Background-Independence

arXiv:1106.0920v1 [gr-qc] 5 Jun 2011

Gordon Belot
University of Michigan
[email protected]
June 7, 2011

Abstract
Intuitively speaking, a classical field theory is background-independent
if the structure required to make sense of its equations is itself subject
to dynamical evolution, rather than being imposed ab initio. The
aim of this paper is to provide an explication of this intuitive notion.
Background-independence is not a not formal property of theories: the
question whether a theory is background-independent depends upon
how the theory is interpreted. Under the approach proposed here, a
theory is fully background-independent relative to an interpretation if
each physical possibility corresponds to a distinct spacetime geometry;
and it falls short of full background-independence to the extent that
this condition fails.

Forthcoming in General Relativity and Gravitation.

Introduction

In in pre-relativistic physics, as in common sense, space and time provide a

fixed scaffolding or stage against which physical systems evolve.
Ask an intelligent man who is not a scholar what space and time
are, and he will perhaps answer as follows. If we imagine all
physical things, all stars, all light taken out of the universe, what
then remains is something like a giant vessel without walls called
space. With respect to what is happening in the world, it plays
the same role as the stage in a theater performance. In this
space, in this vessel without walls, there is an eternally uniformly
occurring tick-tock . . . that is time. Most natural scientists, up
to the present, had this conception about the essence of space
and time . . . . (Einstein [26, item 44a])
It is often suggested that what makes general relativity (and theories of its
ilk) special is their background-independencetheir treatment of space and
time as among the actors in the theory rather than as a fixed stage upon
which the action unfolds.
Background-independence has no precise and fixed meaning. But according to the sense in which the term is used here, general relativity (in
its usual formulation and under its usual interpretation) is a paragon of
background-independence, while the theory of a KleinGordon field in Minkowksi
spacetime is a paragon of background-dependence (whether or not the spacetime metric is treated as a dynamical variable).1
Intuitively, a background-independent theory must be generally covariant
any structure that breaks general covariance is background structure. And if
a theory is generally covariant without being background-independent, then
its general covariance must in some sense be artificial or dispensableone
1

In fact, of course, the term background-independence is used in many ways. The

use in the present discussion is very close to that found in, e.g., [19] and [37]. Sometimes,
however, background-independence is used as a synonym for generally covariant [42]
under this usage, a theory of a KleinGordon field propagating against a Minkowski metric
counts as background-independent if the spacetime metric is treated as a variable in the
equations of motion. Again, background-independence is sometimes used as a name
for a sweeping form of relationalism [41]and then even general relativity is not fully
background-independent, because the topological structure of spacetime and the signature
of the metric are fixed ab initio. For another approach, see [22].

expects general covariance to be broken in any formulation in which the background is made explicit. So, speaking very roughly and intuitively, a theory
is background-independent if and only if its most perspicuous formulation is
generally covariant (but see the discussion of the EinsteinMaxwell theory in
Section 7.4 below).
The question whether future theories of physics can be expected to be
background-independent and the question whether various proposed approaches
to quantum gravity live up to this expectation have been very widely discussed in recent years. The goal of the present discussion is to offer an
explication of the intuitive notion of background-independence as it applies
to classical field theories with the hope that clear standards of backgroundindependence will provide a framework to structure such debates.
It might be thought that the notion of an absolute object [1, 2, 14, 15, 16]
should play a crucial role here. For an absolute object of a theory is a
field that is in a certain sense the same in each solutionand so it might
naturally be suggested that a theory is background-independent if and only
if it features no absolute objects. However, this suggestion runs into serious
difficulties (see Remark 3.1 below). A different one is offered in its place
roughly speaking, that a theory is fully background-independent if alteration
of the physical degrees of freedom always implies an alteration in geometry;
and that a theory falls short of full background-independence to the extent
that this condition fails.
The analysis is intended to apply only to theories of the universe as a
whole in which all interactions between systems are taken into account. The
clearest cases of background-dependent theories are theories featuring fixed
(i.e., solution-independent) fields. Such fields can play a number of roles
in a theorybut in practice, they are typically used to represent geometry,
fixed sources, or external fields. Plausibly, in a theory in which background
structure is present without being encoded in fixed fields, the background
still plays one of these same roles. By restricting attention to theories of
the universe as a whole in which all interactions are taken into account, the
possible roles of background are narrowed down to a single onegeometry.
In Section 2 below, a framework for talking about field theories and
their symmetries is sketched. Section 3 includes a preliminary discussion of
background-independence illustrated by several examples. Sections 4 and 5
develop the notions of geometrical degrees of freedom and of physical degrees
of freedom that are required for the present approach. Various gradations of
background-(in)dependence are defined in Section 6. Features and limits of
3

these notions are discussed in Section 7.

Conventions. All Lie groups are taken to be connected. Where g denotes
a metric, Ein[g], Weyl[g], Riem[g], Ricci[g], and R[g] denote, respectively,
the corresponding Einstein tensor, Weyl tensor, Riemann tensor, Ricci tensor, and scalar curvature.

Field Theories

For present purposes, we can take a classical field theory to consist of the
following elements.
(a) A connected n-dimensional manifold V, the spacetime of the theory.
(b) A set of tensors on V. These are the fixed fields of the theory. is
often the empty set in cases of interest.
(c) A set {1 , . . . , k } of dynamical fields on spacetime. A field is specified
by specifying a type of tensor on V and a particular configuration of the
field by specifying a tensor of that type. A configuration of the complete
set of dynamical fields is denoted = (1 , . . . , k ).
(d) A space K of field configurations, thought of as consisting of the kinematically possible . In typical cases, K is determined by specifying the
smoothness, asymptotic behaviour, etc. of the fields of the theory.
(e) A set of differential equations (; ) that determines the space S K
of solutions of the theory. The fixed fields play the role of parameters
rather than variables in . Any derivative operators appearing in must
be definable in terms of the fixed and dynamical fields of the theory.
The symmetry group, G, of a classical field theory is the group consisting
of diffeomorphisms from K to itself that map solutions to solutions and that
are suitably local on V.2 We call the elements of G the symmetries of the
2

Here is one way to make this notion precise [46]. If G is a one-parameter group of
diffeomorphisms from K to itself, then it makes sense to speak of the infinitesimal generator
of G. will be a vector field on K. In the present setting, a vector T K, K,
can be identified locally with a tensor field on V. We call G local in V if there is some k
such that for any x V the value of () T K when evaluated at x depends only on
the value of and its first k derivatives at x. We call G a one-parameter symmetry group
of the field theory if it is local and maps solutions to solutions. We take G to be the group
generated by all of the one-parameter symmetry groups of the theory.

