GRG manuscript No.

(will be inserted by the editor)

The Superspace of Geometrodynamics

Domenico Giulini

I dedicate this contribution to the scientific legacy of John Wheeler

arXiv:0902.3923v1 [gr-qc] 23 Feb 2009

Received: date / Accepted: date

Abstract Wheeler’s Superspace is the arena in which Geometrodynamics takes place.

I review some aspects of its geometrical and topological structure that Wheeler urged
us to take seriously in the context of canonical quantum gravity.
PACS 04.60.-m · 04.60.Ds · 02.40.-k
Mathematics Subject Classification (2000) 83C45 · 57N10 · 53D20

“The stage on which the space of the Universe moves is certainly not space
itself. Nobody can be a stage for himself; he has to have a larger arena in which
to move. The arena in which space does its changing is not even the space-
time of Einstein, for space-time is the history of space changing with time. The
arena must be a larger object: superspace. . . It is not endowed with three or
four dimensions—it’s endowed with an infinite number of dimensions.” (J.A.
Wheeler: Superspace, Harper’s Magazine, July 1974, p. 9)

1 Introduction

From somewhere in the 1950’s on, John Wheeler repeatedly urged people who were
interested in the quantum-gravity programme to understand the structure of a mathe-
matical object that he called Superspace [80][79]. The intended meaning of ‘Superspace’
was that of a set, denoted by S(Σ), whose points faithfully correspond to all possible
Riemannian geometries on a given 3-manifold Σ. Hence, in fact, there are infinitely
many Superspaces, one for each 3-manifold Σ. The physical significance of this concept
is suggested by the dynamical picture of General Relativity (henceforth abbreviated
by GR), according to which spacetime is the history (time evolution) of space. Ac-
cordingly, in Hamiltonian GR, Superspace plays the rôle of the configuration space
the cotangent bundle of which gives the phase space of 3d-diffeomorphism reduced
states. Moreover, in Canonical Quantum Gravity (henceforth abbreviated by CQG),

Max-Planck-Institute for Gravitational Physics (Albert-Einstein-Institute),

Am Mühlenberg 1, D-14476 Potsdam OT Golm, Germany.
E-mail: [email protected]
2

Superspace plays the rôle of the domain for the wave function which is still subject
to the infamous Wheeler-DeWitt equation. In fact, Bryce De Witt characterised the
motivation for his seminal paper on CQG as follows:

“The present paper is the direct outcome of conversations with Wheeler, dur-
ing which one fundamental question in particular kept recurring: What is the
structure of the domain manifold for the quantum-mechanical state functional?”
([16], p. 115)

More than 41 years after DeWitt’s important contribution I simply wish to give a
small overview over some of the answers given so far to the question: What is the struc-
ture of Superspace? Here I interpret ‘structure’ more concretely as ‘metric structure’
and ‘topological structure’. But before answers can be attempted, we need to define
the object at hand. This will be done in the next section; and before doing that, we
wish to say a few more words on the overall motivation.
Minkowski space is the stage for relativistic particle physics. It comes equipped
with some structure (topological, affine, causal, metric) that is not subject to dynami-
cal changes. Likewise, as was emphasised by Wheeler, the arena for Geometrodynamics
is Superspace, which also comes equipped with certain non-dynamical structures. The
topological and geometric structures of Superspace are as much a background for GR
as the Minkowski space is for relativistic particle physics. Now, Quantum Field Theory
has much to do with the automorphism group of Minkowski space and, in particular,
its representation theory. For example, all the linear relativistic wave equations (Klein-
Gordan, Weyl, Dirac, Maxwell, Proca, Rarita-Schwinger, Dirac-Bargmann, etc.) can
be understood in this group-theoretic fashion, namely as projection conditions onto
irreducible subspaces in some auxiliary Hilbert space. (In the same spirit a charac-
terisation of ‘classical elementary system’ has been given as one whose phase space
supports a transitive symplectic action of the Poincaré group [5][4].) This is how we ar-
rive at the classifying meaning of ‘mass’ and ‘spin’. Could it be that Quantum Gravity
has likewise much to do with the automorphism group of Superspace? Can we under-
stand this group in any reasonable sense and what has it to do with four dimensional
diffeomorphisms? If elementary particles are unitary irreducible representations of the
Poincaré group, as Wigner once urged, what would the ‘elementary systems’ be that
corresponded to irreducible representations of the automorphism group of Superspace?
I do not know any reasonably complete answer to any of these questions. But the
analogies at least suggests the possibility of some progress if these structures and their
automorphisms could be be understood in any depth. This is a difficult task, as John
Wheeler already foresaw forty years ago:

“Die Struktur des Superraumes enträtseln? Kaum in einem Sprung, und kaum
heute!” ([79], p. 61)

Related in spirit is a recent approach in the larger context of 11-dimensional su-

pergravity (see [15, 14] and references therein), which is based on the observation that
the supergravity dynamics in certain truncations corresponds to geodesic motion of a
massless spinning particle on an E10 coset space. Here the Wheeler-DeWitt metric (9c)
appears naturally with the right GR-value λ = 1, which in our context is the only value
compatible with 4-dimensional diffeomorphism invariance, as we will discuss. This may
suggest an interesting relation between E10 and spacetime diffeomorphisms.
 3

2 Defining Superspace

As already said, Superspace S(Σ) is the set of all Riemannian geometries on the 3-
manifold Σ. Here ‘geometries’ means ‘metrics up to diffeomorphisms’. Hence S(Σ)
is identified as set of equivalence classes in Riem(Σ), the set of all smooth (C ∞ )
Riemannian metrics in Σ under the equivalence relation of being related by a smooth
diffeomorphism. In other words, the group of all (C ∞ ) diffeomorphisms, Diff(Σ), has
a natural right action on Riem(Σ) via pullback and the orbit space is identified with
S(Σ):
S(Σ) := Riem(Σ)/Diff(Σ) . (1)
Let us now refine this definition. First, we shall restrict attention to those Σ which
are connected and closed (compact without boundary). We note that Einstein’s field
equations by themselves do not exclude any such Σ. To see this, recall the form of the
constraints for initial data (h, K), where h ∈ Riem(Σ) and K is a symmetric covariant
2nd rank tensor-field (to become the extrinsic curvature of Σ in spacetime, once the
latter is constructed from the dynamical part of Einstein’s equations)
´2 `
kKk2h − Trh (K) − R(h) − 2Λ = − (2κ)ρm ,
` ´
(2a)
` ´
divh K − h Trh (K) = (cκ) j , (2b)

where ρm and jm are the densities of energy and momentum of matter respectively,
R(h) is the Ricci scalar for h, and κ = 8πG/c4 . Now, it is known that for any smooth
function f : Σ → R which is negative somewhere on Σ there exists an h ∈ Riem(Σ)
so that R(h) = f [47]. Given that strong result, we may easily solve (2) for j = 0
on any compact Σ as follows: First we make the Ansatz K = αh for some constant
α and some h ∈ Riem(Σ). This solves (2b), whatever α, h will be. Geometrically this
means that the initial Σ will be a totally umbillic hypersurface in spacetime. Next
we solve (2a) by fixing α so that α2 > (Λ + κ supΣ (ρm ))/3 and then choosing h
so that R(h) = 2Λ + 2κρm − 6α2 , which is possible by the result just cited because
the right-hand side is negative by construction. This argument can be generalised to
non-compact manifolds with a finite number of ends and asymptotically flat data [84].
Next we refine the definition (1), in that we restrict the group of diffeomorphisms to
the proper subgroup of those diffeomorphisms that fix a preferred point, called ∞ ∈ Σ,
and the tangent space at this point:
˘ ¯
DiffF (Σ) := φ ∈ Diff(Σ) | φ(∞) = ∞, φ∗ (∞) = id|T∞ Σ . (3)

The reason for this is twofold: First, if one is genuinely interested in closed Σ, the
space S(Σ) := Riem(Σ)/Diff(Σ) is not a manifold if Σ allows for metrics with non-
trivial isometry groups (not all Σ do; compare footnote 3). At those metrics Diff(Σ)
clearly does not act freely, so that the quotient Riem(Σ)/Diff(Σ) has the structure of a
stratified manifold with nested sets of strata ordered according to the dimension of the
isometry groups [22]. In that case there is a natural way to minimally resolve the sin-
gularities [23] which amounts to taking instead the quotient Riem(Σ) × F(Σ)/Diff(Σ),
where F(Σ) is the bundle of linear frames over Σ. The point here is that the action of
Diff(Σ) is now free since there simply are no non-trivial isometries that fix a frame.
Indeed, if φ is an isometry fixing some frame, we can use the exponential map and
φ ◦ exp = exp ◦φ∗ (valid for any isometry) to show that the subset of points in Σ fixed
by φ is open. Since this set is also closed and Σ is connected, φ must be the identity.
4

Now, the quotient Riem(Σ) × F(Σ)/Diff(Σ) is isomorphic1 to

SF (Σ) := Riem(Σ)/DiffF (Σ) , (4)

albeit not in a natural way, since one needs to choose a preferred point ∞ ∈ Σ. This
may seem somewhat artificial if really all points in Σ are considered to be equally real,
but this is irrelevant for us as long as we are only interested in the isomorphicity class
of Superspace. On the other hand, if we consider Σ as the one-point compactification
of a manifold with one end2 , then (4) would be the right space to start with anyway
since then diffeomorphisms have to respect the asymptotic geometry in that end, like,
e.g., asymptotic flatness. Therefore, from now on, we shall refer to SF (Σ) as defined
in (4) as Superspace. In view of the original definition (1) it is usually called ‘extended
Superspace’ [22].
Clearly, the move from (1) to (4) would have been unnecessary in the closed case
if one restricted attention to those manifolds Σ which do not allow for metrics with
continuous symmetries, i.e. whose degree of symmetry3 is zero. Even though these man-
ifolds are not the ‘obvious’ ones one tends to think of first, they are, in a sense, ‘most’
3-manifolds. On the other hand, in order not to deprive ourselves form the possibility
of physical idealisations in terms of prescribed exact symmetries, we prefer to work
with SF (Σ) defined in (4) (called ‘extended superspace’ in [22], as already mentioned).
Let us add a few words on the point-set topology of Riem(Σ) and SF (Σ). First,
Riem(Σ) is an open positive convex cone in the topological vector space of smooth
(C ∞ ) symmetric covariant tensor fields over Σ. The latter space is a Fréchet space,
that is, a locally convex topological vector space that admits a translation-invariant
metric, d, inducing its topology and with respect to which the space is complete. The
metric can be chosen such that Diff(Σ) preserves distances. Riem(Σ) inherits this
metric which makes it a metrisable topological space that is also second countable
(recall also that metrisability implies paracompactness). SF (Σ) is given the quotient
topology, i.e. the strongest topology in which the projection Riem(Σ) → SF (Σ) is
continuous. This projection is also open since DiffF (Σ) acts continuously on Riem(Σ).
A metric d on SF (Σ) is defined by

d([h1 ], [h2 ]) := sup d(φ∗1 h1 , φ∗2 h2 ) , (5)

φ1 ,φ2 ∈DiffF (Σ)

which also turns SF (Σ) into a connected (being the continuous image of the connected
Riem(Σ)) metrisable and second countable topological space. Hence Riem(Σ) and
1 “Isomorphic as what?” one may ask. The answer is: as ILH (inverse-limit Hilbert) mani-

folds. In the ILH sense the action of Diff(Σ) on Riem(Σ) × F(Σ) is smooth, free, and proper;
see [23] for more details and references.
2 The condition of ‘asymptotic flatness’ of an end includes the topological condition that

the one-point compactification is again a manifold. This is the case iff there exists a compact
subset in the manifold the complement of which is homeomorphic to the complement of a
closed solid ball in R3 .
3 Let I(Σ, h) := {φ ∈ Diff(Σ) | φ∗ h = h} be the isometry group of (Σ, h), then it is

well known that dim I(Σ, h) ≤ 21 n(n + 1), where n = dim Σ. I(Σ, h) is compact if Σ is
compact (see, e.g., Sect. 5 of [62]). Conversely, if Σ allows for an effective action of a compact
group G then it clearly allows for a metric h on which G acts as isometries (just average any
Riemannian metric over G.) The degree of symmetry of Σ, denoted by deg(Σ), is defined by
deg(Σ) := suph∈Riem(Σ) {dim I(Σ, h)}. For compact Σ the degree of symmetry is zero iff Σ
cannot support an action of the circle group SO(2). A list of 3-manifolds with deg > 0 can be
found in [22] whereas [24] contains a characterisation of deg = 0 manifolds.
 5

SF (Σ) are perfectly decent connected topological spaces which satisfy the strongest
separability and countability axioms. For more details we refer to [73][23][22].
The basic geometric idea is now to regard Riem(Σ) as principal fibre bundle with
structure group DiffF (Σ) and quotient SF (Σ):
i p
DiffF (Σ) −→ Riem(Σ) −→ SF (Σ) (6)

where the maps i are the inclusion and projection maps respectively. This is made
possible by the so-called ‘slice theorems’ (see [20][22]), and the fact that the group acts
freely and properly. This bundle structure has two far-reaching consequences regarding
the geometry and topology of SF (Σ). Let us discuss these in turn.

