Вы находитесь на странице: 1из 18

Journal of Philosophy, Inc.

Quantum Mechanics and the Nature of Continuous Physical Quantities

Author(s): Paul Teller
Source: The Journal of Philosophy, Vol. 76, No. 7 (Jul., 1979), pp. 345-361
Published by: Journal of Philosophy, Inc.
Stable URL: http://www.jstor.org/stable/2025451 .
Accessed: 27/10/2014 19:40

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Journal of Philosophy, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Philosophy.


This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

4_. - , .:,


Q ^ UANTUM mechanics does not provide predictions for si-
multaneous exact values of quantities corresponding to
noncommuting operators. Traditionally, the Copenhagen
interpretation deals with this apparent incompleteness by main-
taining that such complementary quantities do not have values
simultaneously. Under what conditions, then, does a quantity have
a value?
The orthodox answer: For the special case of discrete observables,
all students of quantum mechanics learn from the outset that a sys-
tem has a value a for observable A if the system is in an a-eigenstate
of A. Put informally, such a system has a value if it is in some
special state about which the theory says that the value is sure to
be observed if a measurement is made. A system is also said to have
a value for A if A has been (or, sometimes, will in fact be) mea-
sured. And, traditionally, a system is said to have a value for A only
under these conditions. These answers are brought together with
great beauty by the projection postulate: after measurement for
observable A the system is to be represented by an eigenstate or a
mixture of eigenstates for observable A which, if we restrict atten-
tion to eigenstates for A after an A-measurement, can be under-
* My thanks to the many people who helped me with this paper, and for
the grant support from the National Science Foundation (grant #SOC 76-
82113). I would particularly like to call the reader's attention to Arthur Fine's
"Probability in Quantum Mechanics and in Other Statistical Theories" in
M. Bunge, ed., Problems in the Foundations of Physics (Berlin: Springer, 1971),
pp. 79-92, which advances the idea that quantum-mechanical quantities should
be taken to be "spread out" and treats the connection between this idea and
the probabilistic interpretation of quantum mechanics in more detail than I
have done in this paper.

0022-362X/79/7607/0345$01.70 C 1979 The Journal of Philosophy, Inc.


This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

stood along the lines of the ignorance interpretation. The perceived

generality of this answer is encouraged by John von Neumann's
coarse-graining technique for treating the observation of continuous
quantities. Corresponding to a continuous quantity we construct a
discrete one by dividing the continuum of values into intervals,
where each discrete eigenvalue of the constructed observable is
taken to correspond to location of the quantity in the corresponding
This perceived generality is deceptive. We must look at what
quantum mechanics has directly to say about actual values of a
quantity whose operator has a continuous spectrum. For such a
quantity there are no eigenstates and, hence, no states for which
the system has a probability of 1 for having some specific value of
the quantity. The closest the theory can come is description in terms
of what I will call localized state functions, more specifically strictly
localized state functions, which are 0 outside of some interval (a -,
a + c), and highly localized state functions, which have negligible
amplitude outside of some interval (a - E, a+ E). Moreover, a state
function can be simultaneously highly localized for noncommuting
quantities such as the observables for position and momentum. The
dogma about no simultaneous values for conjugate quantities ap-
plies only to point values or values with precision in excess of what
the uncertainty relations allow.' The interpretation of experimental
data makes this obvious: for example, in cloud-chamber tracks the
position of the passing particle is located to the finite width of the
track, and the momentum is localized by the imprecisely deter-
mined curvature of the track together with the experimental con-
ditions. The fact that such experimental situations localize both
position and momentum, but neither with complete precision,
eases the way toward what I will call the orthodox answer to the
question of when a continuous quantity has a value: A system in
state q has a value of the quantity A which is definite or well de-
fined up to the degree of localization of ql as expressed in the A-
representation; and when an A-measurement is made on a system,
the system takes on a value of A as well defined as, or with the
precision corresponding to, the measurement in question. The re-
sult of measurement is expressed in terms of a localized state func-
tion (or a mixture of these) where the degree of localization cor-
responds to the precision of the measurement.
1 This is already clear in von Neumann's treatment of "macroscopic ob-
servables." See John von Neumann, Mathematical Foundations of Quantum
Mechanics (Princeton, N.J.: University Press, 1955), pp. 402-407.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

