PERGAMON

Studies in History and Philosophy of

Modern Physics 34 (2003) 559-577

Studies in History
and Philosophy
of Modern Physics

The cosmological constant, the fate of the

universe, unimodular gravity, and all that
John Earman*
Department of History and Philosophy of Science, University of Pittsburgh, Pittsburgh, PA 15260, USA

Abstract
The cosmological constant is back. Several lines of evidence point to the conclusion that
either there is a positive cosmological constant or else the universe is filled with a strange form
of matter ("quintessence") that mimics some of the effects of a positive lambda. This paper
investigates the implications of the former possibility. Two senses in which the cosmological
constant can be a constant are distinguished: the capital A sense in which lambda is a universal
constant on a par with the charge of the electron, and the lower case A sense in which lambda is
a humble constant of integration. The latter interpretation has been touted as the means to a
solution to various problems in physics. These claims are critically examined with an eye to
discerning the implications for philosophy of science and foundations of physics.
2003 Elsevier Ltd. All rights reserved.

1. Introduction
Einstein (1917) had two motivations for adding the cosmological term to
his gravitational field equations: the first was his desire to harmonize his general
theory of relativity (GTR) with Mach's principle; the second was to make possible
general relativistic cosmological models that are static as well as homogeneous
and isotropic-features which Einstein then thought to describe the large-scale
structure of the actual cosmos. Einstein (1931) officially abandoned the cosmological
constant after Rubble's redshift observations indicated that the universe is in
fact expanding. These observations are accommodated by Friedmann's (1922)
expanding universe solutions of Einstein's field equations without cosmological
constant. Although Einstein came to regard the cosmological constant as his
*This paper was solicited by Rob Clifton while he was Editor. It is dedicated to his memory.
E-mail address: [email protected] (1. Earman).

ARTICLE IN PRESS
560

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

biggest blunder,1 it has been resurrected a number of times in cosmologyto solve

the age of the universe problem, to help with the structure formation problem, and to
explain the redshifts of QSOs.2 Inationary cosmology invokes an effective
cosmological constant to drive the (hypothesized) exponentially-fast expansion in
the very early universe. The simplest of the many versions of inationary cosmology
also require a genuine cosmological constant to make up for the difference between
the value of the density parameter favored by ination and its value as derived from
estimates of the amount of gravitating matter (including so-called dark matter) in
the universe. More recently, several lines of evidence have been pointing to either a
positive value for the cosmological constant or a stand-in for such a beast.
One line of evidence for this conclusion derives from recent measurements of the
red shifts of Type Ia supernovae indicating that the rate of expansion of the universe
is increasing (see Perlmutter et al., 1999). To t these observations to the preferred
FriedmannRobertsonWalker (FRW) cosmological models requires either a L > 0
or else a surrogate for a positive L in the guise of a strange form of matter. This is
seen to follow from two considerations. First, the symmetries of the FRW model
require that the stress-energy tensor, which serves as the source term in Einsteins
eld equations (EFE), arises from a perfect uid.3 All normally encountered forms of
matter are thought to obey what are called the weak and strong energy conditions,4
which, when applied to the stress-energy tensor for a perfect uid, require
respectively that rX0 and r 3pX0; where r and p are, respectively, the density
and pressure of the uid. Second, the application of the EFE to the FRW models
yields the following equation for scale factor a (a.k.a. the radius of the universe):5
1
1
a.  r 3pa La:
6
3

Thus, in order to have an increasing rate of expansion, a. > 0; it is necessary to have

either L > 0 or else r 3po0 (in violation of the strong energy condition). A
number of models with hypothetical forms of abnormal matter satisfying r 3po0
and r > 0 (a.k.a. quintessence6) have been constructed to mimic various effects of
a positive L; making a decision between a genuine cosmological constant and strange
forms of matter difcult to make with present observational technology.
1

This remark is attributed to Einstein by Gamow (1958, pp. 6667; 1970, p. 44).
For a detailed history of the ups and downs of the cosmological constant, see Earman (2001).
3
The stress-energy tensor Tmn for a perfect uid has the form Tmn r pUm Un pgmn ; where U m is the
(normed) four-velocity of the uid, r is the density, and p is the pressure. The signature of the spacetime
metric gmn is taken to be : Greek indices run from 1 to 4, and the summation convention on
repeated indices is in effect.
4
These conditions require, respectively, that for any timelike V m ; Tmn V m V n X0 and Tmn V m V n X  12T
where T : TrTmn : gmn Tmn :
5
The line element for a FRW spacetime has the form ds2 a2 t ds2  dt2 where at is the scale factor
and ds2 is the line element of a Riemann space of constant curvature in one of three forms: k 0 (at
space), k 1 (hyperbolic space), or k 1 (closed space). FRW models are treated in every standard
text on cosmology; see, for example, Roos (1994). In what follows I have suppressed units.
6
Dark energy is the term used in the literature to refer neutrally to either a positive cosmological
constant or quintessence.
2

