ A &^ n̂ B & @ 11cos~ f B2 2 f A2 1 f A1 2 f B1 !#

i ~ f B2 2 f A2 1 f B1 2 f A1 ! i ~ f B1 2 f A1 !
A f B ^ â A â B & e
1f * 2 2 †2 2
A f B ^ â A â A â B & @ e
1c.c.1 u f A u 2 f * †2
1e i ~ f B2 2 f A2 ! # 1c.c.

i ~ f A1 2 f B1 !
* f A ^ â †2
1 u f Bu 2f B B â B â A & @ e 1e i ~ f A2 2 f B2 ! # 1c.c.% , (5)

where ^ :n̂ r : & denotes the rth normally ordered moment two-photon paths, viz., (a) the photon from S A is de-
of n̂. tected by D 1 and the photon from S B is detected by D 2
For illustration, let us focus once again on the special and (b) the photon from S A is detected by D 2 and the
case in which each source consists of a single excited photon from S B is detected by D 1 . Because cases (a)
and (b) are indistinguishable, we have to add the corre-
two-level atom. We have seen that in this case there is
sponding two-photon amplitudes before squaring to ob-
no second-order interference, because the source of tain the probability, and this generates interference
each detected photon is identifiable in principle. But the terms. In this case most terms on the right of Eq. (5)
same is not true for fourth-order interference of the vanish, and we immediately find the result given by (see
photon pair. This time there are two indistinguishable Box A)

P 2 ~ x 1 ,x 2 ! 5 a 1 a 2 2 u f A u 2 u f B u 2 IV. INTERFERENCE EXPERIMENTS WITH A PARAMETRIC

DOWNCONVERTER SOURCE
3 @ 11cos~ f B2 2 f A2 1 f A1 2 f B1 !# . (6)
The first two-photon interference experiment of the
type illustrated in Fig. 1(b), in which each source deliv-
Despite the fact that the two sources are independent, ers exactly one photon simultaneously, was probably the
they exhibit two-photon interference with 100 percent one reported by Ghosh and Mandel in 1987. They made
visibility. use of the signal and idler photons emitted in the split-
Two-photon interference exhibits some striking non- ting of a pump photon in the process of spontaneous
local features. For example, P 2 (x 1 ,x 2 ) given by Eq. (6) parametric downconversion in a nonlinear crystal of
can be shown to violate one or more of the Bell in- LiIO3. The crystal was optically pumped by the
equalities that a system obeying local realism must sat- 351.1-nm uv beam from an argon-ion laser and from
isfy. This violation of locality, which is discussed more time to time it gave rise to two simultaneous signal and
fully in the article by Zeilinger in this issue, has been idler photons at wavelengths near 700 nm. A modified
demonstrated experimentally. [See, for example, Man- and slightly improved version of the experiment was
del and Wolf, 1995.] later described by Ou and Mandel (1989). The signal (s)

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

 L. Mandel: Quantum effects in photon interference S277

FIG. 3. Principle of the interference experiment with two

downconverters in which both one-photon and two-photon in-
terference can be investigated (after Ou et al., 1990).