theory and write for the result of acting on S by a symmetry

G.
We will be especially interested in symmetries corresponding to diffeomorphisms of V. We will denote by Diff(V ) the group of diffeomorphisms
V and by Diffc (V ) the group of compactly supported diffeomorphisms of V
(so d Diff(V ) is in Diffc (V ) if and only if there is a compact set K V
such that d is the identity on V /K). Since we are working with tensor fields,
each field has a well-defined transformation law 7 d under the action
of spacetime diffeomorphisms. Further, the action of diffeomorphisms on the
space of fields is local in V. So the following is a special case of our general
notion of symmetry.
Definition 2.1 (Spatiotemporal Symmetry). A diffeomorphism d : V V
is a spatiotemporal symmetry of a theory if S implies d S.
Definition 2.2 (Local General Covariance). We call a field theory locally
generally covariant if every d Diffc (V ) is a spatiotemporal symmetry of
the theory.
Definition 2.3 (General Covariance). We call a field theory generally covariant if every d Diff(V ) is a spatiotemporal symmetry of the theory.

Examples

In this section, a number of examples are presented as motivation for the

following claims.
(1) Background-independence admits of degrees, with theories falling in between full background-independence and full background-dependence.
(2) A degree of background-dependence can result from the imposition of
asymptotic boundary conditions.
(3) The notion of background-independence is not purely formal: whether a
theory is background-independent depends on how it is interpreted.

3.1

More or Less Background-Dependent Theories

In paradigm background-dependent theories, solution-independent background

structure is encoded in fixed fields.
5

Example 3.1. Consider the theory of a scalar field propagating on Minkowski

spacetime in which the spacetime metric is treated as a fixed field: spacetime is V = Rn (n 2); the only fixed field is ; the only dynamical field is
a scalar field ; the space K of kinematic possibilities is the space of twicedifferentiable ; and the equation of motion is the KleinGordon equation,
 = 0. Here we have full background-dependence: the Minkowski metric
is provides the fixed stage against which the scalar field evolves.
Of course one can have background-dependence even in the absence of
fixed fields.
Example 3.2. Consider the generally covariant analogue of the theory of
the preceding example: Let V = Rn (n 2); let there be no fixed field; let
there be two dynamic fields, a scalar field and a Lorentz-signature metric
g; the space K of kinematic possibilities is the space of twice-differentiable
and g; and the equations of motion are Riem[g] = 0 and g = 0. Each
solution (g, ) just represents a KleinGordon field on a copy of Minkowski
spacetime: the physics is exactly the same as in the preceding example. But
whereas in the preceding example the group of spatiotemporal symmetries
was given by the Poincare group, here it is the group Diff(V ) of spacetime
diffeomorphisms. Despite its general covariance, this theory is backgrounddependent: although the metric g is a dynamical field, the spacetime geometry is solution-independent and plays the role of stage against which the
scalar field evolves.
It possible to cook up variants on this example that, intuitively, fall just
short of full background-dependence.
Example 3.3. Let everything be as in the preceding example, except that
V is taken to have the topology of R Tm , where Tm is the m-dimensional
torus and it is built into the definition of K that g is globally hyperbolic.
Once again, the metric g is flat in any solution. But the theory has
nontrivial global geometric degrees of freedomthe space of metrics modulo
diffeomorphisms is finite-dimensional.3
This theory is of course about as far from being background-independent
as is possibleat the local level, the spacetime geometry is Minkowskian in
every solution, and it is this geometry that determines the behaviour of the
3

For relevant degrees of freedom in the case of three spacetime dimensions, see, e.g.,
[7, 29]. For their operational significance, see [31].

scalar field. But because the global geometry varies from solution to solution,
it is natural to see the theory as falling (just) short of the full background
dependence of the previous example.
The next example is a variation on this theme which raises a point that
will play a role the discussion below.
Example 3.4. Let V R S 3 (S 3 being the three-sphere). We again have
no fixed fields and two dynamic fields, a scalar field and a metric g. The
space of kinematic possibilities is determined by the usual differentiability
conditions together with the condition that g be geodesically complete. The
equations of motion are
Weyl[g] = 0
g
Ein[g] + R[g] = 0
4
g = 0.
This theory describes a scalar field propagating against a de Sitter spacetime
of constant curvature K 2 : if (g, ) is a solution, then (V, g) is isometric to a
timelike hyperboloid of radius K in Minkowski spacetime.4
The equations of motion almost succeed in fixing the spacetime geometry
up to diffeomorphismin this theory, identifying metrics related by a diffeomorphism leaves us with a single geometric degree of freedom, parameterized
by K 2 . If we regard scale transformations as well as diffeomorphisms as relating physically equivalent solutions, then we should regard this theory as fully
background-dependentfor then there are no geometrical degrees of freedom
on the theory. However, if we regard scale transformations as physical, then
we should regard this theory as nearly but not entirely background-dependent
in virtue of possessing a single geometrical degree of freedom.

3.2

More or Less Background-Independent Theories

At the other end of the spectrum from the examples we have been considering
lies spatially compact vacuum general relativity, a paragon of backgroundindependence.
4

The equations of motion force (V, g) to be a space of constant curvature K 2 [23, p.

124]. The assumption that g is geodesically complete then forces (V, g) to be a quotient
of anti-de Sitter spacetime, Minkowski spacetime, or de Sitter spacetime [45, Theorem
2.4.9]. But the topology of V rules out the first two options and forces (V, g) to be de
Sitter spacetime itself.