The problem of interpreting the orthodox answer: It is part of the

orthodox answer that the values of continuous quantities are im-
precise, or not perfectly well defined. But what does this mean?
This is a pressing problem because the answers commonly given
don't work. Focusing on position as an example, often it is said that
when one makes a position measurement the particle is certain to
be somewhere in the interval of localization of the resulting strictly
localized state function, or almost certain to be in the interval of
localization of the resulting highly localized state function. The
orthodox view itself seems to enjoin us from saying this, but it
would be a waste of time to try to reformulate the orthodox view
so as to remove the suspected conflict. This tack leads immediately
to the devastating conclusion that there are states of systems which
cannot even be described within quantum mechanics. I will say
that a theory is descriptively (as opposed to predictively) incomplete
if entities within the domain of the theory can occupy states that
cannot be described within the theory. In quantum mechanics, the
only states that can be described are those representable as normed
vectors (strictly unit rays) in an appropriate Hilbert space, and the
values assigned to systems by the theory are always the eigenvalues
of the representing vector. Continuous quantities have no eigenvec-
tors with eigenvalues in the continuous part of their spectra. Thus,
if imprecise or imperfectly defined values of a continuous quantity
are to be explained in terms of assumed underlying states with exact
point values, these assumed underlying states cannot be described
within the theory, and the theory is descriptively incomplete. Surely
this is paying too high a price.
Another response to the interpretive problem is to say that the
probability is 1 (in the case of strict localization) or nearly 1 (in
the case of high localization) for the particle being in the interval.
This interpretation is almost forced upon us when we describe the
absolute value squared of the state function as the probability
density for position. But unless what is meant is probability for
point-valued positions, the interpretation remains unclear, and if
probability for point-valued positions is intended, this transparently
involves the same problem as before. If we give probability the
relative-frequency interpretation, we require frequencies of actual
point values, hence actual point values, and so the theory again
comes out descriptively incomplete. If we give probability a sub-
jective interpretation, we require belief on the part of the theorist
that the particle has a point value somewhere in the interval, and,
consequently, belief in the theory's descriptive incompleteness.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

Finally, if we analyze probability as a propensity, we are com-

mitted to the particle having a propensity to realize one or another
exact position, a state which cannot be described within the theory
once again yielding descriptive incompleteness.
A last move often made by way of explaining the orthodox posi-
tion is to say that if a second, more refined measurement imme-
diately follows a first measurement, the second is sure (or almost
certain) to yield a result within the interval obtained with the first
measurement. But this answer gets us only as far as our under-
standing of the interpretation of the second measurement, and to
this point we have only the orthodox position to which to appeal.
This approach appeals to precisely the interpretation that needs
to be clarified. So far all attempts to understand the orthodox
position lead to an interpretive dead end.

Approximations and idealizations: This interpretive impasse is press-

ing. Why has it received no attention? I suggest that physicists con-
done what they take to be inexact descriptions as simplifying ap-
proximations. It is part of the art of doing physics to find an ade-
quate approximate or idealized idescription when some practical
or technical difficulty gets in the way of giving an exact description
of an actual situation-indeed this is almost always what physicists
find themselves having to do. In particular, quantum theory pro-
vides no exact descriptions of measurement results which are pre-
supposed to be, at least in principle, of something precise. So physi-
cists use the approximate descriptions of localized state functions,
reassuring themselves that these can be made to approximate exact
values as closely as one likes. Since such use of approximations is
the physicists' normal way of conducting business, they proceed
without giving further thought to the matter. Although there is no
question that this business procedure is profitable, we must ask
whether such an appeal to approximations can help with the in-
terpretive question. To do so requires some reflection on the nature
of approximations and idealizations, and the contrast between them.
Suppose one is trying to describe a system characterized by cer-
tain quantities which in turn are exactly described by an explicitly
given theory. Suppose further that exact calculations are not needed
or are out of reach. In such a situation the physicist makes simplify-
ing assumptions and obtains values or relations which are, strictly
speaking, incorrect but which are close enough for the practical or
theoretical purposes at hand. I have in mind examples like the
derivation of the pendulum law for which one approximates the