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

561

Given the many entrances and exits of the cosmological constant over the last
century, one might be somewhat skeptical about its present appearance on the
cosmological stage. But there are two things that are different about the present
episode. First, cosmology is no longer a data-starved eld with loose or non-existent
constraints on the basic parameters that characterize cosmological models. Second,
in addition to the supernovae evidence for an increasing rate of expansion of the
universe, there is an independent line of evidence for a positive L; or quintessence,
that comes from combining measurements of the cosmic microwave background
anisotropies and data from surveys of galaxy redshifts (see Efstathiou et al., 2002).
The present paper is dedicated to the proposition that, since it remains a live
possibility that the actual cosmos is characterized by a cosmological constant, it is
worth probing the nature and status of L: One cosmic implication is immediate: if a
positive lambda is the cause of the current accelerating expansion, then our universe,
which began in a big bang, will not end in a big crunch but will expand forever.7 If
that sounds reassuring, think again. A positive cosmological constant implies that
the expansion of our universe will continue to accelerate; and given the current
estimates of L; the horizon distance, which marks the furthermost extent of the
visible universe, will never increase beyond B18 billion light years for an observer
comoving with the expansion. The region within the horizon of any such observer
will eventually become increasingly empty and inhospitable to life as stars and
galaxies are carried across the observers horizon by the outward expansion (see
Krauss & Starkman, 2000). In this scenario, the universe itself is eternal, but our
progenyand the progeny of critters anything remotely like uswill not be able to
partake of this eternity.
In the present paper, I will turn from these cosmic matters to examine the
implications of the nature of the cosmological constant both for methodological
issues in the philosophy of science and for foundations of physics issues.8 The plan of
the paper is as follows. Section 2 distinguishes between two senses in which the
cosmological constant can be a constant: the capital L sense, according to which L is
a universal constant, and the lower case l sense, according to which l is the same
throughout spacetime but can have different values in different universes. Section 3
discusses the importance of this distinction for scientic explanation and for the issue
of the underdetermination of theory by empirical evidence. Section 4 examines the
relevance of the lower case l interpretation for what particle physicists call the
problem of the cosmological constant. Section 5 concerns the derivation of the
lower case l version of gravitation from an action principle. What is revealed is that
the appropriate spacetime setting for such a derivation differs from that of standard
GTR. The implications of this fact for the nature of observables and gauge principles
7
Assuming the validity of the FRW model for the actual universe. This implication is true even if we live
in a k 1 (spatially closed) FRW universe which, absent a positive cosmological constant, would
eventually recollapse to a big crunch.
8
Philosophers of science have been strangely silent about the cosmological constant. A search of
Philosophers Index for the last ten years, during which there have been dramatic observational and
theoretical advances in cosmology, does not reveal any relevant references that discuss the reappearance of
the cosmological constant.

ARTICLE IN PRESS
562

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

in general relativistic theories are discussed in Section 6, along with the prospects
that the lower case version of lambda can ameliorate the problem of time in
quantum gravity. Conclusions are presented in Section 7.

2. In what sense is the cosmological constant a constant?

The standard EFE with cosmological constant read9
1
Rmn  Rgmn Lgmn Tmn :
2
2
It is usually assumed that L is a universal constant, on a par with (say) c or the
charge of an electron. It follows from (2), the Ricci identity rm Rmn  12 Rgmn 0;
and the compatibility of covariant differentiation with the metric rm gmn 0 that
rm Tmn 0;

which expresses the local conservation of energy-momentum.

Although it may seem pedantic at this stage, I will rewrite (2) with a lower case l;
viz.
1
Rmn  Rgmn lgmn Tmn ;
4
2
to indicate a weaker sense in which lambda can be a constant: namely, the value of l
can vary from solution to solution (in the philosophers jargon, from physically
possible world to physically possible world) but is the same throughout the
spacetime of any given solution, i.e.
rm l @m l 0:

The combination of (4) and (5) guarantees (3).

In which sense is the cosmological constant a constant: the capital lambda sense
or the lower case lambda sense? And does it matter? Full answers to these questions
will take some explaining. But one thing is clear from the outset: the standard
derivation of the EFE (2) from an action principle presupposes the capital lambda
sense. The standard action is
L LG LM ;
Z
p
gR  2L d4 x;
LG

6a
6b

where LG and LM are, respectively, the gravitational and matter Lagrangians,

g : detgmn ; and M is the spacetime manifold. If the stress-energy tensor Tmn is
dened as the negative of the variational derivative dLM =dgmn ; variation of (6) with
respect to the metric tensor yields (2). But if L is not a universal constant, it too must
be
R varied;
p 4and the consequence of this variation of (6) is the absurd result that
g d x 0; i.e., the volume of spacetime is zero!
M
9

Rmn is the Ricci curvature tensor and R : TrRmn : gmn Rmn is the Ricci scalar.

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

563

How (4) and (5) can be derived from a variational principle will be taken up below
in Section 5. For the present I want to note only that there is an obvious reason for
taking the combination of (3) and (4) seriously. Taking the trace of (4) gives
l R T=4;

where T : TrTmn : Using (7) to eliminate l from (4) produces the trace-free EFE:
1
1
Rmn  Rgmn Tmn  Tgmn :
8
4
4
The trace-free EFE do not entail local energy-momentum conservation (3). But if (3)
is postulated separately, then (3) and (8) together entail
rm R T 0

R T const:

(Alternatively, (3) need not be postulated separately if it follows from the equations
of motion for the matter-eld which gives rise to Tmn ; see Section 4 below for an
example.) The value of the constant in (9) can be taken, by denition, to be 4l: The
upshot is that if we start, not with the standard EFE (2) and capital lambda, but with
(4) and (5), or alternatively with (8) and (3) (or (8) and (9)), we arrive at a new
perspective on lambda: it is not a new universal constant of nature but rather a
humble constant of integration.
For a matter-eld source whose stress-energy tensor is trace-free, e.g. the
electromagnetic eld, the trace-free EFE (8) reduce to
1
Rmn  Rgmn Tmn ;
10
4
which are the eld equations that Einstein considered in his 1919 paper, Do
Gravitational Fields Play an Essential Part in the Structure of Elementary Particles
of Matter? In regions where only gravitational and electromagnetic elds are
present, (7) gives R 4l: In such regions, therefore, R is a constant Ro ; and (4) can
be rewritten in the form
1
1
Rmn  Rgmn Ro gmn Tmn :
40
2
4
Moreover, (10) can be rewritten as
1
1
1
100
Rmn  Rgmn Ro gmn Tmn  gmn R  Ro :
2
4
4
The difference between (40 ) and (100 ) lies in the fact that the right hand side of (100 )
contains a term that can be interpreted as an additional stress-energy term that
represents a negative pressure R  Ro inside of an electric corpuscle and a zero
pressure outside.10 Einstein suggested that this negative pressure might serve to
maintain the electrodynamic forces of a charged corpuscle in equilibrium. This
suggestion came to naught as the mysteries of the composition of matter have to be
plumbed in the quantum theory.
10

The logic of Einsteins procedure is a bit murky since (10) plus the local energy-momentum
conservation law implies that the Ricci scalar R is constant everywhere.

