When Is a Quantum Measurement?

”
ASHER PERES
Department of Physics
Technion-Israel Institute of Technology
32000 Haifa. Israel

INTRODUCTION

In many textbooks on quantum mechanics, “measurements” appear as a kind of

primitive notion and the “projection postulate” is treated as one of the axioms of the
theory. For example, Dirac’ writes:
. . . a measurement always causes the system to jump into an eigenstate of the dynamical
variable that is being measured, [with] the eigenvalue this eigenstate belongs to being
equal to the result of the measurement.
Likewise, Bohm’ writes:
. . . when the position of the electron was observed, the wave function suffered a collapse
from a broad front down to a narrow region. The exact region to which it collapses is not
determined by the state of the wave function before the collapse; only the probability of
collapse to a given region is determined, and this probability is proportional to the value of
1 +I in that region.
More formally, von Neumann’ makes a distinction between
two fundamentally different types of intervention which can occur in a system. . . . First,
the arbitrary changes by measurements which are given by the formula

[ u , , u2,. . . , a complete orthonormal set; P I , P2,. . . , the corresponding projection

operators]. Second, the automatic changes which occur with passage of time. These are

-
given by the formula
p pl = e-‘HtJ/“
PeiH”h (2)
(His the energy operator, t, the time; H i s independent oft).
While the above recipes give, in most cases, satisfactory results, they are
fundamentally unacceptable. A measurement is not a supernatural event. It is a
physical process involving ordinary matter, and whatever happens ought to be
explained by the ordinary physical laws. Performing a measurement (whether classical
or quantal) means the following: Some apparatus is made to interact with the
measured object in such a way that a property of the object is replicated in a property
of the apparatus. This process ought to yield a definite result, namely, the outcome of
the measurement.

aThis work was supported by the Gerard Swope Fund.

438
PERES: WHEN IS A QUANTUM MEASUREMENT? 439

This immediately raises a difficulty if quantum theory is used to describe the

dynamical evolution of the combined system (that is, the measured object and the
apparatus). The Schriidinger equation is linear. Therefore, if the quantum object is
prepared in a superposition of orthogonal states,

such that each u, yields (with certainty) a distinct outcome for the measurement, then
the final state will involve a superposition of different outcomes. Formally, if the initial
state of the apparatus is 4, and if that apparatus is built in such a way that its final state
is v, whenever the initial state of the object is J, = u,, we have the unitary evolution,4

whereby states of the object are correlated with states of the apparatus. Thus, unless a
single c, is different from zero, the final state does not represent a definite outcome for
the measurement.
Then, a miracle occurs. By fiat, the superposition, 2c,u,v,, “collapses” into one of
the u,v,. The corresponding c, jumps to its new value, 1, while the other components
disappear. The purpose of this paper is to explain this “miracle.” (The Everett
interpretation’ does not involve miraculous events, but is beset by other difficulties.6Its
meaning will be analyzed later in the third section of this article.)
The plan of this paper is as follows: In the following section, I discuss the meaning
of a state vector, $, or a density matrix, p. These symbols represent information on the
preparation of quantum That information can evolve because of known
dynamical interactions, as in equation 2. It can also get degraded and become
ultimately obsolete with the passage of time. If, and only if, the unitary correlation in
equation 4 is followed by a degradation of the available information (because of
irreversible coupling with additional, uncontrollable, degrees of freedom), the resulting
degraded information is compatible with the “collapse” postulate. The latter does not
represent a dynamical miracle, but simply the acquisition of fresh information by
means of an actual observation. In the third section, I briefly discuss the role of the
“conscious observer” in quantum theory, and Everett’s interpretation’ of the wave
function. Finally, in the last section, I mention measurements of finite duration. In
particular, this section gives a detailed treatment of the case of a radioactive atom
continuously monitored by a detector. It is shown that this continuous monitoring is not
a measurement.