3 Geometry of Superspace

Elements of the Lie algebra diffF (Σ) of DiffF (Σ) are vector fields on Σ. For any such
vector field ξ on Σ there is a vector field Vξ on Riem(Σ), called the vertical (or funda-
mental) vector field associated to ξ, whose value at h ∈ Riem(Σ) is just the infinitesimal
change in h generated by ξ, that is,

Vξ (h) = −Lξ h , (7)

where Lξ denoted the Lie derivative with respect to ξ. Hence, for each h ∈ Riem(Σ),
the map V (h) : ξ 7→ Vξ (h) is an anti-Lie homomorphism (the ‘anti’ being due to the
fact that we have a right action of Diff(Σ) on Riem(Σ)), that is [Vξ , Vη ] = −V[ξ,η] ,
if the Lie structure on diffF (Σ) is that of ordinary commutators of vector fields. The
kernel of the map V (h) : ξ 7→ Vξ (h) consists of the finite-dimensional subspace of
Killing fields on (Σ, h). The vertical vectors at h ∈ Riem(Σ) therefore form a linear
subspace Verth ⊂ Th Riem(Σ), isomorphic to the vector fields on Σ modulo the Killing
fields on (Σ, h). It is a closed subspace due to the fact that the the operator ξ 7→ Lξ h
is overdetermined elliptic (cf. [11], Appendices G-I).
The family of ultralocal ‘metrics’ on Riem(Σ) is given by
Z
d3 x α det(h) hab hcd kac ℓbd − λ (hab kab )(hcd ℓcd ) ,
p ` ´
G(α,λ) (k, ℓ) = (8)
Σ

for each k, ℓ ∈ Th Riem(Σ). Here α is a positive real-valued function on Σ and λ a real

number. An almost trivial but important observation is that Diff(Σ) is an isometry
group with respect to all G(α,λ) . The ‘metric’ picked by GR through the bilinear term
in the constraint (2a) corresponds to λ = 1. The positive real-valued function α is not
fixed and corresponds to the free choice of a lapse-function. In what follows we shall
focus attention to α = 1.
The pointwise bilinear form (k, ℓ) 7→ (h ⊗ h)(k, ℓ) − λTrh (k)Trh (ℓ) in the integrand
of (8) defines a symmetric bilinear form on the six-dimensional space of symmetric
tensors which is positive definite for λ < 1/3, of signature (1, 5) for λ > 1/3, and
degenerate of signature (0, 5) for λ = 1/3. It defines a metric on the homogeneous
space GL(3)/O(3), where the latter may be identified with the space of euclidean
metrics on a 3-dimensional vector space. Parametrising it by hab , we have

τ2
Gλ = Gab
λ
cd
dhab ⊗ dhcd = −ǫdτ ⊗ dτ + Tr(r −1 dr ⊗ r −1 dr) , (9a)
c2
6

where

rab := [det(h)]−1/3 hab , τ := c [det(h)]1/4 , c2 := 16|λ−1/3|, ǫ = sign(λ−1/3), (9b)

and
Gab cd 1
det(h) hac hbd + had hbc − 2λhab hcd .
p ` ´
λ = 2 (9c)

This is a 1+5 – dimensional warped-product geometry in the standard form of ‘cos-

mological’ models (Lorentzian for λ > 1/3), here corresponding to the 1+5 decomposi-
tion GL(3)/O(3) ∼ = R × SL(3)/SO(3) with scale factor τ /c and homogeneous Rieman-
nian metric on five-dimensional ’space’ SL(3)/SO(3), given by Tr(r −1 dr ⊗ r −1 dr) =
r ac r bd drab ⊗ drcd . τ = 0 is a genuine ‘spacelike’ (‘cosmological’) curvature singular-
ity. An early discussion of this finite-dimensional geometry was given by DeWitt [16].
We stress that the Lorentzian nature of the Wheeler-DeWitt metric in GR (i.e. for
λ = 1) has nothing to do with the Lorentzian nature of spacetime, as we will see below
from the statement of Theorem 1 and formulae (33,34); rather, it can be related to the
attractivity of gravity [38].
As for the infinite-dimensional geometry of Riem(Σ), we remark that an element h
of Riem(Σ) is a section in T ∗ Σ ⊗ T ∗ Σ and so is an element of Th Riem(Σ). The latter
has the fibre-metric (9). It is sometimes useful to use h (for index raising) in order to
identify Th Riem(Σ) with sections in T Σ ⊗ T ∗ Σ ∼ = End(T Σ), also because the latter
has a natural structure as associative- (and hence also Lie-) p algebra. Then the inner
product (8) for α = 1 just reads (here and below dµ(h) = det(h) d3 x)
Z
` ´
Gλ (k, ℓ) = dµ(h) Tr(k · ℓ) − λ Tr(k)Tr(ℓ) . (10)
Σ

For λ = 0 the infinite-dimensional geometry of Riem(Σ) has been studied in [27].

They showed that all curvature components involving one or more pure-trace directions
vanish and that the curvature tensor for the trace-free directions is given by (now
making use of the natural Lie-algebra structure of T Σ ⊗ T ∗ Σ)

R(k, ℓ)m = − 41 [[k, ℓ], m] . (11)

In particular, this implies that the sectional curvatures involving pure trace directions
vanish and that the sectional curvatures for trace-free directions k, ℓ are non-positive:
Z
K(k, ℓ) = − 41
` ´
dµ(h) Tr k · R(k, ℓ)ℓ
ZΣ (12)
1
` ´
= −4 dµ(h) Tr [k, ℓ] · [ℓ, k] ≤ 0 .
Σ

Similar results hold for other values of λ, though some positivity statements cease
to hold for λ > 1/3. We keep the generality in the value of λ for the moment in order
to show that the value λ = 1 picked by GR is quite special. Since, as already stated,
all elements of DiffF (Σ) are isometries of Gλ , it is natural to try to define a bundle
connection on Riem(Σ) by taking the horizontal subspace Horλ h at each Th Riem(Σ) to
be the Gλ –orthogonal complement to Verth , as suggested in [35]. From (8) one sees
that k ∈ Th Riem(Σ) is orthogonal to all Lξ h iff

(Oλ k)a := −∇b (kab − λhab hcd kcd ) = 0 . (13)

 7

But note that orthogonality does not imply transversality if the metric is indefinite, as
for λ = 1. In that case the intersection Verth ∩ Horλh may well be non trivial, which
implies that there is no well defined projection map

horλ λ
h : Th Riem(Σ) → Horh . (14)

The definition of this map would be as follows: Let k ∈ Th Riem(Σ), find a vector field
ξ on Σ such that k − Vξ is horizontal. Then Vξ is the (λ dependent) vertical component
of k and the map k 7→ k − Vξ is the (λ dependent) horizontal projection (14). When
does that work? Well, according to (13), the condition for k − Vξ to be horizontal for
given k is equivalent to the following differential equation for ξ:
` ´
Dλ ξ := δd + 2(1 − λ)dδ − 2Ric ξ = Oh k . (15)

Here we regarded ξ as one-form and d, δ denote the standard exterior differential and
co-differential (δξ = −∇a ξa ) respectively. Moreover, Ric is the endomorphism on one-
forms induced by the Ricci tensor (ξa 7→ Rab ξb ). Note that the right-hand side of (15)
is L2 -orthogonal for all k to precisely the Killing fields.
The singular nature of the GR value λ = 1 is now seen from writing down the
principal symbol of the operator Dλ on the left-hand side of (15),

ζ a ζb
„ «
σλ (ζ)ab = kζk δba + (1 − 2λ) , (16)
kζk

whose determinant is kξk6 2(1−λ). Hence σλ is positive definite for λ < 1 and indefinite
but still an isomorphism for λ > 1. This means that Dλ is elliptic for λ 6= 1 (strongly
elliptic4 for λ < 1) but fails to be elliptic for precisely the GR value λ = 1. This implies
the possibility for the kernel of Dλ=1 to become infinite dimensional.
The last remark has a direct implication as regards the intersection of the horizontal
and vertical subspaces. Recall that solutions ξ to Dλ ξ = 0 modulo Killing fields (which
always solve this equation) correspond faithfully to vertical vectors at h ∈ Riem(Σ)
(via ξ 7→ Vξ (h)) which are also horizontal. Since Killing vectors span at most a finite
dimensional space, an infinite dimensional intersection Verth ∩ Horλ h would be implied
by an infinite dimensional kernel of D1 .
That this possibility for λ = 1 is actually realised for any Σ is easy to see: Take
a metric h on Σ that is flat in an open region U ⊂ Σ, and consider k ∈ Th Riem(Σ)
of the form kab = 2∇a ∇b φ, where φ is a real-valued function on Σ whose support is
contained in U . Then k is vertical since k = Lξ h for ξ = gradφ (a non-zero gradient
vector-field is never Killing on a compact Σ), and also horizontal since k satisfies (13).
Such ξ clearly span an infinite-dimensional subspace in the kernel of D1 .
On the other hand, for λ = 1 there are also always open sets of h ∈ Riem(Σ) (and
of [h] ∈ SF (Σ)) for which the kernel of D1 is trivial (the kernel clearly depends only on
the diffeomorphism class [h] of h). For example, consider metrics with negative definite
Ricci tensor, which exist for any closed Σ [30]. (Note that Ricci-negative geometries
never allow for non-trivial Killing fields.) Then it is clear from the definition of Dλ
that it is a positive-definite operator for λ ≤ 1. Hence the intersection Verth ∩ Horλ h is
trivial.
4 We follow the terminology of Appendix I in [11].
8

In the latter case it is interesting to observe that Gλ restricted to Verth (h Ricci

negative) is positive definite, since
Z
` ´
Gλ Vξ (h), Vξ (h) = 2 dµ(h) h(ξ, Dλ ξ) . (17)
Σ

This means that Gλ restricted to the orthogonal complement, Horλ h , contains infinitely
many negative and infinitely may positive directions and the same (∞, ∞) – signature
is then directly inherited by T[h] SF (Σ) for any Ricci-negative geometry [h].
Far less generic but still interesting examples for trivial intersections Verth ∩ Horλ
h
in case λ = 1 are given by Einstein metrics with positive Einstein constants. Since this
condition implies constant positive sectional curvature, such metrics only exist on man-
ifolds Σ with finite fundamental group, so that Σ must be a spherical space form S 3 /G,
where G is a finite subgroup of SO(4) acting freely on S 3 . Also note that the subspace
in SF (Σ) of Einstein geometries is finite dimensional. Now, any solution ξ to D1 ξ = 0

h′′
vertical

h izon
tal
hor

h′

Riem(Σ)