ARTICLE IN PRESS
564

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

3. Implications for scientic explanation and the underdetermination of theory by

evidence
A positive value for L (or for l) signies that there is a repulsive force that tends
to counterbalance the attractive force of gravity.11 This is most easily seen
by studying the motion of a test body in the gravitational eld of a spherically
symmetric mass distribution, such as our sun. The relevant solution to the standard
EFE (2) is the exterior Schwarzschild solution, generalized to take into account the
cosmological term. In the slow motion, weak eld approximation, a test body
moving in this gravitational eld experiences a Newtonian gravitational potential of
the form
M Lr2
jNew  
;
r
6

11

where M is the mass of the central body and r is the Schwarzschild radial coordinate
(see Adler, Bazin, & Schiffer, 1975, Sec. 12.2). Thus, the test body experiences, as
expected, a 1=r2 force directed towards the central body; but it also experiences a
repulsive force proportional to Lr=3: For small L; the repulsive force will have
negligible effect on the motion of planets around the central body, but at
cosmological distances, the repulsive force eventually dominates the attractive force
of the central body.
This interpretation of L is conrmed in the setting of the FRW cosmologies. In
the case of an expanding universe lled with ordinary matter, the rst term on
the right hand side of (1) represents the slowing down of the expansion rate
due to the attractive force of gravity,12 while the second term represents the speeding up of the expansion rate due to the repulsive force associated with a
positive L: For ordinary matter, the explanation of the fact that at
. > 0 traces
to the dominance at t of the repulsive force due to L over the attractive force of
gravity.
A parallel statement holds for l: But the parallel is not complete. On the upper
case interpretation of lambda, L is not a dynamical variable, so that if L > 0 in
our universe, then L > 0 in all physically possible universes, with the upshot
that there is a lawlike tendency for the rate of expansion of an expanding universe to
slow down. On the lower case interpretation of lambda, by contrast, a positive value
for l in our universe does not signify a lawlike tendency for the expansion to
de-accelerate, and as a result, the explanation of accelerating expansion has a rather
different avor. Consider how the explanation emerges from the application of the
trace-free EFE (8) to the FRW models. The application yields but a single

11
Strictly speaking, this way of talking only makes sense in the Newtonian limit. In the setting of GTR,
talk of gravitational force is out of place since test particles free falling in the gravitational eld feel no
force in that their world lines are geodesics.
12
Note that a positive pressure contributes to the de-accelerationthis is a relativistic effect absent in
the Newtonian case.

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

565

independent equation for the scale factor:13

a 2
1
a.  r pa :
2
a
The supplemental equation (9) gives
 2

a.
a 2
6
2 r 3p const:
a a

12

13

Calling the constant of integration in (13) 4l; and combining (12) and (13), gives the
lower case lambda version of (1):
1
1
10
a.  r 3pa la:
6
3
If matter is normal, the explanation of a. > 0 now lies in the happenstance that for
our universe the value of the integration constant l is sufciently positive. In some
other physically possible universes, similar to our universe in all other respects, the
value of the integration constant l will be sufciently different that the expansion
rate will eventually fall to zero and then turn negative, with the consequence that
such a universe ends in a big crunch instead of expanding forever.
The lower case and capital lambda interpretations are seemingly empirically
indistinguishable; without the use of a magic metaphysical rocket ship to carry us
from the actual cosmos to other physically possible cosmoi, we have no way of
telling whether the value of lambda measured in our cosmos is shared by the other
possible cosmoi. However, a more detailed examination of the case reveals an
interesting sense in which the theories embodying the capital and lower case versions
of lambda are not empirically equivalent. This is a conrmation of the suspicion
shared by some philosophers of science that genuine scientic examples of
underdetermination of theory by evidence are much rarer than the literature on
scientic realism would lead one to believe.14 Before taking up this matter, I need to
comment on an alleged virtue of the lower case version of lambda.

4. The cosmological constant problem

The lower case interpretation of lambda has been touted as a potential solution to
what particle theorists have dubbed the cosmological constant problem.15 The
problem begins with what looks like a bit of word play on the notion of vacuum
13
The standard EFE (2) yield two independent equations for the scale factor, and when combined these
equations imply (1).
14
It is admittedly not easy to give a characterization of the distinction between what is to be counted as
a genuine example of underdetermination and what is to be counted as a philosophers trick. All I can say
here is that (i) the majority of examples in the philosophical literature strike me as belonging to the latter
category, and (ii) a boundary condition on the genuine examples is that they should not be functionally
equivalent to a brain-in-a-vat or a Cartesian Demon example-otherwise, the issue of scientific realism, as
opposed to the reality of the external world, is not reached.
15
For a critical discussion of the literature on this problem, see Rugh and Zinkernagel (2002).

ARTICLE IN PRESS
566

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

energy density. To begin the explanation, move the cosmological constant term in
L
the standard EFE (2) to the right hand side and label it as Tmn
: Lgmn to give
1
L
M
Rmn  Rgmn Tmn
Tmn
;
2

14

where the superscript M has been added to the ordinary stress-energy tensor that
M
arises from matter-elds. In the case of the (classical) vacuum, Tmn
0; (14) can be
read as saying that a positive L represents a positive energy density of the vacuum.
L
More fully, Tmn
has the form of a stress-energy tensor of a perfect uid with mass
density rL L and pressure pL rL :16
Now comes the word play. If instead of energy density of the vacuum one
says vacuum energy density the ears of particle physicists prick up. According
to quantum eld theory, the quantum vacuum, rather than being a state of
nothingness, is seething with activity. That this activity may have cosmological
implications is suggested by the following line of argument. In the case of Minkowski
spacetime, the vacuum state jvacS is picked out by the requirements of Poincare!
Q
invariance and positivity of energy. The expectation value /vacjTmn
jvacS of
the stress-energy tensor associated with quantum elds is necessarily Poincare!
invariant. And as the only symmetric, second rank, Poincare! invariant tensor has
the form const:  Zmn (where Zmn is the Minkowski metric), it follows that
Q
/vacjTmn
jvacS must have the form LQ Zmn of a cosmological constant term.
The trouble is that dimensional considerations lead particle theorists to expect that
LQ should have a value many tens of orders of magnitude in excess of what
observational limits would allow. Steven Weinberg (1989), who believes that crises in
physics are fruitful in provoking new physics, has promoted this problem to the level
of a crisis.
The logic of the cosmological constant problem is open to challenge at various
places. In particular, Poincare! invariance is lost in the cosmological context where
the background spacetimes are not at. Even worse, the spacetimes of standard
cosmological models are not stationarytechnically, they do not admit, even locally,
a timelike Killing vector eld17and there seems to be no natural way to single out a
preferred vacuum state when quantum eld theory is formulated on such a
background spacetime (see Wald, 1994). Nevertheless, there is at least one
circumstance in which there does appear to be a genuine problem in the
neighborhood of the cosmological constant problem.
Q
Consider the case where the stress-energy tensor of quantum elds Tmn
arises from
a scalar eld F: The stress-energy tensor of F takes the form
1
F
rm Frn F  gmn ra Fra F V F;
Tmn
2

15

16
Hence, the strong energy condition is violated for this uid if L > 0: If Lo0 the weak energy condition
is violated.
17
A Killing vector eld V m is one satisfying rn V m 0: This is the necessary and sufcient condition for
the existence of a local coordinate system in which the components of the metric are independent of the
time coordinate.