ON STATE VECTORS AND MEASUREMENTS

A wave function is not something that “exists” in nature? It is only a mathematical

expression, invented by physicists, so that they can compute physical properties of
matter and, in particular, probabilities of events” following specified preparations.
(Those readers who adopt a “realistic” attitude will disagree with my approach.
However, it is their problem then to explain the miraculous events described in the
previous section.)
Schriidinger wave functions represent information on physical ~ y s t e m s .When
~ . ~ we
say that $ = u,, this means that there is a certain physical test (represented by the
440 ANNALS NEW YORK ACADEMY OF SCIENCES

projection operator, P,,)that will always give a positive result. Moreover, the same test
will always give a negative result if fi = tl, (any state orthogonal to u,,). These notions
are readily generalized to statistical mixtures, which are represented by density
matrices.
It is sometimes stated that “quantum theory is deterministic” because the
Schrodinger equation is a well-behaved differential equation of first order in time
derivatives. One could likewise argue that classical statistical mechanics is determinis-
tic because the Liouvilie equation has exactly the same properties. In both cases, it is
the probability distribution that evolves in a deterministic way. Individual events are
unpredictable, except statistically.
As a special case, consider a spin-Y2particle prepared in state (F).
First, though, I
ought to explain the meaning of these terms. The meaning of “spin-Y2particle” is the
following: The most exhaustive test performed on such an object is of the Stern-
Gerlach type. It can yield only two values, which are conventionally called “up” and
“down.” There are no additional variables. (Here, for simplicity, I ignore the
translational degrees of freedom.) The meaning of “state (;)” is the following: If we
submit the particle to a test whereby it passes through a Stern-Gerlach apparatus
oriented along a unit vector,;;, then the particle has probability,
P(7i) = nx(a*B + ab*) - in,(a*P - ab*) + n,(a*a - P*b), (5)
to leave in the “up” beam. The meaning of the word “probability” is the following: The
physicist who performs that experiment (namely, a spin-$ particle passing through a
Stern-Gerlach apparatus) can imagine that the same experiment, with exactly the
same preparation, is repeated many times. This unique experiment thus becomes part
of a conceptual ensemble (a Gibbs ensemble). The probability of a given outcome (e.g.,
“up”) is defined as the ratio of the number of outcomes of that type to the total number
of outcomes, when the size of the ensemble tends to infinity.
How can we measure that probability? This is done by simulating the conceptual
ensemble by a real ensemble (a Maxwell ensemble) of many noninteracting and
uncorrelated systems. As emphasized by Ballentine,” these are two different concepts:
“One must not confuse the [Gibbs] ensemble, which is a conceptual set of replicas of
one particle in its experimental surroundings, with a beam of particles, which is
another kind of (many-particle) system.” It will be seen below that these two notions
are essentially different because it is impossible to exactly duplicate the experimental
surroundings.
Once the terminology has been clearly defined, one can proceed and describe the
experimental details. For simplicity, it may be assumed that the Stern-Gerlach
apparatus has only two states, which are denoted as [A]
and [:I. (I shall use
parentheses and brackets to denote state vectors in the Hilbert spaces of the observed
object and the apparatus, respectively.) Initially, the apparatus is in state [A]. The
interaction with the spin is such that the apparatus stays in state [A]
if the spin is in
state (A), and it ends up in state [y] if the spin is in state (y).
By linearity, we thus
have

);( [A] --+ (A) [A] (y) [y]

which is a special case of equation 4.
a + *
PERES: WHEN IS A QUANTUM MEASUREMENT? 441

The evolution in equation 6 is unitary and can be generated by the Hamilton-

ian,’2.13

H = g ( M l - a,P, (7)
where a, refers to the spin, P i s the projection operator,

P =2 -1 r[ ’][-1I]+,

and g ( t ) is a smooth function of time, with compact support, satisfying j g ( t ) d t =

a h / 2 . Indeed, the unitary evolution generated by H is
u- e-ifHdlIh = - (1 - .JP9 (9)
and the right-hand side of equation 6 is equal to U (;)[:I, which can easily be
verified.
This evolution is reversible.’*.” This is easily seen by noting that in this model, U is
not only unitary, but also Hermitian, so that U -
U-’. Therefore, the same
Hamiltonian that generates the correlation in equation 6 also reverses it:

The right-hand side of equation 6 is a pure state. There is no justification in treating it

otherwise, nor is there any justification, in particular, in “collapsing” it so that either a
or /3 disappear^.'^ For example, the relative phase of a and fl can be measured by
observing the expectation values of operators such

and

The expectation values of equations 11 and 12 are average values, corresponding, as

explained above, to a conceptual (Gibbs) ensemble, whereby all the spins and all the
apparatuses are identically prepared in identical environments. For example, to obtain
the value of the right-hand side of equation 11, we imagine that after many replicas of
the object and the apparatus have been correlated as in equation 6 , we separately
measure in each one of these replicas, (y and A) [y A],
by means of additional
apparatuses. Each one of these measurements yields a result of 2 1. The average value
of the product of these pairs of results is a*fl + a@*.
It is now that we encounter the difficulty. If we want to actually measure the
expectation values of equations 11 and 12, we must use a real (Maxwell) ensemble with
many spins and many apparatuses that can in the best of cases be in almost identical
environments. In other words, the Hamiltonian is not truly given by equation 7,but we
must add to it some small, unknown, and uncontrollable terms, which differ from
442 ANNALS NEW YORK ACADEMY OF SCIENCES

sample to sample. Consequently, our knowledge about the behavior of each sample is
degraded.
Let us evaluate how bad the situation may be. As a grossly oversimplified example,
suppose that [A] and [(i’] represent two states of a ferromagnet, with electrons
aligned in opposite directions. These two states are macroscopically distinguishable, as
it behooves for states of a measuring instrument. The corresponding magnetic
moments are 2 1023eh/2mc,or about k5.8 x 10’ MeV/gauss. If the experiments are
performed in a region where unknown, stray magnetic fields are reduced to gauss,
the energy difference between the two states of the magnet will be uncertain by about 1
MeV. These states will therefore lose their phase coherence within sec or so.
After that time has elapsed, the only description of our knowledge is a statistical
mixture, with probability IaI’for state (A) [A] and probability !@I2for state ((i’) [(i’].
This is true even if there is a single sample. One might be tempted to imagine that
each object + apparatus system has a well-defined wave function that, however, is
unknown to us. This is the “reaIistic” attitude. Whoever adopts it is then faced with the
problem of explaining the wave function “collapse” as a physical phenomenon without
invoking “miracles.” This issue was already discussed at the beginning of this section
and I shall not return to it.
Before I explain the ‘‘collapse,” it is important to notice the crucial difference
between “microscopic” and “macroscopic” systems. A microscopic object is one that
can be perfectly isolated from its environment (or else placed in a perfectly controlled
environment) over a period of time that is long compared to the duration of the
experiment. A macroscopic object cannot. The distinction between the two depends, of
course, on our technical skill,” just as the boundary between reversible and irreversible
processes may be pushed somewhat by the progress of technology. Thus, from the
assumption that the final states of the apparatus are macroscopically distinguishable,
it follows that they will be found in different environments” if we repeat the
experiment (it is not the energy difference between them that is important, but only the
fact that the latter is irreproducible).
I finally come to the “collapse.” Because of the unavoidable noise that mars the
Hamiltonian (equation 7), the final state of the system is not the right-hand side of
equation 6 , but, to the best of our knowledge, is represented by a statistical mixture,

P= lal2(: ;)[: ;]+ IOl$ :)[;3.

as in the right-hand side of equation 1. This means that the expectation values of the
operators such as in equations 11 and 12 are zero. They would indeed yield a zero
average if we performed many similar experiments with a real ensemble in a noisy
environment. There are, however, some dynamical variables having nonvanishing
expectation values, such as

and
PERES: WHEN IS A QUANTUM MEASUREMENT? 443

These results are exactly the same as if we had an ordinary mixture containing a
fraction 1 a I * of

and a fraction IPI2 of

This is true if we consider a real (Maxwell) ensemble of many spins and many
apparatuses (or with the same spin and/or same apparatus used many times).
Therefore, equation 13 is also the correct description of the conceptual (Gibbs)
ensemble representing a unique experiment performed with a single system. Every-
thing that we can predict is included in equation 13.
Having made that statistical prediction, we may now actually perform one
experiment and observe its result. It must be either

Any other result would be incompatible with equations 14 and 15 and other similar
expressions yielding zero expectation values. Experimentally, this means the following:
After that unique observation has been performed, any further tests, as in equation 5,
are always compatible with either

($1 Or (:)[I.
On the other hand, any further tests are never compatible, in the sense of equation 5,
with