prediction of fourth-order interference as the path dif-

ference in one arm is varied. This has been confirmed
FIG. 2. Principle of the Franson (1989) two-photon interfer- experimentally. A different outcome may be encoun-
ence experiment in which signal and idler photons never mix. tered with pulsed rather than continuous excitation of
PDC is the parametric downconverter. D s and D i are photo- the parametric downconverter.
detectors.
V. INTERFERENCE EXPERIMENTS WITH TWO
and idler (i) photons were incident from opposite sides PARAMETRIC DOWNCONVERTERS
on a 50%:50% beam splitter that mixed them at a small
angle u '1 mrad, and the two mixed beams then fell on Next let us consider the experiment illustrated in Fig.
detectors D 1 and D 2 , each of which carried a 0.1-mm- 3, which allows both one-photon and two-photon inter-
wide aperture. The photons counted by each detector ference to be investigated at the same time. Two similar
separately and by the two detectors in coincidence in a nonlinear crystals NL1 and NL2, which both function as
total time of a few minutes were registered for various parametric downconverters, are optically pumped simul-
positions of the detectors. Because of the two-photon taneously by mutually coherent pump beams that we
state, no second-order interference is expected from shall treat classically and represent by the complex field
quantum mechanics, as we have seen, and none was ob- amplitudes V 1 and V 2 . As a result downconversion can
served. However, the two-photon coincidence rate ex- occur at NL1, with the simultaneous emission of a signal
hibited the expected interference in the form of a peri- s 1 and an idler i 1 photon in two slightly different direc-
odic variation of the rate with detector position, because tions, or downconversion can occur at NL2, with the
the photon pair detected by D 1 and D 2 could have origi- simultaneous emission of an s 2 and an i 2 photon, as
nated as signal and idler, respectively, or vice versa. shown. The question we wish to address is whether, in
An ingenious variation on the same theme of two- view of the mutual coherence of the two pump beams,
photon interference was proposed by Franson (1989) the s 1 and s 2 beams from the two downconverters are
and is illustrated in Fig. 2. Signal and idler photons emit- mutually coherent and exhibit interference when they
ted simultaneously in two slightly different directions are mixed, and similarly for the i 1 and i 2 beams. In order
from a parametric downconverter (PDC) fall on two de- to answer the question the experiment illustrated in Fig.
tectors D s and D i , respectively. The two beams never 3 is carried out. s 1 and s 2 are allowed to come together;
mix. On the way to the detector each photon encounters they are mixed at the 50%:50% signal beam splitter
a beam splitter leading to an alternative time-delayed BSA , and the combined beam emerging from BSA falls
path, as shown, and each photon is free to follow either on the photon detector D A . If s 1 and s 2 are mutually
the shorter direct or the longer delayed path. If the time coherent, then the photon counting rate of D A varies
difference T D between the long and short paths is much sinusoidally as the phase difference between the two
longer than the coherence time T C of the downcon- pump beams V 1 and V 2 is slowly increased. Similarly for
verted light, and much longer than the coincidence re- the two idlers i 1 and i 2 , which are mixed by BSB and
solving time T R , no second-order interference is to be detected by D B .
expected, and at first glance it might seem that no In order to treat this problem theoretically we repre-
fourth-order interference would occur either. But the sent the quantum state of the signal and idler photon
signal and idler photons are emitted simultaneously, pair from each crystal by the entangled state u C j &
and, within the coincidence resolving time, they are de- 5M j u vac& s j ,i j 1 h V j u 1 & s j u 1 & i j (j51,2). The combined
tected simultaneously. Therefore in every coincidence state is then the direct product state u C & 5 u C 1 & 3 u C 2 & ,
both photons must have followed the short path or both because the two downconversions proceed indepen-
photons must have followed the long path, but we can- dently. V 1 and V 2 are the c-number complex amplitudes
not tell which. When T C !T D !T R two more path com- of the pump fields. h represents the coupling between
binations are possible. With continuous pumping of the pump modes and the downconverted signal and idler
parametric downconverter the emission time is random modes, such that ^ u h V j u 2 & (j51,2) is the small probabil-
and unknown, and there is no way to distinguish be- ity of downconversion in a short measurement time. M 1
tween the light paths. We therefore have to add the cor- and M 2 are numerical coefficients that ensure the nor-
responding probability amplitudes, which leads to the malization of u C 1 & and u C 2 & , which we take to be real

