The Reduction of the State Vector and Limitations on Measurement

in the Quantum Mechanics of Closed Systems*

arXiv:gr-qc/9301011v3 12 Jan 2014

James B. Hartle
Department of Physics
University of California
Santa Barbara, CA 93106

... persuaded of these principles, what havoc must we make? Hume

ABSTRACT
Measurement is a fundamental notion in the usual approximate quantum mechanics
of measured subsystems. Probabilities are predicted for the outcomes of measurements.
State vectors evolve unitarily in between measurements and by reduction of the state vector at measurements. Probabilities are computed by summing the squares of amplitudes
over alternatives which could have been measured but werent. Measurements are limited
by uncertainty principles and by other restrictions arising from the principles of quantum mechanics. This essay examines the extent to which those features of the quantum
mechanics of measured subsystems that are explicitly tied to measurement situations are
incorporated or modified in the more general quantum mechanics of closed systems in
which measurement is not a fundamental notion. There, probabilities are predicted for
decohering sets of alternative time histories of the closed system, whether or not they
represent a measurement situation. Reduction of the state vector is a necessary part of
the description of such histories. Uncertainty principles limit the possible alternatives at
one time from which histories may be constructed. Models of measurement situations are
exhibited within the quantum mechanics of the closed system containing both measured
subsystem and measuring apparatus. Limitations are derived on the existence of records
for the outcomes of measurements when the initial density matrix of the closed system is
highly impure.

* To appear in the festschrift for Dieter Brill, edited by B.-L. Hu and T. Jacobson, Cambridge
University Press, Cambridge, 1993
e-mail: hartle@cosmic.physics.ucsb.edu

0. Preface
In 1959, then an undergraduate at Princeton in search of a senior thesis topic, I was
introduced by John Wheeler to his young colleague, Dieter Brill. This was fortunate from
my point of view, for Dieter proved to have the patience, time and talent not only to
introduce me to the beauties of Einsteins general relativity but also give me instruction
and guidance in the practice of research. Our subject the method of the self-consistent
field in general relativity and its application to the gravitational geon was also fortunate.
Through it we helped lay the foundations for the short wavelength approximation for
gravitational radiation (Brill and Hartle, 1964). In particular, building on ideas of Wheeler
(1964), we introduced what Richard Isaacson (Isaacson 1968ab) was later kind enough to
call the Brill-Hartle average for the effective stress-energy tensor of short wavelength
radiation, and which was to to prove such a powerful tool when made precise in his hands in
his general theory of this approximation. It would be difficult to imagine a more marvelous
introduction to research. I have not written a paper with Dieter Brill since, but each day I
use the lessons learned from him so long ago. It is a pleasure to thank him with this small
essay on the occasion of his 60th birthday.

I. Introduction
Measurement is central to the usual formulations of quantum mechanics. Probabilities are predicted for the outcomes of measurements carried out on some subsystems of
the universe by others. In a Hamiltonian formulation of quantum mechanics, states of a
subsystem evolve unitarily in between measurements and by reduction of the state vector
at them. In a sum-over-histories formulation, amplitudes are squared and summed over
alternatives which could have been measured but werent to calculate the probabilities
of incomplete measurements. In these and other ways the notion of measurement plays a
fundamental role in the usual formulations of quantum theory.
2

The quantum mechanics of a subsystem alone, of course, does not offer a quantum
mechanical description of the workings of the measuring apparatus which acts upon it, but
it does limit what can be measured. We cannot, for instance, carry out simultaneous ideal
measurements of the position and momentum of a particle to arbitrary accuracies. Ideal
measurements are defined to leave the subsystem in eigenstates of the measured quantities
and there are no states of the subsystem for which position and momentum are specified
to accuracies better than those allowed by the Heisenberg uncertainty principle. Analyses
of the workings of measuring apparatus and subsystem as part of a single quantum system
reveal further quantum mechanical limitations on ideal measurements, as in the work of
Wigner (1952) and Araki and Yanase (1960).
Cosmology is one motivation for generalizing the quantum mechanics of measured
subsystems to a quantum mechanics of closed systems in which measurement plays no
fundamental role. Simply providing a more coherent and precise formulation of quantum
mechanics, free from many of the usual interpretive difficulties, is motivation enough for
many. Today, because of the efforts of many over the last thirty-five years, we have a
quantum mechanics of closed systems. In this formulation, it is the internal consistency
of probability sum rules that determines the sets of alternatives of the closed system for
which probabilities are predicted rather than any external notion of measurement (Griffiths, 1984; Omn`es, 1988abc; Gell-Mann and Hartle, 1990). It is the absence of quantum
mechanical interference between the individual members of a set of alternatives, or decoherence, that is a sufficient condition for the consistency of probability sum rules. It is the
initial condition of the closed system that, together with its Hamiltonian, determines which
sets of alternatives decohere and which do not. Alternatives describing a measurement situation decohere, but an alternative does not have to be part of a measurement situation in