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

567

where V F is the potential of the eld. In a regime where the eld is not changing,
(15) reduces to 12 V Fgmn ; which means that (supposing V F > 0) in this regime
LF : V F=2 functions as an effective positive cosmological constant. Thus, the
questionable condition of Poincare! invariance does not have to be invoked for a
surrogate cosmological constant to emerge. The quantum expectation value of LF ;
whether computed from some preferred vacuum state or not, must be consistent with
observational limits on lambda.
The lower case interpretation of lambda has been touted as a way of solving, or
avoiding, the cosmological constant problem (see Van der Bij, Van Dam, & Ng,
1988; Zee, 1985; Weinberg, 1989). The idea is that this reading of lambda leads to the
F
F
trace-free EFE (8), and the trace-free part Tmn
 14 T F gmn of Tmn
does not contain
V F: This apparent avoidance of the problem is only partially successful since, by
themselves, the trace-free eld equations do not comprise a complete theory of
gravitation. Also needed is the local law of conservation of energy-momentum. In
the present instance, this law can be obtained as follows. The action for the scalar
eld F is
Z
1 p mn
LF 
gg rm Frn F V F d4 x:
16
2
Variation with respect to the metric yields (15). Variation with respect to F yields the
equation of motion of F:
rb rb F 

dV
0;
dF

17

which is the (generalized) KleinGordon equation. Together with the trace-free EFE
and the Bianchi identity, the equation of motion of F gives the conservation law
F
rm Tmn
0: And this conservation law, in combination with the trace-free EFE, gives
the condition (9), which in the present case reads R T F 4l: The upshot is that
the trace term re-enters the picture.
But on the lower case reading of lambda, the trace term reenters in a context where
there is more exibility. Since l is not a universal constant but only a constant of
integration, it can be chosen to be equal to V F=2 for the value of F when the eld
F
is unchanging and when Tmn
gives rise to an effective cosmological constant term. In
this way the effective cosmological constant arising from the eld F is squelched by
the real l: But since l must be the same throughout spacetime, the chosen value of l
F
will manifest itself in regimes when Tmn
does not give rise to an effective cosmological
constant term. Thus, one form of the cosmological constant problem has simply
been traded for another. If there is a real problem to begin with, lower case lambda is
not a panacea.

5. Variational principles for lower case lambda and unimodular gravity

It has become routine in physics to demand that equations of motion be derivable
from an action principle. This demand surely owes much of its allegiance to the

ARTICLE IN PRESS
568

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

desire to keep open the standard route to quantization. Another motivation comes
from the desire to connect symmetries with conservation principles; that connection
is provided by Noethers theorems, but only if an action principle is available and
only if the symmetries are variational symmetries.18 But whatever its motivation, the
derivation from an action principle of the lower case lambda version of general
relativity is the departure point for the discussion of the gauge principle obeyed by
this theory. That discussion will commence in the following section. In the present
section, I will review two approaches to the derivation of lower case lambda from an
action principle. Both approaches reveal the close connection between lower case
lambda and what is called unimodular gravity.
The rst approach aims to modify the standard action (6) in such a way that the
trace-free EFE (8) are the extrema. A crude way to achieve this aim ispto
to
restrict

coordinate systems that satisfy the unimodular coordinate condition g 1 and

to limit variations to ones that respect this condition. Since such variations satisfy
gmn dgmn 0; only the traceless part of dLG =dgmn dLM =dgmn need vanish, and the
upshot is the eld equations (8). If one is made nervous about the loss of general
covariance involved in this procedure, one can adopt the stratagem of Unruh (1989)
and Unruh and Wald (1989) for achieving formal general covariance. The idea is to
use a xed background volume element eabgd eabgd and to require that the
unimodular condition is satised when the determinant of the metric is calculated
according to g : gab ggd gmn grs eagmr ebdns ; where eabgd is determined by eabgd eabgd 4!
The local conservation law (3) is not an automatic consequence of this approach, but
typically (3) does follow when the equations of motion for the matter elds are
includedas is illustrated by the case of the scalar eld treated in the preceding
section. And from (8) and (3), Eqs. (4) and (5) can be recovered.
Although formal general covariance is satised in the UnruhWald approach, the
diffeomorphism group of the spacetime manifold M is not a variational symmetry of
the modied action since such a symmetry must preserve the absolute object
eabgd :19 However, on a compact manifold M the UnruhWald approach is equivalent
to an approach which uses an action principle that does admit the full
diffeomorphism group as a variational symmetry. (For a spacetime that is
diffeomorphically S  R; for some compact three-manifold S; M can be taken to
be the compact portion of spacetime given by, say, S  0; 1 :) On a compact M the
total four-volume computed from eabgd is well-dened and is, of course, a
diffeomorphic invariant. And, in fact, the four-volume is the only diffeomorphic
invariant of a volume form on a compact manifold (Bombelli, 1991). This suggests

18
And only if the symmetries form a Lie group. Any variational symmetry is necessarily a symmetry of
the equations of motion that follow from the action principle; but the converse does not hold in general.
For a discussion of the connection between the Noether theorems and the issues taken up in this and the
following section, see Brading and Brown (2003) and Earman (2002b, 2003).
19
Philosophers of space and time use the term absolute object in a number of ways. In the present
context the distinction between dynamical objects and absolute objects is drawn in terms of what is
and what is not varied in the action principle.