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

S278 L. Mandel: Quantum effects in photon interference

FIG. 4. Results of measurements of the photon counting rate

by D A and D B in Fig. 3 as a function of path difference, show- FIG. 5. Results of coincidence measurements by D A and D B
ing the absence of one-photon interference. in Fig. 3 as a function of path difference, showing two-photon
interference. The continuous curve is theoretical.
for simplicity. Because ^ u h V j u 2 & !1 it follows that M 1
and M 2 are very close to unity. We shall retain the co- [^V* 1 V 2 & / ^ I & is the complex degree of coherence of the
efficients M 1 and M 2 nevertheless, because they provide two classical pump beams. A two-photon coincidence
us with useful insight into the role played by the measurement with both detectors D A and D B is there-
vacuum. Of course the downconverted light usually has fore expected to exhibit interference as the optical path
a very large bandwidth, and treating each signal and difference or the pump phase difference is varied. This is
idler as occupying one monochromatic mode is a gross confirmed by the experimental results shown in Fig. 5. It
oversimplification. However, a more exact multimode is interesting to note that the vacuum contribution to the
treatment leads to very similar conclusions about the state plays an essential role, because of the presence of
interference. the M 1 M 2 coefficients in Eq. (8).
The positive-frequency parts of the signal and idler Finally, we would like to understand in physical terms
fields at the two detectors can be given the two-mode why no second-order interference is registered by detec-
i u s1
expansions Ê (1) s 5â s1 e 1iâ s2 e i u s2 and Ê (1)
i 5â i1 e i u i1 tors D A and D B separately, but fourth-order interfer-
i u i2
1iâ i2 e , where u s1 , u s2 , u i1 , u i2 are phase shifts corre- ence is registered by the two together. Here it is helpful
sponding to the propagation from one of the two sources to recall the relationship between interference and indis-
NL1, NL2 to one of the two detectors D A , D B . Then tinguishability. From the coincidence measurement in
the expectations of the number of photons detected by Fig. 5 it is impossible to determine whether the detected
D A and by D B are ^ C u Ê (2) photon pair originates in NL1 or in NL2, and this indis-
s Ê s u C &
(1)
and
tinguishability is manifest as a fourth-order interference
^ C u Ê i Ê i u c & , and for the quantum state u C &
(2) (1)
pattern. However, if we are interested only in the inter-
5 u C 1 & 3 u C 2 & we obtain immediately ference of, say, the signal photons registered by D A , we
^ C u Ê ~s2 ! Ê ~s1 ! u C & 5 u h u 2 ~ ^ u V 1 u 2 & 1 ^ u V 2 u 2 & ! can use the detection of the idlers as an auxiliary source
of information, to determine where each detected signal
5 ^ C u Ê ~i 2 ! Ê ~i 1 ! u C & . (7) photon originated. This destroys the indistinguishability
of the two sources and kills the interference of the signal
These averages are independent of the interferometric photons, whether or not the auxiliary measurement is
path lengths and of the phases of the two pump beams, actually carried out.
showing that there is no interference and no mutual co- Figure 6 illustrates a one-photon interference experi-
herence between the two signals s 1 , s 2 or between the ment with two downconverters that exhibits interesting
two idlers i 1 , i 2 . These conclusions are confirmed by the nonclassical features (Zou et al., 1991). NL1 and NL2
experimental results presented in Fig. 4, which exhibit are two similar nonlinear crystals of LiIO3 functioning as
no sign of second-order or one-photon interference. parametric downconverters. They are both optically
Next let us look at the possibility of fourth-order or pumped by the mutually coherent uv light beams from
two-photon interference, by measuring the joint prob- an argon-ion laser oscillating on the 351.1-nm spectral
ability of detecting a signal photon and an idler photon line. As a result, downconversion can occur at NL1 with
with both detectors in coincidence. This probability is the simultaneous emission of a signal s 1 and an idler i 1
proportional to P 125 ^ C u Ê (2) Ê i Ê s u C & , and it
(2) (1) (1)
s Ê i photon at wavelengths near 700 nm, or it can occur at
is readily evaluated. If u V 1 u 2 5I5 u V 2 u 2 and u h u 2 I!1, so NL2 with the simultaneous emission of an s 2 and i 2 pho-
that terms of order u h u 4 I 2 can be neglected, we find ton. Simultaneous downconversions at both crystals is
P 1252 u h u 2 ^ I & @ 12M 1 M 2 u g ~121,1! u cos Q # , (8) very improbable. NL1 and NL2 are aligned so as to
make i 1 and i 2 collinear and overlapping, as shown, so
where U[ u s2 1 u i2 2 u s1 2 u i1 1arg(g(1,1)
12 ) and g (1,1)
12 that a photon detected in the i 2 beam could have come

