UCSBTH-94-41

LA-UR-94-2101
CGPG-94/10-1
grqc/9410006

Conservation Laws
in the Quantum Mechanics of Closed Systems
James B. Hartle
Department of Physics, University of California
Santa Barbara, CA 93106
Theoretical Astrophysics, T-6, MSB288, Los Alamos National Laboratory
Los Alamos, NM 87545
Isaac Newton Institute for the Mathematical Sciences, University of Cambridge
20 Clarkson Road, Cambridge, U.K. CB3 0EH
Raymond La ammey
Theoretical Astrophysics, T-6, MSB288, Los Alamos National Laboratory
Los Alamos, NM 87545
Isaac Newton Institute for the Mathematical Sciences, University of Cambridge
20 Clarkson Road, Cambridge, U.K. CB3 0EH
Donald Marolf z
Center for Gravitational Physics and Geometry, Physics Department,
The Pennsylvania State University, University Park, Pennsylvania 16802

Abstract
We investigate conservation laws in the quantum mechanics of closed sys-
tems. We review an argument showing that exact decoherence implies the exact
conservation of quantities that commute with the Hamiltonian including the to-
tal energy and total electric charge. However, we also show that decoherence
severely limits the alternatives which can be included in sets of histories which
 [email protected]
y [email protected]
z [email protected]

1
 assess the conservation of these quantities when they are not coupled to a long-
range eld arising from a fundamental symmetry principle. We then examine
the realistic cases of electric charge coupled to the electromagnetic eld and
mass coupled to spacetime curvature and show that when alternative values of
charge and mass decohere, they always decohere exactly and are exactly con-
served as a consequence of their couplings to long-range elds. Further, while
decohering histories that describe uctuations in total charge and mass are also
subject to the limitations mentioned above, we show that these do not, in fact,
restrict physical alternatives and are therefore not really limitations at all.

PACS number(s):03.65.Bz, 03.65.Ca, 03.65.Db, 98.80.Bp.

2
I Introduction.
Energy is conserved during the unitary evolution
j (t)i = e iht=h j (0)i (1:1)
of a quantum state of an isolated subsystem of the universe because the Hamiltonian, h,
of that subsystem commutes with the unitary time evolution operator. However, energy
is not generally conserved by the \second law" of quantum evolution that describes how
the state of a subsystem evolves when an \ideal" measurement of it is carried out. If j i
is the state before an ideal measurement, the state afterwards is \reduced" to

j i ! kss jj ik
i (1:2)


where s is a Schrodinger-picture projection operator corresponding to the measurement

outcome { one of a set of such operators fs g;  = 1; 2; : : : describing dierent possible
outcomes. Even if j i is an energy eigenstate, the reduced state vector will generally not
be an energy eigenstate except in the special case that s commutes with the Hamiltonian
of the subsystem h.
More generally, if a sequence of measurements is carried out on the subsystem at
times t1 ; :::; tn, with outcomes represented by Heisenberg picture projection operators
fskk (tk )g; k = 1; :::; n, the joint probability of a particular sequence of outcomes  
(1 ; :::; n) is
p(1; : : : ; n) = ksnn (tn) : : : s11 (t1)j ik2 : (1:3)
Energy is conserved if the joint probability vanishes whenever measurements of the energy
at two times disagree. However, if measurements of quantities that do not commute with
h intervene between the two determinations of the energy, then that joint probability will
not be zero. More precisely, if fshl (tl )g and fshm (tm )g project to the same set of ranges of
h-eigenvalues, the joint probability has nonzero entries for l 6= m when the intervening
projections do not commute with h. Energy is thus not necessarily conserved in a sequence
of measurements.
The conservation of energy by the unitary evolution and the general non-conservation
by the reduction of the state vector are not surprising. The unitary law (1.1) describes
the evolution of a subsystem in isolation. The reduction law (1.2) describes the evolution
of a subsystem interacting with another system which is measuring it. Energy may be
exchanged between the measuring apparatus and the measured subsystem.
3
 The familiar \Copenhagen" quantum mechanics of measured subsystems sketched
above is an approximation to a more general quantum mechanics of closed systems [1,2,3],
most generally the universe as a whole. It is an approximation that is appropriate when
certain approximate features of realistic measurement situations may be idealized as exact .
The most general predictions of quantum mechanics are the probabilities for individual
members of sets of alternative histories of a closed system. One kind of alternative history
set may be specied by giving exhaustive sets of exclusive alternatives at a sequence of
times t1; : : : ; tn. The alternatives at one time are represented by a set of Heisenberg-picture
projection operators fPkk (tk )g satisfying
X
Pkk (tk )Pk k (tk ) = k  k Pkk (tk );
0 0 Pkk (tk ) = I (1:4)
k