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

sine of the angle of the pendulum's deflection with the angle. In

such cases one can, at least in principle, calculate the bounds of
introduced error by appealing to the underlying theory; or, again at
least in principle, one can go back and calculate the exact values.
In such cases (and I think in close accord with physicists' actual
usage) I will speak of the simplifying assumptions as approxima-
tions. Approximations introduce a numerical distortion but not an
"ontological" distortion; that is, approximations do not introduce
distortions that seriously misrepresent the kind of thing the system
is. In contrast, I will speak of idealizations to cover convenient mis-
descriptions that go beyond mere numerical misrepresentation.
Here I have in mind cases such as talk of point masses, point
charges, and perfectly rigid measuring bars. In such cases ease of
calculation or theoretical limitations call for description of a sys-
tem which does seriously misrepresent the kind of thing it is. Some-
times we know a theory in terms of which the simplifying idealiza-
tion can be eliminated (point masses in Newtonian mechanics) and
sometimes not (point charges and perfectly rigid measuring bars);
but in all cases of idealization there is a more nearly correct theory
or point of view relative to which one could judge the idealization
to be a distortion, relative to which one could judge that there
really are no such things. This contrast between approximations
and idealizations is a matter of degree, not an all-or-nothing affair;
for sufficiently large numerical distortions always begin to seriously
misrepresent the nature of a system. The fiction of ideal gases (per-
fectly elastic spheres with intermolecular forces neglected) provides
a good case in which numerical simplifications are carried to an
extreme, resulting in serious ontological misrepresentation. But the
distinction, though not sharp, is important inasmuch as we want a
theory to reveal to us the true nature of its subject matter. We
miss this mark when theoretical limitations force us to idealize.
Now let us look at continuous quantum-mechanical quantities
against the background assumption the ("point-value theory") that
these quantities take on exact point values, at least when the quan-
tities are measured. Quantum mechanics provides only descriptions
with spread, not point-valued descriptions. One cannot even ap-
proximate point values arbitrarily closely in real experimental
situations. For example, to get an arbitrarily precise position
measurement one needs arbitrarily large uncertainty in momentum,
and such an experiment would require an arbitrarily large energy
source. And to get an arbitrarily precise momentum measurement
one would require an arbitrarily large uncertainty in position, but

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

one always knows something about the position of the observed

particle, for example, that it is at least within the laboratory. Con-
sequently, against the background of the point-value theory, such
quantum-mechanical descriptions must be judged to be, not ap-
proximations, but idealizations. If quantum-mechanical systems
have exact point values, the dispersed quantum-mechanical descrip-
tions seriously misrepresent the nature of these systems. Point-value
theorists cannot whitewash the problem by appeal to approxima-
tions, since it is idealizations, not approximations, which are in

The status of point values in traditional interpretations of quantum

mechanics: Physicists sometimes deal with this situation by tacitly
giving up the point-value theory and acquiescing in the orthodox
theorist's talk of imprecise or "imperfectly defined" positions and
momenta. But interpreters have shied away from this approach be-
cause it seems so unclear. Furthermore, interpreters and physicists
working within the Copenhagen tradition feel barred from this
alternative by Bohr's dictum that the results of measurement must
be described with the concepts of classical physics, since a classical
interpretation of quantities such as position and momentum means
point values. There is some delicacy here, since at least often when
Bohr insists on description of measurement results with the con-
cepts of classical physics he seems to mean that the macroscopic
laboratory equipment and record of results (e.g., observable spots
on photographic plates) are what must be described classically. This
leaves open the question of how observed microscopic objects are to
be described, and interpreters differ as to how (if at all) Bohr
wanted to fill in here. I don't want to get sidetracked on Bohr
exegesis. What matters for us is that, whatever Bohr thought, a great
many working physicists, thinking realistically about microscopic
objects, apply Bohr's dictum to quantities like position and mo-
mentum of objects like electrons. If observed positions and mo-
menta of electrons are to be described with the concepts of classical
physics, they are to be described as having point values. But quan-
tum mechanics provides only descriptions with spread. Thus, Bohr's
dictum, applied in this (what I will call "popular") manner, leads
in the quickest possible way to the descriptive incompleteness of
quantum mechanics.
It is important to appreciate that this incompleteness comes in
addition to the predictive incompleteness suggested by the un-
certainty relations and to appreciate that Bohr's doctrine of com-

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

plementarity cannot help, as it might with predictive incomplete-

ness. The argument for predictive incompleteness starts from the
belief, perhaps buttressed by Einstein, Podolsky, and Rosen's con-
siderations, that there are simultaneous values for conjugate quan-
tities. But, by the uncertainty relations, quantum mechanics can-
not predict such simultaneous values, above a certain limit of ac-
curacy. Predictive incompleteness follows. Bohr responds by con-
tradicting the premise of this argument. He calls conjugate quan-
tities such as position and momentum complementary, by which he
means that a system can have a value for one or the other, but
never both. Measurement determines which value a system has:
when, but only when, one of these quantities is measured, we can
correctly attribute a value of the quantity to the system. Since it is
physically impossible simultaneously to measure conjugate quanti-
ties with precision in excess of what is permitted by the uncertainty
relations, Bohr concludes that quantum mechanics has not, after
all, been shown to be predictively incomplete. Note that, as I have
summarized the arguments, I have slurred over the issue of whether
continuous quantities take on point values or, in some yet to be
explained sense, inexact values. In so doing I think I am being
faithful to the presentation of these arguments as they have been
given in the literature. (The exchange between Bohr and Einstein,
Podolsky, and Rosen is phrased in terms of point values and delta
functions, but it seems clear that these notions were used as con-
venient expository idealizations, as is made evident by later pres-
entation of the arguments in terms of discrete spin magnitudes.)
To contrast these arguments with the argument for descriptive
incompleteness, let us grant Bohr's views about complementarity.
Suppose the position of a particle has been measured. Then, ac-
cording to Bohr, the particle must have a position, whether it be
precise or, in some sense, imprecise. Now we apply, in the popular
manner, Bohr's dictum that the outcome of measurements must be
described with the concepts of classical physics. It follows that the
position must take on some precise point value. But quantum
mechanics provides no description of the system which attributes to
it any precise point value. We are forced to conclude that quantum
mechanics is descriptively incomplete, and since this argument
make no assumptions about the particle's momentum, comple-
mentarity cannot be used to block the conclusion.