Example 3.5. Let V be an n-manifold of the form R where is compact

and let the only field be a twice-differentiable globally hyperbolic Lorentzian
metric g subject to the vacuum Einstein equation, Ricci[g] = 0. This theory
is generally covariant. It is also background-independent: the geometry of
spacetime is determined dynamically rather than being imposed ab initio.5
Just as there are theories that fall just short of background-dependence,
there are also theories that fall just short of background-independence.
Example 3.6. Consider the sector of general relativity in which solutions
are asymptotically flat at spatial infinity in the sense of [3]. Here V = R4
and the only field is a Lorentzian metric g subject to the vacuum Einstein
equation that is required to be twice-differentiable, to be globally hyperbolic,
and to approximate Euclidean geometry in a suitable sense at spatial infinity.
A helpful, but dispensable, way to think about these asymptotic boundary conditions appeals to an ideal boundary added to V that represents
spatial infinity. In broadest outline, the picture is as follows. We attach a
non-physical boundary H to V that has the topology and metric of H0 the
unit timelike hyperboloid in Minkowski spacetime; for convenience we fix an
isometry d : H H0 . Let g be a Lorentzian metric on V and let be a
Cauchy surface relative to g. We say that has good asymptotic behaviour
relative to g if the initial data that g induces on represents space as asymptotically Euclidean (and satisfies various additional technical conditions) and
if the set of points H, which we call the boundary of , consisting of
the limit points of on H, has the same form as a set on H0 that arises by
taking the intersection of H0 with an spacelike hyperplane (in the standard
model of Minkowksi spacetime). A metric g on V is in K if by its lights it is
possible to partition V by Cauchy surfaces with good asymptotic behaviour.
The resulting theory is locally generally covariantsince locally its dynamics is given by the Einstein equation. But is is not generally covariant:
only certain diffeomorphisms from V to itself preserve the asymptotic boundary conditions. One can picture this as follows. Let 0 be the set consisting
of all the intersections of H0 with spacelike hyperplanes in Minkowski spacetime. And let := d1 0 be the corresponding set of subsets of H. Any
diffeomorphism f : V V induces a diffeomorphism f : H H. f is a
5

Of course, this is not to say that say that there is nothing that is invariant across
solutions: each solution involves a Lorentz-signature metric, each solution satisfies the
field equation of the theory, etc.

symmetry of our theory if and only if under the obvious action f maps to
itself.
The most natural thing to say is that this theory lies between paradigmatic non-background-independent theories like those in in which fields propagate against the backdrop of Minkowski spacetime and paradigmatic backgroundindependent theories like spatially compact general relativity. On the one
hand, there are no fields on spacetime, fixed or dynamical, that encode a
fixed background structure such as a geometryindeed, locally the field of
the theory has all of the freedom of the metric field of ordinary spatially
compact general relativity. On the other, there is also a sense in which the
boundary conditions of the theory ensure that any solution has the structure
of Minkowski spacetime at infinityand this is reflected in the fact that the
theory is not generally covariant. Perhaps the point is put most vividly by
saying that in this theory one has a class of preferred frames at infinity. An
observation along these lines played an important role in motivating Einstein
to investigate spatially compact cosmologies [25].

3.3

Geometrically Ambiguous Theories

In each of the above examples, there was among the fields of the theory an
obvious candidate to represent the geometry of spacetimeand the question
of the extent to which the theory was background-independent turned on
the question of the extent to which the behaviour of that field was solutionindependent. But it is also possible to find theories featuring more than one
field that could plausibly represent the geometry of spacetime. In such cases
the question of background-independence becomes more subtle.
Example 3.7. Consider a theory modelled on Nordstroms scalar theory of
gravity.6 Let V R4 . Let there be no fixed field and let there be three
dynamical fieldstwo metrics of Lorentz signature, and g, and a positive
scalar field , subject to the usual sort of smoothness conditions. In order to
state the equations of motion, we also introduce a scalar field T that arises by
taking the trace relative to g of the ordinary scalar field stress-energy tensor
for phi relative to g. The equations of the theory fall into three groups. The
6

For discussions of Nordstr

oms theory, see [19, 36, 43]. The theory discussed here
is not quite Nordstr
oms, since in his theory the scalar field coupled to the trace of the
stress-energy tensor of particulate matter.

first group,
Riem[] = 0
 = 4G3 T,

(1a)
(1b)

says that is a flat metric and that is a modified KleinGordon field

with respect to with an interesting source term (here and below G denotes
Newtons constant). The second group,
Weyl[g] = 0
R[g] = 24GT,

(2a)
(2b)

says that g is conformally flat and has scalar curvature proportional to T.

The third group,
g = 2
= (detg)

(3a)
1
8

= g(detg)

(3b)
41

(3c)

encodes relations between g on the one hand and and on the other.
These three sets of equations are highly redundant: in the presence of the
third group, the first and second groups are equivalent to one another.
Is this theory background-independent? This depends upon how we understand the theory.7
On the one hand, one could take and to be the fundamental physical
fields of the theory and understand the theory as describing the propagation
of a non-linear scalar field in Minkowski spacetime (g would then just be a
clever but unphysical way of encoding and in a single object obeying an
elegant equation). Understood this way, the theory is as fully backgrounddependent as the theory of a KleinGordon field in Minkowski spacetime.
Alternatively, we could think of g as the fundamental physical variable,
understood as directly representing the geometry of spacetime. In this case
we would view Equations (2a) and (2b) as giving the laws of a non-linear
field theory of spacetime geometry (Equations (3b), (3c), and (1b) would
7

For discussion of options for interpreting such theories, see [39, 3.III and 3.IV].
Note that if one were to couple this theory to particulate matter in the obvious way (as
Norstr
om in fact did) then physical considerations would exert some pressure in favour of
the second option discussed below [32].

10

then tell us how to rewrite the theory as a theory of a scalar field propagating
against a non-physical flat metric). In this case we would presumably want
to consider the theory as being far from background-dependent: just as in
general relativity, here we have a non-linear theory of a field with infinitely
many degrees of freedom describing a spacetime geometry that is in general
variably curved.

3.4

Morals

(1) Background-dependence and independence come in degrees: some theories are fully background-(in)dependent, others only nearly soand others fall somewhere in between.
(2) A theory can fail to be fully background-independent in virtue of asymptotic boundary conditions.
(3) The extent of the background-(in)dependence of a theory is not a strictly
formal one: in particular, it depends on how one thinks of the geometric
structure of each solution and on what sorts differences between solutions
one takes to be unphysical.
Running through the discussion of the above examples was the idea that
a theory is fully background-independent if each physical possibility corresponds to a distinct spacetime geometry, and that it falls short of full
background-independence to the extent that this condition fails. The task
of the next several sections will be to put in place a framework of concepts
that will allow us to make this idea precise.
Remark 3.1 (Absolute Objects). General relativity was the first physical
theory in which space and time did not have the objectionable feature of acting on matter without being acted upon by it.8 The notion of an object that
acts upon others without itself being acted upon is given a precise sense in
the notion of an of an absolute object due to Anderson and Friedman: an absolute object of a theory is one that is locally the same up to diffeomorphism
in every solution [1, 2, 14, 15, 16].9
8

This was a favourite theme of Einsteins; see [33, 3.9].

