VOLUME 55, NUMBER 25 PHYSICAL REVIEW LETTERS 16 DECEMBER 1985

Measurement of the Wigner Function

Antoine Royer
Centre de Recherches Mathematiques, Universite de Montreal, Montreal, Quebec H3C317, Canada
(Received 11 February 1985}
An operational procedure is given for determining experimentally, in principle at least, the
Wigner function for an ensemble of particles. This manner of "measuring" a quantum state,
whether pure or mixed, via its Wigner function, seems the simplest possible, and closely parallels
the method one might use in classical mechanics to determine a (true) phase-space probability den-
sity.

PACS numbers: 03.65.8z, 02. 50. + s, [Link]

The Wigner function, although it is not strictly posi- single measurement cannot, on the one hand, yield
tive, provides a phase-space representation of quantum sufficient information to reconstruct the particle's
mechanics which has many desirable features. ' A state operator, on the other hand it uncontrollably per-
point which does not seem to have been much dis- turbs the state, thereby eliminating the possibility of
cussed is whether the Wigner function has any opera- gathering the remaining information from subsequent
tional significance, i.e. , whether it can be recorded ex- measurements. However, the notion of "measuring a
perimentally. 2 We give here a simple procedure for state" is meaningful when it refers not to an individual
doing so, which seems, in fact, to be the simplest particle, but rather to a specific preparation procedure;
manner of "measuring" a quantum state; moreover, it i. e. , the following question may be posed: Is it possi-
closely parallels the method one might use in classical ble, by performing a sufficient number of measure-
mechanics to determine a (true) phase-space probabili- ments on different members of an ensemble of similar-
ty density, so that the Wigner function simulates a ly prepared particles, to determine the state operator p
phase-space distribution function not only formally, describing the ensemble?
but operationally also. For simplicity, we consider a This question seems to have been seldom addressed
single particle of mass m moving in one dimension. in the literature. A way of measuring states has been
We first explain what we mean by "measuring a indicated by Kemble, which, however, applies only to
quantum state. " In classical mechanics, the state of a cases where it is known a priori that a pure state (al-
particle is specified by its position and momentum, ways the same) is produced. But most preparation
and is directly observable; moreover, the act of obser- procedures yield mixed states, due, e.g. , to unpredict-
vation disturbs the state as little as desired (in princi- able fluctuations in the state of the apparatus used,
ple, at least). and/or to residual quantum correlations of the particle
A quite different situation prevails in quantum with other objects (e.g. , parts of the apparatus).
mechanics: Here, all that can be observed in a mea- D'Espagnat has remarked that measurement of the
surement is into which one of a complete orthonormal j j)
set of observables m;J + = ( ) ( ~i ) ( +( ~(i ), where
~

set of states (P„, n =0, 1, 2, . . .I, determined by the (i),( j),. . . , is a complete orthonormal set, would al-
apparatus, the particle "collapses" as a result of the low deduction of the state operator p, whether pure or
observation itself, which thus uncontrollably disturbs mixed. Although it is not clear that every "observ-
the state of the particle. The probability that the parti- able" can be measured, Lamb7 has given, for the case
cle collapse into the state Q;, or (speaking loosely) that of observables depending only on positions and mo-
it be "found" in the state Q; (which does not mean menta, an explicit prescription for effectively measuring
that it was in that state before the measurement), is them. This, however, requires the use of a large
equal to the transition probability (ttt; p ~Q;), where p is
~ number of different potentials of generally complicat-
the state operator (or density matrix) of the particle ed shapes, so that determining p by measuring the
before the measurement. It is more usual to refer operators m;~+ in this manner would be extremely la-
quantum measurements to observaMes3 (i.e. , Hermi- borious. We will see that a simpler approach, requir-
tian operators whose eigenstates form a complete ing only one or two simple potential shapes, consists in
orthonormal set) thus: If the Q„are the eigenstates determining p by measuring its Wigner function.
(eigenvalues a„) of the Hermitian operator 2, and if Let us first imagine the procedure whereby a classi-
the particle is "found" in the state P;, then one says cal physicist might determine experimentally the prob-
that he has measured the observable A and found the ability f(X,P, T) that a "classical" particle (e.g. , an
value a;. But as emphasized by Wigner, 4 the measure- electron), prepared according to some specific rules
ment basically refers to the set alt„. (but which still leave an element of randomness, due,
It follows from the above that the state of an indivi- e.g. , to thermal fluctuations) be at X with momentum
dual particle is unobservable, even in principle: While a P at time T. We describe two closely related possible
1985 The American Physical Society 2745
VOLUME 55, NUMBER 25 PHYSICAL REVIEW LETTERS 16 DECEMBER 1985

