VOLUME 71, NUMBER 9 PH YSICAL REVIEW LETTERS 30 AUGUST 1993

Interferometric Detection of Optical Phase Shifts at the Heisenberg Limit

M. J. Holland and K. Burnett
Department of Physics, Clarendon Laboratory, Parks Road, Oxford, OX1 9PU, England
(Received 25 January 1993)
We show that the uncertainty in the relative quantum phase of two fields propagating in the
arms of a Mach-Zehnder interferometer can be reduced to the Heisenberg limit by driving the
interferometer with two Fock states containing equal numbers of photons. This leads to a minimum
detectable phase shift far below that of any interferometer driven by a coherent light source.

PACS numbers: 42. 50.Lc, 04.80.+z, 07.60.Ly, 42. 50.Dv

The interferometer is a fundamental apparatus in op- turn phase difference between these input fields is uncer-
tical physics whose output signal is sensitive to the rela- tain. We shall show, however, that the relative phase un-
tive phase shift between two fields traveling down sepa- certainty between the output fields from the first beam
rated paths. The use of interferometers in optical gyro- splitter in the interferometer depends strongly only on
scopes [1] as well as gravitational wave detectors [2] relies the amplitude difference noise of the input fields.
on the ability to resolve extremely small relative shifts in Consider two classical fields with amplitudes n;„
the two path lengths with the smallest detectable shift and P;„and with absolute phases 8;„and P;„ incident
in principle determined by the quantum properties of the on a 50/50 beam splitter as illustrated in Fig. 1. We as-
illuminating field. Zero-point fluctuations in the laser sume the beam splitter introduces a phase shift of 7r/2
and vacuum ports of the input beam splitter produce on reflection. The classical phase difference and intensity
phase difference uncertainty between the fields propagat- difference between the output fields are given by
ing down the two paths. These phase fluctuations are in-
distinguishable from genuine changes in the path length in in
(g QUt oUt)
difference of the two arms. 2 P (P 0 )
In the case of the two port interferometer with a co-
herent laser field and a vacuum field as inputs, the effect (ctout)
2
(Pout)
2
= 2ctinPin sin (4'in ~in) ~

of zero-point fluctuations in the vacuum on the relative

length measurement is amplified by the mean intensity of
If the input fields have equal amplitude ct;„= Pi„, the
beam splitter produces output fields with zero phase dif-
the laser [3—5]. During a measurement interval in which
ference, independent of the absolute and relative phase
the laser supplies an average of n photons, the phase dif-
noise in the input modes. Alternatively, if the input fields
ference uncertainty between the two fields in the inter-
have equal phase P;„= 8;„, the two output ports generate
ferometer arms is 1/~n rad. Increasing the strength of
light of equal intensity.
the incident laser source does therefore increase the res-
In analogy to the classical scheme analyzed above, we
olution of the device. Huge and expensive laser sources
consider two Fock states with the same photon number m
will, however, be required in order to resolve the small
as inputs to a 50/50 beam splitter. We want to examine
perturbations expected by the passage of a gravitational
the nature of the quantum phase distribution of the out-
wave.
A possible mechanism for improving the sensitivity is
to drive the interferometer with nonclassical states of
light as the 1/~n level of relative phase fluctuations for 50/50 p&in
B e am Sp li tter in
a coherent source is well above the Heisenberg limit of
1/n rad [6]. To approach the Heisenberg limit it is neces-
sary to introduce nonlocal quantum correlations between
the photons in the two arms. One proposed scheme
illuminates one of the input ports by a squeezed vac-
uum to reduce the vacuum fluctuations in the appropri-
p+isin &out
ate quadrature [3]. An alternative method is to coher- p f out P
ently drive the optical fields with two strongly correlated
atomic transitions [7].
In this paper, we show that the Heisenberg limit of sen-
sitivity can be realized by driving the interferometer by out
out
two fields with no amplitude difference noise. Since the
amplitude difference and phase difference between opti- FIG. 1. Classical amplitudes and phases for the input and
cal fields are Heisenberg conjugate variables, the quan- output fields of a beam splitter.

0031-9007/93/71(9)/1355(4) $06.00 1355

Q& 1993 The American Physical Society
 VOLUME 71, NUMBER 9 PH YSICAL REVI EW LETTERS 30 AUrUST 1993