SF (Σ)
[h] [h′ ] [h′′ ]

Fig. 1 The rectangle depicts the space Riem(Σ) which is fibred by the orbits of DiffF (Σ)
(curved vertical lines). The metric G1 on Riem(Σ) is such that as we move along Riem(Σ)
transversal to the fibres the “light-cones” tilt relative to the fibre directions. The process here
shows a transition at [h′ ] where some fibre directions are lightlike and no metric can be defined
in T[h′ ] SF (Σ), whereas they are timelike at [h] and spacelike at [h′′ ]. The parallelogram at h
merely indicates the horizontal and vertical components of a vector in Th Riem(Σ).

must be divergenceless (take the co-differential δ of this equation) and hence Killing.
The last statement follows without computation from the fact that D1 ξ = 0 is noth-
ing but the condition that Lξ h is G1 –orthogonal to all vertical vectors in Th Riem(Σ),
which for divergenceless ξ (traceless Lξ h) is equivalent to G0 –orthogonality, but then
positive definiteness of G0 and G0 (Lξ h, Lξ h) = 0 immediately imply Lξ h = 0. This
shows the triviality of Verth ∩ Horλ
h.
The foregoing shows that G1 indeed defines a metric at T[h] SF (Σ) for Ricci-positive
Einstein geometries [h]. How does the signature of this metric compare to the signature
 9

(∞, ∞) at Ricci-negative geometries? The answer is surprising: Take, e.g., for [h] the
round geometry on Σ = S 3 . Then it can be shown that G1 defines a Lorentz geometry on
T[h] SF (Σ), that is with signature (1, ∞), containing exactly one negative direction [35].
This means that the signature of the metric defined at various points in Superspace
varies strongly, with intermediate transition regions where no metric can be defined at
all due to signature change. Figure 1 is an attempt to picture this situation.

4 Intermezzo: GR as simplest representation of symmetry

It is well known that the field equations of GR have certain uniqueness properties
and can accordingly be ‘deduced’ under suitable hypotheses involving a symmetry
principle (diffeomorphism invariance), the equivalence principle, and some apparently
mild technical hypotheses. More precisely, the equivalence principle suggests to only
take the metric as dynamical variable [76] representing the gravitational field (to which
matter then couples universally), whereas diffeomorphism invariance, derivability from
an invariant Lagrangian (alternatively: local energy-momentum conservation in the
sense of covariant divergencelessness), dependence of the equations on the metric up
to at most second derivatives, and, finally, four-dimensionality lead uniquely to the
left-hand side of Einstein’s equation, including a possibly non-vanishing cosmological
constant [54]. Here we will review how this ‘deduction’ works in the Hamiltonian setting
on phase space T ∗ Riem(Σ), which goes back to [40, 74, 51, 41].

4.1 3+1 decomposition

Since the 3+1 split of Einstein’s equations has already been introduced in Claus Kiefer’s
contribution I can be brief on that point. The basic idea is to first imagine a spacetime
(M, g) being given, where topologically M is a product R × Σ. Spacetime is then
considered as the trajectory (history) of space in the following way: Let Emb(Σ,M )
denote the space of smooth spacelike embeddings Σ → M . We consider a curve R ∋
t → Et ∈ Emb(Σ,M ) corresponding to a one-parameter family of smooth embeddings
with spacelike images. We assume the images Et (Σ) =: Et ⊂ M to be mutually disjoint
and moreover that Ê : R × Σ → M , (t, p) 7→ Et (p), is an embedding (it is sometimes
found convenient to relax this condition, but this is of no importance here). The Lorentz
manifold (R × Σ, E∗ g) may now be taken as (E–dependent) representative of M (or at
least some open part of it) on which the leaves of the above foliation simply correspond
to the t = const. hypersurfaces. Let n denote a field of normalised timelike vectors
normal to these leaves. n is unique up to orientation, so that the choice of n amounts
to picking a ‘future direction’.
The tangent vector dEt /dt|t=0 at E0 ∈ Emb(Σ,M ) corresponds to a vector field
over E0 (i.e. section in T (M )|E0 ), given by

dEt (p) ˛ ∂˛
˛ ˛
=: = αn + β (18)
dt t=0 ∂t E0 (p)
˛ ˛

with components (α, β) = (lapse, shift) normal and tangential to Σ0 ⊂ M .

Conversely, each vector field V on M defines a vector field X(V ) on Emb(Σ,M ),
corresponding to the left action of Diff(M ) on Emb(Σ,M ) by composition. In local
10

coordinates y µ on M and xk on Σ it can be written as

Z
δ
X(V ) = d3 x V µ (y(x)) µ . (19)
Σ δy (x)

One easily verifies that X : V →

7 X(V ) is a Lie homomorphism:
ˆ ˜ ` ´
X(V ), X(W ) = X [V, W ] . (20)

In this sense, the Lie algebra of the four-dimensional diffeomorphism group is imple-
mented on phase space of any generally covariant theory whose phase space includes
the embedding variables [44] (so-called ‘parametrised theories’).
Alternatively, decomposing (19) into normal and tangential components with re-
spect to the leaves of the embedding at which the tangent-vector field to Emb(Σ,M ) is
evaluated, yields an embedding-dependent parametrisation of X(V ) in terms of (α, β),
Z “ ” δ
X(α, β) = d3 x α(x)nµ [y](x) + β m (x)∂m y µ (x) , (21)
Σ δy µ (x)

where y in square brackets indicates the functional dependence of n on the embedding.

The functional derivatives of n with respect to y can be computed (see the Appendix
of [74]) and the commutator of deformation generators then follows to be,

X(α1 , β1 ) , X(α2 , β2 ) = − X(α′ , β ′ ) ,

ˆ ˜
(22)

where

α′ = β1 (α2 ) − β2 (α1 ) , (23a)

′
β = [β1 , β2 ] + σα1 gradh (α2 ) − σα2 gradh (α1 ) . (23b)

Here we left open whether spacetime M is Lorentzian (σ = 1) or Euclidean (σ = −1),

just in order to keep track how the signature of spacetime, (−σ, +, +, +), enters. Note
that the h-dependent gradient field for the scalar function α is given by gradh (α) =
(hab ∂b α)∂a . The geometric idea behind (23) is summarised in Figure 2.

Σ21
)
, β1
(α 1

Σ2
)
, β2
(α 2

Σ (α′ , β ′ )
(α
1,β
1)

Σ1
(α
2,β
2)
Σ12
Fig. 2 An (infinitesimal) hypersurface deformation with parameters (α1 , β1 ) that maps
Σ 7→ Σ1 , followed by one with parameters (α2 , β2 ) that maps Σ1 7→ Σ12 differs by one
with parameters (α′ , β ′ ) given by (23) from that in which the maps with the same parameters
are composed in the opposite order.
 11

4.2 Hamiltonian geometrodynamics

The idea of Hamiltonian Geometrodynamics is to realise these relations in terms of a

Hamiltonian system on the phase space of physical fields. The most simple case is that
where the latter merely include the spatial metric h on Σ, so that the phase space is
the cotangent bundle T ∗ Riem(Σ) over Riem(Σ). One then seeks a correspondence

(α, β) 7→ H(α, β) : T ∗ Riem(Σ) → R ,

` ´
(24)

where
Z
d3 x α(x)H[h, π](x) + hab (x)β a (x)D b [h, π](x) ,
` ´
H(α, β)[h, π] := (25)
Σ

with integrands H[h, π](x) and D b [h, π](x) yet to be determined. H should be re-
garded as distribution (here the test functions are α and β a ) with values in real-valued
functions on T ∗ Riem(Σ). Now, the essential requirement is that the Poisson brackets
between the H(α, β) are, up to a minus sign,5 as in (23):

H(α1 , β1 ) , H(α2 , β2 ) = H(α′ , β ′ ) .

˘ ¯
(26)

Once the distribution H satisfying (26) has been found, we can turn around the
arguments given above and recover the action of the Lie algebra of four-dimensional
diffeomorphism on the extended phase space including embedding variables [45]. That
such an extension is indeed necessary has been shown in [64], where obstructions against
the implementation of the action of the Lie algebra of four-dimensional diffeomorphisms
have been identified in case the dynamical fields include non-scalar ones.

4.3 Why constraints

From this follows a remarkable uniqueness result. Before stating it with all its hypothe-
ses, we show why the constraints H[h, π] = 0 and D b [h, π] = 0 must be imposed.
Consider the set of smooth real-valued functions on phase space, ˘F : T ∗ Riem(Σ)
¯ →
R. They are acted upon by all H(α, β) via Poisson bracketing: F 7→ F, H(α, β) . This
defines a map from (α, β) into the derivations of phase-space functions. We require this
map to also respect the commutation relation (26), that is, we require

F, H(α1 , β1 ) , H(α2 , β2 ) − F, H(α2 , β2 ) , H(α1 , β1 ) = F, H (α′ , β ′ ) . (27)

˘˘ ¯ ¯ ˘˘ ¯ ¯ ˘ ¯

The subtle point to be observed here is the following: Up to now the parameters (α1 , β1 )
and (α2 , β2 ) were considered as given functions of x ∈ Σ, independent of the fields
h(x) and π(x), i.e. independent of the point of phase space. However, from (23b) we
see that β ′ (x) does depend on h(x). This dependence may not give rise to extra terms
∝ {F, α′ } in the˘ Poisson ¯bracket, for, otherwise, the extra terms would prevent the
map (α, β) 7→ −, H(α, β) from being a homomorphism from the algebraic structure
of hypersurface deformations into
˘ the derivations
¯ of phase-space functions. This is
necessary in order to interpret −, H(α, β) as a generator (on phase-space functions)
5 Due to the standard convention that the Hamiltonian action being defined as a left action,

whereas the Lie bracket on a group is defined by the commutator of left-invariant vector fields
which generate right translations.
12

of a spacetime evolution corresponding to a normal lapse α and tangential shift β. In

other words, the evolution of observables from an initial hypersurface Σi to a final
hypersurface Σf must be independent of the intermediate foliation (‘integrability’ or
‘path independence’ [74, 40, 41]). Therefore we placed the parameters (α′ , β ′ ) outside
the Poisson bracket on the right-hand side of (27), to indicate that no differentiation
with respect to h, π should act on them.
To see that this requirement implies the constraints, rewrite the left-hand side of
(27) in the form
˘˘ ¯ ¯ ˘˘ ¯ ¯
F, H(α1 , β1 ) , H(α2 , β2 ) − F, H(α2 , β2 ) , H(α1 , β1 )
˘ ˘ ¯¯
= F, H(α1 , β1 ), H(α2 , β2 )
(28)
= F, H(α′ , β ′ )
˘ ¯

= F, H (α′ , β ′ ) + H {F, α′ } , {F, β ′ } ,

˘ ¯ ` ´

where the first equality follows from the Jacobi identity, the second from (26), and the
third from the Leibniz rule. Hence the requirement (27) is equivalent to
H {F, α′ } , {F, β ′ } = 0
` ´
(29)
for all phase-space functions F to be considered and all α′ , β ′ of the form (23). Since
only β ′ depends on phase space, more´ precisely on h, this implies the vanishing of
the phase-space functions H 0, {F, β ′ } for all F and all β ′ of the form (23b). This
`

can be shown to imply H(0, β) = 0, ˘i.e. D[h, π] = 0. ¯Now, in turn, for this to be
preserved
˘ under all
¯ evolutions we need H(α, β̃), H(0, β) = 0, and hence in particular
H(α, 0), H(0, β) = 0 for all α, β, which implies H(α, 0) = 0, i.e. H[h, π] = 0. So we
see that the constraints indeed follow.
Sometimes the constraints H(α, β) = 0 are split into the Hamiltonian (or scalar)
constraints, H(α, 0) = 0, and the diffeomorphisms (or vector) constraints, H(0, β) =
0. The relations (26) with (23) then show that the `vector constraints form a Lie-
subalgebra which, because of {H(0, β), H(α, 0)} = H β(α), 0 6= H(0, β ′ ), is not an
´

ideal. This means that the Hamiltonian vector fields for the scalar constraints are not
tangent to the surface of vanishing vector constraints, except where it intersects the
surface of vanishing scalar constraints. This implies that the scalar constraints do not
act on the solution space for the vector constraints, so that one simply cannot first
reduce the vector constraints and then, on the solutions of that, search for solutions to
the scalar constraints. Also, it is sometimes argued that the scalar constraints should
not be regarded as generators of gauge transformations but rather as generators of
physically meaningful motions whose effect is to change the physical state in a fashion
that is, in principle, observable. See [52] and also [7] and Sect. 2.3 of Claus Kiefer’s
contribution for a recent revival of that discussion. However, it seems inconsistent to
me to simultaneously assume 1) physical states to always satisfy the scalar constraints
and 2) physical observables to exist which do not Poisson commute with the scalar
constraints: The Hamiltonian vector field corresponding to such an ‘observable’ will
not be tangent to the surface of vanishing scalar constraints and hence will transform
physical to unphysical states upon being actually measured.