methods (our overall approach is very much influ- Note that if X were known, then one could deduce
enced by Lamb's classic paper ).
— (deconvolute) f
(x,p): By taking the Fourier
Method 1. We put ourselves in a reference frame
moving with uniform speed u = P/m relative to the
transform of (I), we get I (k, s) = X(k, s) (k, s), where
I (k, s) = dx dp e'
f
f "P I (x,
p), etc. , and we can solve
laboratory, and turn on in that frame at time T a po-
tential well V(x —X) whose minimum lies at X [e.g. ,
f
for (k, s) [provided X(k, s) &0 almost everywhere].
For the classical physicist, the particle is sharply lo-
a harmonic well —, mes (x —X) ]. If the particle is
'
calized at some phase-space point (x,p), so that the
then found to lie at rest at the bottom of the well, as "sharp" phase-space probability density (x,p) f
evidenced, e.g. , by the absence of any emitted radia- ["sharp" in the sense that it is the probability that the
tion, we infer that the particle was at L with momen- particle be at exactly the point (x,p), not just in a fuzzy
tum I' at time T. The relative number of times this neighborhood of it] is physically real, and could in
happens, in many repetitions of this experiment on principle be measured with arbitrary accuracy (if only
different similarly prepared particles, yields the proba- he had good enough instruments), Not so for a quan-
f
bility (X,P, T) .
—
Method 2. An alternative method, avoiding the use
tum physicist, for the uncertainty principle forbids
states which are sharply localized in phase space, so
of moving apparatus, is to first shift the momentum by that it is meaningless to speak of the probability that a
—P at time T by applying an impulsive force particle be at exactly the point (x,p); for him, a
—P 5 ( t —T) derived from a linear potential xP
x 5 (t —T); immediately
f
"sharp" (x,p) does not exist, and he would give the
following analysis of the above experiments.
thereafter, the potential
V(x —X) is turned on (in the laboratory frame this —
Method 1. In the moving frame, the state operator
time); again, if the particle is found at rest, we infer "seen" is8
that its state was (X,P) at time T. — ei(PX —xP)jii
p(t) D ( ) p(t)D &
D
Because of the experimental imprecisions, there is a
nonvanishing probability )((hx, Ap) that even if the where p is the particle's state operator in the laborato-
particle was at (X + hx, P + b, p ) at time T, it was ry frame, and D„P effects phase-space translations (x
nevertheless recorded as (X,P). Thus, what is mea- and p are the position and momentum operators). Let
f
sured in fact is not (X,P), but rather the convolution $" and E" be the eigenstates and energies of the Ham-
iltonian H = p /2m + V(x); then those of Hxo
I (XP) = I dx dp f
x(x —X,p —P) (x,p), = p /2m + V(x —X) are Pxo and E", where we set
i.e. , the relative probability that the particle be in a
"fuzzy" neighborhood of (X,P) [this is, in general,
not expected to be normalized, i. e. , jdX dP I
f
= fdx dp Xe fdx dp = I]. Usually, the function
X(x,p) is unknown, and can only be roughly estimated.
Hence, at times t
on in the moving frame, the
)
V(x —X) has been turned
T, after
state operator in that
frame is (dropping h's for simplicity)
p ( t) = exp [ —i ( t —T) Hxo ]p ( T) exp [i ( t —T ) Hxo ]
.
= X, exp [ —i « —T) (E"—E ) ] @xo) (oxP p ( T)
I I I @xP) (@xo I

(@xo lp ( T) l@xo) = (@xP Ip( T) @xP) .