Two alleged oversights: The reader might object that two oversights
compromise my presentation. First, one might argue that classical

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

physics faced unavoidably imprecise measurements no less than

quantum mechanics, so the problem, if any, is in no way special to
quantum mechanics. I agree that the classical physicist was also
restricted to inexact measurements. But the theory of classical
mechanics includes descriptions of states with exact values, and this
makes all the difference. In the context of classical physics the out-
come of an inexact measurement may be described as a probability
distribution over exact values, and we know how to state what this
means: There is a probability (interpreted as a relative frequency,
subjective degree of belief, or propensity) for the value of the mea-
sured quantity to have one or another of the exact values in the
support of the distribution. We may not know exactly what this
exact value is, but whatever it may be, there is a statement expres-
sible within the language of classical physics which attributes that
value of the quantity to the system. But quantum mechanics pro-
vides no description that attributes any given exact value to the
system. Thus if we believe that systems possess exact values for
continuous quantities, classical theory contains the descriptive
resources for attributing such values to the system, whether or not
measurements are taken to be imprecise in some sense. Quantum
mechanics has no such descriptive resource.
One might claim, however, that, contrary to what I have alleged,
quantum mechanics does have the descriptive resources for describ-
ing exact values for continuous quantities, for example, Dirac delta
functions for position and plain waves for momentum. For many
years interpreters castigated Dirac delta functions for their lack of
a sound mathematical development. This objection no longer holds,
since delta functions have found a rigorous interpretation within
the mathematical theory of distributions. Distribution theory itself
does not provide the required formal relations; so it has been sup-
plemented in turn by more complex mathematical objects called
rigged Hilbert spaces, which include as one component the usual
Hilbert spaces used by quantum mechanics.2 The components of a
rigged Hilbert space which extend standard Hilbert spaces now
include additional mathematical objects which might be thought
of as representing exact values of continuous quantities. The diffi-
culty with this approach is that the physical theory still represents
all actual physical states by unit rays in Hilbert space. The addi-

2See J.-P. Antoine, "Dirac Formalism and Symmetry Problems in Quantum

Mechanics, I: General Dirac Formalism," Journal of Mathematical Physics, x,
1 (January 1969): 53-69; and J. E. Roberts. "The Dirac Bra and Ket Formal-
ism," ibid., VII, 6 (June 1966): 1097-1104.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

tional mathematical objects, however useful for mathematical ma-

nipulation and clarification, represent only idealized states which,
strictly speaking, do not exist according to the theory.3 One who is
convinced that continuous quantities do take on exact values might
argue at this point that our considerations have revealed an in-
adequacy in the physical theory, which should be remedied by
allowing the new mathematical objects to describe actually existing
states. But there are strong physical reasons for not extending the
interpretation of the formalism in this way. Such an extended
physical theory would describe systems as having precise point-
valued positions, but as also having totally indeterminate momen-
tum, not even highly localized to any finite interval. Such systems
would have an infinite expectation value for their kinetic energy.
We never observe such systems, and we have strong theoretical
reasons for thinking that such systems cannot exist. The additional
mathematical objects of rigged Hilbert spaces are idealizations only,
and quantum mechanics, properly construed, provides descriptions
only of systems that do not actually have precise values for con-
tinuous quantities. One who believes that continuous quantities
take on precise values must judge quantum mechanics to be de-
scriptively incomplete.

