Faster than Light?
Experiments in quantum optics show that two
distant events can influence each other faster
than any signal could have traveled between them
by Raymond Y. Chiao, Paul G. Kwiat and Aephraim M. Steinberg
F
or experimentalists studying quantum mechanics, the fantastic often turns into reality. A recent example emerges from the study of a
phenomenon known as nonlocality, or
Òaction at a distance.Ó This concept calls
into question one of the most fundamental tenets of modern physics, the
proposition that nothing travels faster
than the speed of light.
An apparent violation of this proposition occurs when a particle at a wall
vanishes, only to reappearÑalmost instantaneouslyÑon the other side. A reference to Lewis Carroll may help here.
When Alice stepped through the looking glass, her movement constituted in
some sense action at a distance, or nonlocality: her eÝortless passage through
a solid object was instantaneous. The
particleÕs behavior is equally odd. If we
attempted to calculate the particleÕs average velocity, we would Þnd that it exceeded the speed of light.
Is this possible? Can one of the most
famous laws of modern physics be
breached with impunity? Or is there
something wrong with our conception
of quantum mechanics or with the idea
of a Òtraversal velocityÓ? To answer
such questions, we and several other
workers have recently conducted many
optical experiments to investigate some
of the manifestations of quantum nonlocality. In particular, we focus on three
RAYMOND Y. CHIAO, PAUL G. KWIAT
and AEPHRAIM M. STEINBERG have been
using nonlinear optics to study several fundamental features of quantum mechanicsÑnamely, interference, nonlocality and
tunneling. As an undergraduate at Princeton University, Chiao was directed by John
A. Wheeler to quantize gravity. Despite his
failure in this monumental task, Chiao received his bachelorÕs degree in 1961. He
received his Ph.D. from the Massachusetts
Institute of Technology under the tutelage of Charles Townes and since 1967 has
been a professor of physics at the University of California, Berkeley. A fellow of
52
demonstrations of nonlocal eÝects. In
the Þrst example, we ÒraceÓ two photons, one of which must move through
a Òwall.Ó In the second instance, we look
at how the race is timed, showing that
each photon travels along the two different race paths simultaneously. The
Þnal experiment reveals how the simultaneous behavior of photon twins
is coupled, even if the twins are so far
apart that no signal has time to travel
between them.
T
he distinction between locality
and nonlocality is related to the
concept of a trajectory. For example, in the classical world a rolling
croquet ball has a deÞnite position at
every moment. If each moment is captured as a snapshot and the pictures
are joined, they form a smooth, unbroken line, or trajectory, from the playerÕs mallet to the hoop. At each point
on this trajectory, the croquet ball has
a deÞnite speed, which is related to its
kinetic energy. If it travels on a ßat
pitch, it rolls to its target. But if the ball
begins to roll up a hill, its kinetic energy is converted into potential energy.
As a result, it slowsÑeventually to stop
and roll back down. In the jargon of
physics such a hill is called a barrier,
because the ball does not have enough
energy to travel over it, and, classically,
it always rolls back. Similarly, if Alice
the American Physical Society, Chiao is
described by his students as a concertquality pianist to within experimental error. Kwiat is a postdoctoral fellow studying quantum optics at Innsbruck University. He received his B.S. from M.I.T. in 1987
and recently earned his Ph.D. under ChiaoÕs
direction. He is also devoted to the study
of aikido, a Japanese martial art. Steinberg
received his B.S. in physics from Yale University in 1988. He worked at the ƒcole
Normale SupŽrieure for one year before
becoming a Ph.D. student of ChiaoÕs. He
spends most of each day doing physics
and wishing he had more time to ski.
SCIENTIFIC AMERICAN August 1993
were unable to hit croquet balls (or
rolled-up hedgehogs, as Carroll would
have them) with enough energy to send
them crashing through a brick wall,
they would merely bounce oÝ.
According to quantum mechanics,
this concept of a trajectory is ßawed.
The position of a quantum mechanical
particle, unlike that of a croquet ball, is
not described as a precise mathematical point. Rather the particle is best represented as a smeared-out wave packet.
This packet can be seen as resembling
the shell of a tortoise, because it rises
from its leading edge to a certain height
and then slopes down again to its trailing edge. The height of the wave at a
given position along this span indicates
the probability that the particle occupies
that position: the higher a given part of
the wave packet, the more likely the particle is located there. The width of the
packet from front to back represents
the intrinsic uncertainty of the particleÕs
location [see box on page 57]. When the
particle is detected at one point, however, the entire wave packet disappears.
Quantum mechanics does not tell us
where the particle has been before this
moment.
This uncertainty in location leads to
one of the most remarkable consequences of quantum mechanics. If the
hedgehogs are quantum mechanical,
then the uncertainty of position permits the beasts to have a very small
but perfectly real chance of appearing
on the far side of the wall. This process
is known as tunneling and plays a major role in science and technology. Tunneling is of central importance in nuclear fusion, certain high-speed electronic devices, the highest-resolution
microscopes in existence and some
theories of cosmology.