I

What is measured, in the counting of the relative on in the laboratory frame, the state operator p(t) in
number of times that no emission of radiation is ob- that frame is given by (2), but with p(t), p(T), and
served, is the (time-independent) transition probabili- p(T) . replaced by p(t), p(T+0), and p(T —0),
,

ty (@xoolp(t) l@xoo) to the ground state @xo, according respectively. %'hat is measured is
to (2), this is equal to
(@',lp(t)l@', ) =l,, (X,P, T —0)
I'~, (x, P, T) = (@xP lp(» l@xP). (3)
—
Method 2. The state operator p(T+0) just after again. 9

application of the pulse potential xP5(t —T) is related The function (3) is the probability of "finding" the
to p ( T —0) just before by
p(T+0) =D, ,p(T
since the time evolution operator is
0)D— particle in the state gxP, which is only partially local-
ized about (X,P); optimal localization is obtained if
the potential V(x) = —, mes x is harmonic, in which
case
p T+0 l@xP(x) I'= (2vrhx') ' exp[ ——,' (x —x)'/Ax2],
exp —i)l T —0 dt [p /2m+5(t —T)Px] = Do P.
Then, at times t )
T, after V(x —X) has been turned I@xP(p) I'= (2~&p') "'exp[ ——,' (p P)'/~p'],
(4)

2746
V0LUME 55, NUMBER 25 PHYSICAL REVIEW LETTERS 16 DEcEMBER 1985

where b, x = (2m'/5') ' and Ap = (mt'/2)' [the ous energies).

corresponding I 0 is then known as the Husimi distri- Heeding these suggestions, and then being much
bution'0; the more general distributions, correspond- impressed by the fact that he indeed observes the
ing to arbitrary potentials V(x), were first considered discrete energy spectrum predicted by his quantum
by Klauder" ]. For the quantum physicist,
harmonic is the "sharpest"
for V
phase-space distribution
I, colleague, the classical physicist might yet argue that
even though a particle cannot be prepared or observed,
allowed by the uncertainty principle. or even described by conventional quantum rnechan-
Also, he might propose that a more "complete" ics, in a state better localized in phase space than (4),
quantum experiment would be to measure the whole it may still have, in reality, a sharp position and
set of transition probabilities & @xp p ( T) Pxp &, n i i momentum (here he may seek support in
=0, 1, 2, . . . (i.e. , measure Hzp), ' e.g. , by analyzing Heisenberg's famous statement "the uncertainty prin-
the radiation emitted after time T, noticing that by (2), ciple does not refer to the past"), ' so that a "sharp"
the probability that the particle radiate a total energy (in our previous sense) f(x, p) is still a valid concept
E" Eo (via — one or several photons in cascade), i. e. , to
be "found" in the state @xo, is
f
[of course, this "sharp" (x,p) cannot itself be sharp-
ly localized, of the form 5(x —X)5(p —P), say, since
& @xo p ( t) 4xo &
I i
= [Link]'
& i p ( T) 4'xp
i &
it must reflect the uncertainty relations inherent in the
(because the energy is emitted in discrete amounts, preparation process]. Moreover, considering (4), he
the quantum measurements can be made much more might conjecture that the quantum analysis provides a
definite expression for X(x,p) [see Eq. (1)] when
precise than the classical ones which involve continu-
l V(x) = —,'mes'x', to wit,
x(xp) = (2mb ) (47r'bx2hp') 't2
exp[ ——,' (x/bx)2 ——,' (p/Ap)']
[the
= fdx
factor
dpi,
2mll is deduced
=2mil] where he made the assumption,
from f dx dp X %=1 is of degrees of freedom.
the number The L-
quite plausible in view of the Gaussian forms of (4), kets
that the x and p "errors" are uncorrelated; thus, he ixp) = (~a-) -'in„, ), n„, = D„,nD„
can now even calculate f(x,p): Using (5) in (1),
I (x,p) having been measured with V= —, ' mco2x2,
he where II is the parity operator (IIix& = —x& ), form i a
f
solves for (x,p). To his dismay, he finds that (x,p) f complete orthonormal set:
is not everywhere positive in general. %'hat he has ob™ [x'p'ixp) = 5(x' —x) 8(p' —p),
tained is, in fact, the Wigner function F (x,p), which
is not a phase-space probability density (but is tantaliz-
Ji dx Jt dp ixp) [xp i
= 1L.
ingly close to one).
The meaning of f
(x,p) appears most clearly in the
Liouville-space formalism, ' wherein ordinary quan-
The Wigner function is (up to multiplicative constant)
the ixp) representative of the state operator'5 p..
tum operators are viewed as vectors. Introducing a
bra-ket notation in that space, we associate with every f (,p)=(2 &) '[ pip)=( il) 'T II„
operator 2 an L-ket iA) and an L-bra [A i. The scalar The connection with the definition given earlier [viz. ,
product is defined as [3 iB}= (2mtt)~TrA B, where Eq. (1) with X given by (5) and I by (3) for
V = —, m cv x ] is obtained if we note that