This notion has its problemssee [35] for a survey. Foremost among them is the fact
that there are certain types of tensors, such as non-vanishing vector fields or symplectic
forms, that are always locally identical up to diffeomorphismso any field represented by
such an object counts as absolute, no matter what its dynamics.
9

11

Since an absolute object of a theory is a field that is the same in each

solution, it may seem natural to take a theory to be background-independent
if and only if it features no absolute objects. But this suggestion founders
on the morals reached above. For if one identifies background-independence
with the lack of an absolute object then: (i) background-independence will
be an all or nothing affair; (ii) if a theory fails to be background-independent,
this is always in virtue of a the fact that some field of the theory is an absolute
object; and (iii) the question of background-independence is rendered purely
formal.

Geometric Degrees of Freedom

Suppose that one is handed a field theory (in the sense of Section 2 above).
So far it is just a piece of mathematics. In order to endow it with physical
content, one would have to say something about how the fields of the theory
correlate with observable quantities. In particular, if the field theory is to
be understood as an all-encompassing account of the classical world, then
it must be endowed with geometric content: one must have some way of
thinking of each solution of the theory as portraying physical processes in
spacetime.
Let us say that a geometrization for a field theory consists of: (i) a rule
7 g that associates with each solution of the theory a geometric
structure g for the manifold V on which the fields of the theory live; and
(ii) a criterion that tells us when geometries g1 and g2 are geometrically
equivalent.
When we think of a field theory as a relativistic field theory, we are
implicitly thinking of it as endowed with a geometrization that: (i) assigns
to each solution a Lorentzian metric g on V that is among the fields of
the theory or definable in terms of them; and (ii) counts such metrics as
geometrically equivalent if and only they are related to one another by a
diffeomorphism from V to itself.
There are of course other notions of geometrization that arise. Prerelativistic field theories can be thought of as involving a geometrization
that: (i) assigns to each solution a spatial metric, a temporal metric, and
an affine connection; and (ii) counts two such assignments as geometrically
equivalent if and only they are related by a diffeomorphism. One might
also tinker with the standard notion of a relativistic field theory by counting
12

metrics related by scale transformations as geometrically equivalent.

In most of the examples discussed in the preceding section, a default
reading of the theory as a relativistic field theory was available, since one
and only one of the fields of the theory was a Lorentzian metric. In the
final example discussed there were two fields of this typeand the question
arose which one should be taken to describe the spatiotemporal geometry
of solutions. Other situations are of course possible. A field theory may
admit a natural reading as a relativistic field theory even though there are
no Lorentzian metrics among its fields.10 Or a theory that includes one or
more Lorentzian metrics among its fields may admit a natural geometrization
by whose lights the geometry of a solution is given by some further metric
(Nordstroms theory could be put in this form, as could the BransDicke
theory).
Relative to a geometrization for a theory, we can look at the set of spacetime geometries that arise in solutions of the theory and at the set of equivalence classes of such geometries under the relation of geometrical equivalence
of the geometrization. We call any set of variables that parameterize this
latter set the geometrical degrees of freedom of the theory relative to the
geometrization.
Relative to the standard notion of geometrization, the theories of Examples 3.1 and 3.2 above have trivial geometrical degrees of freedom (up to
equivalence, only one geometry occurs); the theories of Examples 3.3 and 3.4
have finitely many geometrical degrees of freedom (the space of geometries
modulo equivalence is finite-dimensional); and the theories of Examples 3.5
and 3.6 have infinitely many geometrical degrees of freedom (the space of
geometries modulo equivalence is the infinite-dimensional space of Ricci-flat
Lorentzian metrics modulo diffeomorphisms).11 Example 3.7, Nordstroms
theory, is more subtle: relative to one way of reading the theory, it has
trivial geometrical degrees of freedom; relative to another, it has infinitely
many.
10

Think here of the formulation of (2 + 1) gravity as a ChernSimons theory [7]. or of

certain parameterized versions of familiar theories [20, 28, 44].
11
Relative to the unorthodox approach that counts metrics related by scale transformations as equivalent, the theory of Section 3.4 has trivial geometrical degrees of freedom.

13

Physical Degrees of Freedom

This section focusses on the way of counting physical degrees of freedom

that is standard among physicists (an alternative system of counting, more
popular among philosophers than among physicists, is mentioned in Remark
5.1 below).
In many cases, one takes the physical degrees of freedom of a theory to
parameterize the space of solutions.12 But there are important exceptions.
The equations of motion of a classical theory are said to be underdetermined if they are not independent of one another; for a precise characterization, see [34, p. 171]. The most prominent types of theories with underdetermined equations include generally covariant theories and theories of
YangMills type. But there are many other examples [24]. Uniqueness of solutions fails radically in theories with underdetermined equations of motion:
the family of solutions corresponding to an admissible set of initial data is
infinite-dimensionalroughly speaking, such a family can be parameterized
by arbitrary functions of the independent variables of the theory.
Since a theory is deterministic if and only if each instantaneous state is
compatible with only one global history, there is usually a tight connection
between a failure of uniqueness of solutions in a theory and the failure of that
theory to be deterministic. But this connection obtains only if we assume that
distinct solutions of our theory always represent physically distinct situations.
Faced with the prospect of a wholesale and dramatic failure of determinism
in the presence of underdetermined equations of motion, one usually prefers
to reject this assumption. Standardly one assumes instead that a theory
featuring underdetermined equations involves gauge freedomone assumes,
that is, that some of the degrees of freedom of the theory are unphysical and
that, except perhaps in special cases, specifying instantaneous values of all
variables suffices to determine the past and future behaviour of the physical
degrees of freedom.
The idea is to introduce introduce an equivalence relationgauge equivalenceon the space of solutions, then to identify the physical degrees of
freedom of the theory with variables that parameterize the quotient space
that results when we identify gauge equivalent solutions.
12

Actually this is non-standardthe space of solutions is isomorphic to the systems

phase space, and the degrees of freedom are usually taken to parameterize the systems
configuration space. But it is convenient here to omit the factor of one-half that should
be inserted at various points below.