A pedagogical introduction to the quantum mechanics of closed systems can be found in

the authors other contribution to these volumes, (Hartle, 1993a), where references to some
of the literature may be found.
3

order to decohere. Thus, for example, with an initial condition and Hamiltonian are such
that they decohere, probabilities are predicted for alternative sizes of density fluctuations
in the early universe or alternative positions of the moon whether or not they are ever
measured.
The familiar quantum mechanics of measured subsystems is an approximation to this
more general quantum mechanics of closed systems. It is an approximation that is appropriate when certain approximate features of measurement situations can be idealized
as exact. These include the decoherence of alternative configurations of the apparatus
in which the result of the measurement is registered, the correlation of these with the
measured alternatives, the short duration of certain measurement interactions compared
to characteristic dynamical time scales of the measured subsystems, the persistence of
the records of measurements, etc. etc. The question naturally arises as to the extent to
which those features of the quantum mechanics of measured subsystems that were tied to
measurement situations are incorporated, modified, or dispensed with in the more general quantum mechanics of closed systems. Are two laws of evolution still needed? Is
there reduction of the state vector, and if so, when? What becomes of a rule like square
amplitudes and sum over probabilities that one could have measured but didnt? What
becomes of the limitations on measurements in a more general theory where measurement
can be described but does not play a fundamental role. This essay is devoted to some
thoughts on these questions.

II. The Reduction of the State Vector

In the approximate quantum mechanics of measured subsystems the Schrodinger picture state of the subsystem is described by a time-dependent vector, |(t)i, in the subsys

For more discussion of ideal measurement models in the context of the quantum mechanics
of closed systems see Section IV and Hartle (1991a)
4

tems Hilbert space. In between measurements the state vector evolves unitarily:
i
h

|(t)i
= h|(t)i .
t

(2.1)

If a measurement is carried out at time tk , the probabilities for its outcomes are
2

p(k ) = skk |(tk )i .

(2.2)

Here, the {skk } are an exhaustive set of orthogonal, Schrodinger picture, projection operators describing the possible outcomes. The index k denotes the set of outcomes at time
tk , for example, a set of ranges of momentum, or a set of ranges of position, etc. The index
k denotes the particular alternative within the set a particular range of momentum, a
particular range of position, etc. If the measurement was an ideal one, that disturbed
the system as little as possible, the state vector is reduced at tk by the projection that
describes the outcome of the measurement:
|(tk )i

skk |(tk )i
kskk |(tk )ik

(2.3)

This is the second law of evolution, which together with the first (2.1), can be used to
calculate the probabilities of sequences of ideal measurements.
The two laws of evolution can be given a more unified expression. For example, in the
Heisenberg picture, the joint probability of a sequence of measured outcomes is given by
the single expression :
p(n , . . . , 1 ) = ksnn (tn ) s11 (t1 )|ik2

(2.4)

where |i is the Heisenberg state vector and

skk (tk ) = eihtk /h skk eihtk /h

(2.5)

The utility of the Heisenberg picture in giving a compact expression for the two laws of
evolution has been noted by many authors, Groenewold (1952) and Wigner (1963), among
the earliest. Similar unified expressions can be given in the sum-over-histories formulation
of quantum mechanics (Caves 1986, 1987 and Stachel 1986)
5

are the Heisenberg picture projection operators with h the Hamiltonian of the subsystem.
Nevertheless, even in such compact expressions one can distinguish unitary evolution from
the action of projections at an ideal measurement.
One gains the impression from parts of the literature that some think the law of state
vector reduction to be secondary in importance to the law of unitary evolution. Perhaps
by understanding the quantum mechanics of large, macroscopic systems that include the
measuring apparatus the second law of evolution can be derived from the first. Perhaps
the law of the reduction of the state vector is unimportant for the calculation of realistic
probabilities of physical interest. No ideas could be further from the truth in this authors
opinion. Certainly the second law of evolution is less precisely formulated that the law of
unitary evolution because the notion of an ideal measurement is vague and many realistic
measurements are not very ideal. However, as shown conclusively by Wigner (1963), the
second law of evolution is not reducible to the first and it is essential for the calculation
of probabilities of realistic, everyday interest as we shall now describe.
Scattering experiments can perhaps be said to involve but a single measurement of
the final state once the system has been prepared in an initial state. Many everyday
probabilities, however are for time sequences of measurements. For instance, in asserting
that the moon moves on a certain classical orbit one is asserting that successions of suitably
crude measurements of the moons position and momentum will be correlated in time
by Newtons deterministic law. Thus, measured classical behavior involves probabilities
for time sequences like (2.4). Successive state vector reductions are essential for their
prediction as well as many other questions of interest in quantum mechanics.
Since the state vector of a subsystem evolves unitarily except when that subsystem
is measured by an external device, some have argued that one could dispense with the
second law of evolution in the quantum mechanics of a closed system. All predictions
would be derived from a state vector, |(t)i, of the closed system that evolves in time only
6

according to the Schr

odinger equation (Everett, 1957; DeWitt, 1970). However, a state
vector is a function of one time and can, therefore, be used to predict only the probabilities
of alternatives that are at one time according to the generalization of (2.2)
2

p(k ) = Pkk |(tk )i .