Summary of the problem and approach to a solution: We are faced

with a dilemma. If one believes that continuous quantum-mechan-
ical quantities take on point values-perhaps only upon measure-
ment and thus not in conflict with the doctrine of complementarity
-then one must conclude that quantum mechanics is descriptively
incomplete. On the other hand, if one rejects point values and
believes that localized state functions with nonzero spread represent
all there is to be said about continuous quantities, one has an in-
terpretive problem. What does it mean to say that a particle has, for
example, a position that is spread out as described by a localized
state function? I propose to investigate how well we can reconcile
ourselves to the second horn of this dilemma by coming to grips
with the interpretive problem. I hope to do this by focusing on the
fact that many theoretical concepts arise through refinement of
inexact prescientific precursors. When we look at the way in which
we make inexact concepts more precise, non-point-valued descrip-
tions of quantities such as position may come to seem quite natural.
3 See Antoine, op. cit., p. 67.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

Refinement of inexact concepts: To carry out my program I need at

least to outline an account of the refinement of inexact concepts. A
concept comes equipped with (or perhaps is constituted by) a set of
rules for its application, which in turn generates the concept's truth
conditions. (This is only a first approximation, which, however,
should be close enough for present purposes.) For most, perhaps all,
rules, there are situations in which they break down-we want to
apply the rules but are unable to do so. This fact is a commonplace
for legal rules, keeping lawyers and judges in business. Or again,
the rules of chess do not tell us what to do if one of the pieces
jumps up and refuses to move to KB4, arguing that it would result
in checkmate in three moves. (Chess is occasionally played with
human chess pieces.) When the rules governing a concept fail to
apply in a situation, they fail to give us truth conditions, and a
borderline case results. Note, incidentally, that there will also be
situations in which it is not clear whether or not a rule applies;
this might be found to correspond to the indefiniteness of boundary
between borderline and clear-cut cases.
When a concept's rules break down, we may try to improve the
situation in one of two ways. We may change the concept by con-
sistently extending its rules, or we may drop the concept in favor
of a somewhat different one that will do the work of the old con-
cept and apply successfully in some situations in which the prior
concept failed. As we will see, either of these kinds of refinement
may be arbitrary, but sometimes there will be objective considera-
tions that guide our choice.
Here are some examples. When the concept tall (expressed with
a one-place predicate) breaks down, we may fill our unmet con-
ceptual needs by introducing the relational taller than. At first this
replacing concept provides only a qualitative comparison, but if
necessary, we can further refine it by appeal to the theory of linear
measure, good enough for all normal comparisons. Or consider the
problem we experience when the concept flat is brought up against
the fact that no physical object is perfectly flat. Again we can go
over to a relational concept or, equivalently, a concept that takes
flatness to be a matter of degree. This is still imprecise because there
are different ways of measuring degrees of flatness or closeness to a
geometrical plane. The concepts of physical geometry give us the
conceptual tools needed for further refinement, but these can be
applied in various ways. We might measure degrees of flatness in
terms of maximal deviation from some average plane or in terms
of the maximum or average curvature of the surface. Which re-

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

finement we choose will be arbitrary, guided only by varying prac-

tical applications.

Refining the concept of location: To focus my development, I will

concentrate on applying the foregoing picture to refining our con-
cept of position, and, to emphasize that it is our prescientific con-
cept we are looking at, I will use the word 'location'. I want to
argue the thesis that our prescientific concepts include a concept of
location which is imprecise and for which the descriptions sug-
gested by quantum mechanics provide a refinement consistent in
practice with all prescientific usage.
No doubt my thesis sounds implausible at first, and it is essential
to diagnose and neutralize the source of this implausibility. Sup-
pose, for a moment, that there is a prescientific concept of location
which is imprecise. Classical physics provides one refinement, a re-
finement which educated people have drummed into them through
their training in traditional geometry and basic physical concepts.
In fact, our educated usage is so shot through with the conceptual
framework of classical physics that it becomes an intellectual feat
to consider any alternative open-mindedly. But to give my thesis
a fair hearing one must put aside the ingrained prejudice that the
concept of location has to be made precise along classical lines.
So let us try to reconstruct the prescientific concept of location
which we could expect a child to attain without the usual classical
prejudices of an educating environment. Most of our early learning
connected with location comes in application to extended objects,
which can hardly be expected to generate a very precise conception.
Relative precision surely comes, however, in a child's experience with
the location of object boundaries. These often seem sharp, but not
always. Boundaries may be markedly vague or indefinite. Think of
the aspect of a smudge shading off from clear portions into lighter
and less distinct areas into an indefinite mingling with the back-
ground. Similarly a puddle may end in an indefinite compromise
with its relatively dry surroundings and a red area may shade
through a vague boundary into a blue area. A tree seen from a
moderate distance may present a visual aspect without any sharp
line separating it from its backdrop. Altogether we should expect
at least one of a child's boundary concepts to correspond to the lack
of sharpness in these examples. I say at least one because we must
acknowledge the evidence for an innate disposition to sharpen per-
ceived boundaries. Gestalt psychology suggests such a sharpening
in its phenomenon of visual cliffs, and experiments on animal

