VOLUME 77, NUMBER 11 PHYSICAL REVIEW LETTERS 9 SEPTEMBER 1996

Fringe Visibility and Which-Way Information: An Inequality

Berthold-Georg Englert*
Max-Planck-Institut für Quantenoptik, Hans-Kopfermann-Strasse 1, D-85748 Garching, Germany
(Received 21 May 1996)
An inequality is derived according to which the fringe visibility in a two-way interferometer sets an
absolute upper bound on the amount of which-way information that is potentially stored in a which-way
detector. In some sense, this inequality can be regarded as quantifying the notion of wave-particle
duality. The derivation of the inequality does not make use of Heisenberg’s uncertainty relation in any
form. [S0031-9007(96)00950-7]

PACS numbers: 03.65.Bz

Bohr’s principle of complementarity [1] states that entering the interferometer the quanton is in an initial
quantum systems (“quantons” [2] for short) possess prop- state that is characterized by the statistical operator
erties that are equally real but mutually exclusive. The
sid 1 1 ssid ? s
best known example is what is colloquially termed wave- rQ ­
particle duality. In a loose manner of speaking it is some- 2
1
times phrased similarly to the following: Depending on ­ s1 1 sxsid sx 1 sysid sy 1 szsid sz d (1)
the experimental situation a quanton behaves either like a 2
particle or like a wave. sid
with an initial Bloch vector ssid ­ trQ hs rQ j. It is
To be more specific, let us consider a two-way inter-
sufficiently general to represent the action of the beam
ferometer such as Young’s double-slit experiment or a
splitter and the beam merger by
Mach-Zehnder setup. The wavelike property is then doc-
umented by well-visible interference fringes, whereas the µ ∂ µ ∂
p p
particlelike property is evident if one can tell along which rQ ! exp 2i sy rQ exp i sy ,
4 4
way the interferometer has been traversed.
The notions of particle and wave are associated with whereas the phase shifter at the central stage effects
mental pictures that are borrowed from classical (i.e., µ ∂ µ ∂
f f
prequantum) physics. These associations are dangerous rQ ! exp 2i sz rQ exp i sz . (2)
2 2
because of their obvious limitations. Therefore, “wave-
particle duality” should perhaps be abandoned in favor of sid
Consequently, the interferometer of Fig. 1(a) turns rQ
a more neutral term, such as “interferometric duality” or into the final state
simply “duality.” The general formulation of this concept
s fd 1
could read as follows. rQ ­ s1 1 ss fd ? sd , (3)
Duality.—The observations of an interference pattern 2
and the acquisition of which-way information are mutu- with
ally exclusive.
The extreme situations “perfect fringe visibility and no ss fd ­ s2sxsid , sysid cos f 1 szsid sin f,
which-way information” and “full which-way information
and no fringes” are familiar from textbook discussions. sysid sin f 2 szsid cos fd .
But intermediate stages deserve further study. After the quanton has passed the beam merger, the
The objective of this Letter is the derivation of an observable sz is measured and the relative frequency
inequality that quantifies duality by stating to which extent with which the value 21 is found reveals the interference
partial fringe visibility and partial which-way knowledge pattern,
are compatible. The quantitative measure of the fringe Ω æ
visibility is the usual one, and which-way knowledge will 1 s fd
pf ­ trQ s1 2 sz drQ
be turned into a number with the aid of an approach that 2
is originally due to Wootters and Zurek [3]. 1
At the intermediate stage of the interferometer— ­ s1 2 sysid sin f 1 szsid cos fd ,
2
between beam splitter and beam merger, see Fig. 1—the
so that
two ways can be labeled by quantum numbers 11 and
21, say. Accordingly, we are invited to describe the V0 ­ fssysid d2 1 sszsid d2 g1y2
relevant degree of freedom of the quanton by analogs
of Pauli’s spin operators s ­ ssx , sy , sz d. Prior to is the corresponding a priori fringe visibility.