(2.6)

Here, the {Pkk } are an exhaustive set of orthogonal, Schrodinger picture, projection operators representing alternatives of the closed system at a moment of time. For instance,
in a description of the system in terms of hydrodynamic variables they might represent
alternative ranges of the energy density averaged over suitable volumes. In a description of
a measurement situation, the P s might represent alternative registrations of that variable
by an apparatus.
The restriction to a unitary law of evolution and the action of projections at a single
time as in (2.6) would rule out the calculation of probabilities for time histories of the
closed system. Some have suggested that probabilities at the single marvelous moment
of time now are enough for all realistic physical prediction and retrodiction. In this
view, for example, probabilities referring to past history are more realistically understood
as the probabilities for correlations among present records. However, just to establish
whether a physical system is a good record, one needs to examine the probability for the
correlations between the present value of that record and the past event it has supposed
to have recorded. That is a probability for correlation between alternatives at two times
the probability of a history. For this and other reasons, probabilities of histories are
just as essential in the quantum mechanics of closed systems as they were in the quantum
mechanics of measured subsystems.
There is a natural generalization of expressions like (2.4) to give a framework for predicting the joint probabilities of time sequences of alternatives in the quantum mechanics

For a recent expression of this point of view, see Page and Wootters (1983).
7

of closed systems (Griffiths, 1984; Omn`es, 1988abc; Gell-Mann and Hartle, 1990). The
joint probability of a history of alternatives is
p(n , . . . , 1 ) = kPnn (tn ) P11 (t1 )|ik2

(2.7)

where the Heisenberg P s evolve according to

Pkk (tk ) = eiHtk /h Pkk (0) eiHtk /h .

(2.8)

and the times in (2.7) are ordered with the earliest closed to |i. Here, projection operators, state vectors, the Hamiltonian H, etc all refer to the Hilbert space of a closed
system, containing both apparatus and measured subsystem if any. This is most generally
the universe, in an approximation in which gross quantum fluctuations in the geometry
of spacetime can be neglected. The Heisenberg state vector |i represents the initial
condition of the closed system, assumed here to be a pure state for simplicity.
The occurrence of the projections in (2.7) can be described by saying that the state
vector is reduced at each instant of time where an alternative is considered. However,
the important point for the present discussion of state vector reduction is that the projections in (2.7) are not, perforce, associated with a measurement by some external system.
This is a quantum mechanics of a closed system! The P s can represent any alternative
at a moment of time. Measurement situations within a closed system of apparatus and
measured subsystem can be described by appropriate P s (see Section IV) but the P s do
not necessarily have to describe measurement situations. They might describe alternative
positions of the moon whether or not it is being observed or alternative values of density
fluctuations in the early universe where ordinary measurement situations of any kind are
unlikely to have existed. Thus, the state vector can be said to be reduced in (2.7) by the
action of the projections and one might even say that there are two laws of evolution

For a generalized quantum mechanics of closed systems that includes quantum spacetime
see Hartle (1993b) and the references therein.
8

present, but those reductions and evolutions have nothing to do, in general, with measurement situations. In the authors view, it is clearer not to use the language of reduction
and two laws of evolution, but simply to regard (2.7) as the law for the joint probability
of a sequence of alternatives of a closed system. Projections occur therein because they
are the way alternatives are represented in the quantum mechanics of closed systems.
There is a good reason why the probabilities (2.4) of a sequence of alternatives of a
subsystem refer only to the results of measured alternatives. It would be inconsistent generally to calculate probabilities of histories that have not been measured because the sum
rules of probability theory would not be satisfied as a consequence of quantum mechanical
interference. In the two-slit experiment, for instance, the probability to arrive at a point
on the screen is not the sum of the probabilities to go through the alternative slits and
arrive at that point unless the alternative passages have been measured and the interference between them destroyed. Thus, probabilities are not predicted for all possible sets of
histories of a subsystem but only those which have been measured.
Probabilities are not predicted for every set of alternative histories of a closed system
either. But it is not an external notion of measured that discriminates those sets for
which probabilities are predicted from those which are not. Rather, it is the internal consistency of the probability sum rules that distinguishes them (Griffiths, 1984). Probabilities
are consistent for a set of histories, when, in a partition of the set of histories into an
exhaustive set of exclusive classes, the probabilities of the individual classes are the sums
of the probabilities of the histories they contain for all allowed partitions. A sufficient
condition for the consistency of probabilities is the absence of interference between the
individual histories in the set as measured by the overlap
h|Pn (tn ) P1 (t1 ) P11 (t1 ) Pnn (tn )|i 1 n n .
n

(2.9)

Sets of histories that satisfy (2.9) are said to decohere. Decoherence implies the consistency

There are several possible decoherence conditions. This is medium decoherence in the
9

of the probability sum rules. In the quantum mechanics of closed systems, probabilities
are predicted for just those sets of alternative histories that decohere according to (2.9) as
a consequence of the systems Hamiltonian and initial quantum state |i.

III. Uncertainty Principles

The state of a single particle cannot be simultaneously an eigenstate of position and

momentum. It follows from their commutation relations that position and momentum
cannot be specified to accuracies greater than those allowed by the Heisenberg uncertainty
principle
xp 21 h .