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

vision indicate that the optic nerve responds to perceived bound-

aries in an all-or-nothing way. But, even conceding a predisposi-
tion to exaggerate perceived boundaries in many situations, there
can be no question that we also have extensive experience with
boundaries that we perceive to be indefinite, and such cases also
must be covered by some prescientific concept of boundary location.
Since boundary location plausibly provides our most precise pre-
scientific concept of location, we may confidently suppose our gen-
eral prescientific concepts of location at least to include an intrinsi-
cally inexact variant, that is, a concept of location for which suffi-
ciently refined questions of the form "Here or there?" receive no
clear answer, just as in certain cases the question "Is he bald?" re-
ceives no clear answer. This imprecise location concept shows itself
clearly in everyday discourse, as when we say things such as "The
chair's place is near the window" or (with Wittgenstein) "Stand
roughly there."
How will the refining tools of theory mold such an imprecise
concept of location? Just as tall may be refined or supplemented
by the more precise and quantized height or taller than, our pre-
scientific concept of location may be refined or supplemented by a
more precise and quantized concept suggested by science. The exact,
point-valued position of classical physics is one way of doing this,
perhaps encouraged by our innate predispositions. But, a priori,
classical position does not constitute the only possible refinement.
Quantum mechanics provides an alternative. It suggests that all
locations are dispersed in a way that admits of exact quantitative
description. Quantum mechanics describes this dispersion of posi-
tion in terms of an underlying continuous variable, but nothing is
held to have a (classical) position corresponding to some exact value
of the underlying variable. Rather, quantum mechanics describes
position quantitatively and exactly in terms of a distribution over
the variable; specifically by the amplitude squared of the state func-
tion representing the object, as expressed in the position represen-
What, after all, provides our best guide to the true nature of a
quantity such as position? We have here an example in which re-
finement is not arbitrary. We start with an inexact prescientific
concept of location or position and then look to our best theory of
the world to inform us as to its real nature. But, insofar as the
theory is wrong, it may misinform us. Thus, we should be prepared
to find that the characterization of classical physics gives us not
the truth about position, but only an idealization, however prac-

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions
ticaL The present discussion has aimed to free us from the grip of
classical thinking so that we may consider open-mindedly the al-
ternative suggested by quantum mechanics.
What quantum mechanics tells us about position might easily
be misunderstood in a way that one must be careful to avoid. If
an actual particle's position is correctly described as being spread
out over some interval, then one might think that if we look any-
where within the interval we should always find some evidence of
the particle. But such an experiment has only a probability of
yielding a positive result. One might conclude that particles do
not, after all, have positions that are spread out in any sense. Such
an argument turns on a misunderstanding of what quantum me-
chanics tells us about position, since it takes a state function to
describe some material substance or space-occupying field spread
out over the state function's interval of localization. Schroedinger
tried in vain to force such a classical reinterpretation on the state
function of the electron, appealing to electric charge as the sub-
stance that was thought to be distributed throughout the interval.
No such classical reconstruals of the state function seem to work.
Quantum mechanics suggests a different refinement of our pre-
scientific concepts. The position a particle has, the manifest property
of position, is a unified whole which no more has parts than did
the point positions of classical physics. This partless whole has
spread in a way which can be formally represented as the state
function's absolute value squared. At the same time, this extended
partless position includes (or perhaps is) a collection of dispositions
or potentialities to manifest a more refined position of the same
nature whenever a more refined measurement interaction takes
place. There is nothing exceptional in thus viewing position as a
manifest property and at the same time an array of dispositions.
Being white is a manifest property, but it also includes an array
of dispositions, such as to appear in various ways in various circum-
stances. Many, perhaps all properties are like this. The dispositions
involved in quantum-mechanical properties are special only in that
they are generally governed by probabilities for manifesting a dis-
played property.4
I want to mention one last surprising feature of continuous
quantities as described by quantum mechanics. What goes for posi-
tion goes for momentum (as the quantity found more technically
4 For a more detailed examination of the relation between quantum-mechan-
ical properties as localized quantities and probabilities understood along the
lines of the propensity interpretation, see Fine, op. cit., pp. 90/1.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