1&o(XP) = (2 ll) '[l@xp& &@xt ip} = (2 tt) 'Jtdx dp [l@ttp& &Pxt Ilxp} [xplp)
= 2m. t J dx dp f, (x — X,p —P) f (x,p), (8)
where f
0(x,p) = (2nf) '(xpii$ & &@ i} is the Wig-
ner function for the pure state @o; we recover (1) by According to (6) and (7), f
(x,p) is a sharp phase-
setting X = 2+k f' „which equals (5) if p is the space representation of p, where "sharp" here refers to
ground state of the harmonic oscillator p /2m the 5-function phase-space orthogonality of the L-kets
+ —,mcu2x . Equation (8) also shows that we may use ixp). However, because n„~ is not a density matrix
an arbitrary potential V(x) in our experiments, and (not being positive definite), the L-kets ixp} are not
still be able to deconvolute f' from [where @ is
now the ground state of p /2m+ V(x)], by using
I, states. Herein lies the fundamental difference between
(x,p) and the classical idealization that was a sharp
f
X = 2mt y00 in (1) [the x and p "errors" are then corre- phase-space probability density f
"(x,p): In contrast to
lated in general, unlike in (5)]. the latter, f
(x,p) is not a density on (or a transition
2747
VOLUME 55, NUMBER 25 PHYSICAL REVIEW LETTERS 16 DECEMBER 1985

probability to) physical states P. hysically, f

(x,p) is
an expectation value, namely that of the parity opera-
complete set of transition probabilities instead of just a
single one [yielding I' 0(x,p)], to measure f' (x,p)
tion 11„p about the phase-space point (x,p). ' By con- directly. This avoids the delicate deconvolution in-
trast, I 0(x,p) is a "fuzzy" phase-space representa- volved above, and seems to be the simplest possible
tion" of lp}, in terms of the nonorthogonal set of states manner of "measuring" a quantum state [measuring
i@op) (@opl], but is a transition probability. (x,p) is equivalent to measuring p, since the form-
l