($1 nor
9I:[):(:

etc. This is the meaning of the “collapse of the wave function.” It is not a physical
process, but simply the acquisition of fresh knowledge about a physical system. It is a
change of our description, whereby we return from a Gibbs ensemble to a single
object.
I now come to the crucial point: There is nothing inconsistent (nor miraculous) in
this “collapse” because the relative phases in the right-hand side of equation 6 have
been blurred so that the expression has been replaced by equation 13. If it were not so,
the collapse assumption would be inconsistent. For example, collapsing the right-hand
side of equation 6 is inconsistent with the possibility of restoring the left-hand side of
equation 6 by applying once more the operator, U, as shown in equation 10. I t is,
therefore, essential that the measuring instrument, together with its environment, have
some noise. In other words, their description must involve some degrees of freedom that
are not accounted for, except statistically. It does not matter whether these are internal
degrees of freedom16 or whether the noise is due to external agents.
444 ANNALS NEW YORK ACADEMY OF SCIENCES

We are naturally free to imagine that some apparatus has all its degrees of freedom
perfectly controlled and isolated from the environment-somewhere in intergalactic
space-as in Albert’s quantum mechanical automaton.Ig However, in that case, there
is no measurement.*’ The fact that some variables of the automaton get correlated to
some others is not a measurement.20*2’ The two electrons in the ground state of the
helium atom are correlated, but no one would say that each electron “measures” its
partner. In general, if we have a piece of hardware that can be used as a measuring
apparatus, we must choose one of the following alternatives: either let it work in a noisy
environment (including its own internal “irrelevant” degrees of freedom), or let it be
perfectly prepared and isolated, and described by the Schrijdinger equation. In the
latter case, that piece of hardware loses its status of “measuring apparatus.”22This is
just a matter of having consistent definitions: A measuring apparatus must have
macroscopically distinguishable states, and the word “macroscopic” has just been
defined as “incapable of being isolated from the environment.” If there is no
irreversibility, there are no measurements.

ON OBSERVERS AND CONSCIOUSNESS

The reader may object that my explanations are anthropocentric and, in particular,
that the observer’s consciousness (see reference 2, p. 587, or reference 3, p. 420) plays a
dominant role. I, therefore, expect some sarcastic questions such as: Should the
observer be highly qualified (with a Ph.D. in physics), or what happens in inhospitable
places such as stellar interiors, let alone black holes?
This sarcasm is not justified. Just as “wave functions” are mathematical expres-
sions invented by physicists, “observers” too are only figments of our imagination that
are used to prove theorems. Fictitious observers are not restricted to quantum theory:
They are also used in relating thermodynamics to information theory?’ and, as is well
known, in the special theory of relativity. For example, when we say that no
information can be transferred faster than by light signals, we imagine a pair of
observers, with one actively making decisions and sending information, and the other
passively receiving that information. This apparently subjective distinction between
cause and effect yields an absolute distinction between the past and future light
cones.
Nobody in his right mind would argue that Lorentz invariance (with all its
consequences) is valid only in those parts of the universe that are inhabited by human
observers. Likewise, quantum theory is valid even if the observers are not there: We can
safely compute cross sections (that is, probabilities of events), energy levels, etc.,
without invoking fictitious observers. It is enough to invoke fictitious “wave functions”
that symbolize the “knowledge” of these observers?” However, once in a while, the
derivation of a formula, the proof of a theorem, or the clarification of a result can be
made easier by introducing the mythical observers. There is no reason then to apologize
for that.
This paper would not be complete without a mention of Everett’s valiant attempt to
eliminate the wave function collapse? In Everett’s formalism, the observer is included
in the wave function and all the terms in the right-hand side of equation 6 always exist.
This can be understood in different ways, which have been called the “relative-states
interpretation” and the “many-worlds interpretation.” None is a satisfactory alterna-
PERES WHEN IS A QUANTUM MEASUREMENT? 445

tive6to the standard formalism with the meaning that was assigned to it in the previous
section.
However, Everett’s work can also be interpreted in a way that is completely
compatible with the standard approach. In the latter, each observer imagines that he or
she is dealing with a Gibbs ensemble of quantum objects and apparatuses. The wave
function represents all the information compatible with that ensemble. The inclusion of
the observer in the wave function simply means that one now has to consider a Gibbs
ensemble of observers, with each one of them experimenting with a single quantum
object and a single apparatus. It is a matter of personal taste which interpretation one
prefers.
There is nothing strange or unusual in being a member of a Gibbs ensemble. This is
indeed the situation of anyone having a life insurance policy. The officers of the life
insurance company only know a wave function (it is the medical report of their
physician who examined the customer) and there is a corresponding Schrodinger
equation (they call ,it a mortality table). According to it, the company collects the
insurance premium, until some day the beneficiaries will collect the amount of the
policy.

