PHYSICAL REVIEW D

VOLUME 27, NUMBER 12

15 JUNE 1983

Evolution without evolution: Dynamics described by stationary observables

Don N. Page* and William

K. Wootters

Center for Theoretical Physics, The Uniuersity of Texas at Austin, Austin, Texas 78712 (Received 17 June 1982)

Because the time parameter in the Schrodinger equation is not observable, energy apparently obeys a superselection rule in the same sense that charge does. That is, observabies must all commute with the Hamiltonian and hence be stationary. This means that it is consistent with all observations to assume that any closed system such as the Universe is in a stationary state. We show how the observed dynamic evolution of a system can be described entirely in terms of stationary observables as a dependence upon internal clock readings.

I. UNOBSERVABILITY OF COORDINATE TIME

It is widely believed that there is a charge superselection rule, ' so that superpositions of states of Indeed it has different charge are unobservable. field been proved in local relativistic quantum theory that the long-range Coulomb field causes the total charge operator to commute with all quasilocal Because energy is coupled to a longobservables. range gravitational field, there should be an analogous superselection rule for it. This would say that only operators that commute with the Hamiltonian can be observables. But such operators are stationary, so how is it that we observe time dependence in the world? We shall argue that the temporal behavior we observe is actually a dependence on some internal clock time, not on an external coordinate time. It is perfectly consistent with our observations to assume that any closed system is in an eigenstate of energy and thus stationary with respect to coordinate time, since coordinate time translations are unobservable. Such a state can be decomposed into states of definite clock time. The dependence of these component states upon the clock time labeling them can then represent the observed temporal behavior of the system. A formula for this dependence in terms of conditional probabilities will be given below. The circumvention of the energy superselection rule is precisely analogous to what Aharonov and Susskind, Mirman, and Lubkin have shown about the charge' and univalence'*' superselection rules. These authors demonstrate that any additively conserved quantity may be considered to have a superselection rule in the sense that all subsystems at all time have density matrices which commute with the conserved quantity, if the density matrix of the corn27

with that quantity. commutes posite system Nevertheless, if the conserved quantity is coherently shared between subsystems, the state of one subsystem relative to a particular state of another can show interference between its eigenstates of the conserved quantity. Indeed, the univalence superselection rule has been experimentally circumvented by the observation of a relative rotation of 2~ for one half of a split beam of neutrons. Our analysis applies to any closed system that may be completely described quantum mechanically Since a closed and has a well-defined Hamiltonian. system by definition interacts only with itself, any observations must be done entirely within the system, so the system must include all its observers. By and elecvirtue of the long-range gravitational tromagnetic interactions, the smallest closed system we observe appears to be astronomically large and is (Of course we generally known as the Universe. cannot rule out other quantum systems which do not interact with ours. ) In order that our Universe will have a well-defined Hamiltonian, we assume that spacetime is asymptotically flat in some apThis is inconsistent with the propriate sense. ' cosmological principle of large-scale homogeneity and isotropy if the average energy density is nonzero, but it is not inconsistent with observations: one can imagine that the energy density drops off beyond the maximum distance we can now see, so that the total energy of the Universe is finite. We do not claim that this model is necessarily correct but only that it is consistent. Alternatively, one may ignore long-range interacclosed systems tions and consider hypothetical within a single spacetime universe. However, then one must restrict attention to the interactions within each system and not invoke the neglected interac-

"'

'

2885

1983

The American Physical Society

2886

DON N. PAGE AND WILLIAM K. WOOTTERS

27

tions between them in discussing the observed temporal behavior of a closed system. That is, an observer cannot be allowed to read clocks outside his own closed system. Although a rigorous demonstration is lacking, one would expect the Strocchi-Wightman proof of the electrocharge supers election rule in quantum dynamics to apply to energy in quantum gravity. That is, one would expect the energy, as determined by the gravitational field at spacelike infinity, to commute with all quasilocal observables, which depend upon fields interior to infinity. But even if we ignore gravity and consider quantum field theory in Minkowski spacetime, the unobservability of a global time translation of the entire Universe would seem to imply that energy obeys a superselection rule. (This is not a trivial consequence of the Wigner-Araki- Yanase theorem, ' ' because we are measurenot assuming precise von Neumann-type ments. ) In other words, a shift in the time coordinate t, which changes the relative phase between two different energy eigenstates, is an unobservable coordinate transformation. Thus only operators which have no explicit or implicit dependence on t can be observables. These operators commute with the
I