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

 L. Mandel: Quantum effects in photon interference S279

FIG. 6. Outline of the one-photon interference experiment

with two downconverters (Zou et al., 1991). See text for de-
scription.

from NL1 or NL2. At the same time the s 1 and s 2 signal

beams come together and are mixed at beam splitter
BS0. The question to be explored is whether, in view of FIG. 7. Results of the one-photon interference experiment
the mutual coherence of the two pump beams, s 1 and s 2 shown in Fig. 6: A, data with i 1 unblocked; B, data with i 1
blocked.
are also mutually coherent and exhibit interference, un-
der the conditions when the downconversions at NL1
and NL2 are spontaneous and random. More explicitly, ity wipes out all interference between s 1 and s 2 . A simi-
if BS0 is translated in a direction normal to its face. will lar conclusion applies when the two idlers i 1 and i 2 do
the photon counting rate of detector D s vary sinusoi- not overlap, so that they can be measured separately.
dally, thereby indicating that interference fringes are However, when i 1 is unblocked and the two idlers over-
passing across the photocathode? lap, this source identification is no longer possible, and
With the experiment in Fig. 3 in mind, one might not s 1 and s 2 exhibit interference. Needless to say, it is not
expect to see one-photon interference at D s , but, as necessary actually to carry out the auxiliary measure-
shown in Fig. 7 (curve A), interference fringes were ac- ment with D i ; the mere possibility, in principle, that
tually observed so long as i 1 and i 2 were well aligned such a measurement could determine the source of the
and overlapped. The relatively small visibility of the in- signal photon is sufficient to kill the interference of s 1
terference is largely due to the incomplete overlap of the and s 2 .
two idlers. However, after deliberate misalignment of i 1 This kind of argument leads to an important conclu-
and i 2 , or if i 1 was blocked from reaching NL2, all in- sion about the quantum state of a system: in an experi-
terference disappeared, as shown by curve B in Fig. 7. ment the state reflects not what is actually known about
Yet the average rate of photon emission from NL2 was the system, but rather what is knowable, in principle,
unaffected by blocking i 1 or by misalignment. In the ab- with the help of auxiliary measurements that do not dis-
sence of induced emission from NL2, how can this be turb the original experiment. By focusing on what is
understood? knowable in principle, and treating what is known as
Here it is instructive again to invoke the relationship largely irrelevant, one completely avoids the anthropo-
between interference and indistinguishability. Let us morphism and any reference to consciousness that some
suppose that an auxiliary perfect photodetector D i is physicists have tried to inject into quantum mechanics.
placed in the path of the i 2 beam equidistant with D s We emphasize here that the act of blocking the path of
from NL2, as shown in Fig. 6. Now the insertion of D i in i 1 between NL1 and NL2 kills the interference between
the path of i 2 does not in any way disturb the interfer- s 1 and s 2 not because it introduces a large uncontrol-
ence experiment involving the s 1 and s 2 beams. How- lable disturbance. After all, the signal photons s 1 and s 2
ever, when i 1 is blocked, D i provides information about are emitted spontaneously and the spontaneous emis-
the source of the signal photon detected by D s . For sions are not really disturbed at all by the act of blocking
example, if the detection of a signal photon by D s is i 1 . In this experiment the disturbance introduced by
accompanied by the simultaneous detection of an idler blocking i 1 is of a more subtle kind: it is only the possi-
photon by D i , a glance at Fig. 6 shows immediately that bility of obtaining information about the source of the
the signal photon (and the idler) must have come from signal photon which is disturbed by blocking i 1 .
NL2. On the other hand, if the detection of a signal If, instead of blocking i 1 completely from reaching
photon by D s is not accompanied by the simultaneous NL2, one merely attenuates i 1 with some sort of optical
detection of an idler by D i , then the signal photon can- filter of complex transmissivity T, then the degree of co-
not have come from NL2 and must have originated in herence and the visibility of the interference pattern
NL1. With the help of the auxiliary detector D i we can formed by s 1 and s 2 are reduced by the factor uT u (Zou
therefore identify the source of each detected signal et al., 1991). This provides us with a convenient means
photon, whenever i 1 is blocked, and this distinguishabil- for controlling the degree of coherence of two light-