put fields from the beam splitter. To do this, we use the generated by two coherent states with an average num-
following basis of s+ 1 states ( 8i) l = 0, . . . , s of well
~ j ber of 50 photons but Poissonian photon statistics. The
defined phase [8,9] to describe the distribution for each width of the resulting phase distribution is much wider
mode, than for the Fock states.
We now want to show how we can exploit this nar-
(2) row phase difFerence distribution produced by Fock states
in an interferometer. To do this we consider our beam
where jp) p = O, . . . , s} denotes the Fock states. The
~
splitter to be the first in a Mach-Zehnder arrangement
rotation between adjacent phase states is e = 2'/(s+ 1). as illustrated in Fig. 3. Providing an equal number of
The corresponding basis for the two mode input field is photons is injected into each of the two input ports, the
then formed from the outer product 8I 8I ) = 8I) 8I ). relative phase uncertainty between the two fields in the
We denote the annihilation operators for the two modes arms of the interferometer will be at the Heisenberg limit.
by a and b. The probability that the phase difference %'e consider input fields from a system which simultane-
between the output fields is A8, where 48 is an inte- ously produces one photon in each mode. In general this
gral multiple of e, can be found by applying the unitary is described by the density operator
transformation for the beam splitter and overlapping the
result with the phase state basis. Finally, tracing over p=)
nn'
„„n)( '.
'
(4)
the possible values for the absolute phase we find
Examples of sources of such states are two photon emis-
S sion and nondegenerate parametric amplification which
P(48) = )
l=O
(8I8I (~e)/, e' i + ) mm)
2
generates the two photon squeezed state.
The coherent mixing of the input fields by the inter-
2 ferometer is a unitary transformation which depends on
—r)! 2r!
1
22m(s + 1)
,
)
r=O
2(m
m rf2 r!2
the path length difference between the two arms. If we
denote the path length difference by z, and let the wave
number of the field be k, then the output state which
is measured by the photodetectors can be calculated by
Figure 2 illustrates this quantum phase distribution for
applying the operator exp[kz(atb —bta)/2] to the in-
m = 50 photons. The distribution is well localized put state. For the mixed state specified in Eq. (4), the
around a phase difference of zero with a width at the
photons divide equally between the two input ports so
Heisenberg limit of 1/(2m) rad. Since no phase origin is
that counting a combined total of 2r photons at the two
defined for either of the inputs, the absolute phase of the
output ports specifies the input field as the dual Fock
output fields must be completely uncertain. I ocalization
state r r). Note that this requires measuring the sum
of the relative phase variable indicates that the beam
current as well as the difference current from the pho-
splitter correlates the phases of the photons at its output
ports. For comparison we have overlaid the distribution
Input
A

Input

B1 M1
(a)

Signal
Processing

0 — ——
—0, 2 0, 0 0, 2
Phase (radians) FIG. 3. The layout of a Mach-Zehnder interferometer. The
sum current from the photon counting detectors Pl and P2
FIG. 2. (a) Overlap of the output field of a beam splitter contains information about the total number of photons ar-
illuminated by two number states containing 50 photons with riving in the input ports A and B. The difFerence current con-
the phase diR'erence states. (b) The much larger phase un- tains information about the relative length of the two possi-
certainty resulting from mixing two coherent states of mean ble paths from the 50/50 beam splitter Bl to the 50/50 beam
50 photons at the same beam splitter. splitter B2. Ml and M2 are reflecting mirrors.
1356
VOLUME 71, NUMBER 9 PHYSICAL REVIEW LETTERS 30 AUGUST 1993

0. 16 0..015 I I I I I IIII I I I I I IIII I I I I 1 1111 I I I I I 111 0..04 I I I I I I I I I I I I I I I II I I I I I I I I I I I I I I I I I I

Modes

0. 12 c5

0
9

nh. Llilll
0 I I I
0 I I I IIII I I I I I I I I II

as 0. 08 10 10 10-' 10 10 10
0
0.. 08 I I I I I IIII I I I I I I IIII I I I I I III 0.. 15 1 1111 I I I I 1 1 111 I I I I 1 1 111 I I I I I 1 11

0. 04 5 Modes 10 Modes

0
C4
0
—30 0 I I I I I IIII I I I I I II ( I I I I I IIII
0 I I IIIII I I I I I I 111 I I I I I IIII I \ I I l Ill

Photon 10 10 10 10 10 10
Number Difference
Phase Difference (radians) Phase Difference (radians)
I IG. 4. The probability distribution for the photon number
difFerence between the two output fields of a Mach-Zehnder FIG. 5. Simulated probability distributions for the path
interferometer. The input field was a dual Pock state with length difFerence for a Mach-Zehnder interferometer contain-
each input port receiving 5000 photons. The relative phase ing 1, 2, 5, and 10 modes. The peaks of the distributions are
shift between the arms was 10 rad. around the actual path length difFerence of 10 rad and be-
come narrower as information from more modes is combined.

