Topological Lensing in Spherical Spaces

Abstract

This article gives the construction and complete classification of all three-dimensional spherical manifolds, and orders them by decreasing volume, in the context of multiconnected universe models with positive spatial curvature. It discusses which spherical topologies are likely to be detectable by crystallographic methods using three-dimensional catalogs of cosmic objects. The expected form of the pair separation histogram is predicted (including the location and height of the spikes) and is compared to computer simulations, showing that this method is stable with respect to observational uncertainties and is well suited for detecting spherical topologies.

Full text

arXiv:gr-qc/0106033v1 11 Jun 2001
Topological Lensing in Spherical Spaces
Evelise Gausmann1, Roland Lehoucq2, Jean-Pierre
Luminet1, Jean-Philippe Uzan3and Jeffrey Weeks4
(1) D´epartement d’Astrophysique Relativiste et de Cosmologie,
Observatoire de Paris – C.N.R.S. UMR 8629, F-92195 Meudon Cedex (France).
(2) CE-Saclay, DSM/DAPNIA/Service d’Astrophysique,
F-91191 Gif sur Yvette Cedex (France)
(3) Laboratoire de Physique Th´eorique – C.N.R.S. UMR 8627, Bˆat. 210,
Universit´e Paris XI, F-91405 Orsay Cedex (France).
(4) 15 Farmer St., Canton NY 13617-1120, USA.
Abstract. This article gives the construction and complete classification of all
three–dimensional spherical manifolds, and orders them by decreasing volume, in
the context of multiconnected universe models with positive spatial curvature. It
discusses which spherical topologies are likely to be detectable by crystallographic
methods using three–dimensional catalogs of cosmic objects. The expected form
of the pair separation histogram is predicted (including the location and height of
the spikes) and is compared to computer simulations, showing that this method is
stable with respect to observational uncertainties and is well suited for detecting
spherical topologies.
PACS numbers: 98.80.-q, 04.20.-q, 02.40.Pc
1. Introduction
The search for the topology of our universe has focused mainly on candidate spacetimes
with locally Euclidean or hyperbolic spatial sections. This was motivated on the one
hand by the mathematical simplicity of three–dimensional Euclidean manifolds, and on
the other hand by observational data which had, until recently, favored a low density
universe. More recently, however, a combination of astrophysical and cosmological
observations (among which the luminosity distance–redshift relation up to z∼1
from type Ia supernovae [1], the cosmic microwave background (CMB) temperature
anisotropies [2], gravitational lensing [3], velocity fields [4], and comoving standard
rulers [5]) seems to indicate that the expansion of the universe is accelerating, with
about 70% of the total energy density Ω0being in the form of a dark component
ΩΛ0with negative pressure, usually identified with a cosmological constant term or
a quintessence field [6]. As a consequence, the spatial sections of the universe would
be “almost” locally flat, i.e. their curvature radius would be larger than the horizon
radius (∼10h−1Gpc, where his the Lemaˆıtre-Hubble parameter in units of 100
km/s/Mpc). Recent CMB measurements [2] report a first Doppler peak shifted by a
few percent towards larger angular scales with respect to the peak predicted by the
standard cold dark matter (CDM) inflationary model, thus favouring a marginally
spherical model [7]. Indeed, under specific assumptions such as a Λ–CDM model and