14

How can we characterize gauge equivalence? Roughly speaking, we want

to consider solutions to be gauge equivalent if the underdetermination of the
theorys equation of motion forces us to consider them physically equivalent,
on pain of taking the theory to be indeterministic. But we cannot just take
two solutions to be gauge equivalent if and only if they induce the same
initial data at some instant of time.
(i) From the fact that solutions 1 and 2 induce the same initial data at
time t1 and the fact that solutions 2 and 3 induce the same initial
data at time t2 , it need not follow that there is a time at which 1 and
3 induce the same initial data. But we need gauge equivalence to be
an equivalence relation, so we need to set things up so that from the
fact that 1 and 2 are gauge equivalent and the fact that 2 and 3
are gauge equivalent, it follows that 1 and 3 are gauge equivalent.
(ii) Simply identifying solutions whenever they induce the same initial data
will efface genuine instances of indeterminism. The Newtonian n-body
problem admits solutions in which a system of particles interacts in
such a way that they all disperse to spatial infinity in finite time [12].
Such a solution matches the trivial empty solution at late times. This
constitutes real physical indeterminism, not an instance of gauge equivalence. Similarly, in the case of spatially compact general relativity, we
will want to consider solutions gauge equivalent if and only if they are
related by a diffeomorphism. Instances in which globally hyperbolic
solutions admit non-isometric extensions are instances of genuine indeterminism, not gauge equivalence; for examples see [8].
There are a number of ways of addressing the problem of characterizing gauge equivalence. For theories in Lagrangian form, one can employ
by-products of the variational procedure to construct a presymplectic form
on the space of solutions of the theory and take two solutions to be gauge
equivalent if and only if they can be connected by a (piecewise) null curve of
this presymplectic form [10, 11, 46]. Alternatively, one can follow the Dirac
constraint algorithm, which leads from a Lagrangian to a notion of gauge
equivalence on the space of initial data of the Hamiltonian formulation of
the theory [13, 21, 24]. These two approaches are closely related to one
another [28].
There is also a low-tech procedure that gives the same answers in standard applications. Let us take for granted that we know, for any solution ,
15

which hypersurfaces in V correspond to initial data surfaces (=instants of

time). Then we can make the following definitions. Let us say that a symmetry of a theory is a spoiler relative to solution if there is some initial
data surface V relative to such that acts as the identity on some
neighbourhood U of .13 We call solutions 1 and 2 spoiler-related if there
is a spoiler relative to 1 such that 2 = 1 . Being spoiler-related is not
an equivalence relation (for the sort of reasons raised in point (i) above). We
call an equivalence relation R an extension of the relation of being spoilerrelated if any two spoiler-related solutions are also R-related. There are
many equivalence relations between solutions that extend the relation of being spoiler-related. One such is the relation that takes any two solutions
to be equivalent. That relation is maximally strong. We are after a much
weaker one (i.e., one according to which fewer pairs of solutions are equivalent). Heuristically, what we want is the weakest equivalence relation that
extends the spoiler-relatedness.14 For technical convenience, we take gauge
equivalence to be the weakest equivalence relation on the space of solutions
that extends spoiler-relatedness and which has equivalence classes that form
submanifolds of the space of solutionsif there is any such relation.
One can show (modulo certain technical questions) that for various sectors
of general relativity this definition is well-formed and gives the same answer
as the high-tech Lagrangian approach [6]. In particular for spatially compact
general relativity, both approaches agree that solutions are gauge equivalent
if and only if they are related by diffeomorphisms. And for well-behaved
asymptotic boundary conditions at spacelike, timelike, or null infinity, both
approaches agree that solutions are gauge equivalent if and only if they are
related by diffeomorphisms that act as the identity at the appropriate asymptotic boundary.15
To recap. Whichever way one handles the details, the idea is that we
take solutions to be gauge equivalent if the formalism of the theory forces us
13
I.e., for any x U, ( )(x) = (x). In vacuum general relativity, the spoilers are
the diffeomorphisms that act as the identity on a neighbourhood of a Cauchy surface; in
Maxwells theory set in Minkowski spacetime (with a fixed metric) the spoilers are the
transformations of the form A 7 A + d where is a scalar function that vanishes on a
neighbourhood of a hyperplane of simultaneity.
14
This the the relation R that takes solutions and to be equivalent if and only if
there are solutions 1 , . . . , n such that 1 = , n = , and each i is spoiler-related
to i+1 (i = 1, . . . , n 1).
15
In some cases, this requires taking initial data surfaces to correspond to (partial)
Cauchy surfaces with good asymptotic behaviour.

16

to consider them physically equivalent on pain of recognizing rampant and

uninteresting indeterminism. We then identify the space of physically distinct
possibilities described by the theory with the space of solutions modulo gauge
equivalence, and identify the physical degrees of freedom of the theory with
any set of variables that parameterizes that quotient space.
Remark 5.1 (Non-Standard Counting). On the standard approach discussed above, symmetries that relate gauge-equivalent solutions have a quite
different status from other symmetries: indeed, the idea of gauge equivalence
is often introduced by saying that whereas global symmetries like Lorentz
symmetries relate physically distinct situations, the symmetries characteristic of YangMills theories are non-physical and their presence an indication
of redundancy in the variables of the theory.
There is an alternative approach on which all symmetries are on equal
footing: in order to find the degrees of freedom of a theory, one has to quotient
out all symmetriesincluding those that are not gauge in the present sense.
Thus, in the case of a theory whose equations of motion are not underdetermined, one must quotient the space of solutions by the spatiotemporal
and other symmetries of the theory in order to arrive at the true degrees of
freedom. For this approach, see [40, 41].

Background-Independence

In paradigm background-dependent theories, the geometry is the same from

solution solution to solution: vary the physical degrees of freedom as you
will, you cannot alter the geometry of spacetime. What is the very opposite
of such behaviour? A theory in which any change in the physical degrees
of freedom makes a difference to the geometry of spacetime. This motivates
the following definitions.
Definition 6.1 (Full Background-Dependence). A field theory is fully backgrounddependent if it has no geometrical degrees of freedom: every solution is assigned the same spacetime geometry as every other solution.
Definition 6.2 (Full Background-Independence). A field theory is fully backgroundindependent if all of its physical degrees of freedom correspond to geometrical
degrees of freedom: two solutions correspond to the same physical geometry
if and only if they are gauge equivalent.
17

Note that both of these definitions presuppose that a notion of geometrization is in place; the second also requires a notion of gauge equivalence.
In effect, these definitions measure the degree of background-independence
by looking at how many of the physical degrees of freedom correspond to
geometrical degrees of freedom. The limiting cases are full backgrounddependence (in which there are no geometrical degrees of freedom) and full
background-independence (in which all physical degrees of freedom are geometrical degrees of freedom). Intermediate cases are also possible.
Definition 6.3 (Near Background-Dependence). A field theory is nearly
background-dependent if it has only finitely many geometrical degrees of freedom: quotienting the space of geometries that arise in solutions of the theory
by the relation of geometrical equivalence yields a finite-dimensional space.
Definition 6.4 (Near Background-Independence). A field theory is nearly
background-independent if it has a finite number of non-geometrical degrees
of freedom: there is some N such that for any geometry arising in a solution
of the theory, the space of gauge equivalence classes of solutions with that
geometry is no more than N-dimensional.
These definitions give the desired verdicts concerning the examples of
Section 3 above.
The theories of Examples 3.1 and 3.2 describe a KleinGordon field in
Minkowski spacetime (with or without the metric as a fixed field). Under
the natural geometrization, these theories are fully background-dependent:
each solution has the geometry of Minkowski spacetime, so there are no
geometrical degrees of freedom.16
The theory of Example 3.3 concerns a scalar field propagating against a
flat but dynamical metric on a spacetime with toroidal spatial topology.
Under the natural geometrization it is nearly background-dependent: each
solution has the local geometry of Minkowski spacetime, but a finite number
of global geometrical degrees of freedom remain.
The theory of Example 3.4 describes a scalar field propagating against a
de Sitter metric. Under the natural geometrization, the theory is nearly
16