In spite of the name Òtunneling,Ó the
barrier is intact at all times. In fact, if a
particle were inside the barrier, its kinetic energy would be negative. Velocity is proportional to the square root of
the kinetic energy, and so in the tunneling case one must take the square
Copyright 1993 Scientific American, Inc.
root of a negative number. Hence, it
is impossible to ascribe a real velocity
to the particle in the barrier. This is
why when looking at the watch it has
borrowed from the White Rabbit, the
hedgehog that has tunneled to the far
side of the wall wearsÑlike most physicists since the 1930sÑa puzzled expression. What time does the hedgehog
see? In other words, how long did it take
to tunnel through the barrier?
Over the years, many attempts have
been made to answer the question of the
tunneling time, but none has been universally accepted. Using photons rather
than hedgehogs, our group has recently
completed an experiment that provides
one concrete deÞnition of this time.
Photons are the elementary particles
from which all light is made; a typical
light bulb emits more than 100 billion such particles in one billionth of a
second. Our experiment does not need
nearly so many of them. To make our
measurements, we used a light source
that emits a pair of photons simultaneously. Each photon travels toward a
diÝerent detector. A barrier is placed
in the path of one of these photons,
whereas the other is allowed to ßy unimpeded. Most of the time, the Þrst photon bounces oÝ the barrier and is lost;
only its twin is detected. Occasionally, however, the Þrst photon tunnels
through the barrier, and both photons
reach their respective detectors. In this
situation, we can compare their arrival
times and thus see how long the tunneling process took.
The role of the barrier was played by
a common optical element: a mirror.
This mirror, however, is unlike the ordinary household variety (which relies on
metallic coating and absorbs as much as
15 percent of the incident light). The
laboratory mirrors consist of thin, alternating layers of two diÝerent types
of transparent glass, through which
light travels at slightly diÝerent speeds.
These layers act as periodic Òspeed
bumps.Ó Individually, they would do little more than slow the light down. But
when taken together and spaced appropriately, they form a region in which
light Þnds it essentially impossible to
travel. A multilayer coating one micron
thickÑone one-hundredth of the diameter of a typical human hairÑreßects
ÒTUNNELINGÓ ALICE moves eÝortlessly through a mirror,
much as photons do in experiments in quantum optics. Although he was not a physicist, Lewis Carroll almost seems to
Copyright 1993 Scientific American, Inc.
99 percent of incident light at the photon energy (or, equivalently, the color
of the light) for which it is designed.
Our experiment looks at the remaining
1 percent of the photons, which tunnel
through this looking glass.
D
uring several days of data collection, more than one million
photons tunneled through the
barrier, one by one. We compared the
arrival times for tunneling photons and
for photons that had been traveling
unimpeded at the speed of light. ( The
speed of light is so great that conventional electronics are hundreds of thousands of times too slow to perform the
timing; the technique we used will be
described later, as a second example of
quantum nonlocality.)
The surprising result: on average, the
tunneling photons arrived before those
that traveled through air, implying an
average tunneling velocity of about 1.7
times that of light. The result appears to
contradict the classical notion of causality, because, according to EinsteinÕs
theory of relativity, no signal can travel
faster than the speed of light. If signals
have anticipated a thorny 20th-century physics problemÑ
that of the tunneling timeÑwhen he had Sir John Tenniel draw
a strange face on the looking-glass clock.
SCIENTIFIC AMERICAN August 1993
53
LOOKING-GLASS CROQUET has Alice hitting rolled-up hedgehogs, each bearing an uncanny resemblance to a young Werner Heisenberg, toward a wall. Classically, the hedgehogs al-
could move faster, eÝects could precede
causes from the viewpoints of certain
observers. For example, a light bulb
might begin to glow before the switch
was thrown.
The situation can be stated more
precisely. If at some deÞnite time you
made a decision to start Þring photons
at a mirror by opening a starting gate,
and someone else sat on the other side
of the mirror looking for photons, how
much time would elapse before the
other person knew you had opened the
gate? At Þrst, it might seem that since
the photon tunnels faster than light
she would see the light before a signal
traveling at the theoretical speed limit
could have reached her, in violation of
the Einsteinian view of causality. Such
a state of aÝairs seems to suggest an array of extraordinary, even bizarre communication technologies. Indeed, the
implications of faster-than-light inßuences led some physicists in the early
part of the century to propose alternatives to the standard interpretation of
quantum mechanics.
Is there a quantum mechanical way
out of this paradox? Yes, there is, although it deprives us of the exciting
possibility of toying with cause and effect. Until now, we have been talking
54
ways bounce oÝ. Quantum mechanically, however, a small
probability exists that a hedgehog will appear on the far side.