showing that they represent a mutually exclusive, exhaustive set of alternatives. An indi-
vidual history corresponds to a particular sequence of alternatives   (1 ; : : : ; n) and is
represented by the corresponding chain of projections. When the theory assigns probabil-
ities to a set of such alternative histories, the probabilities of the individual histories are
given by
p(1; : : : ; n) = kPnn (tn) : : : P11 (t1 )j ik2 (1:5)
assuming (for simplicity) that the initial condition of the closed system j i is pure.
Eq.(1.5), giving the probability of a history of a closed system, has the same form
as eq.(1.2) giving the probability of a history of measurements of a subsystem. The only
dierence between the expressions is that in (1.2) states operators, etc, all act on the Hilbert
space of the measured subsystem, while in (1.5) they act on the Hilbert space of a closed
system including both the measured subsystem and any measurement apparatus. Thus, if
fPHl (tl)g and fPHm (tm )g are projectors onto the same sets of ranges of the total energy
H at two dierent times there is no reason to believe that the probability of histories with
l 6= m will vanish if the intervening projections do not commute with the Hamiltonian.
Eq.(1.5) no more conserves energy than does (1.2). However, in the quantum mechanics of
a closed system there is nothing \external" to cause a uctuation in the total energy. Does
this mean that the quantum mechanics of closed systems predicts non-zero probabilities
for violations of energy conservation? Further, not only conservation of energy is at stake.
Similar remarks hold for any other quantity that commutes with the Hamiltonian such as
 For example, as in [4], Section II.10.

4
the total electric charge. In the following, we shall show that no such violations are in
fact predicted.
In posing the question of possible violations of fundamental conservation laws in the
quantum mechanics of closed systems we should stress that we do not mean violations that
might be revealed by successive measurements of a subsystem. The probabilities of the
outcomes of ideal measurements on a subsystem are described by (1.3) to an accuracy far
beyond the precision available in any experimental check of a conservation law. A sequence
of two measurements that determines whether the value of a quantity a that commutes
with h is in one of a set of ranges f
g is represented by the string of projections
sa2 (t2 )sa1 (t1 ): (1:6)
The Heisenberg equations of motion
sa (t) = eiht=h sa (0)e iht=h (1:7)
together with the analog of (1.4) show that, when a commutes with the Hamiltonian h,
the operator string (1.6) is proportional to 1 2 . The probabilities (1.5) of a measured
uctuation in the value of a quantity commuting with h, including the energy itself, are
therefore zero .
However, the quantum mechanics of closed systems does not only predict probabili-
ties for the outcomes of measurements of a subsystem. We may consider, if we wish, the
probabilities of histories which describe alternative values of the total value of a quantity
commuting with the total Hamiltonian H for the whole closed system at various moments
of time. Such total quantities are unlikely to be accessible to experiment, but their con-
servation, or lack of it, is still of theoretical interest, and it is this question which is the
subject of this paper.
The expression (1.5) for the probabilities of the histories of a closed system would
seem to allow non-zero probabilities for uctuations in a quantity commuting with the
Hamiltonian if projections that appear between two projections associated with this con-
served quantity do not commute with it. However, in the quantum mechanics of closed
systems, probabilities are not predicted for an arbitrary set of alternative histories. They
 Following the usual terminology, when no confusion should result, we will often refer
to quantities that commute with the Hamiltonian as \conserved quantities" even though
it is their conservation that is being investigated!
 This is true for any two times t1 and t2 despite common misconceptions concerning
the energy-time uncertainty principle
5
are predicted only for those sets for which there is negligible quantum mechanical inter-
ference between the individual histories in the set [1, 2, 3]. Such sets of histories are said
to decohere. It would be inconsistent to assign probabilities to sets of histories that did
not decohere because the correct probability sum rules would not be obeyed. Decoherence
of histories implies the validity of the probability sum rules so that decoherent sets of
histories are consistent.
Conservation laws are obeyed by consistent sets of histories. In Section II we review
an argument of Griths [5] that exact decoherence implies exact conservation of quan-
tities that commute with the Hamiltonian. However, as we also show in Section II, in a
closed system of particles interacting by potentials, there are severe limits on the exactly
decohering sets of histories through which the probabilities of uctuations in a quantity
commuting with the Hamiltonian could even be dened. The only other alternatives per-