(3.1)

Following the standard discussion, we infer from the mathematical inequality (3.1) that it
is not possible to simultaneously perform ideal measurements of position and momentum
to accuracies better than that allowed by the uncertainty principle (3.1). There can be no
such ideal measurement because there is no projection operator, s, that could represent
its outcome in (2.3).
The limitations on ideal measurements implied by the uncertainty principle (2.3) are
usually argued to extend to non-ideal measurements as well. Examination of quantum
mechanical models of specific measurement situations have for the most part verified the
consistency of this extension although some have maintained otherwise. No such elaborate
analysis is needed to demonstrate the impossibility of ideal measurements of position and
momentum to accuracies better than those allowed by the uncertainty principle. That
limitation follows from the quantum mechanics of the subsystem alone.
The mathematical derivation of the uncertainty relation (3.1) is, of course, no less valid

terminology of Gell-Mann and Hartle (1990b).

For example, Margenau (1958) and Prugovecki (1967). We are not discussing here, nor do
we discuss later, unsharp observables or effects. For those see e.g., Busch (1987).
10

in the Hilbert space of a closed system than it is for that of a subsystem. In the quantum
mechanics of a closed system, however, the absence of projection operators that specify x
and p to accuracies better than (3.1) is not to be interpreted as a limitation on external
measurements of this closed system. By hypothesis there are none! Rather, in the quantum
mechanics of closed systems, uncertainty relations like (3.1) are limitations on how a closed
system can be described. There are no histories in which position and momentum can be
simultaneously specified to accuracies better than allowed by Heisenbergs principle.
Although there are no projection operators that specify position and momentum simultaneously to accuracies better than the limitations of the uncertainty principle, we
can consider histories in which position is specified sharply at one time and momentum
}
at another time. Let { } be an exhaustive set of exclusive position intervals, {
be an exhaustive set of exclusive momentum intervals, and {P (t )} and {P (t)} be the
corresponding Heisenberg picture projection operators at times t and t respectively. An
at time t and the position
individual history in which the momentum lies in the interval
in interval at a later time t would correspond to a branch of the initial state vector of
the form:
P (t )P (t)|i .

(3.2)

As and range over all values, an exhaustive set of alternative histories of the closed
system is generated. Probabilities are assigned to these histories when the set decoheres,
that is, when the branches (3.2) are sufficiently orthogonal according to (2.9).
Nothing prevents us from considering the case when t coincides with t. If the alternative histories decohere, one would predict the joint probability p(, ) that the momentum
at one time and immediately afterwards the position is in the interval
is in the interval
. That would give a different meaning to the probability of a simultaneous specification
of position and momentum.
11

 } are infinitesimal, corresponding to a sharp specEven if the intervals { } and {

ification of position and momentum there are some states |i for which these alternatives
decohere. Eigenstates of momentum provide one example. However, for no state |i will
the marginal probability distributions of position and momentum have variances that violate the uncertainty principle. That is because decoherence implies the probability sum
rules so that
p()

p(, ) = kP (t)|ik2

(3.3a)

p(, ) = kP (t)|ik2

(3.3b)

and
p()

where, in each case, the last equality follows from decoherence. However, the left-hand
sides of (3.3) are just the usual probabilities for position and momentum computed from
a single state. Their variances must satisfy the uncertainty principle.
One would come closer to the classical meaning of simultaneously specifying the position and momentum if histories of coincident position and momentum projections decohered independently of their order. That is, if the set of histories
Pe (t)P (t)|i

(3.4)

were to decohere in addition to the set defined by (3.2) with t = t. In that case it is
straightforward to show that the joint probabilities p(, ) are independent of the order
of the projections as a consequence of decoherence. Whether states can be exhibited in
which both (3.2) and (3.4) decohere is a more difficult question.

IV. Limitations on Ideal Measurements

While measurement is not fundamental to a formulation of the quantum mechanics
of a closed system, measurement situations can be described within it. That is because
12

we can always consider a closed system consisting of measuring apparatus and measured
subsystem or most generally and accurately the entire universe. Roughly speaking, a
measurement situation is one in which a variable of the measured subsystem, perhaps
not normally decohering, becomes correlated with high probability with a variable of the
apparatus that decoheres because of its interactions with the rest of the universe. The
variable of the apparatus is called a record of the measurement outcome. The decoherence
of the alternative values of this record leads to the decoherence of the measured alternatives
because of their correlation. Measurement situations can be described quantitatively in
the quantum mechanics of closed systems by using the overlap (2.9) to determine when
measured alternatives decohere and using the resulting probabilities to assess the degree
of correlation between record and measured variable. By such means, any measurement
situation, ideal or otherwise, may be accurately handled in the quantum mechanics of the
closed system containing both measuring apparatus and measured subsystem.
Conventional discussions of measurement in quantum mechanics often focus on ideal
measurement models in which certain approximate features of realistic measurement situations are idealized as exact. In particular an ideal measurement is one that leaves a
subsystem that is initially in an eigenstate of a measured quantity in that same eigenstate after the measurement. The subsystem is thus disturbed as little as possible by
its interaction with the apparatus. Of course, not very many realistic measurements are
ideal in this sense. Typically, after a measurement, subsystem and apparatus are not even
in a product state for which it makes sense to talk about the state of the subsystem.
Probably the reason for the focus on ideal measurement models is that they are models of
the sorts of measurements for which the reduction of the state vector could accurately
model the evolution of the measured subsystem interacting with the measuring apparatus.
In particular, a reduction of the state of the apparatus will leave the subsystem in the

Some classic references are von Neumann (1932), London and Bauer (1939), and Wigner
(1963) or see almost any text on quantum mechanics.
13

correlated eigenstate of the measured variable.