f (x,p) and I, Thus,

(x,p) each share only part of the attri- er is a sharp representation of the latter].
f
butes of "(x,p).
As we saw earlier, it is I', (x,p) which is observed
"
"classically, whence f (x,p) can be deduced via Eq.
(8). But f (x,p) is the expectation value of an observ-
able, H„~, and should, in principle, be directly measur-
For recent reviews, see M. Hillery, R. F. O' Connell,
M. O. Scully, and E. P. Wigner, Phys. Rep. 106, 121 (1984);
N. L. Balazs and B. K. Jennings, Phys. Rep. 104, 347
.
able; this can indeed be done, in effect by the same
(1984); V. I. Tatarskii, Usp. Fiz. Nauk. 139, 587 (1983)
experiments as already described: The eigenvalues of [Sov. Phys. Usp. 26, 311 (1983)].
II„p are +1 (since 11„~=1), and a complete set of This question is raised by K. Wodkiewicz [Phys. Rev.
eigenstates @„"p may be gotten by displacing in phase Lett. 52, 1064 (1984)], who seems to imply, however, that
space any complete set of states @" of definite parity the Wigner function is not operational.
about the origin, i.e. , 3See, e.g. , P. A. M. Dirac, The Principles of Quantum
Mechanics (Oxford Univ. Press, Oxford, 1958) 4th ed. ;
J. von Neumann, Mathematical Foundations of Quantum
Mechanics (Princeton Univ. Press, Princeton, 1955).
4E. P. Wigner, in Quantum Optics, Experimental Gravitation
Aware that Hamiltonians are, as stressed by Lamb, and Measurement Theory, edited by P. Meystre and M. O.
ideal observables from an operational point of view, ' Scully (Plenum, New York, 1983).
let us choose the @" as the eigenstates sE. C. Kemble, The Fundamental Principles of Quantum
of H
Mechanics (McGraw-Hill, New York, 1937), p. 71.
=p /2m+ V(x), where V(x) = V( —x) is any con- 6B. d'Espagnat,
venient symmetric potential (e.g. , V = —, m to x ).
Conceptual Foundations of Quantum
Mechanics (Benjamin, Reading, Massachusetts, 1976), 2nd
Measurement of H~p, or its expectation value ed. , p. 57.
7W. E. Lamb, Jr. , Phys. Today 22, No. 5, 23 (1969).
TriIxp p ( T) = X„( I )"(@xp p ( T) @xp )
I I
sSee, e.g. , W. Pauli, General Principles of' Quantum
then amounts to measurement of the set of transition Mechanics (Springer, Berlin, 1980), p. 19.
probabilities (@xplp(T) l@xp) (or Hitp); but this is 9Another manner of measuring I o(x, p), via a scattering
precisely what is done in the "complete" quantum ex- experiment, is given in Ref. 2. Our approach is closer to
periments suggested previously. Lamb's in Ref. 7.
toK. Husimi, Proc. Phys. Math. Soc. Jpn. 22, 264 (1940).
To summarize, we first devised "classical" experi- tt J. R. Klauder, J. Math. Phys. 4, 1055 (1963).
ments to measure a "sharp" phase-space probability
In the three-dimensional case, one might use
density f'"(x,p), a concept idealized from everyday
V(xy, z) = 2 m(~fx +cozy +t03z ) where coi, cu2, and cu3
macroscopic experience. The quantum analysis of
are incommensurate, such as to avoid degeneracies in the
these experiments reveals that what is measured is in set of energies &(night+ n2c02+ 0)n33), n; =0, 1, 2, . . ., thus
fact a fuzzy phase-space density I 0(x,p), subject to allowing each degree of freedom to be analyzed separately.
the uncertainty principle. But if one surmises that a t3W. Heisenberg, The Physical Principles of the Quantum
"sharp" f(x, p) nevertheless exists in reality, and tries Theory (Dover, New York, 1949), p. 20.
to calculate it by deconvoluting I,
(x,p) in a naive
manner, one almost obtains such an idealized function:
t4See, e.g. , U. Pano, Rev. Mod. Phys. 29, 74 (1957), and
in The ManyBody Problem, edited by E. R. Caianiello
One gets the Wigner function f
(x,p), which is
indeed a sharp phase-space representation of p, but
(Academic, New York, 1964).
t5B. R. Mollow, Phys. Rev. 162, 1256 (1967); A. Royer,
Phys. Rev. 15, 449 (1977).
not a probability density on physical states, and is not This is because the associated transition probabilities are
always positive. Remarkably, it is possible, by expand- time independent, so that a long time is available for per-
ing the above "classical" experiments into "com- forming a measurement and "finding" the particle in one of
plete" quantum experiments, i.e. , by measuring a the energy eigenstates.

2748