mitted in such histories are of the values of quantities that eectively commute with the
conserved quantity; i.e., commute when acting on the initial condition of the closed sys-
tem. The probability for any non-trivial evolution of such systems is zero. That limitation
would prohibit, for instance, consideration of a set of histories that contained alternatives
of the total energy as well as the alternatives referring to position and momentum that
would be needed to predict the outcomes of our everyday observations. (Recall that as
observers we are part of this closed system.)
However, a closed system of particles interacting via potentials is not a realistic model
of our universe. The two most important absolutely conserved quantities { electric charge
and mass { are coupled to long-range elds. This fact has two consequences: (1) It allows
decoherent histories that describe possible uctuations in charge or mass together with
other realistic, everyday alternatives. (2) It ensures the exact decoherence of the alternative
values of these uctuations, and that the probability is zero for any non-vanishing value of
a uctuation. That is, total charge and total energy are exactly conserved. The simplest
case of electric charge is discussed in Section III. Section IV discusses the conservation of
total energy.
II Exact decoherence and Exact Conservation.
In this section we review Griths' demonstration that the probabilities for uctuations
in the values of quantities that exactly commute with the Hamiltonian are exactly zero
for exactly decohering sets of alternative histories of a closed system. We also show that,
given a quantity A which commutes with the Hamiltonian, the only alternatives which
 The argument appears well known to a number of people. We learned it from R.
Griths.
6
can occur in an exactly decohering set of histories describing possible uctuations of A are
values of operators which eectively commute with A when acting on the initial condition
of the system.
Let A be any quantity satisfying
[A; H ] = 0 (2:1)
including the Hamiltonian itself. Let f
g;  = 1; 2; : : : be an exhaustive set of non-
overlapping ranges of the eigenvalues of A, and let fPA(t)g be the set of Heisenberg picture
projections onto them. The fPA(t)g obey (1.4). Consider a set of histories (consisting of
alternatives at a sequence of times) in which sets of projections onto ranges of A occur at
two dierent times tl and tm . The individual histories in such a set would be represented
by chains of projections operators
C = Cc c PAm (tm )Cb b PAl (tl )Caa (2:2)
where the fCaa g; fCb b g; fCc c g are the chains of projections representing alternatives be-
fore tl, between tl and tm , and after tm respectively. More generally the Caa ; Cb b ; Cc c
could be sums of chains of projections corresponding to alternative histories dened by
partitions of the chains into exclusive classes, and they could be branch dependent in the
sense of [4] without aecting the subsequent simple argument.
The decoherence functional whose o diagonal elements measure quantum interference
between parts of histories is
D(0 ; ) = Tr(C Cy )
0 (2:3)
where  is the density matrix representing the initial condition of the closed system. When
Re(D) vanishes for 0 6=  the set of histories exactly (weakly) decoheres and the proba-
bilities are given by the diagonal elements, as summarized in the equation
ReD(0 ; ) =  p() :
0 (2:4)
We can now proceed with Griths' argument.
Consider the probabilities p(c ; m; b; l ; a) of the set of histories represented by
(2.2). Exact weak decoherence implies that these probabilities are consistent. That is,
they must obey the probability sum rules and in particular
X
p(c; m ; b; l; a ) = p(m ; l) : (2:5)
a ;b ;c
7
The p(m ; l) are the probabilities for the set of histories represented by the chain
PAm (tm )PAl (tl ) : (2:6)
But the individual operators in the chain are in fact independent of t because A is conserved.
Specically, the Heisenberg equations of motion show that
PA(t) = eiHt=hPA (0)e iHt=h = P A (0)
 (2:7)
because A commutes with H . Thus
PAm (tm )PAl (tl) = l m PAl (tl ) (2:8)
and the probabilities p(m ; l) which follow from (2.4) vanish if m 6= l. Since the left
hand side of (2.5) is the sum of positive numbers, they must vanish individually. We have
p(c; m ; b; l ; a) = 0 ; m 6= l (2:9)
and the probability is zero for any non-vanishing uctuation in the value of a quantity that
commutes with the Hamiltonian. Energy in particular is conserved.
This satisfactory state of aairs is somewhat vitiated by the following result which
shows that exact decoherence permits only alternatives values of quantities that eectively
commute with the conserved quantity A in between times tl and tm.
Suppose the set of histories represented by (2.2) exactly decoheres. Then every coarse
graining of it must also exactly decohere and in particular the set represented by
Cmbl = PAm (tm )Cb b PAl (tl ) : (2:10)
is exactly decoherent. According to the result of Griths derived above, the probability
of a uctuation in the value of A is zero:
p(m ; b; l)  Tr(Cm bl Cy mbl ) = 0 ; l 6= m : (2:11)
Write the density matrix  in the basis in which it is diagonal as
X
= jihj (2:12)