2154 0031-9007y96y77(11)y2154(4)$10.00 © 1996 The American Physical Society

VOLUME 77, NUMBER 11 PHYSICAL REVIEW LETTERS 9 SEPTEMBER 1996

Let us consider symmetric interferometers with maxi-

mal a priori fringe visibility,
sxsid ­ 0, szsid 1 isysid ­ e 2iu ,
which are coupled to another physical system that is
meant to serve as a which-way detector as indicated in
sid
Fig. 1(b). The detector is initially prepared in a state rD
so that the combined system of quanton and detector has
an initial statistical operator
sid sid
r sid ­ rQ rD . (6)
As a result of the interaction between quanton and
detector the detector state is altered in accordance with
y
rD ! U6 rD U6 for the sz ­ 61 way ,
where U1 and U2 are unitary operators that affect solely
the degree(s) of freedom of the detector. Thus what
FIG. 1. Schematic two-way interferometer. (a) The beam
splitter BS distributes the input among the two ways; after happens at the central stage of the interferometer is now
the phase shifter PS has acted in its f dependent manner, the described by the unitary operator
beam merger BM recombines the contributions and produces
the output. Measurements on the output may either reveal the 1 1 sz ify2 1 2 sz 2ify2
e U1 1 e U2 ;
f dependent interference pattern, or alternatively determine 2 2
the way that has been taken. (b) When the interferometer i
is supplemented by a which-way detector WWD, then both it replaces exps 2 fsz d of (2).
the interference pattern may be observed and the which-way The initial state (6) is transformed into the final state
information may be acquired, within the limits set by duality.
1 1 sx y sid 1 2 sx y sid
r s fd ­ U1 rD U1 1 U2 rD U2
The probabilities for taking either one of the two ways 4 4
sz 2 isy 2isf2ud y sid
are 2 e U1 rD U2
Ω µ ∂ µ ∂æ 4
1 p sid p sz 1 isy isf2ud y sid
w6 ­ trQ s1 6 sz d exp 2i sy rQ exp i sy 2 e U2 rD U1 ,
Ω2 æ 4 4 4
1 s fd 1
­ trQ s1 6 sx drQ ­ s1 7 sxsid d . (4) which yields the final states of the quanton and the
2 2 detector upon tracing over the other degree of freedom.
The magnitude of their difference, The quanton state is of the form (3) with
P ­ jw1 2 w2 j ­ jsxsid j , sxs fd ­ 0, szs fd 1 isys fd ­ 2e2isf2ud C ,
is the predictability of the ways through the interferom- y sid
eter. This number, which is in the range 0, . . . , 1, is a where C ­ trD hU1 rD U2 j is a complex contrast factor.
quantitative measure of the a priori which-way knowl- The fringe visibility is therefore given by the magnitude
edge. If one bets on the more probable way, then the of this number,
probability of predicting the way correctly is s1 1 P dy2. V ­ jC j . (7)
In particular, if the interferometer is operated symmetri-
1 The final detector state
cally, w1 ­ w2 ­ 2 so that P ­ 0, then 50% of the pre-
dictions are right and the other 50% are wrong. In short, s fd 1 y sid 1 y sid
there is nothing predictable about the ways in a symmetric rD ­ trQ hr s fd j ­ U1 rD U1 1 U2 rD U2
2 2
interferometer. 1 s1d 1 s2d
Since the length of the Bloch vector ssid must not ­ rD 1 rD
2 2
exceed unity, one immediately establishes the inequality
is the sum of the two final states corresponding to
P 2 1 V 20 # 1 . (5) the two ways, weighted by the fraction of quantons
This observation has been made by Greenberger and taking the respective way—50% each in the symmetric
Yasin [4]; it is implicitly contained in the work of Woot- interferometer under consideration.
ters and Zurek [3] and also in a paper by Mandel [5]. The The extraction of the which-way information stored
measurements by Rauch, Summhammer, and Tuppinger in the detector requires the measurement of a suitable
[6], who introduce an asymmetry into a neutron interfer- observable W with eigenvalues W 0 and eigenkets jW 0 l.
ometer, are consistent with (5). We take for granted that the relevant eigenvalues of W are