Hamiltonian and are stationary, so how can they show any dynamical behavior? In other words, how can the energy superselection rule be reconciled with the fact that we do observe interference effects between states of different energy whenever we see anything change?

II. OBSERVABILITY OF CLOCK TIME

BY STATIONARY OPERATORS

To illustrate how we see change through the use of observables which are necessarily stationary, consider a system of n spin-j particles at fixed positions
but precessing around the z axis with frequency co. For simplicity, we assume there is a surrounding laboratory with respect to which relative positions, orientations, and velocities are defined; such a reference system is necessary since the momentum and angular momentum superselection rules prevent the observation of any absolute positions, orientations, or velocities. However, we assume that in the relative laboratory frame the laboratory is stationary, so that all the time dependence is given by the particle precession; i.e., there is no clock built into the laboratory. The Hamiltonian is taken to be

a=ricoz,

m&

$ j
&

$ j (m, +. . . +m)ficolm, &. m=

J
I

. m&(m,

. &ml,

where m; labels the z component of the angular momentum of the ith particle. That is,

J lm&=elm

Suppose we consider the product state in which in the x each particle has angular momentum direction at t=0. The transformation' to eigenstates of J, allows us to write this as

'j

le&=II
n

e;&

j =II2 ' .X m, = j
n

2j

1/2

lm;&,

m!(n

m)!

'

J;(t=0)lp;(t=0)&=jig, (t=0)&.
This state is nonstationary
l

with respect to t, since

(pl J;+iJ~ g&=je+-'"',

but this t dependence cannot be observed. There is no way to select a particular t at which a measurement of J~ or Jiy is to be performed, so these non-

stationary operators are not observables. The time average of a nonstationary operator is certainly stationary and may therefore be observable, but for J; and J;~ the averages are zero. The time dependence that is actually observable is the dependence of the dynamical variables upon each other, in particular, upon any variable that represents a clock reading. The measurement of time by quantum clocks has been discussed by Peres. Here we wish to emphasize how the dependence of a system upon a clock reading is determined entirely by stationary observables. Consider how the precessing orientation of particle 2 depends upon the orientation of particle 1, which is viewed as a clock. For example, we may ask what value J2 has when particle 1 reads 12 o' clock, by which we mean that J& is measured to in the x direction. In the give angular momentum state (3) this is certain to occur if t=O, but it can occur with lower probability at almost any other t, so t is not precisely determined by the clock reading. If the laboratory apparatus is triggered to measure J2 when J& gives j, the measurement does not necessarily occur at t=O, so J2(t=O) cannot be measured. Instead, what is measurable is the value of J2, given a particular value of J&. The value

27

EVOLUTION WITHOUT EVOLUTION: DYNAMICS DESCRIBED. . .

2887

measured is governed by the conditional probabilities of the various eigenstates of Jqwith respect to a particular eigenstate of Ji. The relative frequency distribution of values measured on an ensemble of identical systems gives a statistical estimate of these conditional probabilities, so we say these are observable. (To determine the conditional probabilities precisely, one would need an infinite ensemble. This is an idealization not realized in nature, so no predictions of quantum mechanics can ever be completely verified by quantum-mechanical observers within the Universe, for whom the theory can only make
I

statistical predictions. One can only verify that the predictions are consistent with observations to some confidence level. Since the theory does predict that all but a relatively small set of conditional probabilities are very small, its consistency with observation is nontrivial. It is only in this consistency sense that the conditional probabilities we shall discuss are observable. ) To determine the quantum-mechanical predictions for the conditional probabilities of J2when J1has the value j, we need the projection operator onto the subspace of states with J and the one and J2 onto the subspace with J1 k:

j j

=2
n
I I

P, ,
where, '