4.4 Uniqueness of Einstein’s geometrodynamics

It is sometimes stated that the relations (26) together with (23) determine the function
H(α, β) : T ∗ Riem(Σ) → R, i.e. the integrands H[h, π] and D[h, π], uniquely up to two
 13

free parameters, which may be identified with the gravitational and the cosmological
constants. This is a mathematical overstatement if read literally, since the result can
only be shown if certain additional assumptions are made concerning the action of
H(α, β) on the basic variables h and π.
The first such assumption concerns the intended (‘semantic’ or ‘physical’) meaning
of H(0, β), namely that the action of H(0, β)} on h or π is that of an infinitesimal
spatial diffeomorphism of Σ. Hence it should be the spatial Lie derivative, Lβ , applied
to h or π. It then follows from the general Hamiltonian theory that H(0, β) is given by
the momentum map that maps the vector field β (viewed as element of the Lie algebra
of the group of spatial diffeomorphisms) into the function on phase space given by the
contraction of the momentum with the β-induced vector field h → Lβ h on Riem(Σ):
Z Z
H(0, β) = d3 x π ab (Lβ h)ab = −2 d3 x(∇a π ab )hbc β c . (30)
Σ Σ

Comparison with (25) yields

D b [h, π] = −2∇a π ab . (31)
The second assumption concerns the intended (‘semantic’ or ‘physical’) meaning of
H(α, 0), namely that {−, H(α, 0)} acting on h or π is that of an infinitesimal ‘timelike’
diffeomorphism of M normal to the leaves Et (Σ). If M were given, it is easy to prove
that we would have Lαn h = 2α K, where n is the timelike field of normals to the leaves
Et (Σ) and K is their extrinsic curvature. Hence one requires
{h, H(α, 0)} = 2α K . (32)
Note that both sides are symmetric covariant tensor fields over Σ. The important fact
to be observed here is that α appears without differentiation. This means that H(α, 0)
is an ultralocal functional of π, which is further assumed to be a polynomial. (Note
that we do not assume any relation between π and K at this point).
Quite generally, we wish to stress the importance of such ’semantic’ assumptions
concerning the intended meanings of symmetry operations when it comes to ‘deriva-
tions’ of physical laws from ‘symmetry principles’. Such derivations often suffer from
the same sort of overstatement that tends to give the impression that the mere require-
ment that some group G acts as symmetries alone distinguishes some dynamical laws
from others. Often, however, additional assumptions are made that severely restrict
the form in which G is allowed to act. For example, in field theory, the requirement
of locality often enters decisively, like in the statement that Maxwell’s vacuum equa-
tions are Poincaré- but not Galilei invariant. In fact, without locality the Galilei group,
too, is a symmetry group of vacuum electrodynamics [29]. Coming back to the case
at hand, I do not know of a uniqueness result that does not make the assumptions
concerning the spacetime interpretation of the generators H(α, β). Compare also the
related discussion in [66, 67].
The uniqueness result for Einstein’s equation, which in its space-time form is spelled
out in Lovelock’s theorem [54] already mentioned above, now takes the following form
in Geometrodynamics [51]:
Theorem 1 In four spacetime dimensions (Lorentzian for σ = 1, Euclidean for σ =
−1), the most general functional (25) satisfying (26) with (23), subject to the conditions
discussed above, is given by (31) and the two-parameter (κ, Λ) family
H[h, π] = σ (2κ) Gab cd π ab π cd − (2κ)−1 det(h) R(h) − Λ ,
p ` ´
(33)
14

where
1
hac hbd + had hbc − 12 hab hcd ,
` ´
Gab cd = √ (34)
2 det(h)

and R(h) is the Ricci scalar of (h, Σ). Note that (34) is just the “contravariant version”
of the metric (9c) for λ = 1, i.e., Gab nm Gnm cd = 21 (δac δbd + δad δbc ).

The Hamiltonian evolution so obtained is precisely that of General Relativity (with-

out matter) with gravitational constant κ = 8πG/c4 and cosmological constant Λ. The
proof of the theorem is given in [51], which improves on earlier versions [74, 41] in that
the latter assumes in addition that H[h, π] be an even function of π, corresponding to
the requirement of time reversibility of the generated evolution. This was overcome in
[51] by the clever move to write the condition set by {H(α1 , 0), H(α2 , 0)} = H(0, β ′ )
(the right-hand side being already known) on H(α, 0) in terms of the corresponding
Lagrangian functional L, which is then immediately seen to turn into a condition which
is linear in L, so that terms with even powers in velocity decouple form those with odd
powers. However, a small topological subtlety remains that is neglected in all these
references and which potentially introduces a little more ambiguity that that encoded
in the two parameters κ and Λ, though its significance is more in the quantum theory.
To see this recall that we can always perform a canonical transformation of the form

π 7→ π ′ := π + Θ (35)

where Θ is a closed one-form on Riem(Σ). The latter condition ensures that all Poisson
brackets remain the same if π is replaced with π ′ . Since Riem(Σ) is an open positive
convex cone in a vector space and hence contractible, it is immediate that Θ = dθ for
some function θ : Riem(Σ) → R. However, π and π ′ must satisfy the diffeomorphism
constraint, which is equivalent to saying that the kernel of π (considered as one-form on
Riem(Σ)) contains the vertical vector fields, which implies that Θ, too, must annihilate
all Vξ so that θ is constant on each connected component of the DiffF (Σ) orbit in
Riem(Σ). But unless the DiffF (Σ) orbits in Riem(Σ) are connected, this does not
mean that θ is the pull back of a function on Superspace, as assumed in [51]. We can
only conclude that Θ is the pull back of a closed but not necessarily exact one-form
on Superspace. Hence there is an analogue of the Bohm-Aharonov-like ambiguity that
one always encounters if the configuration space is not simply connected. Whether this
is the case depends in a determinate fashion on the topology of Σ: One has, due to the
contractibility of Riem(Σ),

πn Riem(Σ)/DiffF (Σ) ∼
` ´ ` ´
= πn−1 DiffF (Σ) (n ≥ 1) . (36)

For n = 1 the right hand side is

π0 DiffF (Σ) := DiffF (Σ)/Diff 0F (Σ) =: MCGF (Σ)

` ´
(37)

where Diff 0F (Σ) is the identity component of DiffF (Σ) and where we introduced the
name MCGF (Σ) (Mapping-Class Group for Frame fixing diffeomorphisms) for the
quotient group of components.
In view of the uniqueness result above, one might wonder what goes wrong when
using (the contravariant version of) the metric Gλ for λ 6= 1 in (34). The answer is
that it would spoil (26). More precisely, it would contradict {H(α1 , 0) , H(α2 , 0)} =
H(0, α1 ∇α2 − α2 ∇α1 ) due to an extra term ∝ (hab π ab )2 in H[h, π], unless the ad-
ditional constraint hab π ab = 0 were imposed, which is equivalent to Trh (K) = 0 and
 15

hence to the condition that only maximal slices are allowed [35]. But this is clearly
unacceptable (cf. Sect. 6 of [8]).
As a final comment about uniqueness of representations of (26) we mention the ap-
parently larger ambiguity—labelled by an additional C-valued parameter, the Barbero-
Immirzi parameter—that one gets if one uses connection variables rather than metric
variables (cf. [6, 42], Sect. 4.2.2 of [75], and Sect. 4.3.1 of [48]). However, in this case
one does not represent (26) but a semi-direct product of it with the Lie algebra of
SU (2) gauge transformations, so that after taking the quotient with respect to the lat-
ter (which form an ideal) our original (26) is represented non locally. Also, unless the
Barbero-Immirzi parameter takes the very special values ±i (for Lorentzian signature;
±1 for Euclidean signature) the connection variable does not admit an interpretation
as a space-time gauge field restricted to spacelike hypersurfaces (cf. [42, 67]). For ex-
ample, the holonomy of a spacelike curve γ varies with the choice of the spacelike
hypersurface containing γ, which would be impossible if the spatial connection were
the restriction of a space-time connection [67]. Accordingly, the dynamics generated by
the constraints does then not admit the interpretation of being induced by appropri-
ately moving a hypersurface through a spacetime with fixed geometric structures on
it. Consequently, the argument provided here for why one should require (26) in the
first place does, strictly speaking, not seem to apply in case of connection variables.
It is therefore presently unclear to me on what set of assumptions a uniqueness result
could be based in this case.