useful than the closely related quantity of speed). As I have already

remarked, the position and momentum of an object may be simulta-
neously highly localized. But quantum theory allows at most one
to be strictly localized. That is to say, at least one of the two
quantities must have nonzero amplitude for arbitrarily large values
of the underlying variable; and so quantum mechanics tells us that
in every case at least one of the two quantities has, in a certain
sense, indefinitely wide spread. Thus, quantum theory tells us that
if the speed (and hence the momentum) of a baseball is completely
bounded by some value, say 10,000 miles per hour, then the spread-
out position of the baseball must extend indefinitely far beyond
the confines of the ball park. Of course this conclusion loses most
of its shock value when we pay attention to the way in which the
baseball's position is spread beyond the ball park. Though not
strictly localized, the ball's position is so highly loCalized that,
beyond even microscopic distances from where we take ourselves
to see the ball, the amplitude of its state function takes on com-
pletely negligible values which can have no practical effect on the
environment. The chances of our noticing the ball in any way at
any significant distance from its "average position" are like the
chances of the air in a room beating the odds against entropy and all
rushing out the window. It is just because such remote possibilities
have no effect on our lives that the classical idealizations for posi-
tion and momentum work so superbly in our everyday macroscopic
experience. Still, when we look beyond everyday experience, the
unavailability of strict localization means that the refinement of
quantum mechanics conflicts not only with classical idealizations
but also with our prescientific conceptions, which surely place some
limits on even indefinite boundary location-that is why, at the
beginning of this section on refining the concept of location, I
claimed only that quantum mechanics' refinement of this concept
would be consistent in practice with all prescientific usage. The
conflict in principle, however, should not surprise us too much. As
we already know from cases such as the relativity of simultaneity,
a prescientific characterization of a quantity can seem ever so right
in its application to everyday experience and still turn out to be
wrong in ways that reveal themselves only when science probes
much further.5
5 The referee, who urged me to discuss the foregoing issue and suggested the
ball-park example, also worried that the indeterminateness of positions and
momenta that are not strictly localized would run afoul of arguments like the
argument I gave at the end of section I (p. 353) for thinking that there are no

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

In sum, we should allow a theory to speak for itself. The thought

that an interpretive problem arises when we let quantum mechan-
ics speak for itself about the nature of position is an illusion bred
by the prejudice that position has to be understood in terms of
the antecedently given classical conception of point positions. But
we obtain the conception of point positions no less by letting
theories speak to us-the theories of classical geometry and physics
-and we have no reason to suppose that the point positions of
classical geometry and physics provide a more accurate refinement
of our prescientific concepts than the spread-out positions of quan-
tum mechanics. Indeed, since we know classical physics to be less
nearly correct than quantum mechanics, we should take quantum
mechanics to be the greater authority.

University of Illinois at Chicago Circle

There is a technical literature on "fuzzy"quantum observablesbased on
a notion of "fuzzy"events which might seem relevant to the program
developed in this paper. I am including a short summaryto introduce this
formal approachto readersinterested in exploring it and to indicate the
ways in which it does and does not seem to fit with the views I have
This approach replaces a point value a of quantity A with a fuzzy
sample point (a, fa)pwhere fa is the confidence function characteristic of
the measuringinstrument used to measure A. That is, when the instru-
ment yields the value a, fa is taken to be the probabilitydensity character-

states with point-valued positions and totally indeterminate momenta. But

the problem I mentioned there, infinite expected kinetic energy, arises only for
states in which the momentum is "totally indeterminate" in the very strong
sense of not even being highly localized on any finite interval. This and similar
problemsdo not arise when quantities like position and momentum are simulta-
neously highly localized on finite intervals, which conventional quantum
mechanics allows up to the limits of the uncertainty principle.
B The material summarized here is drawn from the following articles: S.
Twareque Ali and H. D. Dobner, "On the Equivalenceof Non-relativisticQuan-
tum MechanicsBased upon Sharp and Fuzzy Measurements,"Journal of Math-
ematical Physics, xvii, 7 (July 1976): 1105-1111;Ali and Gerard G. Emch, "Fuzzy
Observablesin Quantum Mechanics,"ibid., xv, 2 (February1974): 176-182; Ali
and E. Prugovecki,"Systemsof Imprimitivity and Representationsof Quantum
Mechanicson Fuzzy Phase Spaces,"ibid., xviii, 2 (February1977): 219-228; E. B.
Davies, "On the Repeated Measurementof Continuous Observablesin Quantum
Mechanics,"Journal of Functional Analysis, vi (1970):318-346; Davies and J. T.
Lewis, "An Operational Approach to Quantum Probability," Communications
in Mathematical Physics, xvii (1970): 239-260; and Prugovecki, "Probability
Measures on Fuzzy Events in Phase Space," Journal of Mathematical Physics,
xvii, 4 (April 1976): 517-523.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

istic of the instrument for the actual value of A, and the instrument is
taken to give us not the point value a but the fuzzy value (a, fa). Given
this notion of fuzzy sample points, we also introduce the idea of fuzzy
events E = {(a, fa):aeE, E a borel set of reals} and confidence measures
va(E) = fEfa(x)dx. The approach then proceeds by replacing projection-
valued measures in the analysis of observables with the more general no-
tion of positive operator-valued measures. To give an example: where a
projection-valued measure takes on projectors as values,