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

569

looking for an action principle that xes the volume to some constant value K:
A suitable action is
Z

Z
p 4
p 4
0
gR d x  2l
g d x  K
LG
18
M

where l is now a single Lagrange multiplier independent of x (see Sorkin, 1987, 1994;
Bombelli, 1991). Variation with respect to l gives
Z
p 4
g d x K:
19
M

R p
R
The unimodular condition implies that M g d4 x M d4 x; and if the latter is
identied with K; (19) is satised.
In effect, the transition has already been made to the second approach which aims
to produce (4) and (5) directly; for variation of L0G LM with respect to gmn gives
(4), and (5) is automatic since l in (18) is not a function of x: However, the approach
using (18) has the drawback of being conned to manifolds of the form S  R with
compact S; and for this reason I will ignore it in what follows. For the general case it
appears necessary to introduce some additional object elds in order that the second
approach can succeed. Henneaux and Teitelboim (1989) introduce an additional
dynamical eld in the form of a vector density Jm ; and take the gravitational action
to be
Z
p
L00G
gR  2l 2l@m Jm d4 x:
20
M

Variation of L00G LM with respect to gmn produces (4), while variation with respect
to Jm gives (5). Finally variation with respect to l gives
p
@m Jm g:
21
The meaning of (21) is illuminated by supposing that the spacetime manifold M is
diffeomorphically S  R; with S a compact three-manifold. Then integrating (21)
over the spacetime region VCM between two time slices St1 and St2 gives
Z
Z
p 4
g d x four-volume between St1 and St2 ;
@m Jm d4 x
V

Tt2  Tt1
Z

J4 d3 x:

Tt :

22

St

Further illumination is provided by a variant of the second approach due to

Kucha$r (1991). Here, four scalar elds X A X a ; X 4 ; A 1; 2; 3; 4 and a 1; 2; 3
(unimodular coordinates) are introduced, subject to the conditions that the
hypersurfaces X 4 const: are spacelike, and the reference lines X a const: are
timelike. The gravitational action is rewritten as
Z
p
* d4 x;
L000

gR  2l 2lX
23
G
M

ARTICLE IN PRESS
570

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

where X* given by
1
X* : dABCD XaA XbB XgC XdD dabgd ;
4!
XmE : @m X E ;

24

is the Jacobian of the transformation xa -X A : Variation of (22) with respect to l

now gives
p
X* g;
25
p

which says that, in coordinates that coincide with the X A ; g 1; justifying the
moniker unimodular coordinates. Kucha$r (1991) shows that X* can be written as
the divergence of a vector density so that the present formalism is intertranslatable
with that of HenneauxTeitelboim.
The exercise of deriving the lower case lambda eld equations from an action
principle turns out to have an important payoff for the interpretation of the resulting
theory of gravity. What is revealedif one believes in the efcacy of action
principlesis that the proper setting for lower case lambda is not a standard general
relativistic spacetime, as characterized exclusively by a four-dimensional manifold M
and a Lorentz signature metric gmn : Rather, the proper setting is revealed to be
a general relativistic spacetime enriched by some additional structure, consisting
either of a xed background volume form eabgd or else by some additional dynamical
elds, Jm or X A : This revelation in turn raises questions about the general
covariance of the lower case version of gravity. The eld equations that are derived
from the Unruh or the HenneauxTeitelboim or the Kucha$r action principles are
all formally generally covariant in the sense that they hold in all coordinate systems.
And yet, intuitively speaking, there is some sense in which the spirit of general
covariance is violated, since the additional object elds have the effect of privileging
unimodular coordinates. Thus, the lower case interpretation of lambda and the
associated unimodular theory of gravity provide an interesting example which needs
to be taken into account in the (seemingly never ending) debate about the meaning
and status of general covariance.20 I will pursue this matter in the following section.
To lay the groundwork, note that what physicists regard as important about general
covariance is not the demand that equations be written in a form valid in
all coordinate systemsa demand that can be achieved even for Newtonian and
special relativistic theoriesbut, rather, the demand that the spacetime diffeomorphism group be the (or a subgroup of the) gauge group of the equations
of motion.21 In the Lagrangian formalism, the demand takes the form of requiring
that the equations of motion be derivable from an action principle that admits the
diffeomorphism group as a variational symmetry. This requirement rules against
formulations of unimodular gravity that use absolute background objects like eabgd
20

For a comprehensive overview of the history of this debate, see Norton (1993).
Philosophers of science are beginning to take a more active interest in gauge principles; see, for
example, the PSA 2000 symposium papers on this topic by Earman (2002a) and Martin (2002), and also
Belot (2003).
21

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

571

and in favor of those that use the dynamical objects Jm or X A : The implications of
diffeomorphism invariance for the ofcial account of gauge symmetries, which is
formulated in terms of the Hamiltonian formalism, will be taken up in the next
section.
When lower case lambda is situated in the enriched spacetime setting suggested
by the action principle derivation, the issue of empirical indistinguishability of
theories has to be revisited. In this new setting, the following dilemma seems to
arise: Either the additional elds needed for the derivation of the lower case version
of lambdaeabgd or Jm or X A are empirically detectable or they are not. If
they are, then theories that embody capital and lower case lambda respectively
are empirically distinguishable. If the additional elds are not detectable, then the
allegedly new spacetime setting for lower case lambda is a sham because it uses
the sleight of hand of introducing ghost variables. The second horn of this
dilemma is much too crude. In general, it is implausible to require of a physical
theory that the quantities it deems to be physically signicant must be connectable,
one by one, to observation or measurement. But even apart from this crudity,
the dilemma misses a point that is crucial in the present context. Let us agree to
use the term observable as a term of art, not to denote a quantity whose values
can be read off by direct observation, but rather to denote a genuine physical
magnitude that is supposed to correspond to a natural feature of reality and
that connects, perhaps quite indirectly, to observation and measurement. Then
it would seem that a necessary condition for a quantity to qualify as an observable
in the intended sense is that the quantity be gauge invariant. And if diffeomorphism invariance of the theory is to be treated as a gauge symmetry, then not
even scalar eldsexcept the trivial ones that are constantwill qualify as
observables.
This naturally raises the question of what the observables of such a theory are.
When the question is pursued within the Dirac formalism for constrained
Hamiltonian systems, it is found that the set of observables recognized by standard
GTR is different from the set of observables recognized by unimodular gravity.22
Moreover, the difference in the two sets of observables is signicant enough that
unimodular gravity was for a time touted as a possible solution to the problem of
time and change in canonical quantum gravity.