= J,=J'&&J,=J Ig J,=j k k)&J,=J

I

(8)
l

'

=3

in the Heisenberg picture,

J J,=j)=2-i X

2J

1/2

im

cot

Imi),
1/2
k

2J
m2

+ 1/2

2J

k 2j

J
.

j+m2

1=0

i( 1)'e l 2

' Imp),

I;=
m

= J

~m;)(m;~

These projection operators depend upon the undetermined t and hence are not observables, but their time averages, Pz and Pzz , may be (though we do not claim to have found what the sufficient conditions are for an operator to be an observable):

The conditional
is then

probability

for

J2 j given J1

Pg(J2

=j Ji. =j)=2
I

1)!! (8j 1)"1' I:(4j

2J

Pj

2'i
m&

2J .

Im, )(m,

e gI;,
f

=2

(4l 2) j

~2
(15)

(12)
and P& J k is a more complicated stationary observable which shows correlations between the two particles (i.e. , the time averaging breaks the product structure of Pz z ). Then the conditional probabiliis k given J1 ty for J2

so there is always at least a 70% probability that the second particle will have all its spin in the x direction if the first one does. Indeed, for large

j k Jix=j)= ( P(Jzj P~~ )I(P~) .

I

P&(J

k =j

I,=j)=2 '
j.

"
I

2k

k,

(16)

(13)

In the product state (3), the conditional ty (13) turns out to be

probabili-

which is very strongly concentrated around k=O. Another value of interest is the conditional expectation value of J2given J1, This is

k Py(J~. =J
4j

Jim

=j )
Sj Sj
2k

4j 2j

Ey(Jz

Ji. =j)=&0 Jz.~g

I

S&~&PI P, 0&
I

2j
k

4j
k
(14)
which

4j

2j+1
again shows the strong agreement

(17)
between

2888

DON N. PAGE AND WILLIAM K. WOOTTERS

the two spins for large One may also consider conditional probabilities in any given that particle 1 has angular momentum other direction. The agreement between the particle angular momenta in any direction in the x-y plane in which the spins are all precessing will be as good as it is in the x direction. If one considers particle 1 to be a clock, so that different directions of its spin correspond to different clock readings, then the dependence of the angular momenta of the other particles on the clock readings constitutes the observable time behavior of the system. A sequence of clock readings along with the corresponding states of the other particles forms an evolution of the system, with no reference to the dependence on the unobservable coordinate time t. As a technical point, since angular momenta in different directions do not commute, making more than one clock reading on the same particle will alter its precession and make the dependence of the other particles on it more complicated. Therefore it may be preferable to form a clock out of many particles in identical spin states and make each reading on a different one. Then the projection operators for all clock readings will commute, so the dependence of the rest of the system on each reading will be the same as if only that one reading were taken. Analogously, checking the clock-time dependence of the rest of the system without destroying what is to be checked could be done by making many copies and measuring a different copy at each different clock reading. Of course, this procedure requires an ability to make copies, a discussion of which would take us too far afield for this paper. Since the observable conditional probabilities deoperators (the timepend only upon stationary averaged projection operators), they are not at all sensitive to terms in the density operator of the system that connect different energy eigenstates. That is, the results are the same for all density operators that are identical in their matrix elements between states of equal definite energy. In particular, the

j.