5 Topology of configuration space

Much of the global topology of SF (Σ) is encoded in its homotopy groups, which, in turn
are given by those of DiffF (Σ) according to (36). Their structures were investigated in
[83, 33, 36]. Early references as to their possible relevance in quantum gravity are [28,
43, 69, 71].
We start by remarking that topological invariants of SF (Σ) are also topological
invariants of Σ, which need not be homotopy invariant of Σ even if they are homotopy
invariants of SF (Σ). This is, e.g., the case for the mapping-class group of homeomor-
phisms [55] and hence (in 3 dimensions) also for the mapping-class group MCGF (Σ).
Remarkably, this means that we may distinguish homotopy equivalent but non homeo-
morphic 3-manifolds by looking at homotopy invariants of their associated Superspaces.
Examples for this are given by certain types of lens spaces. First recall the definition
of lens spaces L(p, q) in 3 dimensions: L(p, q) = S 3 /∼, where (p, q) is a pair of pos-
itive coprime integers with p > 1, S 3 = {(z1 , z2 ) ∈ C2 | |z1 |2 + |z2 |2 = 1}, and
(z1 , z2 ) ∼ (z1′ , z2′ ) ⇔ z1′ = exp(2πi/p)z1 , and z2′ = exp(2πi q/p)z2 . One way to picture
them is to take a solid ball in R3 and identify each point on the upper hemisphere with
a points on the lower hemisphere after a rotation by 2πq/p about the vertical symmetry
axis. (Usually one depicts the ball in a way in which it is slightly squashed along the
vertical axis so that the equator develops a sharp edge and the whole body looks like
a lens; see e.g. Fig. in [68].) In this way each set of p equidistant points on the equator
is identified to a single point. The fundamental group of L(p, q) is Zp , independent of
q, and the higher homotopy groups are those of its universal cover, S 3 . Moreover, for
connected closed orientable 3-manifolds the homology and cohomology groups are also
determined by the fundamental group in an easy fashion: If A denotes the operation
of abelianisation of a group, F the operation of taking the free part of a finitely gen-
16

erated abelian group, then the first four (zeroth to third, the only non-trivial ones)
homology and cohomology groups are respectively given by H∗ = (Z, Aπ1 , F Aπ1 , Z)
and H ∗ = (Z, F Aπ1 , Aπ1 , Z) respectively. Hence, if taken of L(p, q), all these standard
invariants are sensitive only to p. However, it is known that L(p, q) and L(p, q ′ ) are
1. homotopy equivalent iff qq ′ = ±n2 (mod p) for some integer n,
2. homeomorphic iff (all four possibilities) q ′ = ±q ±1 (mod p), and
3. orientation-preserving homeomorphic iff q ′ = q ±1 (mod p).
The first statement is Theorem 10 in [82] and the second and third statement follow,
e.g., from the like combinatorial classification of lens spaces [65] together with the
validity of the ‘Hauptvermutung’ (the equivalence of the combinatorial and topological
classifications) in 3 dimensions [60]. So, for example, L(15, 1) is homotopy equivalent
but not homeomorphic to L(15, 4). On the other hand, it is known that the mapping-
class group MCGF (Σ) for L(p, q) is Z × Z if q 2 = 1 (mod p) with q 6= ±1 (mod p),
which applies to p = 15 and q = 4, and that in the remaining cases for p > 2 it is
just Z (see Table IV on p. 591 of [83]). Hence MCGF (Σ) ∼ = Z × Z for Σ = L(15, 4)
and MCGF (Σ) ∼ = Z for Σ = L(15, 1), even though L(15, 1) and L(15, 4) are homotopy
equivalent!
Quite generally it turns out that Superspace stores much information about the
topology of the underlying 3-manifold Σ. This can be seen from the table in Figure 3,
which we reproduced form [32], and where properties of certain prime manifolds (see
below for an explanation of ‘prime’) are listed. There is one interesting observation
from that list which we shall mention right away: From gauge theories it is known that
there is a relation between topological invariants of the classical configuration space
and certain features of the corresponding quantum-field theory [46], in particular the
emergence of certain anomalies which represent non-trivial topological invariants [1].
By analogy one could conjecture similar relations to hold quantum gravity. An inter-
esting question is then whether there are preferred manifolds Σ for which all these
invariants are trivial. From those represented on the table there is indeed a unique pair
of manifolds for which this is the case, namely the 3-sphere and the 3-dimensional real
projective space. To understand more of the information collected in the table we have
to say more about general 3-manifolds.
Of particular interest is the fundamental group of Superspace. Experience with
ordinary quantum mechanics (cf. [34] and references therein) already suggests that
its classes of inequivalent irreducible unitary representations correspond to a supers-
election structure which here might serve as fingerprint of the topology of Σ in the
quantum theory. The sectors might, e.g., correspond to various statistics (in the pres-
ence of diffeomorphic primes) that preserve or violate a naively expected spin-statistics
correlation [3, 2, 17, 18] (see also below).

5.1 General three-manifolds and specific examples

The way to understand general 3-manifolds is by cutting them along certain embedded
two manifolds so that the remaining pieces are simpler in an appropriate sense. Here we
shall only consider those simplifications that are achieved by cutting along embedded
2-spheres. (Further decompositions by cutting along 2-tori provide further simplifica-
tions, but these are not directly relevant here.) The 2-spheres should be ‘essential’ and
‘splitting’. An essential 2-sphere is one which does not bound a 3-ball and a splitting
 17

Prime Π HC S C N H1 (Π) π0 (DF (Π)) π1 (DF (Π)) πk (DF (Π))

S 3 /D8∗ + + + − Z2 × Z2 O∗ 0 πk (S 3 )
S 3 /D8n
∗ + + + − Z2 × Z2 ∗
D16n 0 πk (S 3 )
S 3 /D4(2n+1)
∗ + + + + Z4 ∗
D8(2n+1) 0 πk (S 3 )
S 3 /T ∗ ? + + − Z3 O∗ 0 πk (S 3 )
S 3 /O∗ w + + + Z2 O∗ 0 πk (S 3 )
S 3 /I ∗ ? + + − 0 I∗ 0 πk (S 3 )

S 3 /D8∗ × Zp + + + − Z2 × Z2p Z2 × O∗ Z πk (S 3 ) × πk (S 3 )
S 3 /D8n
∗ ×Z
p + + + − Z2 × Z2p ∗
Z2 × D16n Z πk (S 3 ) × πk (S 3 )
S 3 /D4(2n+1)
∗ × Zp + + + + Z4p ∗
Z2 × D8(2n+1) Z πk (S 3 ) × πk (S 3 )
S 3 /T ∗ × Zp ? + + − Z3p Z2 × O∗ Z πk (S 3 ) × πk (S 3 )
S 3 /O∗ × Zp w + + + Z2p Z2 × O∗ Z πk (S 3 ) × πk (S 3 )
S 3 /I ∗ × Zp ? + + − Zp Z2 × I∗ Z πk (S 3 ) × πk (S 3 )
S 3 /D2′ k (2n+1) × Zp + + + + Zp × Z2k Z2 × ∗
D8(2n+1) Z πk (S 3 ) × πk (S 3 )
S 3 /T8·3
′
m × Zp ? + + − Zp × Z3m O∗ Z πk (S 3 ) × πk (S 3 )

L(p, q1 ) w − + (−)p Zp Z2 Z πk (S 3 )
L(p, q2 ) w+ − + (−)p Zp Z2 × Z2 Z×Z πk (S 3 ) × πk (S 3 )
L(p, q3 ) w − − (−)p Zp Z2 Z×Z πk (S 3 ) × πk (S 3 )
L(p, q4 ) w − + (−)p Zp Z2 Z×Z πk (S 3 ) × πk (S 3 )

RP 3 + − − + Z2 1 0 0
S3 + − − − 1 1 0 0

S2 × S1 / − − + Z Z2 × Z2 Z πk (S 3 ) × πk (S 2 )

R3 /G1 / + − + Z ×Z ×Z St(3, Z) 0 πk (S 3 )
R3 /G2 / + − + Z × Z2 × Z2 AutZ2
+ (G2 ) 0 πk (S 3 )
R3 /G3 / + + + Z × Z3 AutZ2
+ (G3 ) 0 πk (S 3 )
R3 /G 4 / + + − Z × Z2 AutZ2
+ (G4 ) 0 πk (S 3 )
R3 /G 5 / + + + Z AutZ2
+ (G5 ) 0 πk (S 3 )
R3 /G 6 / + + − Z4 × Z4 AutZ2
+ (G5 ) 0 πk (S 3 )

S 1 × Rg / + − − Z × Z2g AutZ2
+ (Z × Fg ) 0 πk (S 3 )

K(π, 1)sl / + ∗ ∗ Aπ AutZ2

+ (π) 0 πk (S 3 )

Fig. 3 This table, taken from [32], lists various properties of certain prime 3-manifolds. The
manifolds are grouped into those of finite fundamental group, which are of the form S 3 /G,
the exceptional one, S 1 × S 2 , which is prime but not irreducible (π2 (S 1 × S 2 ) = Z), those six
which can carry a flat metric and which are of the form R3 /G, and so-called Haken manifolds
(sufficiently large K(π, 1) primes). For the lens spaces q1 stands for q = ±1, q2 for q 6= ±1 and
q 2 = 1, q3 for q 2 = −1, and q4 for the remaining cases, where all equalities are taken mod p. The
3rd and 4th column list spinoriality and chirality, the last three columns the homotopy groups
of their corresponding superspace. We refer to [32] for the meanings of the other columns.

2-sphere is one whose complement has two (rather than just one) connected compo-
nents. Figure 4 is intended to visualise the analogues of these notions in two dimensions.
Given a closed 3-manifold Σ, consider the following process: Cut it along an essential
splitting 2-sphere and cap off the 2-sphere boundary of each remaining component by
a 3-disk. Now repeat the process for each of the remaining closed 3-manifolds. This
18

B C
A
A
B
C

Fig. 4 A Riemann surface of genus 3 with three pairs of embedded 1-spheres (circles) of type
A, B, and C. Type A is essential and splitting, type B is essential but not splitting, and type
C is splitting but not essential. Any third essential and splitting 1-sphere can be continuously
deformed via embeddings into one of the two drawn here.

process stops after a finite number of steps [50] where the resulting components are
uniquely determined up to diffeomorphisms (orientation preserving if oriented mani-
folds are considered) and permutation [58]; see [39] for a lucid discussion. The process
stops at that stage at which none of the remaining components, Π1 , · · · , Πn , allows for
essential splitting 2-spheres, i.e. at which each Πi is a prime manifold. A 3-manifold
is called prime if each embedded 2-sphere either bounds a 3-disc or does not split; it
is called irreducible if each embedded 2-sphere bounds a 3-disc. In the latter case the
second homotopy group, π2 , must be trivial, since, if it were not, the so-called sphere
theorem (see, e.g., [39]) ensured the existence of a non-trivial element of π2 which
could be represented by an embedded 2-sphere. Conversely, it follows from the validity
of the Poincaré conjecture that a trivial π2 implies irreducibility. Hence irreducibility
is equivalent to a trivial π2 . There is precisely one non-irreducible prime 3-manifold,
and that is the handle S 1 × S 2 . Hence a 3-manifold is prime iff it is either a handle or
if its π2 is trivial.
Given a general 3-manifold Σ as connected sum of primes Π1 , · · · , Πn , there is a
general method to establish MCGF (Σ) in terms of the individual mapping-class groups
of the primes. The strategy is to look at the effect of elements in MCGF (Σ) on the
fundamental group of Σ. As Σ is the connected sum of primes, and as connected sums
in d dimensions are taken along d − 1 spheres which are simply-connected for d ≥ 3, the
fundamental group of a connected sum is the free product of the fundamental groups
of the primes for d ≥ 3. The group MCGF (Σ) now naturally acts as automorphisms of
π(Σ) by simply taking the image of a based loop that represents an element in π(Σ)
by a based (same basepoint) diffeomorphism that represents the class in MCGF (Σ).
Hence there is a natural map
` ´
dF : MCGF (Σ) → Aut π1 (Σ) . (38)

The known presentations6 of automorphism groups of free products in terms of pre-

sentations of the automorphisms of the individual factors and additional generators
(basically exchanging isomorphic factors and conjugating whole factors by individual
elements of others) can now be used to establish (finite) presentations of MCGF (Σ),
provided (finite) presentations for all prime factors are known.7 Here I wish to stress
that this situation would be more complicated if Diff(Σ) rather than DiffF (Σ) (or at
least the diffeomorphisms fixing a preferred point) had been considered; that is, had
6 A (finite) presentation of a group is its characterisation in terms of (finitely many) gener-

ators and (finitely many) relations.

7 This presentation of the automorphism group of free products is originally due to Fouxe-

Rabinovitch [25, 26]. Modern forms with corrections are given in [57] and [31]
 19

we not made the transition from (1) to (4). Only for DiffF (Σ) (or the slightly larger
group of diffeomorphisms fixing the point) is it generally true that the mapping-class
group of a prime factor injects into the mapping-class group of the connected sum in
which it appears. For more on this, compare the discussion on p. 182-3 in [37]. Clearly,
one also needs to know which elements are in the kernel of the map (38). This will be
commented on below in connection with Fig 7.