6. Observables, gauge freedom, and the problem of time in canonical quantum

gravity
For theories whose equations of motion are derivable from an action principle
(and, thus, are in the form of generalized EulerLagrange equations), the standard
way to get a x on the gauge freedom of the theory is to shift from a Lagrangian to a
Hamiltonian formulation and then to apply Diracs algorithm for constrained
22

Other differences between unimodular gravity and standard GTR with L emerge in the context of
quantum cosmology; see Daughton, Louko, and Sorkin (1998).

ARTICLE IN PRESS
572

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

Hamiltonian systems.23 When non-trivial gauge freedom is present, it announces

itself in the Hamiltonian formulation in terms of constraints. First, the denition of
the canonical momenta in terms of derivatives of the Lagrangian density with respect
to the time rate of change of the conguration variables imposes constraints on the
momentathe primary constraints. Next, secondary constraints may emerge when it
is demanded that the primary constraints be preserved by the motion. Then tertiary
constraints may emerge as the price for preserving the secondary constraints, etc.
Eventually, after a nite number of steps, this algorithm terminates. The constraint
surface is then dened as the subspace of the Hamiltonian phase space where all the
constraints are satised. The first class constraints are identied as those that
commute with all the constraints in the sense that the Poisson bracket of any rst
class constraint with any constraint vanishes on the constraint surface (or, in the
jargon, is weakly zero). Finally, points of the constraint surface are counted as
gauge equivalentin the sense that they correspond to the same physical state of
affairsjust in case they are connected by a transformation generated by the rst
class constraints. A dynamical variablea function from the phase space to Ris
deemed to be an observable if and only if it is gauge independent in the sense that
it commutes with the rst class constraints or, equivalently, is constant along the
gauge orbits.24
When applied to standard GTR, with or without L; the Dirac procedure yields the
following verdict. The Hamiltonian phase space for GTR consists of pairs
hab y; pab y where hab is the Riemann three-metric induced on an arbitrary time
slice S by the spacetime metric, pab is the conjugate momentum, and yAS: (For
future reference, note that the spacetime metric gmn is decomposed into hab ; the shift
vector N a [which is tangent to S], and the lapse function N [which measures the
distance along an orthogonal to S] as follows: gab hab ; Na N a hab gb4 ; and
g44 Na N a  N 2 :) The rst class constraints are comprised of the super-momenta
and the super-Hamiltonian constraints:
Ha y 0 Hy;


1
H : h1=2 pab pab  p2  h1=2 R3 super-Hamiltonian;
2
Ha : 2pbajb super-momenta;
23

26

A detailed exposition of Diracs method is to be found in Henneaux and Teitelboim (1992).

A key reason for seeing gauge freedom at work is to overcome the apparent failure of determinism
that results from the presence of arbitrary functions of time in solutions to the Hamiltonian equations of
motion. The apparent failure is blamed on the fact that not all the canonical variablesthe qs (the
conguration variables) and ps (the conjugate momenta)are observable ( gauge invariant). As
Henneaux and Teitelboim put it: [A]lthough the physical state is uniquely dened once the set of ps and
qs is given, the converse is not truei.e., there is more than one set of values of the canonical variables
representing a given physical state. To see how this conclusion comes about, we notice that if we give an
initial set of canonical variables at time t1 and thereby completely dene the physical state at that time, we
expect the equations to fully determine the physical state at other times. Thus, by denition any ambiguity
in the value of the canonical variables at t2 at1 should be a physically irrelevant ambiguity (1992,
pp. 1617).
24

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

573

where h : dethab ; R3 is the Ricci curvature scalar of the three-metric hab ; and the
vertical bar indicates covariant differentiation with respect to hab : When the
momentum
functions Ha are smeared with an arbitrary shift vector N a
R a constraint
4
(i.e., N Ha d x) they generate the gauge change in a dynamical variable F hab ; pab
that corresponds to change generated by performing an arbitrary diffeomorphism on
S: And when the Hamiltonian
constraint function H is smeared with an arbitrary
R
lapse function N (i.e., NH d4 x), it generates the gauge change in a dynamical
variable that corresponds to evolving the initial data via the equations of motion a
distance N normal to S:25
The interpretational puzzle raised by this verdict arises from the fact that
dynamical motion in standard GTR is seen as pure gauge. Put in terms of the
concept of an observable, the dynamics appear to be frozen in that a dynamical
variable F hab ; pab is an observable in the ofcial sense only if it is a constant of the
motion. This puzzle becomes a real problem when quantization is attempted along
#a
the Dirac route by promoting the constraint functions Ha and H to operators H
#
and H on a suitable Hilbert space and identifying the physical sector of this space as
the subspace spanned by the state vectors c that are annihilated by the constraints:
# a c 0 Hc:
#
H
27
The second of the equations (27) (called the WheelerDeWitt equation) is a kind of
.
degenerate Schrodinger
equation in which there is no time dependence, a feature that
is thought to pose insuperable technical and interpretational problems for the
quantum formalism (see Kucha$r, 1992).
In the case of unimodular gravity written, say, in the Kucha$r form, the verdict of
the Dirac procedure is interestingly different. The phase space is now enlarged to
X A ; PA ; hab ; pab where the PA are the momenta conjugate to X A : In standard GTR
the X A play no role in the action and are pure gauge, so the conjugate momenta
satisfy PK 0: Unimodular gravity implies the weaker constraints
& 0;
28
Pa : X A PA 0; P4 y  lXy
a

where X& : det@b X a : In addition, the constraints (26) of standard GTR are
replaced by
Ha y 0 Hy lh1=2 y:

29

A non-zero value for lower case l unfreezes the dynamics because there are then
dynamical variables that commute with the new constraints (28) and (29) but do not
commute with Hy and, thus, are not constants of the motion. And Dirac
.
constraint quantization of the present formalism leads to a Schrodinger
equation in
which there is a non-trivial time dependence. This last feature led a number of
25

These remarks do not sufce to explain the precise sense in which the spacetime diffeomorphism
invariance of GTR is expressed in the Dirac constraint formalism. Isham and Kuchar (1986a, b) show that
when the Hamiltonian phase space of GTR is enlarged by appropriate embedding variables and their
conjugate momenta, there is a natural homomorphism of the Lie algebra of the diffeomorphism group
into the constraint algebra. This result justies calling the spacetime diffeomorphism group a gauge group
of GTR.