The puzzle facing quantum physicists: How long does it take
about the tunneling velocity of photons
in a classical context, as if it were a
directly observable quantity. The Heisenberg uncertainty principle, however, indicates that it is not. The time of
emission of a photon is not precisely
deÞned, so neither is its exact location
or velocity. In truth, the position of a
photon is more correctly described by a
bell-shaped probability distributionÑ
our tortoise shellÑwhose width corresponds to the uncertainty of its location.
A relapse into metaphor might help
to explain the point. The nose of each
tortoise leaves the starting gate the instant of opening. The emergence of the
nose marks the earliest time at which
there is any possibility for observing a
photon. No signal can ever be received
before the nose arrives. But because of
the uncertainty of the photonÕs location,
on average a short delay exists before
the photon crosses the gate. Most of the
tortoise (where the photon is more likely to be detected) trails behind its nose.
For simplicity, we label the probability distribution of the photon that travels unimpeded to the detector as Òtortoise 1Ó and that of the photon that
tunnels as Òtortoise 2.Ó When tortoise 2
reaches the tunnel barrier, it splits into
two smaller tortoises: one that is reßect-
SCIENTIFIC AMERICAN August 1993
ed back toward the start and one that
crosses the barrier. These two partial
tortoises together represent the probability distribution of a single photon.
When the photon is detected at one position, its other partial tortoise instantly disappears. The reßected tortoise is
bigger than the tunneling tortoise simply because the chances of reßection are
greater than that of transmission (recall
that the mirror reßects a photon 99 percent of the time).
We observe that the peak of tortoise
2Õs shell, representing the most likely
position of the tunneling photon, reaches the Þnish line before the peak of tortoise 1Õs shell. But tortoise 2Õs nose arrives no earlier than the nose of tortoise
1. Because the tortoisesÕ noses travel at
the speed of light, the photon that signals the opening of the starting gate
can never arrive earlier than the time
allowed by causality [see illustration on
opposite page].
In a typical experiment, however, the
nose represents a region of such low
probability that a photon is rarely observed there. The whereabouts of the
photon, detected only once, are best
predicted by the location of the peak.
So even though the tortoises are nose
and nose at the Þnish, the peak of tor-
Copyright 1993 Scientific American, Inc.
to go through the wall? Does the traversal time violate Albert EinsteinÕs famous speed limit?
toise 2Õs shell precedes that of tortoise
1Õs (remember, the transmitted tortoise
is smaller than tortoise 1). A photon
tunneling through the barrier is therefore most likely to arrive before a photon traveling unimpeded at the speed
of light. Our experiment conÞrmed this
prediction.
But we do not believe that any individual part of the wave packet moves
faster than light. Rather the wave packet gets ÒreshapedÓ as it travels, until the
peak that emerges consists primarily of
what was originally in front. At no
point does the tunneling-photon wave
packet travel faster than the free-traveling photon. In 1982 Steven Chu of Stanford University and Stephen Wong, then
at AT&T Bell Laboratories, observed a
similar reshaping eÝect. They experimented with laser pulses consisting of
many photons and found that the few
photons that made it through an obstacle arrived sooner than those that could
move freely. One might suppose that
only the Þrst few photons of each pulse
were ÒallowedÓ through and thus dismiss the reshaping eÝect. But this interpretation is not possible in our case, because we study one photon at a time.
At the moment of detection, the entire
photon ÒjumpsÓ instantly into the transmitted portion of the wave packet, beating its twin to the Þnish more than half
the time.
Although reshaping seems to account
for our observations, the question still
lingers as to why reshaping should occur in the Þrst place. No one yet has any
physical explanation for the rapid tunneling. In fact, the question had puzzled
investigators as early as the 1930s,
when physicists such as Eugene Wigner
of Princeton University had noticed that
quantum theory seemed to imply such
high tunneling speeds. Some assumed
that approximations used in that prediction must be incorrect, whereas others held that the theory was correct but
required cautious interpretation. Some
researchers, in particular Markus BŸttiker and Rolf Landauer of the IBM Thomas
J. Watson Research Center, suggest that
quantities other than the arrival time of
the wave packetÕs peak (for example,
the angle through which a ÒspinningÓ
RACING TORTOISES help to characterize tunneling time. Each
represents the probability distribution of the position of a photon. The peak is where a photon is most likely to be detected.
The tortoises start together (left). Tortoise 2 encounters a barrier and splits in two (right). Because the chance of tunneling is
low, the transmitted tortoise is small, whereas the reßected one
Copyright 1993 Scientific American, Inc.
particle rotates while tunneling) might
be more appropriate for describing the
time ÒspentÓ inside the barrier. Although
quantum mechanics can predict a particleÕs average arrival time, it lacks the
classical notion of trajectories, without
which the meaning of time spent in a
region is unclear.