Topological Lensing in Spherical Spaces 2
a nearly scale invariant primordial power spectrum, the value of the total energy–
density parameter is given by Ω0≡Ωm0+ ΩΛ0= 1.11+0.13
−0.12 to 95% confidence [8].
Note that while flat models still lie well inside the 95% confidence level, the relation
between the angular diameter distance and the acoustic peak positions in the angular
power spectrum makes the peak positions in models with a low matter content very
dependent on small variations of the cosmological constant.
As a consequence of these observable facts, spherical spaceforms are of increasing
interest for relativistic cosmology, in the framework of Friedmann–Lemaˆıtre solutions
with positive spatial curvature. Due to the current constraints on the spatial curvature
of our universe and to the rigidity theorem [9], hyperbolic topologies may be too
large to be detectable by crystallographic methods. For example, if Ωm0= 0.3
and ΩΛ0= 0.6, then even in the smallest known hyperbolic topologies the distance
from a source to its nearest topological image is more than a half of the horizon
radius, meaning that at least one member of each pair of topological images would
lie at a redshift too high to be easily detectable. When using statistical methods for
detecting a hyperbolic topology, the topological signature falls to noise level as soon
as observational uncertainties are taken into account.
On the other hand, spherical topologies may be easily detectable, because for
a given value of the curvature radius, spherical spaces can be as small as desired.
There is no lower bound on their volumes, because in spherical geometry increasing a
manifold’s complexity decreases its volume, in contrast to hyperbolic geometry where
increasing the complexity increases the volume. Thus many different spherical spaces
would fit easily within the horizon radius, no matter how small Ω0−1 is.
In order to clarify some misleading terminology that is currently used in the
cosmological literature, we emphasize the distinction between spherical and closed
universe models. A spherical universe has spatial sections with positive curvature.
The volume of each spatial section is necessarily finite, but the spacetime can be open
(i.e. infinite in time) if the cosmological constant is high enough. On the other hand,
a closed universe is a model in which the scale factor reaches a finite maximum value
before recollapsing. It can be obtained only if the space sections are spherical and the
cosmological constant is sufficiently low.
As emphasized by many authors (see e.g. [10, 11, 12, 13] for reviews), the key idea
to detect the topology in a three–dimensional data set is the topological lens effect, i.e.
the fact that if the spatial sections of the universe have at least one characteristic size
smaller than the spatial scale of the catalog, then different images of the same ob ject
shoud appear in the survey. This idea was first implemented in the crystallographic
method [14], which uses a pair separation histogram (PSH) depicting the number of
pairs of the catalog’s objects having the same three–dimensional spatial separations in
the universal covering space. Even if numerical simulations of this method showed the
appearance of spikes related to characteristic distances of the fundamental polyhedron,
we proved that sharp spikes emerge only when the holonomy group has at least one
Clifford translation, i.e. a holonomy that translates all points the same distance [15]
(see also [16]). Since then, various generalisations of the PSH method have been
proposed [17, 18, 19] (see [13, 20] for a discussions of these methods) but none of them
is fully satisfactory. The first step towards such a generalisation was to exploit the
property that, even if there is no Clifford translation, equal distances in the universal
covering space appear more often than just by chance. We thus reformulated the
cosmic crystallographic method as a collecting–correlated–pairs method (CCP) [21],
where the topological signal was enhanced by collecting all distance correlations in a

Topological Lensing in Spherical Spaces 3
single index. It was proven that this signal was relevant to detect the topology.
The goal of the present article is twofold. First we give a mathematical description
of all spherical spaces, which have been overlooked in the literature on cosmology. This
provides the required mathematical background for applying statistical methods to
detect the topology: crystallographic methods using three–dimensional data sets such
as galaxy, cluster and quasar catalogs, and methods using two–dimensional data sets
such as temperature fluctuations in the Cosmic Microwave Background (CMB), and
we investigate their observational signature in catalogs of discrete sources in order to
complete our previous works [14, 15, 20, 21] on Euclidean and hyperbolic manifolds.
We first review in §2 the basics of cosmology and topology in spherical universes,
including the basic relations to deal with a Friedmann–Lemaˆıtre universe of constant
curvature and the basics of cosmic topology. We then describe and classify three–
dimensional spherical manifolds in §3 and §4 and explain how to use them, with
full details gathered in Appendix A and Appendix B. One prediction of our former
works [15, 21] was that the PSH method must exhibit spikes if there exists at least one
Clifford translation. In §5 we discuss the spaceforms that are likely to be detectable.
We show that the location and height of the spikes can be predicted analytically and
then we check our predictions numerically. We also study the effect of observational
errors on the stability of the PSH spectra, and briefly discuss the status of the CCP
method for spherical topologies.
2. Cosmology in a Friedmann–Lemaˆıtre spacetime with spherical spatial
sections
2.1. Friedmann–Lemaˆıtre spacetimes of constant positive curvature
In this section we describe the cosmology and the basics of the topology of universes
with spherical spatial sections. The local geometry of such a universe is given by a
Friedmann–Lemaˆıtre metric
ds2=−dt2+a2(t)dχ2+ sin2χdω2.(1)
where ais the scale factor, tthe cosmic time and dω2≡dθ2+sin2θdϕ2the infinitesimal
solid angle. χis the (dimensionless) comoving radial distance in units of the curvature
radius RCof the 3–sphere S3.
The 3–sphere S3can be embedded in four–dimensional Euclidean space by
introducing the set of coordinates (xµ)µ=0..3related to the intrinsic coordinates
(χ, θ, ϕ) through (see e.g. [22])
x0= cos χ
x1= sin χsin θsin ϕ
x2= sin χsin θcos ϕ
x3= sin χcos θ, (2)
with 0 ≤χ≤π, 0 ≤θ≤πand 0 ≤ϕ≤2π. The 3–sphere is then the submanifold of
equation
xµxµ≡x2
0+x2
1+x2
2+x2
3= +1,(3)
where xµ=δµν xν. The comoving spatial distance dbetween any two points xand y
on S3can be computed using the inner product xµyµ. The value of this inner product
is the same in all orthonormal coordinate systems, so without loss of generality we