ARTICLE IN PRESS
574

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

authors to think that unimodular gravity offers a possible solution to the problem of
time in canonical quantum gravity (see Unruh, 1989; Unruh & Wald, 1989; Brown &
York, 1989; Bombelli, 1991).
Unfortunately a detailed analysis indicates that the quantization of gravity via
Dirac constraint quantization applied to unimodular gravity is unsatisfactory (see
Kucha$r, 1991). At rst sight, unimodular gravity seems to provide a global measure
t of cosmological timenamely, t is the four-volume between a duciary
hypersurface X 4 x 0 and the embedding X A x: This measure is gauge invariant
since the change generated by (28)(29) does not affect the four-volume. However, a
value of t does not label a particular spacelike hypersurface since a given value of t
may correspond to innitely many hypersurfaces. Nor do hypersurfaces corresponding to different values of t provide a causal ordering since a hypersurface
corresponding to t1 can intersect a hypersurface corresponding to t2 at1 : Thus
t cannot set the conditions for quantum measurements.
Despite its inability to solve the problem of time in quantum gravity, lower case
lambda and unimodular gravity pose interesting issues for philosophers who are
interested in gauge principles. And unimodular gravity appears to serve as a template
for what happens when one tries to nesse the requirement of general covariance.26
Start with a theory that is generally covariant, not only in the sense that its equations
of motion are formally generally covariant, but also in the sense that these equations
are derivable from an action principle that admits the spacetime diffeomorphism
group as a variational symmetry. Add a coordinate condition that breaks general
covariance (the unimodular condition or some other). Then restore formal general
covariance by introducing four scalar elds FA ; A 1; 2; 3; 4; that express the special
covariance-breaking coordinates as functions of arbitrary label coordinates xi :
Restore diffeomorphism invariance by providing a modied action principle that
incorporates the expression of the coordinate condition in terms of the FA and
admits the diffeomorphism group as a variational symmetry. This demand, together
with the requirement that the new action reduces to the original action when the FA
are used as label coordinates, generally uniquely determine the new action (see
Kucha$r, 1991). Then starting from this parameterized action, the Dirac constraint
formalism can be run for the parameterized theory to nd the new rst class
constraints and the new set of Dirac observables. Generally the constraints are
weaker and, thus, the set of observables is larger than for the non-parameterized
theory. At this juncture, it does seem appropriate to ask for an account of how the
dynamical variables that are classied as observables in the parameterized but not in
the non-parameterized theory are connected to observation and measurement. If no
satisfactory account is forthcoming, then the attempted nesse of general covariance
can be dismissed as a mere technical trick. (If this be positivism, lets make the most
of it.)

26
The concern being raised here is related to but not the same as the widely discussed Kretschmann
worry that any theory can be re-engineered so as to satisfy general covariance; see Sorkin (2002) and
Earman (2002b) for a discussion of the latter worry.

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

575

What would be perplexing for the status of general covariance is a nding that the
identication of observables given by the parameterized theory incorporating the
elds FA combines with the favored theory of measurement27 of observables to yield
results that are empirically correct. For then there would be an empirically adequate
theory that both giveth and taketh away general covariance: the theory satises the
usual standards for general covariance; and yet the theory entails a condition on the
FA which, when the FApare
used as label coordinates, reduces to a restriction on
coordinate systems (e.g. g 1). One way to resolve the perplexity is to claim that
since, by hypothesis, the theory is empirically adequate, it gives a satisfactory
explanation of why the FA behave in such a way that it is as ifbut only as if
general covariance is broken and, thus, that the spirit as well as the letter of general
covariance is fullled. Those who nd this response unsatisfying are obligated to nd
a better and stronger formal criteria for general covariance that will disbar the
parameterized theory.

7. Conclusion
The shape of spacetime and the fate of the universe literally turn on whether or not
there is a positive cosmological constant, in either the L or the l sense.
Unfortunately, an observational decision on this matter may be far off since the
effects of a positive lambda on the currently observed accelerating expansion of the
universe can be mimicked in models of dark energy that do not imply, as does a
positive lambda, that the expansion of the universe will continue to accelerate ad
infinitum. While waiting for a decision, we can try to understand better the
implications of lambda, especially the lower case sense. In this paper I have
examined claims to the effect that lower case lambda offers solutions to various
problems in physics-in particular, the problem of the stability of matter, the
cosmological constant problem, and the problem of time in quantum gravity. All of
the claims have been found to be wanting. This is a disappointment for physicists,
but not for philosophers of science since understanding why the attempted solutions
fail provides illumination on the foundations of general relativistic theories of
gravity. And the lower case lambda version of GTR serves as a fascinating case study
for probing the concept of observables and the nature and status of the requirement
of general covariance. Nor are these issues merely hypothetical. Although physicists
may have lost interest in unimodular gravity as a solution to various problems, they
will have to face up to a choice between the upper case and lower case versions of
lambda if cosmological observations continue to indicate that a positive
cosmological constant is with us.
27

What complicates the situation is that if the theory we are dealing with aspires to be a fundamental
theory, then it must, as it were, provide its own theory of measurement; that is, measurement instruments
cannot be treated either as primitives or as objects whose behavior is left to be explained by some auxiliary
theory, but must be analyzed within the theory itself. As a rst step, however, it may be permissible to take
the observables of the theory to refer to the deliverances of measurement instruments whose behavior is
antecedently understood.

ARTICLE IN PRESS
576

J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

Acknowledgements
I am grateful to John Norton, Rafael Sorkin, and Roberto Torretti for helpful
comments on an earlier draft of this paper.