S1

S2

Fig. 5 The connected sum of two real projective spaces may be visualised by the shaded
region obtained from the spherical shell that is obtained by rotating the shaded annulus about
the vertical symmetry axis as indicated. Antipodal points on the outer 2-sphere boundary
S1 , as well as on the inner 2-sphere boundary S2 , are pairwise identified. This results in the
connected sum of two RP 3 along the connecting sphere S. Due to the antipodal identifications,
the two thick horizontal segments in the shaded region become a single loop, showing that the
entire space is fibred by circles over RP 2 . Remarkably, the handle S 1 × S 2 , which is prime, is
a double cover of this reducible manifold.

a b

Fig. 6 The shown loops represent the generators a and b of the fundamental group Z2 ∗
Z2 = ha, b | a2 = b2 = 1i. The product loop c = ab is then seen to be one of the fibres
mentioned in the caption of the previous Figure. In terms of a and c we have the presentation
ha, c | a2 = 1, aca−1 = ci, showing the isomorphicity Z2 ∗ Z2 ∼
= Z2 ⋉ Z.
20

5.2 The connected sum of two real-projective spaces

In some (in fact many) cases the map dF is an isomorphism. For example, this is
the case if Σ is the connected sum of two RP 3 , so that π1 (Σ) is the free product
Z2 ∗ Z2 , a presentation of which is ha, b | a2 = b2 = 1i. For the automorphisms we
have Aut(Z2 ∗ Z2 ) ∼= Z2 ∗ Z2 = hE, S | E 2 = S 2 = 1i, where E : (a, b) → (b, a) and
−1
S : (a, b) → (a, aba ). In this sense the infinite discrete group Z2 ∗ Z2 is a quotient
of the automorphism group of Superspace SF (Σ) for Σ being the connected sum of
two real projective spaces. It is therefore of interest to study its unitary irreducible
representations. This can be done directly in a rather elementary fashion, or more sys-
tematically by a simple application of the method of induced representations (Mackey
theory) using the isomorphicity Z2 ∗ Z2 ∼ = Z2 ⋉ Z (cf. caption to Fig. 6). The result is
that, apart form the obvious four one-dimensional ones, given by (E, S) 7→ (±id, ±id),
there is a continuous set of mutually inequivalent two-dimensional ones, given by

„ « „ «
1 0 cos τ sin τ
E 7→ , S 7→ , τ ∈ (0, π) . (39)
0 −1 sin τ − cos τ

Already in this most simple example of a non-trivial connected sum we have an inter-
esting structure in which the two ‘statistics sectors’ corresponding to the irreducible
representations of the permutation subgroup (here just given by the Z2 subgroup gen-
erated by E) get mixed by S, where the ‘mixing angle’ τ uniquely characterises the rep-
resentation. This behaviour can also be studied in more complicated examples [32][72].
For a more geometric understanding of the maps representing E and S, see [37].

S2 S1

S1

S2
Fig. 7 Both pictures show rotations parallel to spheres S1 and S2 : On the left, a rotation of
a prime manifold in a connected sum parallel to the connecting sphere, on the right a rotation
parallel to two meridian spheres in a ‘handle’ S 1 × S 2 . The support of the diffeomorphism is
on the cylinder bound by S1 and S2 . In either case its effect is depicted by the two curves
connecting the two spheres. The two-dimensional representation given here is deceptive insofar,
as in two dimensions the original and the mapped curves are not homotopic (keeping their
endpoints fixed), due to the one-sphere not being simply connected, whereas they are in three
(and higher) higher dimensions they are due to the higher-dimensional spheres being simply
connected.
 21

5.3 Spinoriality

I also wish to mention one very surprising observation that was made by Rafael Sorkin
and John Friedman in 1980 [28] and which has to do with the physical interpretation
of the elements in the kernel of the map (38), leading to the conclusion that pure
(i.e. without matter) quantum gravity should already contain states with half-integer
angular momenta. The reason being a purely topological one, depending entirely on
the topology of Σ. In fact, given the right topology of Σ, its one-point decompactifi-
cation used in the context of asymptotically flat initial data will describe an isolated
system whose asymptotic (at spacelike infinity) symmetry group is not the ordinary
Poincaré group [10] but rather its double (= universal) cover. This gives an intriguing
answer to Wheeler’s quest to find a natural place for spin 1/2 in Einstein’s standard
geometrodynamics (cf. [59] Box 44.3).
I briefly recall that after introducing the concept of a ‘Geon’ (‘gravitational-
electromagnetic entity’) in 1955 [77], and inspired by the observation that electric
charge (in the sense of non-vanishing flux integrals of ⋆F over closed 2-dimensional
surfaces) could be realised in Einstein-Maxwell theory without sources (‘charge with-
out charge’), Wheeler and collaborators turned to the Einstein-Weyl theory [13] and
tried to find a ‘neutrino analog of electric charge’ [49]. Though this last attempt failed,
the programme of ‘matter as geometry’ in the context of geometrodynamics, as out-
lined in the contributions to the anthology [78], survived in Wheeler’s thinking well
into the 1980s [81].
Back to the ‘spin without spin’ topologies, the elements of the kernel of (38) can
be pictured as rotation parallel to certain spheres, as depicted in Figure 7. (In many—
and possibly all—cases the group generated by such maps actually exhaust the kernel;
compare Theorem. 1.5 in [56] and footnote 21 in [37]). The point we wish to focus
on here is that for some prime manifolds the diffeomorphism depicted on the left in
Figure 7 is indeed not in the identity component of all diffeomorphisms that fix a frame
exterior to the outer (S2 ) 2-sphere. Such manifolds are called spinorial. For each prime
it is known whether it is spinorial or not, and the easy-to-state but hard-to-prove
result is, that the only non-spinorial manifolds8 are the lens spaces L(p, q), the handle
S 1 × S 2 , and connected sums amongst them. That these manifolds are not spinorial
is, in fact, very easy to visualise. Hence, given the proof of the ‘only’ part and of the
fact that a connected sum is spinorial iff it contains at least one spinorial prime, one
may summarise the situation by saying that that the only non-spinorial manifolds are
the ‘obvious’ ones.
Even though being a generic property in the sense just stated, spinoriality is gen-
erally hard to prove in dimensions three or greater. This is in marked contrast to
two dimensions, where the corresponding transformation shown in the left picture of
Figure 7 acts non trivially on the fundamental group. Indeed, consider a base point
outside (below) S2 in the left picture in Figure 7, then the rotation acts by conjugating
each of the 2g generators (a1 · · · ag , b1 · · · b2 ) that Π adds to π1 (Σ) by the element
Qg −1 −1
i=1 ai bi ai bi , which is non-trivial in π1 (Σ) if other primes exist or otherwise if
the point with the fixed frame is removed (as one may do, due to the restriction to
diffeomorphisms fixing that point).

8 We remind the reader that ‘manifold’ here stands for ‘3-dimensional closed orientable

manifold’.
22

An example of a spinorial manifold is the spherical space form S 3 /D8∗ , where S 3 is

thought of as the sphere of unit quaternions and D8∗ is the subgroup in the group of
unit quaternions given by the eight elements {±1, ±i, ±j, ±k}. The coset space S 3 /D8∗
may be visualised as solid cube whose opposite faces are identified after a 90-degree
rotation by either a right- or a left-handed screw motion; see Figure 8. Drawings of
fundamental domains in form of (partially truncated) solid polyhedra with suitable
boundary identifications for spaces S 3 /G are given in [19]. Let us take the basepoint

A b B

d c

B a A
a
c

b
B d A

a b

c B
A

Fig. 8 Fundamental domain for the space S 3 /D8∗ .

Opposite faces are identified after a right-
handed screw motion with 90-degree rotation, as indicated by the coinciding labels for the
edges and vertices. The corresponding space with a left-handed identification is not orientation-
preserving diffeomorphic to this one.

∞ ∈ S 3 /D8∗ to be the centre of the cube in Figure 8. The two generators, a and b, of
the fundamental group

π1 (S 3 /D8∗ ) ∼ ∗ 2 2 2
= D8 = ha, b | a = b = (ab) i (40)

are then represented by two of the three oriented straight segments connecting the
midpoints of opposite faces. The third corresponds to the product ab. A rotation of
the cube about its centre by any element of its crystallographic symmetry group de-
fines a diffeomorphism of S 3 /D8∗ fixing ∞ since it is compatible with the boundary
identification. It is not in the identity component of ∞-fixing diffeomorphisms since it
obviously acts non-trivially on the generators of the fundamental group. Clearly, each
such rigid rotation may be modified in an arbitrarily small neighbourhood of ∞ so
as to also fix the tangent space at this point. That S 3 /D8∗ is spinorial means that in
going from the point-fixing to the frame-fixing diffeomorphisms one acquires more dif-
feomorphisms not connected to the identity. More precisely, the mapping-class group
of frame-fixing diffeomorphisms is a Z2 -extension of the mapping-class group of merely
point-fixing diffeomorphisms. The generator of this extending Z2 is a full 360-degree
rotation parallel to two small concentric spheres centred at ∞. In this way, the spino-
riality of S 3 /D8∗ extends the crystallographic symmetry group O ⊂ SO(3) of the cube
to its double cover O∗ ⊂ SU (2) and one finally gets

MCGF (Π) ∼
=O
∗
for Π = S 3 /D8∗ . (41)
 23

This is precisely what one finds at the intersection of the 2nd row and 7th column of
the table in Figure 3, with corresponding results for the other spherical space forms.
Coming back to the previous example of the connected sum of two (or more [32])
real projective spaces, we can make the following observation: First of all, real projec-
tive 3-space is a non-spinorial prime. This is obvious once one visualises it as a solid
3-ball whose 2-sphere boundary points are pairwise identified in an antipodal fashion,
since this identification is compatible with a rigid rotation. A full rotation about the,
say, centre-point of the ball may therefore be continuously undone by a rigid rota-
tion outside a small ball about the centre, suitably ‘bumped off’ towards the centre.
Second, as we have seen above, the irreducible representations of the mapping-class
group of the connected sum contains both statistics sectors independently for one-
dimensional representations and in a mixed form for the continuum of two-dimensional
irreducible representations. This already shows [2] that there is no general kinematical
spin-statistics relation as in other non-linear theories [21, 70]. Such a relation may at
best be re-introduced for some manifolds by restricting the way in which states are
constructed, e.g., via the sum-over-histories approach [17].

5.4 Chirality

There is one last aspect about diffeomorphisms that can be explained in terms of
Figure 8. As stated in the caption of this figure, there are two versions of this space:
one where the identification of opposite faces is done via a 90-degree right-handed
screw motion and one where one uses a left-handed screw motion. These spaces are
not related by an orientation preserving diffeomorphism. This is equivalent to saying
that, say, the first of these spaces has no orientation-reversing self-diffeomorphism.
Manifolds for which this is the case are called chiral. There are no examples in two
dimensions. To see this, just consider the usual picture of a Riemannian genus g surface
embedded into R3 and map it onto itself by a reflection at any of its planes of symmetry.
So chiral manifolds start to exist in 3 dimensions and continue to do so in all higher
dimensions, as was just recently shown [61]. If one tries to reflect the cube in Figure 8
at one of its symmetry planes one finds that this is incompatible with the boundary
identifications, that is, pairs of identified points are not mapped to pairs of identified
points. Hence this reflection simply does not define a map of the quotient space. This
clearly does not prove the nonexistence of orientation reversing maps, since there could
be others than these obvious candidates.
In fact, following an idea of proof in [83], the chirality of S G /D8∗ (and others of
the form S 3 /G) can be reduced to that of L(4, 1). That the latter is chiral follows
from the following argument: Above we have already stated that L(p, q) and L(p, q ′ )
are orientation preserving diffeomorphic iff q ′ = q ±1 (mod p). Since taking the mirror
image in R3 of the lens representing L(p, q) gives the lens representing L(p, −q), L(p, q)
admits an orientation reversing diffeomorphism iff L(p, q) and L(p, −q) are orientation
preserving diffeomorphic. But, as just stated, this is the case iff −q = q ±1 (mod p), i.e. if
either p = 2, q = 1 (recall that p and q must be coprime) or q 2 = −1 (mod p).9 Hence,
in particular, all L(p, 1) are chiral. Now, G = D8∗ has three subgroups isomorphic
to Z4 , the fundamental group of L(4, 1), namely the ones generated by i, j, and k.