probabilities are the same for the corresponding block-diagonal density matrix in which all terms connecting states of different energy are set equal to zero. The expectation values of all operators have no dependence on t for this density matrix, so it represents a stationary state. As a result, all of the observable clock-time dependence of any state, stationary or not, is precisely the same as that of the statistical state. Thus stationary corresponding there is no way to tell whether or not the Universe is in a stationary state. In general, the stationary state that gives the correct conditional probabilities will be a mixed state, with nonzero probabilities for different energies. The presence of more than one energy in a system would be detectable to an observer who had access to a large number of copies of the system so that he could determine the conditional probabilities. But if the system is closed and therefore contains the observer, he will not be able to determine the precise conditional probabilities. Since the conditional probabilities for his measurement results are an average over the energy eigenstates, weighted by the diagonal elements of the density matrix, his observations are always consistent with at least one energy eigenstate. Indeed, for the purpose of describing the observed dynamics inside the closed system there is no loss of generality in assuming the system is in a pure state. Thus the Universe may not only be stationary, it may be in an energy eigenstate. The two-particle conditional probabilities given above for the nonstationary state (3) would be the same in any state that agrees with the state (3) in its probability distribution for m =m&+m2 and has the same density matrix over m~ and m2 within each two-particle energy eigenspace of fixed m. The corresponding two-particle stationary density matrix which gives the same conditional probabilities is
p~~q~

g 2j P(m) ~2j, m)(2j, m m=

m2

2j

~,

(18)

a mixed state composed of pure states

2J
1/2
m&

'

1/2

'

1/2

m~ j+m
I

m m&

(19)

weighted with probabilities

P(m)

=2

4j

2j+m

result of a complete measurement system in this state would be consistent with at least one of the eigenstates (19) of the density matrix (18), so a measurement within the

The individual

of the two-particle

system could not distinguish the mixed state from an energy eigenstate of the system. But even if one contemplated making measurements on an ensemble of identical systems and wished to maintain the precise values of the conditional probabilities, one could do this by combining each two-particle system with another system having the reversed energy spectrum. One simply takes the pure energy eigenstate of the composite system which has the appropriate proba-

27

EVOLUTION WITHOUT EVOLUTION: DYNAMICS DESCRIBED. . .

gy spectrum

2889

bility distribution for the energy of the two-particle subsystem (whose energy is now precisely anticorrelated with the energy of the other subsystem, so that each subsystem is in a mixed state). Since the enerI

of the two-particle subsystem is symmetric about zero, it may simply be combined with another two-particle subsystem. Suppose the latter has spin-2j states
1/2

~2j,

')=

4j

1/2

J+m

X
' 1/2

2J

1/2

m3 J +m
I

. 2J

~m, ) ~m4

m' m, )

(21)

Then the pure zero-energy eigenstate

2J'

4j

12jm)12jm =

(22)

unobservable t needed to make P, and PQP, (which is not, in general, the same as PQP, ) stationary in order that they might be observable, and p is the density operator representing the state of the system. If A is a projection operator onto some eigenspace then labeled by a, 3

=P,

has the first two-particle subsystem in the stationary mixed state (18), so for measurements confined to that subsystem the conditional probabilities are exactly the same as they are for the nonstationary state (3). This example illustrates that even if one knows that a system is not in an energy eigenstate, it may be part of a larger system which is. It might be objected that if one knows that a system is in a pure Hilbert-space superposition of different energy eigenstates, it cannot possibly be part of a larger system which is in an energy eigenstate. This is because the joint state must be a product state in order that the first system be in a pure state, and such a product state containing a superposition of different energies cannot be an energy eigenstate. (Compare Refs. 9 and 5, which show that if an additive quantity commutes with the density matrix of a joint system, it also commutes with the density matrix of each subsystem obtained by Landau tracing. ) The rebuttal is that one cannot know that any system is in a pure superposition state of different energies, since the t dependence that would show this is What one often naively completely unobservable. calls an observable t dependence is actually a dependence on a clock reading, and we have argued that this can always be expressed in terms of stationary operators which cannot distinguish energy superpositions from mixtures. Although the mathematical details might be more cumbersome, we could apply our analysis to any other system that contains some dynamical variable that could serve as a clock. If the projection operathen the dependence tor onto a clock reading ~ is upon ~ of some other dynamical variable 3 is given by its conditional expectation value

E(P /r)=P(a/r),

(24)

the conditional probability of the eigenspace a given the clock reading ~. Note that the conditional expectation values would be the same if p were replaced by p, the time average of e ' 'pe' ' (rather than of p itself, which is constant in the Heisenberg picture we are using). This p is the stationary state corresponding to p, with no nonzero terms connecting states of different energy. Thus the observable conditional expectation values give no way to tell whether or not a system is in a stationary state.