Quantum mechanics severely restricts the possible ideal measurement situations. Wigner
(1952) and Araki and Yanase (1960) showed that, even given arbitrary latitude in the choice
of Hamiltonian describing the combined system of apparatus and measured subsystem, only
quantities that commute with additive, conserved quantities can be ideally measured. This
is a very restrictive conclusion. It rules out, for example, precise, ideal measurements of the
position and momentum of a particle with a realistic Hamiltonian. (They do not commute
with the additive, conserved angular momentum.) Araki and Yanase showed that, in a
certain sense, ideal measurements were approximately possible for such quantities, but in
a strict sense quantum mechanics prohibits them.
Impurity of the initial state of a closed system limits ideal measurements in another way.
To derive this limitation it is necessary to discuss more precisely ideal measurement models
in the quantum mechanics of closed systems. We will use the more of the formulation of
the quantum mechanics of closed systems than has been developed here. The reader can
find the necessary background in the authors other contribution to these volumes, (Hartle
1993a).
We consider a closed system in which we can identify alternatives of a subsystem
that are to be measured. Let {Sk k (tk )}, k = 1, 2, 3, be the Heisenberg projection
operators corresponding to these alternatives at a set of times {tk }, k = 1, 2, 3 . In a more
detailed ideal measurement model we might assume that the Hilbert space of the closed
system is a tensor product of a Hilbert space Hs defining the subsystem and a Hilbert space
Hr defining the rest of the universe outside the subsystem. In the Schrodinger picture,
projection operators representing the alternatives of the subsystem would have the form
Sk k = skk I r where the s s act on Hs alone. However, such specificity is not needed for
the result that we shall derive.

For more detail than can be offered here see Hartle (1991a), Section II.10.
14

Let us consider how a sequence of ideal measurements of alternatives of the subsystem

Sk k (tk ) at times t1 < < tn is described. A history of specific alternatives (1 , , n )
is represented by the corresponding chain of projections:
C = Snn (tn ) S1 1 (t1 ) .

(4.1)

One defining feature of an ideal measurement situation is that there should exist at a
time T > tn a record of the outcomes of the measurement that is exactly correlated with
the measured alternatives of the subsystem. That is , there should be a set of orthogonal,
commuting, projection operators {R (T )} with R (T ) Rn1 (T ) which are always
exactly correlated with the measured alternatives C in histories that contain them both,
as a consequence of the systems initial condition. The degree of correlation is defined by
the decoherence functional D( ; ) which measures the interference between a history
consisting of a sequence of measured alterhatives = (1 , , n ) followed by a record
= (1 , , n ) at time T and a similar history with alternatives and . If the records
are exactly correlated with the measured alternatives.


D , ; , = T r[R (T )C C R (T )]

(4.2)

where is the Heisenberg picture initial density matrix of the closed system of apparatus
and measured subsystem and means 1 1 2 2 n n , etc.
The existence of exactly correlated records as described by (4.2) ensures the decoherence of the histories of the subsystem and permits the prediction of their probabilities.
That is because the records are orthogonal and exhaustive:
R (T )R (T ) = R (T ),

R (T ) = I .

(4.3)

These properties together with the cyclic property of the trace are enough to show that
T r[C C ]
15

(4.4)

follows from (4.2). This is the generalization of the decoherence condition (2.9) for an
initial density matrix. The measurement correlation thus effects the decoherence of the
measured alternatives.
Of course, much more is usually demanded of an ideal measurement situation than
just decoherence of the measured alternatives. There is the idea that ideal measurements
disturb the measured subsystem as little as possible and, in particular, that values of
measured quantities are not disturbed. These are described in more detail in the context of
the quantum mechanics of closed systems in Hartle (1991a). For our discussion, however,
we need only the feature that ideal measurements assume exactly correlated records of
measurement outcomes, for we shall now show that if the density matrix is highly impure
such records cannot exist for non-trivial sets of measured histories.
We begin by introducing bases of complete sets of states in which the density matrix
and the commuting set of projection operators {R (T )} are diagonal, viz. :
=
R (T ) =

|rir hr| ,

(4.5)

|, nih, n| .

(4.6)

where r are the diagonal elements of . When diagonal elements of the condition (4.2)
(those with = , = ) are written out in terms of these bases they take the form
X

r,n

r |h, n|C |ri|2 .

(4.7)

The left-hand side is a sum of positive numbers so that this implies

hr|C |, ni = 0 , when 6= ,

(4.8)

for all r for which r 6= 0.