 The argument for conservation depends only on the consistency of the set of histories.
Although we introduced it by discussing weak decoherence which implies consistency, the
argument could proceed directly from (2.5).
8
for positive probabilities  . In that basis, (2.11) reads
X
 jhjCmb l j0 ij2 = 0 ; l 6= m (2:13)

0

so that
Cmbl ji = 0; if l 6= m and  6= 0: (2:15)
Thus, Cmbl for l 6= m must vanish on the subspace S of initial states with non-
vanishing probabilities in the initial density matrix. In particular
Cmbl  = 0 ; l 6= m : (2:16)
The result (2.15) can be used to show that Cb b must commute with A when acting on
the subspace S. Suppose f
m m g and f
l g are sets of ranges of uniform, innitesimal
l
size
centered on eigenvalues am . From (2.15) and (2.10) we can write
(am al )PAm (tm )Cb b PAl (tl )ji = 0; when  6= 0 (2:17)
now holding for all values of m and l. In the limit of innitesimal intervals
, we have
X
am PAm = A(tm ): (2:18)
m

Thus, by summing (2.17) over m and l we have

[A; Cb b ]ji = 0; when  6= 0 (2:19)
and in particular
[A; Cb b ] = 0 : (2:20)
Thus the only permissible alternatives Cb b in a set of exactly decohering histories of the
form (2.2) that test the conservation of a quantity A which commutes with the Hamiltonian
are alternatives of quantities that eectively commute with A in the initial condition .
The Hamiltonian corresponding to the total energy of course commutes with itself.
The result (2.20) is sucient to show that histories of the form (2.2) that test conservation
of energy can only exhibit trivial dynamics. Consider the case when fCb b g is a set of
histories of alternative ranges of values of a quantity B at a time t such that tl < t < tm .
Then,
Cb b = PB (t) : (2:21)
9
From (2.20) we conclude
[H; PB (t)] = 0 : (2:22)
Eq.(2.22) and the Heisenberg equation of motion
PB (t) = eiHt=hPB (0)e iHt=h (2:23)
are enough to show that for tl < t1 < t2 < tm
PB2 (t2)PB1 (t1) = 0 ; 1 6= 2 (2:24)
and this implies that there is zero probability of any change in the value of B for the
histories in which uctuations in the energy can be dened.
To appreciate the strength of this result, imagine a model universe consisting of a
large box of non-relativistic particles interacting by potentials. Sets of histories describing
just uctuations in the total energy of the model universe, say,
C = PH2 (t2 )PH1 (t1 ) (2:25)
always exactly decohere. That is because the PH (t)'s are constant in time [cf (2.7)] so
that the fCg are a set of orthogonal projections. The cyclic property of the trace in (2.3)
shows that histories consisting of orthogonal projections always decohere exactly. Further,
as a consequence of the orthogonality of the projections in (2.25) for dierent ranges of
the total energy, the C representing non-vanishing uctuations in the total energy vanish
identically. The probability of a uctuation in the total energy is exactly zero.
However, the result (2.24) shows that it is not possible to ne grain the histories (2.25)
to include alternatives of any quantity other than the total energy without the values of
that quantity being constant in time with probability one. Thus, were we part of such a
system, we could not consider a set of histories that describe the changes in our everyday
lives and at the same time describe the uctuations in the total energy of the closed system.
The above discussion considered the conservation of precisely dened values of quan-
tities commuting with the Hamiltonian in exactly decohering sets of histories in a model
of particles interacting via potentials. It would be possible to consider the conservation of
quantities dened imprecisely by a range of values but the results would depend on the
details of the dynamics of the system studied. It might also be thought that it would be
more realistic to consider approximately decohering sets of histories and the approximate
violations of conservation laws that could be expected to occur. However, more important
consequences can be derived for interesting quantities like electric charge and mass by
including the long-range elds coupled to them that we have neglected until now. We do
this in the next two sections.
10
III Electric Charge.
The most rmly established examples of absolutely conserved quantities are electric
charge and total energy. Both are coupled to long-range elds and the conservation of
each is connected with a fundamental symmetry principle. These fundamental symmetry
principles limit the sets of decoherent histories which can describe uctuations in quantities
that commute with the Hamiltonian and the values of the probabilities of these uctuations.
The simplest case is electric charge which we discuss in this section.
The quantum theory of the electromagnetic eld can be conveniently studied in tem-
poral, A0 (x) = 0, gauge where A (x) are the components of the potential. The states may
be represented by vectors in the fermion Fock space with components that are functionals
of the vector potential A~ (x). Thus formulated on a space which contains electromagnetic
degrees of freedom beyond the true physical ones, the theory has a constraint represented
by the operator Z
C () = d3x(x)[r
E~ (x) (x)] (3:1)
where (x) is the charge density and x denotes the three spatial coordinate of some par-
ticular Lorentz frame. The function (x) is the parameter of the gauge transformation
which is generated by C () via
Ai(x) = hi [C (); Ai (x)] = @i (x) (3:2)
More generally, C () generates gauge transformations for an arbitrary quasilocal (see, e.g.
[6]) operator O that vanishes suciently fast outside some bounded region of space (which
we will call the eective support) according to
O = hi [C (); O] : (3:3)
Physical, gauge invariant, quasilocal operators must commute with the constraint
[C (); Ophys ] = 0 (3:4)
and physical, gauge invariant states are annihilated by the constraint.
As a particular case of (3.4) we may take  = const: Then Gauss' law may be applied
to reduce the r
E~ term in (3.1) to a surface term with spacelike separation from the
eective support of Ophys yielding the result
[Q; Ophys] = 0 (3:5)
 This derivation is naive form a rigorous point of view. For a better one see [7].

11
where Z
Q = d3x(x) (3:6)
is the total charge operator. Quasilocal physical quantities therefore commute with the
total charge.
An exhaustive set f
g;  = 1; 2; : : : of ranges of a gauge invariant quantity O dene
a set of alternatives for a closed system at a moment of time. These are represented by
a set of Heisenberg picture projection operators fPO (t)g. Sets of histories for the closed
system may be dened by giving a series of such sets at a sequence of times t1; : : : ; tn. The
individual histories correspond to particular sequences of alternatives (1 ; : : : ; n ) and are
represented by the corresponding chains of projectors as in (1.5). More general examples
of histories can be obtained by partitioning such sequences into classes fcg represented
by a set of class operators fCg that are sums of the chains in the class. Thus, a gauge
invariant set of histories is generally represented by a set of class operators of the form
X
C = POnn (tn ) : : : PO11 (t1 ) : (3:7)
(1:::n )2

When electromagnetism is formulated in this way, with redundant as well as true

physical degrees of freedom, the decoherence functional is not given by a formula like
(2.2) in which the Hilbert space is a space of wave functionals of the vector potential
A~ (x). Rather, it is given by that formula utilizing the Hilbert space of functionals of the
true physical degrees of freedom; that is, just the transverse part AT (x) of the vector
potential. The class operators (3.7) must rst be reduced to class operators fCT g on the
Hilbert space of physical degrees of freedom by integrating their matrix elements over the
redundant (longitudinal) degrees of freedom with appropriate gauge xing conditions. The
details of this process are not important for us; it suces to note that the decoherence
functional may be written
D(0 ; ) = TrT (CT CT y) 0 (3:8)
where  is a density matrix in the Hilbert space of the matter degrees of freedom and the
true physical degrees of freedom of the electromagnetic eld.
 In representing physical quantities by operators in this way we are considering, as
usual, quantities dened at one moment of time. For more general spacetime alternatives
see [8], Chap.VI.
 The construction is standard, but for more details in the present notation see [8].