2155
VOLUME 77, NUMBER 11 PHYSICAL REVIEW LETTERS 9 SEPTEMBER 1996

not degenerate. Suppose that the eigenvalue W 0 has been The equal sign holds in (10) always if the detector is
found. This happens with a relative frequency of prepared in a pure state,
sfd 1 s1d sid
kW 0 jrD jW 0 l ­ kW 0 jrD jW 0 l rD ­ jdl kdj :
2
y
1 s2d D ­ s1 2 jkdjU2 U1 jdlj2 d1y2 , (11)
1 kW 0 jrD jW 0 l ,
2 y
where the two summands refer to the two ways. Extend- V ­ jkdjU2 U1 jdlj .
ing an idea by Wootters and Zurek [3], we note that the The general proof of the duality relation (10) is based on
best guess about the way one can thus make is to opt for the triangle inequality for the trace-class norm,
s fd
the way that contributes most to kW 0 jrD jW 0 l. In many trhjA 1 Bjj # trhjAjj 1 trhjBjj ,
repeated experiments, this yields a “likelihood for guess-
ing the way right” that is given by which holds for all trace-class operators A and B. Upon
sid
X Ω æ inserting the spectral decomposition of rD ,
1 0 s1d 0 1 0 s2d 0 sid
X
LW ­ Max kW jrD jW l, kW jrD jW l rD ­ Dk jdk lkdk j
W0 2 2
k
1 1 X s1d s2d P
­ 1 jkW 0 jsrD 2 rD djW 0 lj . (with Dk $ 0, k Dk ­ 1, and kdj jdk l ­ djk , of course),
2 4 W0
into (9) yields
Its calculated value can be checked experimentally if the
interferometer is modified such that the actual way is 1 X y y
D # Dk trD hjU1 jdk lkdk jU1 2 U2 jdk l kdk jU2 jj
determined rather than the fringe pattern. This can be 2 k
done, for example, either by removing the beam merger X y
and measuring sz or by leaving the beam merger in ­ Dk s1 2 jkdk jU2 U1 jdk lj2 d1y2 ,
place and measuring sx in the final state r s fd ; these k

two possibilities correspond to the two traces in (4). where the last step makes use of the pure-state result (11).
Since such a measurement yields also the probabilities In conjunction with
s6d É É
kW 0 jrD jW 0 l, the numerical value of the likelihood LW X y
can be inferred from experimental data. V ­ Dk kdk jU2 U1 jdk l ,
k
This value depends on the observable W that is mea-
sured. An unfortunate choice could result in LW ­ 2 , in
1 this leads to
" q
which case one could just as well throw dice. The largest X q
2 2
value of LW is obtained if the (relevant) eigenkets jW 0 l of D 1V # Dj Dk 1 2 juj j2
1 2 juk j2
s1d s2d j,k #
W are also eigenkets of the difference rD 2 rD . Accor-
dingly, there is an absolute optimum for LW , viz. 1 p 1 p
1 u uk 1 u uj ,
1 2 j 2 k
LW # Lopt ­ s1 1 D d , (8) (12)
2
y
with where uk ­ kdk jU2 U1 jdk l is a convenient abbreviation.
1 s1d s2d The magnitude of these complex numbers does not exceed
D ­ trD hjrD 2 rD jj . (9) unity, juk j # 1, and therefore 0 # f· · ·g # 1 holds for the
2
In mathematical terms, this number D is the distance be- square brackets in (12). Accordingly, one gets
s1d s2d
tween rD and rD in the trace-class norm; its physical X sid
significance is, however, more important: D is a quanti- D2 1 V 2 # Dj Dk ­ ftrD hrD jg2 ­ 1 ,
j,k
tative measure of the distinguishability of the ways, i.e.,
of the amount of which-way information that has become and this closes the case.
available. The ways cannot be distinguished at all if D ­ It must be emphasized that this proof of the duality
0, and they can be held apart completely if D ­ 1. relation (10) does not rely on an uncertainty relation of
The stage is now set for reporting the central result the Heisenberg-Robertson kind [7], i.e.,
of this Letter. The fringe visibility V of (7) and the 1
distinguishability of (9) obey the inequality dXdY $jkfX, Yglj (13)
2
D2 1 V 2 # 1, (10) for the spreads of two observables X and Y and the
which is a fundamental quantitative statement about expectation value of their commutator. One understands
duality. Of course, it comprises the extreme situations why (10) cannot be an uncertainty relation in disguise
mentioned above inasmuch as V ­ 1 implies D ­ 0 when noting that really only one observable is involved,
y
and D ­ 1 implies V ­ 0. not two. This observable X is identified by U2 U1 ­