One hint to explain fast tunneling time
stems from a peculiar characteristic of
the phenomenon. According to theory,
an increase in the width of the barrier
does not lengthen the time needed by
the wave packet to tunnel through. This
observation can be roughly understood
using the uncertainty principle. SpeciÞcally, the less time we spend studying
a photon, the less certain we can be of
its energy. Even if a photon Þred at a
barrier does not have enough energy
to cross it, in some sense a short period initially exists during which the particleÕs energy is uncertain. During this
time, it is as though the photon could
temporarily borrow enough extra energy to make it across the barrier. The
length of this grace period depends
only on the borrowed energy, not on
the width of the barrier. No matter how
wide the barrier becomes, the transit
time across it remains the same. For a
suÛciently wide barrier, the apparent
traversal speed would exceed the speed
of light.
O
bviously, for our measurements
to be meaningful, our tortoises
had to run exactly the same distance. In essence, we had to straighten
the racetrack so that neither tortoise
had the advantage of the inside lane.
Then, when we placed a barrier in one
path, any delay or acceleration would
be attributed solely to quantum tunneling. One way to set up two equal lanes
would be to determine how much time
is nearly as tall as the original. On those rare occasions of
tunneling, the peak of tortoise 2Õs shell on average crosses the
Þnish line ÞrstÑimplying an average tunneling velocity of 1.7
times the speed of light. But the tunneling tortoiseÕs nose never
travels faster then lightÑnote that both tortoises remain Ònose
and noseÓ at the end. Hence, EinsteinÕs law is not violated.
SCIENTIFIC AMERICAN August 1993
55
a
b
DETECTOR
BOTH PHOTONS TRANSMITTED
BARRIER
MIRROR
BEAM
SPLITTER
DETECTOR
c
BOTH PHOTONS REFLECTED
PHOTON
PHOTON
MIRROR
DOWN-CONVERSION
CRYSTAL
TWIN-PHOTON INTERFEROMETER (a) precisely times racing
photons. The photons are born in a down-conversion crystal
and are directed by mirrors to a beam splitter. If one photon
beats the other to the beam splitter (because of the barrier),
both detectors will be triggered in about half the races. Two
possibilities lead to such coincidence detections: both photons
are transmitted by the beam splitter (b), or both are reßected
it takes for a photon to travel from the
source to the detector for each path.
Once the times were equal, we would
know the paths were also equal.
But performing such a measurement
with a conventional stopwatch would require one whose hands went around
nearly a billion billion times per minute.
Fortunately, Leonard Mandel and his
co-workers at the University of Rochester have developed an interference technique that can time our photons.
MandelÕs quantum stopwatch relies
on an optical element called a beam
splitter [see illustration above]. Such a
device transmits half the photons striking it and reßects the other half. The
racetrack is set up so that two photon
wave packets are released at the same
time from the starting gate and approach the beam splitter from opposite
sides. For each pair of photons, there are
four possibilities: both photons might
pass through the beam splitter; both
might rebound from the beam splitter;
both could go oÝ together to one side;
or both could go oÝ together to the other side. The Þrst two possibilitiesÑthat
both photons are transmitted or both
reßectedÑresult in what are termed
coincidence detections. Each photon
reaches a diÝerent detector (placed on
either side of the beam splitter), and
both detectors are triggered within a
56
SCIENTIFIC AMERICAN August 1993
(c). Aside from their arrival times, there is no way of determining which photon took which route; either could have traversed
the barrier. (This nonlocality actually sustains the performance
of the interferometer.) If both photons reach the beam splitter
simultaneously, for quantum reasons they will head in the
same direction, so that both detectors do not go oÝ. The two
possibilities shown are then said to interfere destructively.
billionth of a second of each other. Unfortunately, this time resolution is about
how long the photons take to run the
entire race and hence is much too
coarse to be useful.
So how do the beam splitter and the
detectors help in the setup of the racetrack? We simply tinker with the length
of one of the paths until all coincidence
detections disappear. By doing so, we
make the photons reach the beam splitter at the same time, eÝectively rendering the two racing lanes equal. Admittedly, the proposition sounds peculiarÑ
after all, equal path lengths would seem
to imply coincident arrivals at the two
detectors. Why would the absence of
such events be the desired signal?
The reason lies in the way quantum
mechanical particles interact with one
another. All particles in nature are either bosons or fermions. Identical fermions (electrons, for example) obey the
Pauli exclusion principle, which prevents any two of them from ever being
in the same place at the same time. In
contrast, bosons (such as photons) like
being together. Thus, after reaching the
beam splitter at the same time, the two
photons prefer to head in the same direction. This preference leads to the detection of fewer coincidences (none, in
an ideal experiment) than would be the
case if the photons acted independent-
ly or arrived at the beam splitter at different times.
Therefore, to make sure the photons
are in a fair race, we adjust one of the
path lengths. As we do this, the rate of
coincident detections goes through a
dip whose minimum occurs when the
photons take exactly the same amount
of time to reach the beam splitter. The
width of the dip (which is the limiting
factor in the resolution of our experiments) corresponds to the size of the
photon wave packetsÑtypically, about
the distance light moves in a few hundredths of a trillionth of a second.