P(E): (P(E)p)x = X.E(X),(X)

a positive operator-valued measure takes on as values operators

(a(E): ((a(E)41)x= V,(E)4t(x)
where the va are the confidence measures associated with a measuring
instrument Finally, positive operator-valuated measures give rise to fuzzy
observables, A-=fRXa(i(dX). One very nice result is that there are fuzzy
position and momentum observables, Q, P, which are (under the assump-
tion of invariance under Galilean transformations) unitarily equivalent to
the position and momentum observables Q and P, under the same uni-
tary transformation. Furthermore, Q and P are informationally equivalent
to Q and P, respectively, in the sense that knowledge of the probability
distributions for observed values gives the same information about the
state function. In addition, Q and Pl are together informationally maximal
in the sense that the distributions for outcomes of both completely deter-
mine the state function. Finally, Q and P are simultaneously observable in
the sense that the state function representing the outcome of such a
measurement operation has probabilistic spread of position and momen-
tum in the conventional sense corresponding to the spread in the ob-
served combined fuzzy position and momentum sample point.
In one area, the theory of fuzzy events and observables provides an at-
tractive formal complement to the interpretive considerations discussed
in this paper. The standard treatment of measurement of continuous
quantities appeals to coarse-graining and projection operators. But de-
scription of measurement outcomes in terms of projection operators means
that the spread in the measured quantities is characterized as being strictly
localized, and conjugate quantities cannot both be strictly localized. Thus
it would be good to have an alternative, and an alternative is available in
the treatment of measurement transformations which appeals to positive
operators instead of projection operators. This approach provides a uni-
fied treatment of changes of state associated with measurement results for
both discrete and continuous quantities. As in the traditional treatment,
the statistical operator p is described as changing to a new statistical op-
erator p' with a probability of obtaining a result for the fuzzy quantity
A being in E given by tr (G,(E)p). Abandoning appeal to projection op-
erators involves abandoning von Neumann's assumption on the repeat-
ability of measurements, but a formal sense can be given to the require-

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions

ment that measurement be very nearly repeatable, and in practice one can
continue to use projection-operator-based descriptions of measurements,
knowing that they are not idealizations but approximations.
In other respects, however, it is not at all dear that the work on fuzzy
events and observables can help us with the present interpretive program.
Our program is to interpret outcomes of measurements as partlessly spread-
out values, not as distributions over actual precise values. Once we see
this, it is not clear what is gained by distinguishing between fuzzy sample
points, as we would have to reinterpret them, and the amplitude squared
of state functions as the proper description of observed or prepared posi-
tions or other quantities. There also may be technical difficulties in re-
describing fuzzy sample points as partlessly extended values; for the inter-
pretations of the fuzziness of sample points and of state functions'
amplitude squared are very explicitly taken in this literature to be
probability distributions over exact values.


DEALLY, there is a close connection between confirmation and
truth. Theories that are well confirmedtend to be true, or at
least approximately true. Theories that are disconfirmed tend
to be false. It would seem that some such nice connection between
confirmation and truth must exist if scientific method is to be justi-
fied as a rational activity. For one of the important ends of scientific
method is the construction of theories that are true, or at least ap-
proximately true. Obviously, then, scientific method will be an
effective means to this end only if it does indeed tend to produce
true, or at least approximately true, theories. Of course, this is not
sufficient to justify scientific method as a rational activity, if only
because scientific method has other important ends besides the
construction of true theories: for example, to produce theories that
have a lot of explanatory power. However, assuming (as I will here)
that the construction of true theories is one important end of
scientific method, such a connection between confirmation and truth
is clearly necessary.
Now I don't know whether there really is this kind of relationship
between confirmation and truth, whether scientific method really
* An earlierversion of this paperwas presentedat a Tufts University philosophy
of science colloquium in April 1976. I am indebted to Hartry Field and Hilary
Putnam for helpful conversations and general inspiration. I am also grateful to
the editors of the Journal of Philosophy for forcing me to clarify and (I hope)
improve my argument.
0022-362X/79/7607/0361$02.20 ? 1979 The Journal of Philosophy, Inc.

This content downloaded from on Mon, 27 Oct 2014 19:40:09 PM

All use subject to JSTOR Terms and Conditions