III. EVOLUTION IN

A STATIONARY UNIVERSE

E(A

~r)=(PQP, )/(P, )

= tr(PQP, p) /tr(P, p),

where the overbar

(23)
over the

denotes the averaging

If there is no observable difference between a stationary state of the Universe and a nonstationary state, is it really necessary to have a law of evolution? In the above example the spin direction of each particle precesses with angular frequency co according to clock time, clock time being defined by the state of a particular particle which has been chosen as the clock. Thus the clock-time evolution is the same as what one would normally predict by applying the equation of motion to the spin of each particle. But as we have shown, the state of the whole system, including the clock, could with no observable differences be chosen to be a stationary state, and for such a state the evolution is trivial. In this stationary state, therefore, the clock-time evolution of the system is not being dictated by the usual law of evolution, but rather is determined by correlations between the clock and the rest of the system. This raises the question whether this sort of evolution (evolution by correlations) always imitates evolution as obtained from the equation of motion. To answer this question let us consider a (not necessarily pure) stationary state p of the closed sys-

2890

DON N. PAGE AND WILLIAM K. WOOTTERS

27

consists of two parts: a clock with Hamiltonian H and the rest, which has Hamiltonian H, . We assume the clock does not interact dynamically with the rest of the system, so that H =HcI+&cH, I, and Ibeing the identity operators in their respective spaces. We now define clock time as follows. Choose a special state l(, (0) ) of the clock to mark the zero of clock time (an optimal choice for this state may be very difficult, but the optimization is inessential for our discussion), and then with each of the states
~

tern (Q, H]=0), and assume that the closed system

'

~g, (r)) ~g, (0))

associate the value v of clock time. Other states of the clock are assumed not to be associated with any definite value of r. For any stationary observable A =IcAof the rest of the system, the conditional expectation value of A given that the clock reads r is [cf. Eq. (23) with A and P, stationary and commuting]

E(A r) =tr(AP, p)/tr(P,

~

p),

(25)

where

We wish to show that this expectation value evolves in clock time in accordance with the Heisenberg equation of motion for the rest of the system. Using the fact that [H, p] =0, we have

' pe i(H, sl, )c = trjAPoe i(H,' si)~ ] i(I' sH)r i(l sH)r] " = trjAPoe pe
&

quires that the density matrix of the entire system commute with the Hamiltonian, then it follows automatically that expectation values can be computed from operators obeying the Heisenberg equations of motion with respect to clock time. Examples of dynamical variables that can serve as clock times are the size or expansion rate of the observed part of the Universe, the cosmic radiation density, the relative abundances of radioactive elements, the orientation of the Earth-Sun combination or of the Earth itself relative to distant stars, the vertical angle of a pendulum, the orientation of the hands of a watch, etc. In each case one may consider the conditional expectation values of other variables with respect to these. These realistic clocks do interact with the rest of the Universe so the dependence on clock time will not be so simple as in the noninteracting case analyzed above. Indeed, such an interaction is necessary if the clock is to be read. The determination of a very precise time resolution entails a large energy exchange, so if the total energy available is finite, there is a limit to the precision of any clock. ' The theoretical limit is usually so short that it is ignored. For a good clock the interaction can thus have a practically negligible effect. One may idealize the behavior of actual clocks to form the abstract notion of a preferred coordinate time. ("Time is defined so that motion looks simple. ) However, this abstraction, useful though it is, is not directly observable. Only the dependence on an internal clock time can be observable. No one can tell whether something happens at t=12, but one can tell whether it happens at 12 o' clock.