Only when we knew that the two
path lengths were equal did we install
the barrier and begin the race. We then
found that the coincidence rates were
no longer at a minimum, implying that
one of the photons was reaching the
beam splitter Þrst. To restore the minimum, we had to lengthen the path taken
by the tunneling photon. This correction
indicates that photons take less time to
cross a barrier than to travel in air.
E
ven though investigators designed
racetracks for photons and a clever timekeeping device for the
race, the competition still should have
been diÛcult to conduct. The fact that
the test could be carried out at all constitutes a second validation of the prin-
Copyright 1993 Scientific American, Inc.
ciple of nonlocality, if not for which precise timing of the race would have been
impossible. To determine the emission
time of a photon most precisely, one
would obviously like the photon wave
packets to be as short as possible. The
uncertainty principle, however, states
that the more accurately one determines the emission time of a photon,
the more uncertainty one has to accept
in knowing its energy, or color [see box
below ].
Because of the uncertainty principle, a
fundamental trade-oÝ should emerge in
our experiments. The colors that make
up a photon will disperse in any kind of
glass, widening the wave packet and reducing the precision of the timing. Dispersion arises from the fact that different colors travel at various speeds in
glassÑblue light generally moves more
slowly than red. A familiar example of
dispersion is the splitting of white light
into its constituent colors by a prism.
As a short pulse of light travels
through a dispersive medium (the bar-
rier itself or one of the glass elements
used to steer the light), it spreads out
into a ÒchirpedÓ pulse: the redder part
pulls ahead, and the bluer hues lag behind [see illustration on next page]. A
simple calculation shows that the width
of our photon pulses would quadruple
on passage through an inch of glass. The
presence of such broadening should
have made it well nigh impossible to
tell which tortoise crossed the Þnish
line Þrst. Remarkably, the widening of
the photon pulse did not degrade the
precision of our timing.
Herein lies our second example of
quantum nonlocality. Essentially both
twin photons must be traveling both
paths simultaneously. Almost magically, potential timing errors cancel out as
a result.
To understand this cancellation effect, we need to examine a special property of our photon pairs. The pairs are
born in what physicists call Òspontaneous parametric down-conversion.Ó The
process occurs when a photon travels
into a crystal that has nonlinear optical
properties. Such a crystal can absorb a
single photon and emit a pair of other photons, each with about half the
energy of the parent, in its place (this
is the meaning of the phrase ÒdownconversionÓ). An ultraviolet photon, for
instance, would produce two infrared
ones. The two photons are emitted simultaneously, and the sum of their energies exactly equals the energy of the
parent photon. In other words, the colors of the photon pairs are correlatedÑ
if one is slightly bluer (and thus travels more slowly in glass), then the other must be slightly redder (and must
travel more quickly).
One might think that diÝerences
between siblings might aÝect the outcome of the raceÑone tortoise might
be more athletic than the other. Yet because of nonlocality, any discrepancy
between the pair proves irrelevant. The
key point is that neither detector has
any way of identifying which of the
photons took which path. Either photon
Wave Packets
A
good way to understand wave packets is to construct
one, by adding together waves of different frequencies. We start with a central frequency (denoted by the
green curve), a wave with no beginning and no end. If we
now add two more waves of slightly lower and higher frequency (orange and blue curves, respectively), we obtain
a pulselike object (white curve). When enough frequencies
are added, a true pulse, or wave packet, can be formed,
which is confined to a small region of space. If the range
of frequencies used to make the pulse were decreased
(for example, by using colors only from yellow to green,
instead of from orange to blue), we would create a longer
pulse. Conversely, if we had included all colors from red
to violet, the packet could have been even shorter.
Mathematically speaking, if we use Dn for the width of
the range of colors and Dt for the duration of the pulse,
then we can write
Copyright 1993 Scientific American, Inc.
Dn Dt ≥ 1/4p,
which simply expresses the fact that a wider color range
is needed to make a shorter wave packet. It holds true for
any kind of wave—light, sound, water and so on.
The phenomenon acquires new physical significance
when one makes the identification of electromagnetic frequency, n, with photon energy, E, via the Planck-Einstein relation E = h n, where h is Planck’s constant. The particle aspect of quantum mechanics enters at this point. In other
words, a photon’s energy depends on its color. Red photons have about three fifths the energy of blue ones. The
above mathematical expression can then be rewritten as
DE Dt ≥ h /4p.
Physicists have become so attached to this formula that
they have named it: Heisenberg’s uncertainty principle.
(An analogous and perhaps more familiar version exists
for position and momentum.) One consequence of this
principle for the experiments described in the article is
that it is strictly impossible, even with a perfect apparatus, to know precisely both the time of emission of a photon and its energy.