The argument we shall give is a straightforward extension of that used by M. Gell-Mann

and the author to analyze the possibility of strong decoherence in Gell-Mann and Hartle
(1993a). Thanks are due to M. Gell-Mann for permission to publish here what is essentially
a joint result.
16

If the density matrix is highly impure, so that r 6= 0 for a complete set of states
{|ri}, the relation (4.8) implies the operator condition
C |, ni = 0 ,

6= .

(4.9)

Therefore, C is non-zero only on the subspace defined by R (T ) where R (T ) is effectively

unity. Thus we have
R (T )C = C .
Summing this relation over C and utilizing the fact that

(4.10)
P

C = R (T ) .

= I, we find
(4.11)

which says that the string of projections is itself a projection. This can happen only if the
string consists of a single projection or if all the projections in the string commute with
each other. To see the latter fact write (4.11) in detail as
Snn (tn ) S1 1 (t1 ) = Rn 1 (T ) .

(4.12)

(4.13)

Summation implies
Sk k (tk ) =

Rn 1 (T )

j 6=k

but since the Rs commute with each other the Ss must also. Even in the case that C
consists of a single projection, (4.11) shows that record and projection are indentical. If
C consists of projections that refer to a subsystem defined by a Hilbert space as described
above then the records cannot be elsewhere in the universe. Thus, if the initial density
matrix is highly impure, in the sense that it has non-zero probabilities for a complete set of
states, there cannot be exactly correlated records of measurement outcomes. In particular
there cannot be ideal measurements.
Of course, in realistic measurement situations we do not expect to find records that
are exactly correlated with measured variables of a subsystem. Neither do we necessarily
17

expect exact decoherence of measured alternatives or many of the other idealizations of

the ideal measurement situation as very experimentalist knows! It, therefore, becomes an
interesting question to investigate quantitatively the connection between the {r } of the
density matrix and the degree to which approximate records defined by a relaxed (4.2)
exist.

V. Interfering Alternatives
The starting point for Feynmans sum-over-histories formulation of quantum mechanics
is the prescription of the amplitude for an elementary (completely fine-grained) history of
a measured subsystem as
exp[iS(history)/h]

(5.1)

where S is the action functional summarizing the subsystems dynamics. As an example,

we may think of a non-relativistic particle moving in one dimension. In this case the
elementary histories are the possible paths of the particle, x(t), and the action is the usual
"
#
 2
Z
dx
S[x( )] = dt 21 m
V (x) .
(5.2)
dt
We will use this example for all illustrative purposes in what follows.
A given experimental situation determines some parts of the subsystems path but
leaves undetermined many other parts. For instance, consider a measurement that determines whether or not a particle is in a position interval at time t. In that case the
measurement leaves undetermined the positions at times other than t and the relative position within at time t. Given an initial state at time t0 represented by a wave function
(x0 ), we may compute the probabilities for the outcomes that are determined by the
measurement as follows: We first divide the undetermined alternatives into interfering
and non-interfering (or exclusive) alternatives according to the experimental situation.
We sum amplitudes for histories weighted by the initial wave function over the interfering
18

alternatives, square that, and sum the square over the non-interfering alternatives. The
result is the probability for the measured determination. For example, in the case of the
measurement mentioned above that localized a particle to an interval at time t, the
probability of this outcome is:
Z
2




iS[x( )]/h
x e
(x0 ) .
p() =
dxf 
 xf


(5.3)

The path integral is over all paths in the time interval [t0 , t] that end in xf , and includes an
integral over the initial position x0 . These are the interfering alternatives. The square
of the amplitude is summed over the final position within . These positions are the
non-interfering alternatives.
What determines whether an undetermined alternative is interfering or not? Certainly
it is not whether it is measured in the experimental situation. In the above example,
positions at times other than t were not measured and they were interfering. But
the experiment also did not measure the relative position within and this was noninterfering. According to Feynman and Hibbs (1965):
It is not hard, with a little experience, to tell what kind of alternatives is involved.
For example, suppose that information about alternatives is available (or could be
made available without altering the result) [authors italics], but this information is
not used. Nevertheless, in this case a sum of probabilities (in the ordinary sense)
must be carried out over exclusive alternatives. These exclusive alternatives are
those which could have been separately identified by the information.
Thus, in the above example, the value of x at a time other than t is an interfering alternative
because we could not have acquired information about it without disturbing the later
probability that x is in at t. By contrast, the precise value of x within is a noninterfering or exclusive alternative because we could have measured it precisely and left
the probability for the particle to lie in undisturbed. Indeed, one way to determine
19

whether the particle is in is simply to measure the position at t precisely.