References
Adler, R., Bazin, M., & Schiffer, M. (1975). Introduction to general relativity (2nd ed). New York:
McGraw-Hill.
Belot, G. (2003). Symmetry and gauge freedom. Studies in History and Philosophy of Modern Physics, 34B,
189225.
Bombelli, L. (1991). Unimodular relativity, general covariance, time, and the Ashtekar variables. In R.
Mann, & P. Wesson (Eds.), Gravitation: A Banff Summer Institute (pp. 221232). Singapore: World
Scientic.
Brading, K. A., & Brown, H. R. (2003). Symmetries and Noethers theorems. In K. Brading, & E.
Castellani (Eds.), Symmetries in physics: Philosophical perspectives. Cambridge: Cambridge University
Press, in press.
Brown, J. D., & York, J. W. (1989). Jacobis action and the recovery of time in general relativity. Physical
Review D, 40, 33123318.
Daughton, A., Louko, J., & Sorkin, R.D. (1998). Instantons and unitarity in quantum cosmology with
xed four-volume. Physical Review D, 98 084008-1-17.
Earman, J. (2001). Lambda: The constant that refused to die. Archive for History of the Exact Sciences, 55,
189220.
Earman, J. (2002a). Gauge matters. In J. A. Barrett, & J. McKenzie Alexander (Eds.), PSA 2000, Part II
(S209-S220), Philosophy of Science, Supplement to Volume 69, Number 3.
Earman, J. (2002b). Once more general covariance. Pre-print.
Earman, J. (2003). Getting a grip on gauge: An ode to the constrained Hamiltonian formalism. In K.
Brading, & E. Castellani (Eds.), Symmetries in physics: Philosophical perspectives. Cambridge:
Cambridge University Press, in press.
Efstathiou, G., et al. (2002). Evidence for a non-zero L and a low matter density from a combined analysis
of the 2dF galaxy redshift survey and cosmic microwave background anisotropies. Monthly Notices of
the Royal Astronomical Society, 330, L25L35.
Einstein, A. (1917). Kosmologische Betrachtungen zur allgemeinen Relativit.atstheorie. Koniglich
.
Preussische Akademie der Wissenschaften (Berlin) Sitzungsberichte (pp. 142152). [English translation
as Cosmological considerations on the general theory of relativity, in Perrett & Jeffrey, 1952,
pp. 77188.]
Einstein, A. (1919). Spielen Gravitationsfelder im Aufbau der materiellen Elementarteilchen eine
wesentliche Rolle? Koniglich
Preussische Akademie der Wissenschaften (Berlin) Sitzungsberichte (pp.
.
349356). [English translation as Do gravitational elds play an essential part in the structure of
elementary particles? in Perrett & Jeffrey, 1952, pp. 191198.]
Einstein, A. (1931). Zum kosmologischen Problem der allgemeinen Relativit.atstheorie. Preussische
Akademie der Wissenschaften (Berlin), Sitzungsberichte (pp. 235237).
.
Friedmann, A. (1922). Uber
die Krummung
.
des Raumes. Zeitschrift fur
. Physik, 10, 377386.
Gamow, G. (1958). The evolutionary universe. In The universe. A scientific American book (pp. 5976).
London: Bell.
Gamow, G. (1970). My world line. New York: Viking Press.
Henneaux, M., & Teitelboim, C. (1989). The cosmological constant and general covariance. Physics
Letters B, 222, 195199.
Henneaux, M., & Teitelboim, C. (1992). Quantization of gauge theories. Princeton, NJ: Princeton
University Press.

ARTICLE IN PRESS
J. Earman / Studies in History and Philosophy of Modern Physics 34 (2003) 559577

577

Isham, C. J., & Kuchar, K. (1986a). Representations of spacetime diffeomorphisms. I. Canonical

parametrized eld theories. Annals of Physics, 164, 316333.
Isham, C. J., & Kuchar, K. (1986b). Representations of spacetime diffeomorphisms. II. Canonical
geometrodynmamics. Annals of Physics, 164, 288315.
Krauss, L. M., & Starkman, G. D. (2000). Life, the universe, and nothing: Life and death in an everexpanding universe. Astrophysical Journal, 531, 2230.
Kucha$r, K. (1991). Does an unspecied cosmological constant solve the problem of time in quantum
gravity? Physical Review D, 43, 33323344.
Kucha$r, K. (1992). Time and the interpretation of quantum gravity. In G. Kunsatter, D. Vincent, & J.
Williams (Eds.), Proceedings of the 4th Canadian conference on general relativity and relativistic
astrophysics (pp. 211314). Singapore: World Scientic.
Martin, C. (2002). Gauge principles, gauge arguments, and the logic of nature. In J. A. Barrett & J.
McKenzie Alexander (Eds.), PSA 2000, Part II (S221S234), Philosophy of Science, Supplement to
Volume 69, Number 3.
Norton, J. D. (1993). General covariance and the foundations of general relativity: Eight decades of
dispute. Reports on Progress in Physics, 56, 781858.
Perlmutter, S., et al. (1999). Measurements of O and L from 42 high redshift supernovae. Astrophysical
Journal, 517, 565586.
Perrett, W., & Jeffrey, G. B. (Eds.) (1952). The principle of relativity. New York: Dover Publications.
Roos, M. (1994). Introduction to cosmology. New York: Wiley.
Rugh, S. E., & Zinkernagel, H. (2002). The quantum vacuum and the cosmological constant problem.
Studies in History and Philosophy of Modern Physics, 33, 663705.
Sorkin, R. D. (1987). A modied sum-over-histories for gravity. Talk given in the workshop, Quantum
Gravity and New Directions at the International Conference on Gravitation and Cosmology, Goa, India,
1419 December 1987.
Sorkin, R. D. (1994). The role of time in the sum-over-histories framework for gravity. International
Journal of Theoretical Physics, 33, 523534.
Sorkin, R. D. (2002). An example relevant to the Kretschmann-Einstein debate. Modern Physics Letters A,
17, 695700.
Unruh, W. G. (1989). Unimodular theory of canonical quantum gravity. Physical Review D, 40, 1048
1051.
Unruh, W. G., & Wald, R. M. (1989). Time and the interpretation of canonical quantum gravity. Physical
Review D, 40, 25982614.
Van der Bij, J. J., Van Dam, H., & Ng, Y. J. (1988). The exchange of massless spin-two particles. Physica
A, 116, 307320.
Wald, R. M. (1994). Quantum field theory on curved spacetime and black hole thermodynamics. Chicago:
University of Chicago Press.
Weinberg, S. (1989). The cosmological constant problem. Reviews of Modern Physics, 61, 123.
Zee, A. (1985). Remarks on the cosmological constant problem. In B. Kursunoglu, S. L. Mintz, & A.
Perlmutter (Eds.), High energy physics (pp. 211230). New York: Plenum Press.