12
 We next consider ne-graining a set of histories (3.7) by including alternative values
of the total electric charge Q at a sequence of times t0k ; k = 1; : : : ; m. We consider, for
simplicity the same set of ranges f
 g;  = 1; 2; 3; : : : of Q at each of these times and let
fPQk (t0k )g be the projections of the total charge operator onto them. The class operators
for such a ner grained set are
X
C = POnn (tn) : : : PQm (t0m ) : : : : : : PQ1 (t01 ) : : : PO11 (t1 ) (3:9)
(:::n )2

where the PQk (t0k ) have been inserted at the positions dictated by time ordering.
The projections fPQ (t0 )g have two important properties: First, they commute with
all gauge invariant quantities as a consequence of (3.5), and therefore, in particular
[POkk (tk ); PQl (t0l)] = 0 : (3:10)
Second, they are conserved
[H; PQl (t0l )] = 0 (3:11)
and therefore are independent of the times t0l . Eq.(3.10) means all P Q's may be commuted
to the right or left in (3.9) and (3.11) means that the class operator is zero unless all the
l are the same
C = n1 : : : 21 PQ1 C
(3:12)
= n1 : : : 21 CPQ1
The rst of the relationships (3.12) shows that, for any set of histories, the alternative values
of the charge always decohere exactly. That is because, as a consequence of the cyclic
property of the trace, the decoherence functional D(0 ;  0; ;  ) is always proportional
to 11 . The -functions in (3.12) thus ensure that histories in which the total charge
0

uctuates have probability zero. Total charge decoheres exactly and is exactly conserved.
Allowing approximate decoherence does not permit non-zero probabilities for uctuations
in Q.
The restrictions derived in Section II on histories that include alternative values of the
total charge are still valid. The alternatives in a decohering set of histories must commute
with the total charge. However, all quasilocal physical alternatives satisfy this condition
as a consequence of gauge invariance. It is therefore no restriction at all.
In general, if a set of histories fCg decoheres, then the ner-grained set (3.9) that
includes alternatives values of the charge does not necessarily decohere. However, it does
13
in one interesting and natural case. That is when the initial condition has a denite, xed
total charge q.
Q = Q = q : (3:13)
Then when C , in the form of the second of (3:12), acts on  there will be a non zero
result only for that interval
 , which contains q. D(0  0 ;  ) is thus non-zero only when
both 10 and 1 have this value and is therefore diagonal. The ner-grained set fC g
decoheres if the set fCg does. It follows that when the universe has a denite value of
the total charge, we may always ne-grain any set of decoherent histories to ask about
the total charge without disturbing decoherence and receive from the quantum mechanics
of closed systems the reassuring answer that it is conserved with probability one.
Finally, note that all of our results follow directly from (3.5). Any operator with the
property satised by Q in this equation is said to be superselected (see, e.g. [6]). In specic
restricted models, this occurs for quantities like baryon number, lepton number, and non-
Abelian charges as well as electric charge. Thus, alternative values of superselected charges
always decohere exactly and are exactly conserved. Furthermore, when the total state has
a denite value of such a charge, projections onto its eigenvalues may be added to any set
of histories without aecting decoherence.
IV Total Mass.
Energy universally couples to spacetime curvature which itself can carry energy in the
form of gravitational waves. As a consequence, a realistic classical discussion of the total
energy of a closed system, which in relativity is the same thing as its total mass, must
be carried out in the context of general relativity and a discussion of possible quantum
uctuations in the total energy in the context of quantum gravity.
There is no local denition of mass-energy in general relativity because a general
spacetime does not exhibit a time-translation symmetry. Neither is it possible to dene the
total mass of a spatially closed cosmology except by assigning it the value zero in which
case its conservation is trivial. Conservation of energy becomes an interesting issue in
asymptotically at spacetimes possessing asymptotic time translation symmetries enabling
the total mass of the system to be dened.
For asymptotically at spacetimes, the mass on a spacelike surface can be determined
from the asymptotic behavior of the spatial metric on that surface. Using coordinates
which asymptotically become rectangular Minkowski coordinates at spatial innity the
 As it does for instance in the \no-boundary" [9] initial condition where the total charge
is zero because the universe is spatially closed.
14
deviations from at space must, at the very least, fall o as

g =  + M (rt; ; ) + O r12 :


(4:1)

Here,  is the at metric in rectangular Minkowski coordinates,  = diag( 1; 1; 1; 1; ),

t = x0 and the polar coordinates (r; ; ) are connected to the rectangular coordinates
(x1 ; x2 ; x3 ) by the usual relations, e.g. x1 = r sin  cos .
Consider a spacelike surface that asymptotically is a surface of constant t. The ADM
total mass [10] is dened by evaluating the following integral on a two-surface at large r
Z
1
M (t) = 16 rlim
!1 t dSj (@k gkj @j gkk ) : (4:2)