The author has always found this distinction between types of alternatives confusing.
He did not doubt Feynmans ability to tell what kind of alternative is involved, but
he was less sure of his own. This was especially the case since the distinction seemed to
involve analyzing, not only the particular experiment in question, but also many others
that might have been carried out. No precise rules for analyzing a given experimental
situation seemed to be available. This situation is considerably clarified in the quantum
mechanics of closed systems.
In the quantum mechanics of closed systems, we cannot have a fundamental distinction between interfering and non-interfering alternatives based on different types of
measurement situations, because alternatives are not necessarily associated with measurement situations. Whether alternatives interfere with one another, or do not, depends on
the boundary conditions and Hamiltonian that define the closed system. A quantitative
measure for the degree of interference is provided by the dechoherence functional. To
illustrate this idea, let us consider the single particle model we have been discussing on
the time interval [t0 , t]. The fine-grained histories are the particle paths on this interval.
Sets of alternatives correspond to partitions of these paths into an exhaustive set of exclusive classes {c }, = 1, 2, . The classes are coarse-grained alternatives for the closed
system. For example, one could partition the paths by which of an exhaustive set of position intervals they pass through at one time, which of a different set of position intervals
they pass through at another time, etc. There are many more general possibilities (see,
e.g. Hartle, 1991). The decoherence functional is a complex valued functional on pairs
of coarse-grained alternatives defined in a sum-over-histories formulation of the quantum
mechanics of a closed system by:
D( , )

=N

x
c




x f (xf , xf )exp i S[x ( )] S[x( )] /h i (x0 , x0 ).
20

(5.4)

The first sum is over paths x (t) in the class c and includes a sum over their initial
endpoints x0 and final endpoints xf . The sum over paths x(t) is similar. The normalization
factor is N = 1/T r(f i ) where the s are the operators whose matrix elements appear in
(5.4). We have written the decoherence functional for a general, time-neutral, formulation
of quantum mechanics* in which both an initial and a final condition enter symmetrically,
represented by density matrices i (x0 , x0 ) and f (xf , xf ) respectively. The final condition
which seems to best represent our universe and ensures causality is a final condition of
indifference with respect to final state in which the final density operator is f I.
The off-diagonal elements of the decoherence functional ( 6= ) are a measure of
the degree of interference between pairs of alternatives. When the interference is negligible
between all pairs in an exhaustive set, the diagonal elements ( = ) are the probabilities of the alternatives and obey the correct probability sum rules as a consequence
of the absence of interference. The orthogonality of the branches in (2.9) is an operator
transcription of this condition in the special case that f I.
The important point for a discussion of interfering and non-interfering alternatives is that all alternatives are potentially interfering in the quantum mechanics of closed
systems. For this reason amplitudes are summed over them in the construction of the
decoherence functional (5.4). Whether alternatives are interfering or not depends on the
measure of interference provided by (5.4), but in its construction all sets of alternatives
are treated the same. When interference between each pair is negligible the probabilities
for coarser-grained alternatives may be constructed either directly from (5.4) by summing amplitudes, or by summing the probabilities for the finer-grained alternatives in the
coarser-grained ones. The equivalence between the two is the content of decoherence.
Thus, there is no distinction between kinds of alternatives generally in the formalism,
* See, e.g. Aharonov, Bergmann and Lebovitz (1964) in the quantum mechanics of measured subsystems, and Griffiths (1984) and Gell-Mann and Hartle (1993) in the quantum
mechanics of closed systems.
21

but distinctions may emerge between different kinds of alternatives because of particular
properties of i and f . In particular, if f I any alternatives at the last time will
decohere. Thus, indifference with respect to final states is, in a time-neutral formulation of
quantum mechanics, the origin of the usual rule that final alternatives are non-interfering
rather than an analysis of whether one could have measured them but didnt.

VI. Conclusion
In the quantum mechanics of closed systems, projections act on states in the formula
for the probabilities of histories, but those reductions are not necessarily associated with
a measurement situation within the system and certainly not with one from without.
Uncertainty principles limit what kinds of alternatives a set of projections can describe,
but these limitations need not be of our ability to carry out a measurement. Interfering
alternatives can be distinguished from non-interfering ones, not by analyzing what might
have been measured, but by using the decoherence functional as a quantitative measure
of interference. Probabilities can be consistently assigned only to non-interfering sets of
alternative histories but decoherence as a consequence of a particular initial condition and
Hamiltonian rather than measurement decides which sets these are.
The fundamental role played by measurement in formulating a quantum mechanics
of subsystems is replaced by decoherence in the quantum mechanics of a closed system.
In the opinion of the author, the result is not only greater generality so that the theory
can be applied to cosmology, but also greater clarity. An important reason for this is the
disassociation of the notion of alternative from an ideal measurement. As we saw from the
work of Wigner, Araki and Yanase, and the argument of Section IV, ideal measurements
are almost impossible to realize exactly within quantum mechanics, and are therefore
of limited value as approximations to realistic measurement situations. But in the usual
quantum mechanics of measured subsystems, the second law of evolution is stated for ideal
22

measurements, not realistic ones. To discuss the evolution under realistic alternatives it
appears necessary to consider more and more of the universe beyond the subsystem of
interest until one obtains a subsystem large enough such that measurements of it may be
approximated as ideal. By contrast, the alternatives used in the quantum mechanics of
closed systems are general enough to describe realistic measurement situations. The theory
can provide quantitative estimates of their closeness to ideal and therefore to how closely
the quantum mechanics of measured subsystems approximates the more general quantum
mechanics of closed systems.
Thus, little havoc needs be made to achieve a quantum mechanics of closed systems. All
that is needed is a more general formulation in which decoherence rather than measurement
is fundamental, but in which most features of the approximate quantum mechanics of
subsystems that were tied to measurement reemerge in a more general and conceptually
clearer light.