Here and below, the natural geometrization of a theory treats the theory as a relativistic field theory in the sense of Section 4 above, with the geometry of each solution
given by the metric tensor that appears as one of the fields of the theory.

18

background-dependent: the theory has a single geometrical degree of freedom (the value of the curvature constant). Note, however, that the theory
would count as fully background-dependent if one chose to count metric
tensors related by scaling transformations as geometrically equivalent.
The theory of Example 3.5 is spatially compact vacuum general relativity.
Under the natural geometrization and notion of gauge equivalence, the
theory is of course fully background-independent: two solutions share the
same geometry if and only if they are related by a diffeomorphism if and
only if they are gauge equivalent.
The the theory of Example 3.6 is vacuum general relativity with asymptotic
flatness imposed at spatial infinity. Under the natural geometrization and
the standard notion of gauge equivalence, the theory is nearly backgroundindependent: two solutions correspond to the same geometry if and only
they are related by a diffeomorphism; two solutions are gauge equivalent if
and only if they are related by a diffeomorphism asymptotic to the identity
at spatial infinity; if one quotients the family of solutions sharing a given geometry by the relation of gauge equivalence, the result is a ten-dimensional
space.17 Note, however, that if one were to adopt the non-standard method
of counting physical degrees of freedom discussed in Remark 5.1 above, then
this theory would count as fully background-independent.
The theory of Example 3.7 is inspired by Nordstroms bimetric theory of
gravity: its fields include a metric tensor that is flat in every solution
and a metric tensor g that varies from solution to solution. If the first
of these fields is taken to give the spacetime geometry of the theory, then
the theory is fully background-dependent (spacetime is R4 , so there are
no global degrees of freedom). If instead the metric g is taken to give
the spacetime geometry, then the theory is not background-dependent, nor
even nearly background-dependent: there are infinitely many geometrical
degrees of freedom.
17

The quotient of the group of diffeomorphisms by the group of diffeomorphisms acting

as the identity at spatial infinity is the Poincare group. This group acts as a symmetry
group on the space of gauge equivalence classes. See [3].

19

Discussion

The account of background-(in)dependence advanced above embodies the

three morals developed in Section 3. (i) Background dependence and independence come in degrees. (ii) Background-independence can be spoiled by
imposition of asymptotic boundary conditions, as well as by the presence of
solution-invariant fields in the theory. (iii) The status of a theory vis-`a-vis
background-dependence and independence can depend on the interpretation
given to the theory as well as on its formalism.
In this final section, a number of loose ends and further topics are taken
up: the possibility of translating the framework into a setting that allows
more general sorts of field theories; further niceties concerning the dependence of background-(in)dependence on how theories are understood; further points about asymptotic boundary conditions; and how adding matter
to general relativity affects the question of background-independence.

7.1

More General Field Theories

So far, a relatively narrow notion of a classical field theory has been employed, according to which all fields are tensor fields. More generally, one
might allow configurations of fields to be given by sections of arbitrary fibre bundles over spacetime. The notions of geometrization, spoilers, and
gauge equivalence of Sections 4 and 5 above carry over to this more general
settingand so the notions of background-(in)dependence given in Section
6 continue make sense. Whether these definitions still lead to reasonable
results is a question for further investigation. Of course, not all fields have
a well-defined transformation law under diffeomorphisms. So the notion of
general covariance only makes sense for certain types of fieldsthose fields
whose configurations are sections of a natural bundle over spacetime (see
[27]).

7.2

Dependence on Interpretation

We have already seen that the notions of background-dependence, backgroundindependence, etc., are not strictly formal notions: the status of given theory
may depend on how the formalism of a theory is interpreted. This fact is
reflected in the definitions given in Section 6 above: the degree of background(in)dependence of a theory may depend on which of the fields of a theory are
20

taken to represent geometrical structure, on the standards used to determine

whether two geometrical structures are the same, and on whether one views
solutions represented by (non-gauge) symmetries as always corresponding to
the same physical situation.18
Here is another illustration. In setting up a model of an elastic continuum
(a beam, a shell, a string . . . ) in a fixed spacetime background one can work
with so-called material variables [30]: one begins with a space B whose
points correspond to the material points of the continuum and identifies a
history of the continuum with a map from B := B R to a Lorentzian
manifold (V0 , g0 ) (with a diffeomorphism onto its image and the image of
each {b} R, b B, a timelike curve). Such a theory can be considered
a classical field theory in more or less the sense of Section 2 above, with
B being the spacetime, and the dynamical field of the theory being the
V0 -valued field . There is an obvious way to geometrize such a theory by
viewing each solution as endowing B with the structure of a relativistic
spacetimejust pull g0 back to B via .19 If one now applies the definitions
of Section 6, one typically finds that so-considered the theory is far from
being background-dependentin general distinct embeddings and lead
to non-isometric pull-backed metrics on B (unless boundary conditions have
been imposed that fix the shape of the boundary of (B)). But of course
this is all based on a misreading of the theory. The physical spacetime of the
theory is V0 , not Band the theory counts as fully background-dependent
when reformulated of as a classical field theory with V0 as its spacetime and
with physical geometry given by the fixed field g0 .

7.3

Asymptotic Boundary Conditions

An important feature of the apparatus of Section 6 is that it allows one to

isolate the sense in which the imposition of asymptotic boundary conditions
can spoil background-independence. In the case of general relativity, the
sorts of asymptotic boundary conditions usually discussed can be thought of
in the following terms. One adjoins to the n-dimensional physical spacetime
manifold V a non-physical (n 1)-dimensional boundary V endowed with
18

For illustrations, see the discussions in the preceding section of Nordstr

oms theory,
of the de Sitter-based theory, and of general relativity with asymptotic flatness imposed
at spatial infinity.
19
The NambuGoto action for string theory takes g to be the fundamental dynamical
variable. One can do something similar in continuum mechanics [9].