todetectors. An alternative method with similar results

is to use a pulsed scheme with very nearly the same pho-
photon number difference, P(kz~2q), is given by
ton number from pulse to pulse [10]. The minimum de-
tectable phase shift is determined primarily by the mean 1
illuminating field and is not sensitive to small fluctua- P(kz~2q) = P(2q~kz)P(kz) . (7)
P 2q
tions in the input intensity.
The probability distribution for the difference in the The fraction 1/P(2q) is a normalization factor and P(kz)
number of photons 2q measured at the two output ports denotes the prior knowledge about kz. Note that in-
for a phase difference kz is given by formation from several spatially independent modes in
the interferometer can be combined. In Fig. 5, we illus-
P(2q~kz) = (r-qr+q e~E"~-~'-l „) (5) trate possible probability distributions for kz after com-
bining the results of 1, 2, 5, and 10 modes. The proce-
If the path length difference is zero the interferometer dure adopted is as follows. We consider a Mach-Zehnder
transmits the input states without modification which interferometer with a phase difference corresponding to
leads to a zero photon number difference for the dual 10 rad. Each input mode is taken to be in a Fock state
Fock state input. An imbalance in the photon number containing 5000 photons. Using the probability distribu-
difference indicates a nonzero path length difference for tion in Eq. (6) and a random number we then simulate a
particular photon number difference between the detec-
the two arms. For q ((
r the probability distribution for
tors at the output ports. Applying Eqs. (6) and (7) we
the difference count reduces to
generate a probability distribution for kz. For simplicity
P(2q~kz) = J,'(kzr), (6) we have assumed a flat prior for the first mode. Infor-
mation from subsequent modes is combined by updating
where J
denotes the Bessel function. This is illustrated the prior.
in Fig. 4 for r = 5000 photons and a phase difference In Fig. 6(a), we simulate the measurement of a phase
of kz = 10 s rad. The width of this distribution is of shift of 10 rad by a five mode interferometer for various
order ~q~ (
kzr This refiects th. e fact that the relative
phase noise in the arms is only limited by the Heisenberg
amplitudes of the input dual Fock states. For each choice
of the input field strength, we select a particular photon
uncertainty principle. We are therefore able to detect number difference for each mode. Using these simulated
phase shifts close to the Heisenberg limit. To see how measurements we construct the probability distribution
this might be done in a practical device, we shall now for the relative path length and calculate the mean value
consider the dependence of the photon number differ- and standard error. Providing the total photon number
ence distribution on the relative path length between the is greater than the Heisenberg limit of 1000 photons, the
two arms. We shall see that we can derive information phase shift can be distinguished from the zero position
about the phase shift from the difference current between of the interferometer. In Fig. 6(b), we show a similar
the photodetectors. The probability distribution for the simulation for a classical interferometer. Injected into
relative path length after measurement of a particular the input ports are a vacuum field and a coherent laser

1357
VOLUME 71, NUMBER 9 PH YSICAL REVI EW LETTERS 30 AUGUST 1993
-3
x10 ing less than 10 photons during the detection period,
most measurements record no photons in the dark port
05 (a)
2 and the phase shift cannot be distinguished from zero.
We have presented a new scheme for measuring phase
V)
shifts at the Heisenberg limit. The important property of
Q5

CL
the input field is found to be equal numbers of photons
0 —— i I
in the two ports. If this is satisfied nonlocal quantum
I I

10 10 1O' 1O' 10
Tota[ Photon Number correlations are generated between the fields in the two
x10 arms of the interferometer. This allows greater resolution
acd(0 than for any interferometer driven by a coherent field.
(b)
a5
Although the two mode squeezed state satisfies this re-
V)
quired input field criteria, the most energy eKcient state
CO
G5
is the dual Fock state.
CL
I I
This work was supported by the Royal Society.
10 1O' 106 107
Average Photon Number

FIG. 6. (a) Simulations of the measurement of a phase shift

of 10 rad for a five mode interferometer driven by dual Fock
states of various amplitudes. We plot the mean value of the
1 W. W. Chou, Rev. Mod. Phys. 57, 61 (1985).
probability distribution for the phase shift derived from the
[2 K. Thorne, Rev. Mod. Phys. 52, 285 (1980).
simulated measurements and use error bars to illustrate the
[3] C. M. Caves, Phys. Rev. D 28, 1693 (1981).
standard error. Note the phase shift is distinguishable from
[4] C. M. Caves, Phys. Rev. Lett. 45, 75 (1980).
zero for input photon numbers down to the Heisenberg limit
[5] R. S. Bondurant and J. H. Shapiro, Phys. Rev. D 80,
(dotted line). (b) As for (a) but with the interferometer driven 2548 (1984).
by a coherent field (see text). The phase shift is resolved only
[6] W. Heitler, The Quantum Theory of Radiation (Oxford
if the input intensity is above the classical limit (dotted line).
University Press, Oxford, 1954), 3rd ed. , p. 65.
[7] M. O. Scully, Phys. Rev. Lett. 55, 2802 (1985).
field. The zero position of the interferometer is adjusted [8] R. Loudon, Quantum Theory of Light (Oxford University
Press, Oxford, 1973), 1st ed.
so that no photons exit from one of the output ports for
[9] D. T. Pegg and S. M. Barnett, Europhys. Lett. 6, 483
zero phase shift. The phase shift of 10 rad is then (1988); D. T. Pegg and S. M. Barnett, Phys. Rev. A 89,
detected by the presence of photons in this port. The 1665 (1989); S. M. Barnett and D. T. Pegg, J. Mod. Opt.
result of the simulation shows that the classical driving 86, 7 (1989).
field must be significantly stronger in order to detect the [10] D. T. Smithey, M. Beck, M. Belsley, and M. G. Raymer,
relative path length change. For a coherent field supply- Phys. Rev. Lett. 69, 2650 (1992).

1358