Here @i is the at-space gradient and we have followed the usual convention of indicating a
summation in asymptotic expressions by repeated lower indices. The asymptotic behavior
of the metric (4:1) ensures that M (t) is nite.
Whether total mass-energy is conserved in a quantum theory of asymptotically at
spacetimes depends on the probabilities of decoherent histories that describe diering val-
ues of M (t) on dierent spacelike surfaces. There are, of course, a variety of approaches to a
quantum theory of spacetime. We shall analyze the question in the sum-over-histories gen-
eralized quantum theory of spacetime geometry. A generalized quantum theory is specied
by three elements: (1) The ne-grained histories, which here are a class of four-dimensional
metrics and matter eld congurations. The metrics g (x) are asymptotically at at least
in the sense of (4.1) but with possibly more restrictive conditions to be discussed below and
dwell on a manifold with two spacelike boundaries 0 and 00 representing the \endpoints"
of the history. To keep the notation manageable we shall indicate only a single matter eld
(x). (2) The allowed coarse-grainings, which here are dieomorphism invariant partitions
of the ne-grained histories into exclusive classes fcg;  = 1; 2;


called coarse-grained
histories. (3) A decoherence functional dening the measure of interference between pairs
of coarse-grained histories. The precise details of the construction of this decoherence
functional will not be important for us. Its form is similar to (2.3) but with notions of
, Tr, etc. appropriate to gravity. It is the form of the class operators corresponding to
coarse-grained histories that is important for the present discussion of the conservation of
the total mass. These class operators act on the space of wave-functionals dened on the
space of three-metrics hij (x) and spatial matter eld congurations (x) on a spacelike
surface. The matrix elements of the class operator corresponding to a dieomorphism in-
variant class c of asymptotically at four geometries and eld congurations are dened
15
by the sum-over-ne-grained-histories:
Z
hh00ij ; 00 jjCjjh0ij ; 0 i = g expfiS [g(x); (x)]=hg : (4:3)
[(h0 ;0 );c;(h00 ;00 )]

Here, h0ij (x) and 0 (x) are the induced metrics and matter eld congurations on the
boundary 0 . There are similar denitions on 00 . S [g; ] is the action for geometry
coupled to matter elds. The sum is over four-metrics g (x) and four dimensional eld
congurations (x) which are in the dieomorphism invariant class c and match the
prescribed conditions on 0 and 00 . Of course, the expression (4.3) is only formal and
must be augmented by gauge xing machinery, regularization procedures, etc to make
sense, but its form will be sucient for the level of argument we are able to give. Further
details can be found in, for example [8].
With these preliminaries in hand we may return to the issues of the conservation of
total ADM mass at spatial innity and whether histories that dene uctuations in the
total mass are limited to trivial dynamics as they were in the simple model of Section
II which neglected gravitation. To calculate the probability of a uctuation in the mass,
we must consider partitions of the set of ne-grained histories into classes by ranges of
the value of the total mass M () on at least two dierent spacelike surfaces 1 and 2 ,
in addition to whatever other alternatives dene the classes under consideration. Such
histories are the analog of those represented by (2.2) when A is the total energy, H , in
the non-gravitational case. When such sets decohere, the issue of conservation of ADM
mass is then the question of whether the probability is zero for those with M (1 ) = 6
M (2 ). For this case, the arguments of Section III are not satisfactory as quasilocal
dieomorphism invariant operators are dicult to construct { in fact, because there are
no local dieomorphism invariant operators for gravity, strict use of the denition in [6]
shows that there are no quasilocal invariant operators at all. To nd a more satisfying
argument we must look more closely at what is meant by \asymptotically at".
Penrose's notion of conformal completion [11,12] gives a standard denition of a space-
time which is asymptotically at . A consequence of this denition is that asymptotically
at metrics have a more restricted asymptotic behavior than that given by (4.1). In par-
ticular the Riemann tensor must decay at large r as


R  = O 1=r3 : (4:4)
It is not dicult to show that, in any 3 + 1 decomposition of spacetime into space and
time, this implies
M_ ij = 0 (4:5)
 For a lucid review see Ashtekar[13]