CONTINUOUS MONITORING IN QUANTUM THEORY

Until now, we have considered each “measurement” as an instantaneous event. In
other words, it was assumed that the interaction of the measuring instrument with the
quantum system was so strong that it could be made arbitrarily brief, and thus one
could neglect Ht/h during the measurement. This drastic simplification may not
always be justified. Coupling constants occurring in nature are finite, and sometimes
very small.24It may, therefore, be necessary to couple the measured system and the
apparatus during a finite, possibly long time.
This problem is not specific to quantum theory. It may arise in everyday life, for
example, when a photographer takes a snapshot of a moving object. However, quantum
theory introduces some novel features because a measurement is not only a passive
observation, but also the preparation of a new state. The detailed dynamical theory of
such quantum measurements of finite duration can be worked out e~plicitly.~’ It turns
out that the result of the measurement is not, in general, the time average of the
observed quantity. It may not be one of the eigenvalues of the operator representing the
dynamical variable being measured. Finally, some outcomes, those which do corre-
spond to the eigenvalues, appear frozen by the “Zen0 e f f e ~ t . ” ~I”shall
~ ~ not repeat
here the details of the calculations because they can be found in the literat~re.’~
As a final example, consider a radioactive atom in the presence of a perfect
detector. It should be emphasized that the mere presence of a detector capable of
registering the decay is not a measurement (in the technical sense of this term3v4).A
complete quantum mechanical treatment3’ of the atom + detector system shows that
the final state of the detector (that is, after many mean lifetimes of the atom) is a
superposition of states corresponding to different decay times that are distributed
according to the familiar exponential decay law. Obviously, the final wave function
only gives statistical information on what may happen: quantum theory is unable to
predict when an individual atom will decay.
The point is that the continuous interaction between the detector and the decay
446 ANNALS NEW YORK ACADEMY OF SCIENCE§

products is not a measurement (let alone the equivalent of a large number of

consecutive measurements). A measurement of the type capable of causing the Zeno
effect is a very brief and intense interaction between the observed system (the
radioactive atom) and a macroscopic apparatus. This interaction causes different
states of the atom to be correlated to macroscopically distinguishable states of the
apparatus. As explained above, it is essential to the consistency of the von Neumann
f ~ r m a l i s mthat
~ . ~these final states be macroscopically distinguishable, that is, incoher-
ent.I3Otherwise, there may be no Zen0 effect.29

Particle / detected
x2

R+c

R+b
R

0 0 20 X1
FIGURE 1. Topographic map of the potential energy in equation 19. Blank V = 0; hatched: V =
-
Vo;crosshatched: V 2V0;dotted: V = - V,. See text for details.

To conclude, let me present a very simple one-dimensional model for the radioac-
tive atom and its detector. The “atom” is represented by a particle of mass, m, and
position, x,, in a piecewise constant potential,
V , ( x , ) = V, if u < x , < 2a,
=m ifx,<O,
= 0 otherwise. (16)
Here, V, >> h2/ma2so that there are long-lived metastable states in the potential well.
Likewise, the “detector” is represented by a particle of mass, M , and position, x,, in
PERES WHEN IS A QUANTUM MEASUREMENT? 447

a piecewise constant potential,

V2(x2)= V, if 0 < x2 < R or R + b < x2 < R + c,
= m if x2 < 0,

= OifR < x 2 < R + b o r x , > R + c. (17)

Here, R >> u is the distance between the atom and the detector, and, for simplicity, I
have chosen the same Voas before. Moreover, M(c - b)’ B mu2 so that the detector
would have a much longer lifetime than the atom if the two did not interact.
Finally, there is an interaction term,
I/in,(x,- x,) = -Vo if (xI- x21< d,
= 0 otherwise. (18)
FIGURE1 shows a topographic map of the potential,

Initially, the wave function is localized in region 1, inside both potential wells; that is,
the atom is excited and the detector is in its “ready” state. Tunneling then proceeds via
region 2 (this is the radioactive decay), and then the ejected particle freely travels
through region 3 until it reaches the detector in region 4. It will then continue toward
larger xI,either detected or undetected (the efficiency of the detector depends on the
value of c - b versus that of d).
Although this model is grossly oversimplified, it has all the features to describe how
the decay of an unstable particle triggers the decay of a detector prepared in a
metastable state. Their interaction is continuous and yet it is not subject to any Zen0
effect.