"'

IV. DISCUSSION
We have shown that although the dependence of a system upon coordinate time is completely unobservable, one can observe a dependence upon an internal clock time. Such a description of dynamics is given entirely in terms of stationary operators, which indeed are the only operators that can be observables. We now consider briefly a number of consequences and extensions of this idea. The observed arrow of time described by the second law of thermodynamics must be formulated in terms of the dependence of some observable representing entropy upon certain clock times. It is completely unobservable whether or not entropy increases with coordinate time. There is the additional question of what observables can be used to define entropy. Most functionals of the density operator are not observable, even if they are invariant under such as those representing unitary transformations time translations. For example, the usual quantummechanical formula for the entropy of a system,
I

( Ic

8 Hr

&A

( Ic 3Hr ) rp
&

In the special case where A is the identity operator, this equation tells us that tr(P, p)=tr(Pop), a constant. From these results and from Eq. (25) it follows finally that

E (A r) = tr[A(r)p, ],
f

(27)

where Ar(r) is the Heisenberg-evolved

operator
(28)

and p is the relative density matrix

p=tr(P()p) ltr(Pop) .

(29)

Thus it is not necessary to assume, a priori, any equation of motion for the rest of the closed system with respect to coordinate time. If one instead re-

27

EVOLUTION WITHOUT EVOLUTION: DYNAMICS DESCRIBED. . .

2891

tr(plnp}, is an invariant but depends upon the off-diagonal elements connecting eigenstates of different charge or energy, which are completely unobservable. However, one should be able to formulate an entropy based on coarse-grained properties that one actually observes. Such a quantity need not be zero even for pure states. ' In an asymptotically flat spacetime, our analysis for energy would apply equally well for momentum and angular momentum. There should be superselection rules for all of these quantities, either because they are determined by the asymptotic gravitational field which should commute with all quasilocal observables, or because asymptotically defined Poincare transformations are unobservable: there is no way to determine the absolute position, orientation, or velocity of any system. We thus expect only Poincare-invariant operators to be observable. Of course, not all components of the fourmomentum P& and the angular momentum tensor J&commute, so the Universe cannot be in an eigenstate of all of them. However, it would be consistent with observations to assume the Universe is in an eigenstate of any combination that do commute, though which combination is chosen is completely arbitrary. For example, one might assume for simplicity that the Universe has a definite energy, zero spatial momentum, zero spatial angular momentum J;J, but completely undetermined time-space angular momentum components Jok (which represent the energy multiplied by the position of the center of mass choice would be to assume all J at t=0} Another . are zero but that only the magnitude PPI' of the four-momentum is definite. Then the Universe would have a definite rest mass and zero spin, and its center of mass would pass through the coordinate origin at t=O with a completely undetermined velocity. No observations could contradict either assignment or any other, but no observations could confirm any choice either. If one goes to curvilinear coordinates, only gen-

erally covariant operators can be observable. In flat spacetime the Minkowski metric reduces the symmetry to the Poincare group, but in quantum gravity in which the metric is treated as a dynamical variable, the spacetime intervals that in flat spactime are invariant under the Poincare group would have to be defined relative to intrinsic geometrical variables. For example, the distribution of fields over a spatial hypersurface cannot be uniquely specified by some functional dependence on an arbitrary choice of coordinates but can only be unambiguously defined relative to some three-geometry of the surface which gives invariant distance relations between the points. Many of the ideas of this paper have been expressed before, particularly in the context of quantum gravity where after much initial confusion it has now long been recognized that observable time can only be defined intrinsically. ' However, it has not generally been recognized that the same concepts also apply in quantum field theory in flat or even in nonrelativistic spacetime quantum mechanics. For example, the axiomatic approach to quantum field theory in terms of the algebra of operators in definite regions of Minkowski space o should be reformulated in terms of Poincareinvariant observables. No operator that depends explicitly or implicitly upon coordinate time is observable.

ACKNOWLEDGMENTS

The ideas of this paper were stimulated and refined by discussions with Ron Dickman, Ted Jacobson, Dilip Kondepudi, Claudio Teitelboim, and John A. Wheeler. D.N. P. is grateful for the hospitality of Professor Wheeler at the Center for Theoretical Physics and the receipt of an Alfred P. Sloan Research Fellowship. This work was supported in part by NSF Grant No. PHY-7826592. We are indebted to Asher Peres for his detailed critique of the original manuscript and many useful suggestions.