16
where a dot denotes a time derivative and Roman indices range over spatial directions.
This means that the ADM mass, as dened by (4.2) is constant in time. The conservation
of ADM mass in this context does not follow from the equation of motion, but from the
denition of an asymptotically at spacetime. Of course, the asymptotically at context
would be uninteresting except that solutions of this form do exist. The nite propagation
velocity of gravitational radiation ensures that any solution with suitably localized initial
data will be asymptotically at.
The Penrose diagram for the conformally completed asymptotically at spacetime
makes the reason for this \conservation" intuitively clear. Spacelike innity is a single
two-sphere where all spacelike surfaces terminate. A common value of the ADM mass is
therefore shared by all.
Were we to use conformal completion to dene the asymptotically at metrics which
enter into the sum-over-histories (4.1) the question of conservation of total mass would be
trivial. Only geometries with constant total mass contribute to the sum, therefore parti-
tions into classes with dierent masses on dierent spacelike surfaces would be vacuous.
However, while the conservation of total mass at spatial innity is trivial, the dynamics
permitted in histories that dene this conservation is not. In the model without long-range
elds discussed in Section II, only alternatives of quantities that eectively commuted
with the total energy were permitted in exactly decohering sets of histories which also
described uctuations in the total energy. However, in the presence of the gravitational
eld, the analog of (2.17) which led to that result is trivially satised for any dieomorphism
invariant partition of the ne-grained histories, regardless of whether it is associated with
projections onto eigenvalues of quasi-local operators. That is because there are no ne-
grained histories at all with uctuations in the total mass. We therefore expect that in
generalized quantum theory we are permitted arbitrary sets of physical histories that also
describe uctuations in the total mass. If any set of alternatives decoheres, we may always
consider the ner graining which in addition describes uctuations in the total mass. If that
ner graining continues to decohere, the alternatives referring to the total mass decohere
exactly. Total mass, or total energy which is the same thing, is conserved with probability
one.
The above discussion was carried out using the conformal completion denition of
asymptotic atness. However, from the perspective of quantum gravity it appears more
natural to dene asymptotic atness from a property of the action rather than from a
notion of conformal completion. We now show that if the sum-over-histories in (4.3) is
restricted to a class of metrics with the fall-o (4.1) that (1) have nite action and (2) are
invariant under dieomorphisms, then the ADM mass is conserved. To understand this it
17
is sucient to look at the action for pure gravity.
The action for gravity on a domain of spacetime M is
Z
p Z
p
(16G)SE [g] = d x gR + 2
4
d3x hK (4:6)
M @M
where R is the scalar curvature and K is the extrinsic curvature scalar of the boundary of
M. In order to discuss the properties of metrics at spatial innity, it is useful to consider
a standard 3 + 1 decomposition of the metric
ds2 = N 2 dt2 + hij (dxi + N i dt)(dxj + N j dt): (4:7)
which need only hold near spatial innity for our purposes. Consider the action for a
region of spacetime lying between two spacelike surfaces of constant t and bounded by a
timelike surface @ Ms near innity. This may be written as
Z Z
(16G)SE [g] = dt d3xNh1=2 [Kij K ij K 2 +3 R]
Z Z p (4:8)
di ( 2KN i + 2Di N ) + 2 d x h~ (K~ K~ 0 )
3
@ Ms @ Mt

where Kij is the extrinsic curvature of surfaces of constant t,

Kij = 21N h_ ij + D(iNj) ;

h i
(4:9)

Di being the derivative in the surface, K~ ij is the extrinsic curvature of @ Mt , and the tilde
indicates that a quantity is to be evaluated for a timelike surface.
The general form of the metric (4.1) is not sucient to ensure the convergence of
the action (4.8). Eq.(4.1) implies that the asymptotic behavior of 3R is O(1=r3 ) but the
asymptotic behavior of Kij is

Kij = 21r M_ ij + O( r12 ) : (4:10)

If we evaluate the volume term in the action (4.8) out to a large radius rl , the coecient
of the leading term as rl ! 1 is
Z Z Z Z
rl
2
dt d
[(h_ ij ) 2
(h_ kk ) ] =
2
dt d
[(M_ ij )2 (M_ kk )2] (4:11)

18
where d
is an element of solid angle at innity. Thus metrics in the class (4.1) must be
further restricted so that the right hand side of (4.11) vanishes in order to ensure nite
action. If we assume that this should hold for any choice of the time interval between the
boundary surfaces we must have
Z Z
rl
2 d
[(h_ ij )2 (h_ kk )2 ] = d
[(M_ ij )2 (M_ kk )2 ] = 0 : (4:12)

This is not enough to show that the ADM mass is constant in time, but when coupled
with the requirements of dieomorphism invariance it will be.
The asymptotic behavior of (4.1) refers to a particular decomposition of the spacetime
into space and time, and the condition (4.12) ensures that there is no linear divergence
of the action when evaluated between two constant time surfaces in that decomposition.
However, since the class operators (4.3) are to be dened by integrals over dieomorphism
invariant partitions, the notion of asymptotic atness and of niteness of the action must
be independent of the 3 + 1 decomposition. In particular it must be invariant under
dieomorphisms which Lorentz transform the asymptotic slices. This leads to stronger
conditions than (4.12) as we shall show.
Consider innitesimal dieomorphisms (gauge-transformations)
g (x) ! g + r()(x) ; (4:13)
and in particular those which asymptotically correspond to Lorentz boosts
i  vi t + di(t; ; ) + O(1=r) ;
(4:14)
0  vi xi + O(1) :
Lorentz boosts preserve the asymptotic behavior (4.1). The supertranslations di , however,
must be independent of time to preserve (4.1), as substitution into (4.7) will show. The
covariant components of i(x) relevant for the transformation of the hij (x) are thus
i  vi t + di(; ) + Ni(x)(vj xj ) + O(1=r) (4:15)
displaying explicitly the O(r) and O(1) terms. Since Ni (x)  si(t; ; )=r the third term
is of O(1). The spatial part of the metric important for the asymptotic form of the metric
thus transforms as
hij ! ij + Mrij + @(ij) + O( r12 ) (4:16)
with j (x) of the form (4.15).
19
 Dieomorphism invariance requires that the condition (4.11) be enforced for hij (x) of
the form (4.16) with i(x) given by (4.15). The rst term in (4.15) does not change the
spatial metric. In determining the eect of the rest of (4.15) on hij (x) the boost parameter
vi is arbitrary. But it is also important to note that the si (t; ; ) determining Ni (x) is
arbitrary; si merely denes how the spacetime is sliced internally, consistent with a given
asymptotic slicing. We must therefore enforce the condition (4.12) for hij (x) of the form
(4.16) with arbitrary i (x) of O(1). One further invariance should be enforced. In (4.11)
we evaluated the action inside spheres of constant rl and considered the limit rl ! 1. The
same results should hold for arbitrary shaped surfaces rl = rl (; ) = Rf (; ) as R ! 1.
To rst order in i(x), the gauge transformed condition (4.12) becomes the condition that
the linear divergence in R of
Z rl (;) Z
r dr d
[uij (x)@i j (x)]
2
(4:17)