9 Remarkably, this result was already stated in footnote 1 of p. 256 of [50]. An early published

proof is that in Sect. 77 of [68]

24

p
They are normal so that we have a regular covering L(4, 1)→S 3 /G. Now suppose
f : S 3 /G → S 3 /G were orientation reversing, i.e. a diffeomorphism with deg(f ) = −1.
Consider the diagram
f˜
L(4,1) / L(4,1)
KK
KK
KK
p f ◦p K p (42)
KK
KK
 K% 
S 3 /G / S 3 /G
f

If the lift f˜ existed we would immediately get a contradiction since from commutativity
of (42) we would get deg(f˜) · deg(p) = deg(p) · deg(f ) and hence deg(f˜) = −1, which
contradicts chirality of L(4, 1). Now, according ´ to the theory of covering spaces the
lift f˜ of f ◦ p exists iff the image of π1 `L(4, 1) ´ under (f ◦ p)∗ , which is a subgroup
`

Z4 ⊂ D8∗ , is conjugate to the image of π1 L(4, 1) under p∗ . This need not be the case,
however, as different subgroups Z4 in D8∗ are normal and hence never conjugate (here
we deviate from the argument in [83] which seems incorrect). However, by composing
a given orientation reversing f with an orientation preserving diffeomorphism that
undoes the Z4 subgroup permutation introduced by f , we can always create a new
orientation reversing diffeomorphism that does not permute the Z4 subgroups. That
new orientation reversing diffeomorphism—call it again f —now indeed has a lift f˜, so
that finally we arrive at the contradiction envisaged above.
As prime manifolds, the two versions of a chiral prime corresponding to the two
different orientations count as different. This means the following: Two connected sums
which differ only insofar as a particular chiral prime enters with different orientations
are not orientation-preserving diffeomorphic; they are not diffeomorphic at all if the
complement of the selected chiral prime also chiral (i.e. iff another chiral prime ex-
ists). For example, the connected sum of two oriented S 3 /D8∗ is not diffeomorphic to
the connected sum of S 3 /D8∗ with S 3 /D8∗ , where the overbar indicates the opposite
orientation. Note that the latter case also leads to two non-homeomorphic 3-manifolds
whose classic invariants (homotopy, homology, cohomology) coincide. This provides an
example of a topological feature of Σ that is not encoded into the structure of SF (Σ).

6 Summary and outlook

Superspace is for geometrodynamics what gauge-orbit space is for non-abelian gauge

theories, though Superspace has generally a much richer topological and metric struc-
ture. Its topological structure encodes much of the topology of the underlying 3-
manifold and one may conjecture that some of its topological invariants bear the same
relation to anomalies and sectorial structure as in the case of non-abelian gauge the-
ories. Recent progress in 3-manifold theory now allows to make more complete state-
ments, in particular concerning the fundamental groups of Superspaces associated to
more complicated 3-manifolds. Its metric structure is piecewise nice but also suffers
from singularities, corresponding to signature changes, whose physical significance is
unclear. Even for simple 3-manifolds, like the 3-sphere, there are regions in superspace
where the metric is strictly Lorentzian (just one negative signature and infinitely many
pluses), like at the round 3-sphere used in the FLRW cosmological models, so that the
 25

Wheeler-DeWitt equation becomes strictly hyperbolic, but there are also regions with
infinitely many negative signs in the signature.
Note that the cotangent bundle over Superspace is not the fully reduced phase
space for matter-free General Relativity. It only takes account of the vector con-
straints and leaves the scalar constraint unreduced. However, under certain condi-
tions, the scalar constraints can be solved by the ‘conformal method’ which leaves
only the conformal equivalence class of 3-dimensional geometries as physical con-
figurations. In those cases the fully reduced phase space is the cotangent bundle
over conformal superspace, whose analog to (1) is given by replacing Diff(Σ) by the
semi-direct product C(Σ) ⋊ Diff(Σ), where C(Σ) is the abelian group of conformal
rescalings that acts on Riem(Σ) via (f, h) 7→ f h (pointwise multiplication), where
f : Σ → R+ . The right action of (f, φ) ∈ C(Σ) ⋊ Diff(Σ) on h ∈ Riem(Σ) is then
given by by R(f,φ) (h) = f φ∗ h, so that, using R(f2 ,φ2 ) R(f1 ,φ1 ) = R(f1 ,φ1 )(f2 ,φ2 ) , the
` ´
semi-direct product structure is seen to be (f1 , φ1 )(f2 , φ2 ) = f2 (f1 ◦ φ2 ), φ1 ◦ φ2 .
Note that because of (f1 f2 ) ◦ φ = (f1 ◦ φ)(f2 ◦ φ) Diff(Σ) indeed acts as automor-
phisms of C(Σ). Conformal superspace and extended conformal superspace would
then, in analogy to (1) and (4), be defined as CS(Σ) := Riem(Σ)/C(Σ) ⋊ Diff(Σ)
and CSF (Σ) := Riem(Σ)/C(Σ) ⋊ DiffF (Σ) respectively. The first definition was used
in [24] as applied to manifolds with zero degree of symmetry (cf. footnote 3). In any
case, since C(Σ) is contractible, the topologies of C(Σ) ⋊ Diff(Σ) and C(Σ) ⋊ DiffF (Σ)
are those of Diff(Σ) and DiffF (Σ) which also transcend to the quotient spaces anal-
ogously to (36) whenever the groups act freely. In the first case this is essentially
achieved by restricting to manifolds of vanishing degree of symmetry, whereas in the
second case this follows almost as before, with the sole exception being (S 3 , h) with h
conformal to the round metric.10 Hence the topological results obtained before also ap-
ply to this case. In contrast, the geometry for conformal superspace differs insofar from
that discussed above as the conformal modes that formed the negative directions of the
Wheeler-DeWitt metric (cf. (9a) are now absent. The horizontal subspaces (orthogonal
to the orbits of C(Σ) ⋊ DiffF (Σ)) are now given by the transverse and traceless (rather
than just obeying (13)) symmetric two-tensors. In that sense the geometry of confor-
mal superspace, if defined as before by some ultralocal bilinear form on Riem(Σ), is
manifestly positive (due to the absence of trace terms) and hence less pathological than
the superspace metric discussed above. It might seem that its physical significance is
less clear, as there is now no constraint left that may be said to induce this particular
geometry; see however [9].
Whether it is a realistic hope to understand superspace and conformal superspace
(its cotangent bundle being the space of solutions to Einstein’s equations) well enough
to actually gain a sufficiently complete understanding of its automorphism group is
hard to say. An interesting strategy lies in the attempt to understand the solution
space directly in a group- (or Lie algebra-) theoretic fashion in terms of a quotient
G∞ /H∞ , where G∞ is an infinite dimensional group (Lie algebra) that (locally) acts
transitively on the space of solutions and H∞ is a suitable subgroup (algebra), usually
the fixed-point set of an involutive automorphism of G. The basis for the hope that

10 Let CI(Σ, h) := {φ ∈ Diff(Σ) | φ∗ h = f h, f : Σ → R } be the group of conformal

+
isometries. For compact Σ it is known to be compact except iff Σ = S 3 and h conformal to the
round metric [53]. Hence, for Σ 6= S 3 , we can average h over the compact group CI(Σ, h) and
obtain a new Riemannian metric h′ in the conformal equivalence class of h for which CI(Σ, h)
acts as proper isometries. Therefore, by the argument presented in Sect.2, it cannot contain
non-trivial elements fixing a frame.
26

this might work in general is the fact that it works for the subset of stationary and
axially symmetric solutions, where G∞ is the Geroch Group; cf. [12]. The idea for
generalisation, even to d = 11 supergravity, is expressed in [63] and further developed
in [14].

Acknowledgements I thank Hermann Nicolai and Stefan Theisen for inviting me to

this stimulating 405th WE-Heraeus-Seminar on Quantum Gravity and for giving me
the opportunity to contribute this paper. I also thank Hermann Nicolai for pointing
out [63] and Ulrich Pinkall for pointing out [53].

References

1. Luis Alvarez-Gaumé and Paul Ginsparg. The structure of gauge and gravitational anoma-
lies. Annals of Physics, 161:423–490, 1985.
2. Charilaos Aneziris et al. Aspects of spin and statistics in general covariant theories.
International Journal of Modern Physics A, 14(20):5459–5510, 1989.
3. Charilaos Aneziris et al. Statistics and general relativity. Modern Physics Letters A,
4(4):331–338, 1989.
4. Richard Arens. Classical Lorentz invariant particles. Journal of Mathematical Physics,
12(12):2415–2422, 1971.
5. Henri Bacry. Space-time and degrees of freedom of the elementary particle. Communica-
tions in Mathematical Physics, 5(2):97–105, 1967.
6. Fernando Barbero. Real Ashtekar variables for Lorentzian signature space-times. Physical
Review D, 51(10):5507–5510, 1995.
7. Julian Barbour and Brendan Z Foster. Constraints and gauge transformations: Dirac’s
theorem is not always valid. arXiv:0808.1223v1, 2008.
8. Julian Barbour, Brendan Z. Foster, and Niall Ó Murchadha. Relativity without relativity.
Classical and Quantum Gravity, 19(12):3217–3248, 2002.
9. Julian Barbour and Niall Ó Murchadha. Classical and quantum gravity on conformal
superspace. arXiv:gr-qc/9911071v1, 1999.
10. Robert Beig and Niall Ó Murchadha. The Poincaré group as symmetry group of canonical
general relativity. Annals of Physics, 174:463–498, 1987.
11. Arthur L. Besse. Einstein Manifolds. Ergebnisse der Mathematik und ihrer Grenzgebiete,
3. Folge Band 10. Springer Verlag, Berlin, 1987.
12. Peter Breitenlohner and Dieter Maison. On the Geroch group. Annales de l’Institut Henri
Poincaré, Section A, 46(2):215–246, 1987.
13. Dieter R. Brill and John A. Wheeler. Interaction of neutrinos and gravitational fields.
Review of Modern Physics, 29(3):465–479, 1957.
14. Thibault Damour, Axel Kleinschmidt, and Hermann Nicolai. Constraints and the E10
coset model. Classical and Quantum Gravity, 24(23):6097–6120, 2007.
15. Thibault Damour and Hermann Nicolai. Symmetries, singularities and the de-emergence
of space. arXiv:0705.2643v1, 2007.
16. Bryce Seligman DeWitt. Quantum theory of gravity. I. The canonical theory. Physical
Review, 160(5):1113–1148, 1967. Erratum, ibid. 171(5):1834, 1968.
17. Fay Dowker and Rafael Sorkin. A spin-statistics theorem for certain topological geons.
Classical and Quantum Gravity, 15:1153–1167, 1998.
18. Fay Dowker and Rafael Sorkin. Spin and statistics in quantum gravity. In R.C. Hilborn
and G.M. Tino, editors, Spin-Statistics Connections and Commutation Relations: Experi-
mental Tests and Theoretical Implications, pages 205–218. American Institute of Physics,
New York, 2000.
19. Patrick Du Val. Homographies, Quaternions, and Rotations. Clarendon Press, Oxford,
1964.
20. David G. Ebin. On the space of Riemannian metrics. Bulletin of the American Mathe-
matical Society, 74(5):1001–1003, 1968.
21. David Finkelstein and Julio Rubinstein. Connection between spin, statistics, and kinks.
Journal of Mathematical Physics, 9(11):1762–1779, 1968.
 27