Acknowledgment
The work described in this essay is an outgrowth of the authors program with M. GellMann to explore and clarify quantum mechanics. It is a pleasure to thank him for many
discussions that are reflected, in part, in the work presented here.

23

References
Aharonov, Y., Bergmann, P., and Lebovitz, J. (1964) Phys. Rev. B134, 1410.
Araki, H. and Yanase, M. (1960) Phys. Rev. 120, 622.
Brill, D. and Hartle, J.B. (1964) Phys. Rev. B 135, 271.
Busch, P. (1987) Found. Phys. 17, 905.
Caves, C. (1986) Phys. Rev. D 33, 1643.
(1987) Phys. Rev. D 35, 1815.
DeWitt, B. (1970) Physics Today 23, no. 9.
Everett, H. (1957) Rev. Mod. Phys. 29, 454.
Feynman, R.P. and Hibbs, A. (1965) Quantum Mechanics and Path Integrals, McGrawHill, New York.
Gell-Mann, M. and Hartle, J.B. (1990a) in Complexity, Entropy, and the Physics of
Information, SFI Studies in the Sciences of Complexity, Vol. VIII, ed. by W.
Zurek, Addison Wesley, Reading or in Proceedings of the 3rd International
Symposium on the Foundations of Quantum Mechanics in the Light of New
Technology ed. by S. Kobayashi, H. Ezawa, Y. Murayama, and S. Nomura,
Physical Society of Japan, Tokyo.
Gell-Mann, M. and Hartle, J.B. (1990b) in the Proceedings of the 25th International
Conference on High Energy Physics, Singapore, August, 2-8, 1990, ed. by
K.K. Phua and Y. Yamaguchi (South East Asia Theoretical Physics Association and Physical Society of Japan) distributed by World Scientific, Singapore.
Gell-Mann, M. and Hartle, J.B. (1993a) Classical Equations for Quantum Systems
(to be published in Phys. Rev. D).
Gell-Mann, M. and Hartle, J.B. (1993b) in Proceedings of the 1st International A. D.
Sakharov Conference on Physics, USSR, May 27-31, 1991 and in Proceedings of
the NATO Workshop on the Physical Origins of Time Assymmetry, Mazagon,
Spain, September 30-October4, 1991 ed. by J. Halliwell, J. Perez-Mercader,
and W. Zurek, Cambridge University Press, Cambridge.
Griffiths, R. (1984) J. Stat. Phys. 36, 219.
Groenewold, H.J. (1952) Proc. Akad. van Wetenschappen, Amsterdam, Ser. B, 55,
219.
Hartle, J.B. (1991a) The Quantum Mechanics of Cosmology, in Quantum Cosmology
and Baby Universes: Proceedings of the 1989 Jerusalem Winter School for Theoretical Physics, ed. by S. Coleman, J.B. Hartle, T. Piran, and S. Weinberg,
World Scientific, Singapore, pp. 65-157.
Hartle, J.B. (1991b) Phys. Rev. D 44, 3173.
Hartle, J.B. (1993a) The Quantum Mechanics of Closed systems in the Festschrift for
C.W. Misner, ed. by B.-L. Hu, M.P. Ryan, and C.V. Vishveshwara, Cambridge
University Press, Cambridge.
24

Hartle, J.B. (1993b) Spacetime Quantum Mechanics and the Quantum Mechanics of
Spacetime in Proceedings of the 1992 Les Houches Summer School Gravitation
and Quantizations, ed. by B. Julia, North Holland, Amsterdam.
Isaacson, R. (1968a) Phys. Rev. 166, 1263.
(1968b) Phys. Rev. 166, 1272.
London, F. and Bauer, E. (1939) La theorie de lobservation en mecanique quantique,
Hermann, Paris.
Margenau, H. (1958) Phil. of Sci. 25, 23.
Omn`es, R. (1988a) J. Stat. Phys. 53, 893.
(1988b) J. Stat. Phys. 53, 933.
(1988c) J. Stat. Phys. 53, 957.
Page, D. and Wootters, W. (1983) Phys. Rev. D 27, 2885.
Prugovcki, E. (1967) Can. J. Phys. 45, 2173.
Stachel, J. (1986) in From Quarks to Quasars, ed. by R.G. Colodny, University of
Pittsburg Press, Pittsburgh, p. 331ff.
von Neumann, J. (1955) Mathematische Grundlagen der Quantenmechanik, J. Springer,
Berlin (1932). [English trans. Mathematical Foundations of Quantum Mechanics, Princeton University Press, Princeton].
Wheeler, J.A. (1964) in Relativity Groups and Topology, ed. by B. DeWitt and
C. DeWitt, Gordon and Breach, New York.
Wigner, E.P. (1952) Zeit. f. Phys. 131, 101.
(1963) Am. J. Phys. 31, 6.

25