vanish. Here we have abbreviated

uij (x) = h_ ij (x) ij h_ kk (x) (4:18)
and assumed, as discussed above, that the O(1) part of i (x) is an arbitrary function of
(t; ; ). Integrating (4.17) by parts and retaining only the leading terms in large R in the
resulting condition following from the arbitrary form of i (x) we have
R[ Rf (; )@i uij (x) + ni(; )uij (x)] = 0 (4:19)
for large R where ni is the normal to the bounding surface proportional to @i f . Since f is
arbitrary (4.19) can be satised only if both terms vanish and, in particular, if Ruij (x) = 0.
That implies
M_ ij = ij M_ kk : (4:20)
This condition together with (4.12) is enough to guarantee
M_ ij (t; ; ) = 0 : (4:21)
The result (4.21) is enough to guarantee that the Riemann tensor falls o as O(1=r3 ) as
in (4.4) and ensure the conservation of the ADM mass. To see that, simply note that from
(4.2), the ADM mass is determined by the coecients Mij all of whose time derivatives
vanish.
20
 Thus restricting attention to metrics with the minimal asymptotic behavior (4.1)
for which the action is nite and dieomorphism invariant means restricting to metrics for
which the Riemann tensor falls o as O(1=r3 ) and for which the ADM mass is constant. It is
an interesting question whether the above rather clumsy argument could be pushed further
and whether there is full equivalence between asymptotic atness dened by niteness and
dieomorphism invariance of the gravitational action and asymptotic atness dened by
conformal completion. It would be of special interest to investigate and compare the
conditions necessary to dene total angular momentum at spatial innity, a quantity that
we have not touched upon.
Thus, whether asymptotic atness is dened by conformal completion or by the be-
havior (4.1), niteness of the action, and covariance, the result is the same for the ADM
mass. It is conserved in each history. The set of ne-grained histories includes histories
with diering values of the ADM mass but within each history it does not vary. Arbitrary
dieomorphism invariant alternatives may therefore be considered in addition to those nec-
essary to describe the conservation of total mass of a closed system and, if all alternatives
decohere, total mass, which is the same as the total energy, is conserved.
Acknowledgments.
We would like to thank the Aspen Center for Physics where this work was initiated
and also the Isaac Newton Institute for Mathematical Sciences. We would also like to
thank Bob Griths, Jonathan Halliwell, Karel Kuchar, Roger Penrose, John Stewart and
Wojciech Zurek for useful conversations. The work of J.H. was supported in part by NSF
grant PHY90-08502. R.L. thanks Los Alamos National Laboratory for support. D.M. was
supported in part by NSF grant PHY93-96246 and funds provided by the Pennsylvania
State University.
References.
1. R. Griths, J. Stat. Phys. 39, 219 (1984).
2. R. Omnes, J. Stat. Phys. 53 893, (1988); 53, 957 (1988); 53, 993 (1988); 57, 357
(1989); Rev. Mod. Phys. 64, 339 (1992).
3. M. Gell-Mann and J.B. Hartle in Complexity, Entropy and the Physics of Information,
SFI Studies in the Sciences of Complexity, Vol. VIII, ed by W. Zurek (Addison-Wesley,
Reading, 1990).
4. J. B. Hartle, in Quantum Cosmology and Baby Universes, Proceedings of the 1989
Jerusalem Winter School for Theoretical Physics, edited by S. Coleman, J. B. Hartle,
T. Piran, and S. Weinberg (World Scientic, Singapore, 1991).
21
 5. R. Griths, Private communication.
6. R. Haag, Local Quantum Physics (Springer-Verlag, New York, 1992).
7. F. Strocchi and A. S. Wightman, J. Math. Phys. 15, 2198 (1974).
8. J. B. Hartle in Gravitation and Quantization, Proceedings of the 1992 Les Houches
Summer School, ed. by B. Julia and J. Zinn-Justin, Les Houches Summer School
Proceedings Vol LVII (North Holland, Amsterdam, 1995), gr-qc/9304006.
9. J. B. Hartle and S. W. Hawking, Phys. Rev. D 28, 2960 (1983).
10. R. Arnowitt, S. Deser, and C.W. Misner in Gravitation: An Introduction to Current
Research, ed. by L. Witten (Wiley, New York, 1962) and references to earlier work
therein.
11. R. Penrose, Phys. Rev. Lett. 10, 66 (1963).
12. R. Penrose, Proc. Roy. Soc. London, A284, 159 (1965).
13. A. Ashtekar, in General Relativity and Gravitation, Vol. II, ed. by A.Held (Plenum
Press, N.Y., 1980) p.37.
14. M. Gell-Mann and J. B. Hartle, Phys. Rev. D 47, 3345 (1993), gr-qc/9404013.

22