ACKNOWLEDGMENTS

Many of the ideas reported here originated during visits at the University of
California (Santa Barbara), the University of Texas (Austin), and the Centre
International de Sciences M6caniques (Udine). I am grateful to all of these institutions
for their hospitality.

REFERENCES
1. DIRAC,P. A. M. 1947. The Principles of Quantum Mechanics, p. 36. Oxford Univ. Press.
Oxford.
2. BOHM,D. 1951. Quantum Theory, p. 120. Prentice-Hall. New York.
3. VON NEUMANN, J. 1955. Mathematical Foundations of Quantum Mechanics. Princeton
Univ. Press. Princeton, New Jersey. (Translated from: Mathematische Grundlagen der
Quantenmechanik. 1932. Springer-Verlag.Berlin.)
4. WHEELER,J. A. & W. H. ZUREK. 1983. Quantum Theory and Measurement. Princeton
Univ. Press. Princeton, New Jersey.
5. EVERETT, H. 1957. Rev. Mod. Phys. 2 9 454.
6. WHITAKER, M. A. B. 1985. J. Phys. A: Math. Gen. 18: 253.
7. ROTHSTEIN, J. 1951. Science 1 1 4 171; 1951. Am. J. Phys. 2 5 510.
448 ANNALS NEW YORK ACADEMY OF SCIENCES

8. PERFS,A. 1984. Am. J. Phys. 5 2 644.

9. FINKELSTEIN, D. 1971. The physics of logic. In Paradigms and Paradoxes (vol. V). R. C.
Colodry, Ed. Univ. of Pittsburgh Press. Pittsburgh, Pennsylvania. (Reprinted in Logico-
Algebraic Approach to Quantum Mechanics, vol. 11. 1975. C. A. Hooker, Ed.: 141-160.
Reidel. Dordrecht.)
10. STAPP,H. P. 1972. Am. J. Phys. 4 0 1098.
11. BALLENTINE, L. E. 1970. Rev. Mod. Phys. 4 2 358.
12. PERFS,A. 1974. Am. J. Phys. 4 2 886.
13. PERES,A. 1984. Found. Phys. 1 4 1131.
14. BELL,J. S. 1975. Helv. Phys. Acta 48:93.
15. MOLDAUER, P. A. 1972. Phys. Rev. M:1028.
16. PERFS,A. 1980. Phys. Rev. D22 879.
17. ZUREK,W. H. 1981. Phys. Rev. D24: 1516; 1982. Phys. Rev. D26 1862.
18. LEGGETT,A. J. 1980. Suppl. Prog. Theor. Phys. 6 9 80.
19. ALBERT,D. Z. 1983. Phys. Lett. 9 8 A 249.
20. PERES,A. 1984. Phys. Lett. 101A: 249.
21. WHEATHER, J. & R. PEIERLS.1983. Phys. Rev. Lett. 51: 1601.
22. PERFS,A. & W. H. ZUREK.1982. Am. J. Phys. 50: 807.
23. SZILARD, L. 1929. Z. Phys. 5 3 840.
24. CAVES,C. M. et al. 1980. Rev. Mod. Phys. 52: 341.
25. PERES,A.& W. K. WOOTTeRS. 1985. Phys. Rev. D32: 1968.
26. MISRA,B. & E. C. G. SUDARSHAN. 1977. J. Math. Phys. 18: 756.
27. CHIU,C. B., E. C. G. SUDARSHAN & B. MISRA.1977. Phys. Rev. D16 520.
28. PERFS, A. 1980. Am. J. Phys. 4 8 93 1.
29. SINGH,I. & M. A. B. WHITAKER. 1982. Am. J. Phys. 5 0 882.
30. PERES,A. 1980. Am. J. Phys. 48: 552.