Although we arrived at the uncertainty principle by considering the construction of wave packets, its application
is remarkably far more wide-reaching and its connotations far more general. We cannot overemphasize that
the uncertainty is inherent in the laws of nature. It is not
merely a result of inaccurate measuring devices in our
laboratories. The uncertainty principle is what keeps electrons from falling into the atomic nucleus, ultimately limits the resolution of microscopes and, according to some
astrophysical theories, was initially responsible for the nonuniform distribution of matter in the universe.
SCIENTIFIC AMERICAN August 1993
57
might have passed through the barrier.
Having two or more coexisting possibilities that lead to the same Þnal
outcome results in what is termed an
interference eÝect. Here each photon
takes both paths simultaneously, and
these two possibilities interfere with
each other. That is, the possibility that
the photon that went through the glass
was the redder (faster) one interferes
with the possibility that it was the bluer (slower) one. As a result, the speed
diÝerences balance, and the eÝects of
dispersion cancel out. The dispersive
widening of the individual photon pulses is no longer a factor. If nature acted locally, we would have been hardpressed to conduct any measurements.
The only way to describe what happens
is to say that each twin travels through
both the path with the barrier and the
free path, a situation that exempliÞes
nonlocality.
T
hus far we have discussed two
nonlocal results from our quantum experiments. The Þrst is the
measurement of tunneling time, which
requires two photons to start a race at
exactly the same time. The second is
the dispersion cancellation eÝect, which
relies on a precise correlation of the racing photonsÕ energies. In other words,
the photons are said to be correlated in
energy (what they do) and in time (when
they do it). Our Þnal example of nonlocality is eÝectively a combination of the
Þrst two. SpeciÞcally, one photon ÒreactsÓ to what its twin does instantaneously, no matter how far apart they are.
Knowledgeable readers may protest
at this point, claiming that the Heisenberg uncertainty principle forbids precise speciÞcation of both time and energy. And they would be right, for a
single particle. For two particles, however, quantum mechanics allows us
to deÞne simultaneously the diÝerence
between their emission times and the
sum of their energies, even though neither particleÕs time or energy is speciÞed. This fact led Einstein, Boris Podolsky and Nathan Rosen to conclude
that quantum mechanics is an incomplete theory. In 1935 they formulated
a thought experiment to demonstrate
what they believed to be the shortcomings of quantum mechanics.
If one believes quantum mechanics,
the dissenting physicists pointed out,
then any two particles produced by a
process such as down-conversion are
coupled. For example, suppose we measure the time of emission of one particle. Because of the tight time correlation between them, we could predict
with certainty the emission time of the
other particle, without ever disturbing
it. We could also measure directly the
energy of the second particle and then
infer the energy of the Þrst particle.
Somehow we would have managed to
determine precisely both the energy
and the time of each particleÑin eÝect,
beating the uncertainty principle. How
can we understand the correlations
and resolve this paradox?
There are basically two options. The
Þrst is that there exists what Einstein
called Òspooklike actions at a distanceÓ
(spukhafte Fernwirkungen). In this scenario, the quantum mechanical description of particles is the whole story. No
particular time or energy is associated
with any photon until, for example, an
energy measurement is made. At that
point, only one energy is observed.
Because the energies of the two photons sum to the deÞnite energy of the
parent photon, the previously undetermined energy of the twin photon, which
we did not measure, must instantaneously jump to the value demanded by
energy conservation. This nonlocal ÒcollapseÓ would occur no matter how far
away the second photon had traveled.
The uncertainty principle is not violated, because we can specify only one
variable or the other : the energy measurement disrupts the system, instantaneously introducing a new uncertainty in the time.
Of course, such a crazy, nonlocal
model should not be accepted if a
simpler way exists to understand the
correlations. A more intuitive explanation is that the twin photons leave the
source at deÞnite, correlated times, carrying deÞnite, correlated energies. The
fact that quantum mechanics cannot
specify these properties simultaneously would merely indicate that the theory is incomplete.
Einstein, Podolsky and Rosen advocated the latter explanation. To them,
there was nothing at all nonlocal in the
observed correlations between particle
pairs, because the properties of each
particle are determined at the moment
of emission. Quantum mechanics was
only correct as a probabilistic theory, a
kind of photon sociology, and could
not completely describe all individual
particles. One might imagine that there
exists an underlying theory that could
predict the speciÞc results of all possible measurements and show that particles act locally. Such a theory would be
based on some hidden variable yet to
be discovered. In 1964 John S. Bell of
CERN, the European laboratory for particle physics near Geneva, established a
theorem showing that all invocations
of local, hidden variables give predictions diÝerent from those stated by
quantum mechanics.
S
ince then, experimental results
have supported the nonlocal
(quantum mechanical) picture and
contradicted the intuitive one of Einstein, Podolsky and Rosen. Much of the
credit for the pioneering work belongs
to the groups led by John Clauser of
the University of California at Berkeley
and Alain Aspect, now at the Institute
of Optics in Orsay. In the 1970s and
GLASS
DISPERSION of a light pulse occurs because each color travels at a diÝerent speed. A short light pulse passing through a
58
SCIENTIFIC AMERICAN August 1993
piece of glass will broaden into a ÒchirpedÓ wave packet : the
redder colors pull ahead while the bluer hues lag behind.