21

some sort of geometrical structure. Kinematically possible are required to

exhibit good behaviour as they approach V ; likewise initial data surfaces
should be asymptotic to subsets of V that are geometrically nice. We denote
by D the group of spatiotemporal symmetries of our theory and by D0 the
group of spatiotemporal symmetries asymptotic to the identity at V. In
typical examples D0 is a proper subgroup of D which is a proper subgroup of
Diff(V ). Further, one typically finds that D0 is a normal subgroup of D and
that the quotient group G = D/D0 can be thought of as acting geometrically
at V (i.e., every diffeomorphism in D induces an isomorphism from V
to itself, with diffeomorphisms that differ by an element of D0 inducing the
same isomorphism). When this obtains, G is called the asymptotic symmetry
group.
For the most common sorts of asymptotic boundary conditions for vacuum general relativity, one knows or expects that (relative to a suitable
notion of initial data surface) solutions are gauge equivalent if and only if
they are related by a diffeomorphism asymptotic to the identity at V and
hence that (when all goes well) the asymptotic symmetry group acts on the
space of gauge equivalence classes of solutions (for several examples and one
counter-example, see [6]).
If all of this obtains, then under its natural reading the theory is nearly
background-independent so long the asymptotic symmetry group G is finitedimensionalas it was in Example 3.6 above (where G was the Poincare
group). But even when this picture holds, strange things can happen: when
one imposes asymptotic flatness at null infinity, the asymptotic symmetry
group is the infinite-dimensional BondiMetznerSachs group (see, e.g., [18]);
and one expects that this group acts on the space of gauge equivalence
classes of solutions [6, example 4]. This means that the theory comes out as
not nearly background-independent: even after the geometry of spacetime is
fixed, an infinite number of physical degrees of freedom remain free. This
is an unsettling conclusionbut perhaps it can be accepted as being of a
piece with the strangeness of the symmetry group of this sector of general
relativity.20
One might also worry about a problem at the other end of the spectrum.
What if one were to impose asymmetric asymptotic boundary conditions?
E.g., what if one were to modify the standard framework for asymptotically
20

It is sometimes suggested that the complicated asymptotic structure of this sector is

a reflection of the fact that radiation can reach conformal infinity in this setting [4].

22

anti-de Sitter spacetimes in such a way that the asymptotic symmetry group
became trivial [5]? According to the present approach, the resulting theory
would be fully background-independent, despite the presence of asymptotic
boundary conditions! Presumably, however, this sort of theory is ruled out
by our decision to focus on theories of the entire universeit is difficult to
see how the need for asymmetric asymptotic boundary conditions could arise
for theories of this type.

7.4

General Relativity with Matter

So far, in considering general relativity, we have limited attention to the

vacuum case. What happens when matter is introduced? That depends.
Consider first what happens if one includes a scalar field that sees but
is not seen by the metric g (i.e., obeys the KleinGordon equation relative
to g but g obeys the vacuum Einstein equation). Under the obvious way of
understanding this theory, it has the same geometrical degrees of freedom as
vacuum general relativityso it is certainly not anything like backgrounddependent according to the present approach. But corresponding to each
way that the geometry of spacetime could be, there is a vast number of ways
of arranging the physical degrees of freedom inherent in the scalar field
so the theory is not anything like background-independent according to the
present approach. This is an intuitively attractive verdict: this theory is
a sort of hybrid of vacuum general relativity (the paradigm of backgroundindependence) and a theory in which the spacetime metric affects but is not
affected by matter (which feature has often been taken to be a hallmark of
background-dependence).
Of course, the theory of the preceding paragraph is a pathological example. At the other end of the spectrum lies general relativity coupled to
dust. Given a solution to the Einstein-dust equations, one can reconstruct
the dust trajectories and energy density from knowledge of the spacetime
metric and the equations of motion (see, e.g., [38, 3.14]). So if two solutions
agree in their geometry, they agree with respect to all physical degrees of
freedomthe theory is fully background-independent.
More typical are cases that lie somewhere in the vast middle ground. The
Einstein-Maxwell theory is an interesting example. In effect it splits into a
fully background-independent sector and a sector that exhibits the sort of
behaviour seen in the hybrid theory above: in generic solutions, one can reconstruct the electromagnetic field from knowledge of the geometry and the
23

equations of motion; but there exist exceptional solutionsthose involving

null electromagnetic fieldsfeaturing geometries that are compatible with a
huge range number of configurations of the electromagnetic field [17]. Here
we have a theory that falls short of full background-independence in an illuminating way (and which serves to illustrate the inadequacy of the naive
equivalence between a theorys being background-independent and its having
as its most natural formulation a generally covariant one). It would be desirable to extend the framework of Section 6 to provide comparative measures
of background-independence for theories like this one that fall short of near
background-independence.

Acknowledgements
Thanks to Harvey Brown, John Earman, Michael Friedman, Bob Geroch,
David Malament, Brian Pitts, Oliver Pooley, Tom Ryckman, Laura Ruetsche,
Rob Rynasiewicz, Ryan Samaroo, Philip Stamp, and Steve Weinstein for
helpful comments and conversations. Work on this paper was supported
during 20062007 by the American Council of Learned Societies and by the
Center for Advanced Study in the Behavioral Sciences.

References
[1] Anderson, J.: Relativity principles and the role of coordinates in physics.
In: H.Y. Chiu, W. Hoffman (eds.) Gravitation and Relativity, pp. 175
194. W.A. Benjamin, Inc., New York (1964)
[2] Anderson, J.: Principles of Relativity Physics. Academic Press, New
York (1967)
[3] Andersson, L.: Momenta and reduction in general relativity. Journal of
Geometry and Physics 4, 289314 (1987)
[4] Anninos, D., Ng, G.S., Strominger, A.: Asymptotic symmetries and
charges in de Sitter space (2010). ArXiv: 1009.4730v1 [gr-qc]
[5] Ashtekar, A., Magnon, A.: Asymptotically anti-de Sitter space-times.
Classical and Quantum Gravity 1, L39L44 (1984)