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

S280 L. Mandel: Quantum effects in photon interference

FIG. 8. Outline of the two-photon interference experiment to

measure the time separation between signal and idler photons
(Hong et al., 1987). See text for description.

beams s 1 and s 2 with a variable filter acting on i 1 , with- FIG. 9. Results of the two-photon interference experiment
out affecting the light intensities of s 1 and s 2 . Finally, shown in Fig. 8. The measured coincidence rate is plotted as a
insofaras i 1 falling on NL2 may be said to induce coher- function of beam-splitter displacement in m m or differential
ence between s 1 and s 2 from the two sources NL1 and time delay in fsec. The continuous curve is theoretical.
NL2, we have here an example of induced coherence
without induced emission.

VI. MEASUREMENT OF THE TIME INTERVAL BETWEEN

on the right vanishes, which implies the destructive in-
TWO PHOTONS BY INTERFERENCE
terference of the photon pair in arms 1 and 2, and both
The same fourth-order two-photon interference effect photons emerge together either in arm 1 or in arm 2.
has been used to measure the time separation between Therefore no coincidence counts (other than acciden-
two photons with time resolution millions of times tals) between detectors D 1 and D 2 are registered. The
shorter than the resolution of the detectors and the elec- reason for this can be understood by reference to Fig. 8.
tronics (Hong et al., 1987). Let us consider the experi- A coincidence detection between D 1 and D 2 can occur
ment illustrated in Fig. 8. Here the signal and idler pho- only if the two incoming signal and idler photons are
tons emitted from a uv-pumped crystal of potassium either both reflected from the beam splitter or are both
dihydrogen phosphate (KDP) serving as parametric transmitted through the beam splitter. Because these
downconverter are sent in opposite directions through a two possible photon paths are indistinguishable, we have
symmetric 50%:50% beam splitter (BS) that mixes to add the corresponding two-photon probability ampli-
them. The emerging photon pair is allowed to impinge tudes before squaring to obtain the probability. But be-
on two similar photon detectors D 1 and D 2 , whose out- cause of the phase change that occurs on reflection from
put pulses are counted both separately and in coinci- the beam splitter, as compared with that on transmis-
dence as the beam splitter is translated in the direction sion, the two-photon probability amplitude for two re-
shown though a distance of a few wavelengths. The co- flections from BS is 180° out of phase with the corre-
herence time T c of the downconverted light is made sponding two-photon probability amplitude for two
about 10213 sec with the help of the interference filters transmissions through BS. When these two amplitudes
IF1 , IF 2 . are added they give zero.
Let us consider the quantum state uc& of the photon Needless to say, this perfect destructive interference
pair emerging from the beam splitter. With two photons of the photon pair requires two identical incident pho-
impinging on BS from opposite sides there are really tons, and their description goes well beyond our over-
only three possibilities for the light leaving BS: (a) one simplified two-mode treatment. If we think of the in-
photon emerges from each of the outputs 1 and 2; (b) coming entangled photon pair as two identical wave
two photons emerge from output 1 and none emerges packets that overlap completely in time, then it should
from output 2; (c) two photons emerge from output 2 be obvious that if one wave packet is delayed even
and none emerges from output 1. The quantum state of slightly relative to the other, perfect destructive interfer-
the beam-splitter output is actually a linear superposi- ence is no longer possible, and the apparatus in Fig. 8 no
tion of all three possibilities in the form longer yields zero coincidences. The greater the relative
time delay t D , the greater is the two-photon coinci-
dence rate R c , and by the time the delay t D exceeds the
u c & 5 ~ u Ru 2 2 u T u 2 ! u 1 & 1 u 1 & 2 time duration of the wave packet, the coincidence rate
R c becomes constant and independent of the time delay
1&i u RT u @ u 2 & 1 u 0 & 2 1 u 0 & 1 u 2 & 2 ], (9)
t D between the wave packets. For wave packets of
where R and T are the complex beam-splitter reflectivity Gaussian shape and bandwidth Dv, and with a 50%:50%
and transmissivity. When u Ru 51/&5 u Tu , the first term beam splitter, one finds that R c is given by (see Box B)