Copyright 1993 Scientific American, Inc.
early 1980s they examined the correlations between polarizations in photons. The more recent work of John G.
Rarity and Paul R. Tapster of the Royal Signals and Radar Establishment in
England explored correlations between
the momentum of twin photons. Our
group has taken the tests one step further. Following an idea proposed by
James D. Franson of Johns Hopkins University in 1989, we have performed an
experiment to determine whether some
local hidden variable model, rather than
quantum mechanics, can account for
the energy and time correlations.
In our experiment, photon twins from
our down-conversion crystal are separately sent to identical interferometers
[see illustration at right]. Each interferometer is designed much like an interstate highway with an optional detour.
A photon can take a short path, going
directly from its source to its destination. Or it can take the longer, detour
path (whose length we can adjust) by
detouring through the rest station before continuing on its way.
Now watch what happens when we
send the members of a pair of photons through these interferometers.
Each photon will randomly choose the
long route (through the detour) or the
shorter, direct route. After following one
of the two paths, a photon can leave
its interferometer through either of two
ports, one labeled ÒupÓ and the other
Òdown.Ó We observed that each particle
was as likely to leave through the up
port as it was through the down. Thus,
one might intuitively presume that the
photonÕs choice of one exit would be unrelated to the exit choice its twin makes
in the other interferometer. Wrong. Instead we see strong correlations between which way each photon goes
when it leaves its interferometer. For
certain detour lengths, for example,
whenever the photon on the left leaves
at the up exit, its twin on the right ßies
through its own up exit.
One might suspect that this correlation is built in from the start, as when
one hides a white pawn in one Þst and
a black pawn in the other. Because their
colors are well deÞned at the outset,
we are not surprised that the instant
we Þnd a white pawn in one hand, we
know with certainty that the other
must be black.
But a built-in correlation cannot account for the actual case in our experiment, which is much stranger: by
changing the path length in either interferometer, we can control the nature
of the correlations. We can go smoothly from a situation where the photons
always exit the corresponding ports
(both use the up port, or both use the
Copyright 1993 Scientific American, Inc.
COINCIDENCE COUNTER
DETECTOR
DETECTOR
UP
PORT
UP
PORT
MIRRORS
MIRRORS
DOWN
PORT
DOWN
PORT
BEAM
SPLITTERS
BEAM
SPLITTERS
INTERFEROMETER
MIRRORS MOVE
TO VARY PATH LENGTH
PHOTON
PHOTON
DOWN-CONVERSION CRYSTAL
NONLOCAL CORRELATION between two particles is demonstrated in the so-called
Franson experiment, which sends two photons to separate but identical interferometers. Each photon may take a short route or a longer ÒdetourÓ at the Þrst beam
splitter. They may leave through the upper or lower exit ports. A detector looks at
the photons leaving the upper exit ports. Before entering its interferometer, neither photon knows which way it will go. After leaving, each knows instantly and
nonlocally what its twin has done and so behaves accordingly.
down port) of their respective interferometers to one in which they always
exit opposite ports. In principle, such a
correlation would exist even if we adjusted the path length after the photons
had left the source. In other words, before entering the interferometer, neither
photon knows which way it is going to
have to goÑbut on leaving, each one
knows instantly (nonlocally) what its
twin has done and behaves accordingly.
To analyze these correlations, we look
at how often the photons emerge from
each interferometer at the same time
and yield a coincidence count between
detectors placed at the up exit ports of
the two interferometers. Varying either
of the long-arm path lengths does not
change the rate of detections at either
detector individually. It does, however,
aÝect the rate of coincidence counts, indicating the correlated behavior of each
photon pair. This variation produces
ÒfringesÓ reminiscent of the light and
dark stripes in the traditional two-slit
interferometer showing the wave nature
of particles.
In our experiment, the fringes imply a peculiar interference eÝect. As alluded to earlier, interference can be expressed as the result of two or more indistinguishable, coexisting possibilities
leading to the same Þnal outcome (recall our second example of nonlocality, in which each photon travels along
two diÝerent paths simultaneously, producing interference). In the present case,
there are two possible ways for a coincidence count to occur: either both
photons had to travel the short paths,
or both photons had to travel the long
paths. (In the cases in which one photon
travels a short path and the other a long
path, they arrive at diÝerent times and
SCIENTIFIC AMERICAN August 1993
59
OBSERVED
COINCIDENCE COUNTING RATE
MAXIMUM
PREDICTED
BY LOCAL
THEORIES
0
90
180
PHASE (DEGREES)
270
360
RATE OF COINCIDENCES between left and right detectors in the Franson experiment (red dots, with best-Þt line) strongly suggests nonlocality. The horizontal axis
represents the sum of the two long path lengths, in angular units known as phases.