24

[6] Belot, G.: An elementary notion of gauge equivalence. General Relativity and Gravitation 40, 199215 (2008)
[7] Carlip, S.: Quantum Gravity in 2+1 Dimensions. Cambridge University
Press, Cambridge (1998)
[8] Chrusciel, P., Isenberg, J.: Nonisometric vacuum extensions of vacuum
maximal globally hyperbolic spacetimes. Physical Review D 48, 1616
1628 (1993)
[9] Ciarlet, P., Gratie, L., Mardare, C.: Intrinsic methods in elasticity: A
mathematical survey. Discrete and Continuous Dynamical Systems 23,
133164 (2009)
[10] Crnkovic, v., Witten, E.: Covariant description of canonical formalism
in geometrical theories. In: S. Hawking, W. Israel (eds.) Three Hundred Years of Gravitation, pp. 676684. Cambridge University Press,
Cambridge (1987)
[11] Deligne, P., Freed, D.: Classical field theory. In: P. Deligne, P. Etingof,
D. Freed, L. Jeffrey, D. Kazhdan, J. Morgan, D. Morrison, E. Witten
(eds.) Quantum Fields and Strings: A Course for Mathematicians, vol. 1,
pp. 137225. American Mathematical Society, Providence, RI (1999)
[12] Diacu, F.: Singularities of the N-Body Problem. Les Publications CRM,
Montreal (1992)
[13] Dirac, P.A.M.: Lectures on Quantum Mechanics. Dover, New York
(2001)
[14] Earman, J.: Covariance, invariance, and the equivalence of frames.
Foundations of Physics 4, 267289 (1974)
[15] Friedman, M.: Relativity principles, absolute objects, and symmetry
groups. In: P. Suppes (ed.) Space, Time, and Geometry, pp. 296320.
Reidel, Dordrecht (1973)
[16] Friedman, M.: Foundations of Spacetime Theories. Princeton University
Press, Princeton (1983)
[17] Geroch, R.: Electromagnetism as an aspect of geometry? Already unified field theorythe null case. Annals of Physics 36, 147187 (1966)
25

[18] Geroch, R.: Asymptotic stucture of space-time. In: F.P. Esposito,

L. Witten (eds.) Asymptotic Stucture of Space-Time, pp. 1105. Plenum
Press, New York (1977)
[19] Giulini, D.: Some remarks on the notions of general covariance and background independence. In: E. Seiler, I.O. Stamatescu (eds.) Approaches
to Fundamental Physics: An Assessment of Current Theoretical Ideas,
pp. 105120. Springer-Verlag, Berlin (2007)
[20] Gotay, M., Castrillon Lopez, M.: Covariantizing classical field theories
(2010). ArXiv:1008.3170v1 [math-ph]
[21] Gotay, M., Nester, J., Hinds, G.: Presymplectic manifolds and the
DiracBergmann theory of constraints. Journal of Mathematical Physics
19, 23882399 (1978)
[22] Gryb, S.: A definition of background independence. Classical and Quantum Gravity 27, 215018 (2010)
[23] Hawking, S., Ellis, G.: The Large Scale Structure of Space-Time. Cambridge University Press, Cambridge (1973)
[24] Henneaux, M., Teitelboim, C.: Quantization of Gauge Systems. Princeton University Press, Princeton (1992)
[25] Janssen, M.: No success like failure . . . : Einsteins quest for general
relativity, 19071920 (2008). Phil-Sci archive: 00004377
[26] Janssen, M., Schulmann, R., Illy, J., Lehner, C., Kormos-Buchwald, D.
(eds.): The Collected Papers of Albert Einstein. Volume Seven. The
Berlin Years: Writings, 1918-1921. Princeton University Press, Princeton (2002)
[27] Kolar, I., Michor, P., Slovak, J.: Natural Operations in Differential
Geometry. Springer-Verlag, Berlin (1993)
[28] Lee, J., Wald, R.: Local symmetries and constraints. Journal of Mathematical Physics 31, 725743 (1990)
[29] Louko, J., Marolf, D.: The solution space of 2 + 1 gravity on R T 2 in
Wittens connection formulation. Classical and Quantum Gravity 11,
311330 (1994)
26

[30] Marsden, J., Hughes, T.: Mathematical Foundations of Elasticity.

Dover, New York (1994)
[31] Meusburger, C.: Cosmological measurements, time, and observables in
(2 + 1)-dimensional gravity. Classical and Quantum Gravity 26, 055006
(2009)
[32] Norton, J.: Einstein, Nordstrom, and the early demise of Lorentzcovariant, scalar theories of gravity. Archive for History of the Exact
Sciences 45, 1794 (1992)
[33] Norton, J.: General covariance and the foundations of general relativity:
Eight decades of dispute. Reports on Progress in Physics 56, 791858
(1993)
[34] Olver, P.: Applications of Lie Groups to Differential Equations, second
edn. Springer-Verlag, Berlin (1993)
[35] Pitts, B.: Absolute objects and counterexamples: JonesGeroch dust,
Toretti constant curvature, tetrad-spinor, and scalar density. Studies in
History and Philosophy of Modern Physics 37, 347371 (2006)
[36] Ravndal, F.: Scalar gravitation and extra dimensions. In: C. Cranstrom,
C. Montonen (eds.) Proceedings of the Gunnar Nordstrom Symposium
of Theoretical Physics, pp. 151164. The Finnish Society of Sciences and
Letters, Helsinki (2004)
[37] Rovelli, C.: Quantum Gravity. Cambridge University Press, Cambridge
(2004)
[38] Sachs, R., Wu, H.H.: General Relativity for Mathematicians. SpringerVerlag, Berlin (1977)
[39] Sklar, L.: Philosophy and Spacetime Physics. University of California
Press, Berkeley (1985)
[40] Smolin, L.: The present moment in quantum cosmology: Challenges
to the argument for the elimination of time. In: R. Durie (ed.) Time
and the Instant : Essays in the Physics and Philosophy of Time, pp.
112143. Clinamen, Manchester (2001)

27

[41] Smolin, L.: The case for background independence. In: D. Rickles,
S. French, J. Saatsi (eds.) The Structural Foundations of Quantum Gravity, pp. 196239. Oxford University Press, Oxford (2006)
[42] Sorkin, R.: An example relevant to the KretschmannEinstein debate.
Modern Physics Letters A 17, 695700 (2002)
[43] Straumann, N.:
ph/0006423v1

Reflections on gravity (2000).

ArXiv:astro-

[44] Torre, C.: Covariant phase space formulation of parameterized field

theories. Journal of Mathematical Physics 33, 38023811 (1992)
[45] Wolf, J.: Spaces of Constant Curvature, fifth edn. Publish or Perish,
Wilmington (1984)
[46] Zuckerman, G.: Action principles and global geometry. In: S.T. Yau
(ed.) Mathematical Aspects of String Theory, pp. 259284. World Scientific (1987)

28