2156
VOLUME 77, NUMBER 11 PHYSICAL REVIEW LETTERS 9 SEPTEMBER 1996

expsiXd. Then both the distinguishability D and the holds, it is clear that (15) is more stringent than (14).
visibility V involve X, Consequently, (14) applies to asymmetric interferometers
1 as well.
sid sid
D ­ trD hjrD 2 e2iX rD eiX jj, In setups of the kind depicted in Fig. 1(b), the same
2 physical mechanism can be used both for the phase
sid
V ­ jtrD hrD eiX jj , shifting and the which-way detection. There are other
schemes, such as Einstein’s recoiling-slit proposal [8] or
but no second observable shows up. In conclusion, the quantum-optical Ramsey interferometer [9], in which
the duality relation (10) is logically independent of the the beam splitter also acts as the which-way detector. For
uncertainty relation (13). them, the bounds of (14) hold too, but the analysis is
It is equally important to realize that the inequalities somewhat more involved [10].
(5) and (10) convey utterly different messages despite This work was begun while I was a visitor at the Lab-
their great similarity, because the predictability P and oratoire de Physique des Lasers, Université Paris-Nord, in
the distinguishably D represent pieces of which-way Villetaneuse. I am very grateful for the hospitable en-
knowledge of very different kinds. Furthermore, the two vironment provided by Christian Miniatura and Jacques
inequalities concern different degrees of freedom. In (5) Baudon as well as the other members of the group, and
one meets an immediate consequence of the positivity of I extend my sincerest thanks to the Centre National de la
the statistical operator (1) that specifies the initial state Recherche Scientifique (C.N.R.S.) for supporting my stay
of the quanton. In marked contrast, (10) originates in so generously.
the quantum properties of the detector. The quantum
aspects of the which-way detection enforce duality and
thus make sure that the principle of complementarity is
not circumvented.
Upon combining (10) with (8) one gets the inequality *Presently at Arbeitsgruppe “Nichtklassisches Licht” der
Max-Planck-Gesellschaft an der Humboldt-Universität,
1 1 1p Rudower Chausee 5, Gebäude 10.16, D-12484 Berlin,
# Lopt # 1 1 2 V2, (14) Germany.
2 2 2
according to which the fringe visibility limits the experi- [1] N. Bohr, Naturwissenschaften 16, 245 (1928); Nature
(London) 121, 580 (1928). A formulation in mathematical
menter’s ability of guessing the way right. To begin with,
terms that is based on the concept of complementary
this applies to symmetric interferometers with predictabil- observables is given by M. O. Scully, B.-G. Englert, and
ity P ­ 0 and a priori visibility V0 ­ 1. This is no real H. Walther, Nature (London) 351, 111 (1991)
restriction, however, because asymmetric interferometers [2] According to J.-M. Lévy-Leblond, Physica (Amsterdam)
can be analyzed in an analogous manner. The fringe visi- 151B, 314 (1988), this useful noun, which avoids the
bility is then usage of either “particle” or “wave,” has been coined by
M. Bunge.
V ­ jC jV0 [3] W. K. Wootters and W. H. Zurek, Phys. Rev. D 19, 473
rather than (7), and the optimal likelihood is given by (1979).
[4] D. M. Greenberger and A. Yasin, Phys. Lett. A 128, 391
1 1 s1d s2d
Lopt ­ 2 1 2 trD hjw1 rD 2 w2 rD jj , (1988).
[5] L. Mandel, Opt. Lett. 16, 1882 (1991). Mandel’s
p “degree
which involves the a priori probabilities of the ways of intrinsic indistinguishability” equals V0 y 1 2 P 2 in
[cf. Eq. (4)] and generalizes (8) with (9). The resulting the present notation.
generalization of (14) reads [6] H. Rauch and J. Summhammer, Phys. Lett. 104A, 44
(1984); J. Summhammer, H. Rauch, and D. Tuppinger,
1 1
s1 1 P d # Lopt # Phys. Rev. A 36, 4447 (1987).
2 2 [7] W. Heisenberg, Z. Phys. 43, 172 (1927); H. P. Robertson,
q Phys. Rev. 34, 163 (1929).
1
1 1 2 s1 2 P 2 d sV yV0 d2 (15) [8] N. Bohr, in Albert Einstein: Philosopher-Scientist,
2
edited by P. A. Schilpp (Library of Living Philosophers,
with V # V0 . The reasoning that justifies the upper Evanston, 1949), pp. 200–241.
bound is essentially the same as the one for the upper [9] B.-G. Englert, H. Walther, and M. O. Scully, Appl. Phys.
bound in (14), and the lower bound is a consequence B 54, 366 (1992).
of trhjAjj $ jtrhAjj. Since P is non-negative and (5) [10] B.-G. Englert, Acta Phys. Slov. 46, 249 (1996).

2157