The Òcontrast,Ó or the degree of variation in these rates, exceeds the maximum allowed by local, realistic theories (blue line), implying that the correlations must be
nonlocal , as shown by John S. Bell of CERN.
so do not interfere with each other; we
discard these counts electronically.)
The coexistence of these two possibilities suggests a classically nonsensical picture. Because each photon arrives at the detector at the same time
after having traveled both the long and
short routes, each photon was emitted
ÒtwiceÓÑonce for the short path and
once for the long path.
To see this, consider the analogy in
which you play the role of one of the
detectors. You receive a letter from a
friend on another continent. You know
the letter arrived via either an airplane
or a boat, implying that it was mailed a
week ago (by plane) or a month ago (by
boat). For an interference eÝect to exist,
the one letter had to have been mailed
at both times. Classically, of course, this
possibility is absurd. But in our experiments the observation of interference
fringes implies that each of the twin
photons possessed two indistinguishable times of emission from the crystal. Each photon has two birthdays.
More important, the exact form of the
interference fringes can be used to differentiate between quantum mechanics
and any conceivable local hidden variable theory (in which, for example,
each photon might be born with a deÞnite energy or already knowing which
60
SCIENTIFIC AMERICAN August 1993
exit port to take). According to the constraints derived by Bell, no hidden variable theory can predict sinusoidal fringes that exhibit a ÒcontrastÓ of greater
than 71 percentÑthat is, the diÝerence
in intensity between light and dark
stripes has a speciÞc limit. Our data,
however, display fringes that have a
contrast of about 90 percent. If certain
reasonable supplementary assumptions
are made, one can conclude from these
data that the intuitive, local, realistic
picture suggested by Einstein and his
cohorts is wrong: it is impossible to explain the observed results without acknowledging that the outcome of a
measurement on the one side depends
nonlocally on the result of a measurement on the other side.
S
o is EinsteinÕs theory of relativity
in danger? Astonishingly, no, because there is no way to use the
correlations between particles to send
a signal faster than light. The reason is
that whether each photon reaches its
detector or instead uses the down exit
port is a random result. Only by comparing the apparently random records
of counts at the two detectors, necessarily bringing our data together, can
we notice the nonlocal correlations. The
principles of causality remain inviolate.
Science-Þction buÝs may be saddened
to learn that faster-than-light communication still seems impossible. But several scientists have tried to make the
best of the situation. They propose to
use the randomness of the correlations
for various cipher schemes. Codes produced by such quantum cryptography
systems would be absolutely unbreakable [see ÒQuantum Cryptography,Ó by
Charles H. Bennett, Gilles Brassard and
Artur D. Ekert; SCIENTIFIC AMERICAN,
October 1992].
We have thus seen nonlocality in
three diÝerent instances. First, in the
process of tunneling, a photon is able
to somehow sense the far side of a barrier and cross it in the same amount of
time no matter how thick the barrier
may be. Second, in the high-resolution
timing experiments, the cancellation of
dispersion depends on each of the two
photons having traveled both paths in
the interferometer. Finally, in the last experiment discussed, a nonlocal correlation of the energy and time between two
photons is evidenced by the photonsÕ
coupled behavior after leaving the interferometers. Although in our experiments the photons were separated by
only a few feet, quantum mechanics predicts that the correlations would have
been observed no matter how far apart
the two interferometers were.
Somehow nature has been clever
enough to avoid any contradiction with
the notion of causality. For in no way is
it possible to use any of the above effects to send signals faster than the
speed of light. The tenuous coexistence
of relativity, which is local, and quantum mechanics, which is nonlocal, has
weathered yet another storm.
FURTHER READING
QED: THE STRANGE THEORY OF LIGHT
AND MATTER. Richard P. Feynman.
Princeton University Press, 1985.
SPEAKABLE AND UNSPEAKABLE IN QUANTUM MECHANICS. J. S. Bell. Cambridge
University Press, 1988.
HIGH-VISIBILITY INTERFERENCE IN A BELLINEQUALITY EXPERIMENT FOR ENERGY
AND TIME. P. G. Kwiat, A. M. Steinberg
and R. Y. Chiao in Physical Review A ,
Vol. 47, No. 4, pages R2472ÐR2475;
April 1, 1993.
THE SINGLE-PHOTON TUNNELING TIME.
A. M. Steinberg, P. G. Kwiat and R. Y.
Chiao in Proceedings of the XXVIIIth
Rencontre de Moriond. Edited by Jean
^ Thanh Van.
^
Tran
Editions Fronti•res,
Gif-sur-Yvette, France ( in press).
SUPERLUMINAL (BUT CAUSAL) PROPAGATION OF WAVE PACKETS IN TRANSPARENT MEDIA WITH INVERTED ATOMIC POPULATIONS. Raymond Y. Chiao in Physical Review A ( in press).
Copyright 1993 Scientific American, Inc.