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

 L. Mandel: Quantum effects in photon interference S281

2 2 niques of controlling the degree of coherence of two

R c }K @ 12e 2 t D ~ D v ! # . (10)
light beams without change of intensity.
The two-photon coincidence rate R c is therefore ex-
pected to vary with the time delay t D as in Fig. 9. This
ACKNOWLEDGMENTS
has indeed been observed in an experiment in which the
differential time delay t D was introduced artificially by This article is based on research resulting from col-
translating the beam splitter BS in Fig. 8 (Hong et al., laboration with numerous former graduate students and
1987). It is worth noting that the measurement achieved postdocs whose help is gratefully acknowledged. The re-
a time resolution of a few femtoseconds, which is a mil- search was supported by the National Science Founda-
lion times shorter than the time resolution of the photon tion and by the U.S. Office of Naval Research.
detectors and the associated electronics. This is possible
because the measurement was really based on optical
interference. In some later experiments the resolution REFERENCES
time was even shorter than the period of the light. The
same principle has been used by Chiao and co-workers Badurek, G., H. Rauch, and A. Zeilinger, 1988, Eds., Matter
to measure photon tunneling times through a barrier. Wave Interferometry (North-Holland, Amsterdam).
Buzek, V., and P. L. Knight, 1995, in Progress in Optics
XXXIV, edited by E. Wolf (Elsevier, Amsterdam), p. 1.
Forrester, A. T., R. A. Gudmundson, and P. O. Johnson, 1955,
Phys. Rev. 99, 1691.
VII. CONCLUSIONS
Franson, J. D., 1989, Phys. Rev. Lett. 62, 2205.
Ghosh, R., and L. Mandel, 1987, Phys. Rev. Lett. 59, 1903.
We have seen that quantum effects can show up in
Grangier, P., G. Roger, and A. Aspect, 1986, Europhys. Lett.
both one-photon and two-photon interference. The
1, 173.
analysis of some interference experiments confronts us
Greenberger, D. M., M. A. Horne, and A. Zeilinger, 1993,
with fundamental questions of interpretation and brings Phys. Today 46, 22.
out that the quantum state reflects not what we know Greenstein, G., and A. G. Zajonc, 1997, The Quantum Chal-
about the system, but rather what is knowable in prin- lenge (Jones and Bartlett, Sudbury, MA).
ciple. This avoids any reference to consciousness in the Hong, C. K., Z. Y. Ou, and L. Mandel, 1987, Phys. Rev. Lett.
interpretation of the state. Finally, quite apart from their 59, 2044.
fundamental interest, quantum interference effects have Kwiat, P. G., W. A. Vareka, C. K. Hong, H. Nathel, and R. Y.
led to some valuable practical applications, such as the Chiao, 1990, Phys. Rev. A 41, 2910.
new method for measuring the time separation between Magyar, G., and L. Mandel, 1963, Nature (London) 198, 255.
two photons on a femtosecond time scale, and new tech- Mandel, L., 1991, Opt. Lett. 16, 1882.

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999

S282 L. Mandel: Quantum effects in photon interference

Mandel, L., and E. Wolf, 1995, Optical Coherence and Quan- Pfleegor, R. L., and L. Mandel, 1967, Phys. Rev. 159, 1034.
tum Optics, 1st Ed. (Cambridge University Press, Cambridge, Pfleegor, R. L. and L. Mandel, 1968, J. Opt. Soc. Am. 58, 946.
UK). Richter, Th., 1979, Ann. Phys. (Leipzig) 36, 266.
Ou, Z. Y., and L. Mandel, 1989, Phys. Rev. Lett. 62, 2941. Zou, X. Y., L. J. Wang, and L. Mandel, 1991, Phys. Rev. Lett.
Ou, Z. Y., L. J. Wang, X. Y. Zou, and L. Mandel, 1990, Phys. 67, 318.
Rev. A 41, 566.

Rev. Mod. Phys., Vol. 71, No. 2, Centenary 1999