'Permanent address: Department of Physics, Pennsylvania State University, University Park, Pennsylvania 16802. Present address: Department of Physics and &stronoMassachusetts my, Williams College, Williamstown,

01267.

G. C. Wick, A. S. Wightman, and E. P. Wigner, Phys.

J. M.

Rev. 88, 101 (1952). Jauch and B. Misra, Helv. Phys. Acta 34, 699

(1961).

3R. Haag, Ann. Phys. (Leipzig) 11, 29 (1963). 4R. F. Streater and A. S. Wightman, PCT, Spin and Statistics, and All That (Benjamin, New York, 1964). 5G. C. Wick, A. S. Wightman, and E. P. Wigner, Phys. Rev. D 1, 3267 (1970). F. Strocchi and A. S. Wightman, J. Math. Phys. 15, 2198 (1974). 7Y. Aharonov and L. Susskind, Phys. Rev. 155, 1428 {1967);158, 1237 (1967). R. Mirman, Phys. Rev. 186, 1380 (1969); Phys. Rev. D

2892

DON N. PAGE AND WILLIAM K. WOOTTERS

27

E.

1, 3349 (1970). Lubkin, Ann. Phys. (N. Y.) 56, 69 (1970); Bull. Am. Phys. Soc. 16, 67 (1971). G. C. Hegerfeldt, K. Kraus, and E. P. Wigner, J. Math. Phys. 9, 2029 (1968). 'H. Rauch, A. Zeilinger, G. Badurek, A. Wilfing, W. Bauspiess, and U. Bonse, Phys. Lett. 54A, 425 (1975). S. A. Werner, R. Colella, A. W. Overhauser, and C. F. Eagen, Phys. Rev. Lett. 35, 1053 (1975). J. Weber and J. A. Wheeler, Rev. Mod. Phys. 29, 509

and J. A. Wheeler, Phys. Rev. 126, 1864 (1962). R. Arnowitt, S. Deser, and C. W. Misner, in Gravitation: An Introduction to Current Research, edited by L. Witten (Wiley, New York, 1962), pp. 227 265 J. A. Wheeler, in Battelle Rencontres: 1967 Lectures in Mathematics and Physics, edited by C. M. DeWitt and J. A. Wheeler (Benjamin, New York, 1968), pp.
~

R. F. Baierlein, D. H. Sharp,

242 307.

(1957).

25A. Peres, Phys. Rev. 171, 1335 (1968). C. W. Misner, Phys. Rev. Lett. 22, 1071 (1969); Phys.

C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation (Freeman, San Franciso, 1973). E. P. Wigner, Z. Phys. 133, 101 (1952). H. Araki and M. M. Yanese, Phys. Rev.

Rev. 186, 1319 (1969); 186, 1328 (1969).

120, 622 (1960). ' M. M. Yanese, Phys. Rev. 123, 666 (1961). ' A. R. Edmonds, Angular Momentum in Quantum Mechanics (Princeton University, Princeton, New Jersey, 1957). ' M. E. Rose, Elementary Theory of Angular Momentum (Wiley, New York, 1957). A. Peres, Am. J. Phys. 48, 552 (1980). B. S. DeWitt, Phys. Rev. 160, 1113 (1967).

Phys. Rev. Lett. 26, 1656 (1971); 28, 1082 (1972). J. A. Wheeler, in Problems in the Foundations of Phys
ics, proceedings of the, International School of Physics "Enrico Fermi, Course 72, edited by G. Toraldo di Francia (North-Holland, Amsterdam, 1979), pp. 395 497. R. Haag and D. Kastler, J. Math. Phys. 5, 848 (1964).

J. W. York,

K. Kuchar", J. Math. Phys. 11, 3322 (1970); Phys. Rev. D 4, 955 (1971); in Quantum Gravity 2: A Second Oxford Symposium, edited by C. J. Isham, R. Penrose, and D. W. Sciama (Clarendon, Oxford, 1981), p. 329 376.

"