22. Arthur E. Fischer. The theory of superspace. In M. Carmeli, S.I. Fickler, and L. Witten,
editors, Relativity, proceedings of the Relativity Conference in the Midwest, held June 2-6,
1969, at Cincinnati Ohio, pages 303–357. Plenum Press, New York, 1970.
23. Arthur E. Fischer. Resolving the singularities in the space of Riemannian geometries.
Journal of Mathematical Physics, 27:718–738, 1986.
24. Arthur E. Fischer and Vincent E. Moncrief. The structure of quantum conformal su-
perspace. In S. Cotsakis and G.W. Gibbons, editors, Global Structure and Evolution
in General Relativity, volume 460 of Lecture Notes in Physics, pages 111–173. Springer
Verlag, Berlin, 1996.
25. David Izrailevich Fouxe-Rabinovitch. Über die Automorphismengruppen der freien Pro-
dukte I. Matematicheskii Sbornik, 8(50):265–276, 1940.
26. David Izrailevich Fouxe-Rabinovitch. Über die Automorphismengruppen der freien Pro-
dukte II. Matematicheskii Sbornik, 9(51):297–318, 1941.
27. Daniel S. Freed and David Groisser. The basic geometry of the manifold of Riemannian
metrics and of its quotient by the diffeomorphism group. Michigan Mathematical Journal,
36(3):323–344, 1989.
28. John Friedman and Rafael Sorkin. Spin 1/2 from gravity. Physical Review Letters, 44:1100–
1103, 1980.
29. Wilhelm Fushchich and Vladimir Shtelen. Are Maxwell’s equations invariant under the
Galilei transformations? Doklady Akademii Nauk Ukrainy A, 3:22–26, 1991. In Russian.
30. L. Zhiyong Gao and Shing-Tung Yau. The existence of negatively Ricci curved metrics on
three-manifolds. Inventiones Mathematicae, 85(3):637–652, 1986.
31. Nick D. Gilbert. Presentations of the automorphims group of a free product. Proceedings
of the London Mathematical Society, 54:115–140, 1987.
32. Domenico Giulini. 3-manifolds for relativists. International Journal of Theoretical Physics,
33:913–930, 1994.
33. Domenico Giulini. On the configuration-space topology in general relativity. Helvetica
Physica Acta, 68:86–111, 1995.
34. Domenico Giulini. Quantum mechanics on spaces with finite fundamental group. Helvetica
Physica Acta, 68:439–469, 1995.
35. Domenico Giulini. What is the geometry of superspace? Physical Review D, 51(10):5630–
5635, 1995.
36. Domenico Giulini. The group of large diffeomorphisms in general relativity. Banach Center
Publications, 39:303–315, 1997.
37. Domenico Giulini. Mapping-class groups of 3-manifolds in canonical quantum gravity.
In Bertfried Fauser, Jürgen Tolksdorf, and Eberhard Zeidler, editors, Quantum Gravity:
Mathematical Models and Experimental Bounds. Birkhäuser Verlag, Basel, 2007. Online
available at harxiv.org/pdf/gr-qc/0606066i.
38. Domenico Giulini and Claus Kiefer. Wheeler-DeWitt metric and the attractivity of gravity.
Physics Letters A, 193(1):21–24, 1994.
39. Allen E. Hatcher. Notes on basic 3-manifold topology. Online available at
www.math.cornell.edu/∼hatcher/3M/3Mdownloads.html.
40. Sergio A. Hojman, Karel Kuchař, and Claudio Teitelboim. New approach to general
relativity. Nature Physical Science, 245:97–98, October 1973.
41. Sergio A. Hojman, Karel Kuchař, and Claudio Teitelboim. Geometrodynamics regained.
Annals of Physics, 96:88–135, 1976.
42. Giorgio Immirzi. Real and complex connections for canonical gravity. Classical and Quan-
tum Gravity, 14(10):L117–L181, 1997.
43. Christopher J. Isham. T heta–states induced by the diffeomorphism group in canonically
quantized gravity. In J.J. Duff and C.J. Isham, editors, Quantum Structure of Space and
Time, Proceedings of the Nuffield Workshop, August 3-21 1981, Imperial College London,
pages 37–52. Cambridge University Press, London, 1982.
44. Christopher J. Isham and Karel V. Kuchař. Representations of spacetime diffeomorphisms.
I. Canonical parametrized field theories. Annals of Physics, 164:288–315, 1985.
45. Christopher J. Isham and Karel V. Kuchař. Representations of spacetime diffeomorphisms.
II. Canonical geometrodynamics. Annals of Physics, 164:316–333, 1985.
46. Roman Jackiw. Topological investigations of quantized gauge theories. In Bryce DeWitt
and Raymond Stora, editors, Relativity, Groups, and Topology II, Les Houches, Session
XL, pages 37–52. North Holland, Amsterdam, 1984.
47. Jerry L. Kazdan and Frank W. Warner. Scalar curvature and conformal deformation of
Riemannian structure. Journal of Differential Geometry, 10(1):113–134, 1975.
28

48. Claus Kiefer. Quantum Gravity, volume 124 of International Series of Monographs on
Physics. Clarendon Press, Oxford, second edition, 2007.
49. John Klauder and John A. Wheeler. On the question of a neutrino analog to electric
charge. Reviews of Modern Physics, 29(3):516–517, 1957.
50. Hellmuth Kneser. Geschlossene Flächen in dreidimensionalen Mannigfaltigkeiten. Jahres-
berichte der deutschen Mathematiker Vereinigung, 38:248–260, 1929.
51. Karel Kuchař. Geometrodynamics regained: A Lagrangian approach. Journal of Mathe-
matical Physics, 15(6):708–715, 1974.
52. Karel Kuchař. Canonical quantum gravity. In R.J. Gleiser, C.N. Kosameh, and O.M.
Moreschi, editors, General Relativity and Gravitation, pages 119–150. IOP Publishing,
Bristol, 1993.
53. Jacqueline Lelong-Ferrand. Transformation conformes et quasiconformes des variétés rie-
manniennes compacts (démonstration de la conjecture de A. Lichnerowicz). Mémoires de
la Classe des Sciences de l’Académie royale des Sciences, des Lettres et des Beaux-Arts
de Belgique, 39(5):3–44, 1971.
54. David Lovelock. The four-dimensionality of space and the Einstein tensor. Journal of
Mathematical Physics, 13(6):874–876, 1972.
55. George Shultz McCarty. Homeotopy groups. Transactions of the American Mathematical
Society, 106:293–303, 1963.
56. Darryl McCullough. Topological and algebraic automorphisms of 3-manifolds. In Renzo
Piccinini, editor, Groups of Homotopy Equivalences and Related Topics, volume 1425 of
Springer Lecture Notes in Mathematics, pages 102–113. Springer Verlag, Berlin, 1990.
57. Darryl McCullough and Andy Miller. Homeomorphisms of 3-manifolds with compressible
boundary. Memoirs of the American Mathematical Society, 61(344), 1986.
58. John W. Milnor. A unique decomposition theorem for 3-manifolds. American Journal of
Mathematics, 84(1):1–7, 1962.
59. Charles W. Misner, Kip S. Thorne, and John Archibald Wheeler. Gravitation. W.H.
Freeman and Company, New York, 1973.
60. Edwin Evariste Moise. Affine structures in 3-manifolds V. The triangulation theorem and
Hauptvermutung. Annals of Mathematics, 56(1):96–114, 1952.
61. Daniel Müllner. Orientation Reversal of Manifolds. PhD thesis, Friedrich-Wilhelms-
Universität Bonn, October 2008.
62. Sumner B. Myers and Norman E. Steenrod. The group of isometries of a Riemannian
manifold. Annals of Mathematics, 40(2):400–416, 1939.
63. Hermann Nicolai. On M-theory. Journal of Astrophysics and Astronomy, 20(3-4):149–164,
1999.
64. Josep M. Pons. Generally covariant theories: The Noether obstruction for realizing certain
space-time diffeomorphisms in phase space. Classical and Quantum Gravity, 20(15):3279–
3294, 2003.
65. Kurt Reidemeister. Homotopieringe und Linsenräume. Abhandlungen aus dem Mathema-
tischen Seminar der Universität Hamburg, 11(1):102–109, 1935.
66. Joseph Samuel. Canonical gravity, diffeomorphisms and objective histories. Classical and
Quantum Gravity, 17(22):4645–4654, 2000.
67. Joseph Samuel. Is Barbero’s Hamiltonian formulation a gauge theory of Lorentzian grav-
ity? Classical and Quantum Gravity, 17(20):L141–L148, 2000.
68. Herbert Seifert and William Threlfall. A Textbook of Topology. Academic Press, Orlando,
Florida, 1980. Translation of the 1934 german edition.
69. Rafael Sorkin. Introduction to topological geons. In P.G. Bergmann and V. De Sabbata,
editors, Topological Properties and Global Structure of Space-Time, volume B138 of NATO
Advanced Study Institutes Series, page 249. D. Reidel Publishing Company, Dordrecht-
Holland, 1986.
70. Rafael Sorkin. A general relation between kink-exchange and kink-rotation. Communica-
tions in Mathematical Physics, 115:421–434, 1988.
71. Rafael Sorkin. Classical topology and quantum phases: Quantum geons. In S. De Filippo,
M. Marinaro, G. Marmo, and G. Vilasi, editors, Geometrical and Algebraic Aspects of
Nonlinear Field Theory, pages 201–218. Elsevier Science Publishers B.V., Amsterdam,
1989.
72. Rafael Sorkin and Sumati Surya. An analysis of the representations of the mapping
class group of a multi-geon three-manifold. International Journal of Modern Physics A,
13(21):3749–3790, 1998.
 29

73. Michael D. Stern. Investigations of the Topology of Superspace. PhD thesis, Department
of Physics, Princeton University, April 28th 1967.
74. Claudio Teitelboim. How commutators of constraints reflect the spacetime structure.
Annals of Physics, 79(2):542–557, 1973.
75. Thomas Thiemann. Modern Canonical Quantum General Relativity. Cambridge Mono-
graphs on Mathematical Physics. Cambridge University Press, Cambridge, 2007.
76. Kip S. Thorne, David L. Lee, and Alan P. Lightman. Foundations for a theory of gravita-
tion theories. Physical Review D, 7(12):563–3578, 1973.
77. John A. Wheeler. Geons. Physical Review, 97(2):511–536, 1955.
78. John A. Wheeler. Geometrodynamics. Academic Press, New York, 1962.
79. John A. Wheeler. Einsteins Vision. Springer Verlag, Berlin, 1968.
80. John A. Wheeler. Superspace and the nature of quantum geometrodynamics. In Cecile M.
DeWitt and John A. Wheeler, editors, Battelle Rencontres, 1967 Lectures in Mathematics
and Physics, pages 242–307. W.A. Benjamin, New York, 1968.
81. John A. Wheeler. Particles and geometry. In P. Breitenlohner and H.P. Dürr, editors,
Unified Theories of Elementary Particles, volume 160 of Lecture Notes in Physics, pages
189–217. Springer Verlag, Berlin, 1982.
82. John Henry Constantine Whitehead. On incidence matrices, nuclei and homotopy types.
Annals of Mathematics, 42(5):1197–1239, 1941.
83. Donald Witt. Symmetry groups of state vectors in canonical quantum gravity. Journal of
Mathematical Physics, 27(2):573–592, 1986.
84. Donald Witt. Vacuum space-times that admit no maximal slices. Physical Review Letters,
57(12):1386–1389, 1986.