Topological Lensing in Spherical Spaces 4
may assume x= (1,0,0,0) and y= (cos d, sin d, 0,0), giving xµyµ= cos d. Hence,
the comoving spatial distance between two points of comoving coordinates xand yis
given by
d[x, y] = arc cos [xµyµ],(4)
The volume enclosed by a sphere of radius χis, in units of the curvature radius,
Vol(χ) = π(2χ−sin 2χ).(5)
A convenient way to visualize the 3–sphere is to consider S3as composed of two
solid balls in Euclidean space R3, glued together along their boundaries (figure 1):
each point of the boundary of one ball is the same as the corresponding point in the
other ball. To represent a point of coordinates xµin a three–dimensional space we
consider only the coordinates (xi)i=1..3, which are located in the interior of a ball, and
discard the nearly redundant coordinate x0. However, the two points of coordinates
(χ, θ, φ) and (π−χ, θ, φ), corresponding respectively to the points (x0, x1, x2, x3) and
(−x0, x1, x2, x3) in the four–dimensional Euclidean space, have the same coordinates
(x1, x2, x3) and we thus have to use two balls, one corresponding to 0 ≤χ≤π/2 (i.e.
x0≥0) and the other one to π/2≤χ≤π(i.e. x0≤0). Each ball represents half the
space.
φ φ
θ = 0
θ = π/2
χ=0 χ=π/6
χ=π/3
χ=π/2
χ=π χ=5π/6
χ=2π/3
χ=π/2
Figure 1. Representation of S3by two balls in R3glued together. Top: The θ
and φcoordinates are the standard ones. Bottom: The χcoordinate runs from 0
at the center of one ball (the “north pole” of S3) through π/2 at the ball’s surface
(the spherical “equator” of S3) to πat the center of the other ball (the “south
pole” of S3).
All three–dimensional observations provide at least the position of an object on
the celestial sphere and its redshift z≡a/a0−1 (the value a0of the scale factor today
may be chosen arbitrarily; a natural choice is to set a0equal to the physical curvature
radius today). To reconstruct an object’s three–dimensional position (χ, θ, ϕ) we need

Topological Lensing in Spherical Spaces 5
to compute the relation between the radial coordinate χand the redshift z, which
requires the law of evolution of the scale factor obtained from the Friedmann equations
H2=κ
3ρ−k
a2R2
C
+Λ
3(6)
¨a
a=−κ
6(ρ+ 3P) + Λ
6(7)
where ρand Pare the energy density and pressure of the cosmic fluid, Λ the
cosmological constant, κ≡8πG with Gthe Newton constant, k= +1 is the curvature
index and H≡˙a/a is the Hubble parameter. As a first consequence, we deduce from
these equations that the physical curvature radius today is given by
Rphys
C0≡a0RC0=c
H0
1
p|ΩΛ0+ Ωm0−1|(8)
where the density parameters are defined by
ΩΛ≡Λ
3H2Ωm≡κρ
3H2.(9)
As emphasized above, we can choose a0to be the physical curvature radius today, i.e.
a0=Rphys
C0, which amounts to choosing the units on the comoving sphere such that
RC0= 1, hence determining the value of the constant a0. As long as we are dealing
with catalogs of galaxies or clusters, we can assume that the universe is filled with a
pressureless fluid. Integrating the radial null geodesic equation dχ= dt/a leads to
χ(z) = Zz
0pΩm0+ ΩΛ0−1dx
pΩΛ0+ (1 −Ωm0−ΩΛ0)(1 + x)2+ Ωm0(1 + x)3.(10)
2.2. Basics of cosmic topology
Equations (1)–(10) give the main properties that describe the local geometry, i.e. the
geometry of the universal covering space Σ, independently of the topology. Indeed it
is usually assumed that space is simply connected so that the spatial sections are the
3–sphere S3. To describe the topology of these spatial sections we have to introduce
some basic topological elements (see [10] for a review). From a topological point of
view, it is convenient to describe a three–dimensional multi–connected manifold Mby
its fundamental polyhedron, which is convex with an even number of faces. The faces
are identified by face–pairing isometries. The face–pairing isometries generate the
holonomy group Γ, which acts without fixed points on the three–dimensional covering
space Σ (see [22, 23, 24] for mathematical definitions and [10, 13] for an introduction
to topology in the cosmological context). The holonomy group Γ is isomorphic to the
first fundamental group π1(M).
To illustrate briefly these definitions, let us consider the particularly simple case
of a two–dimensional flat torus T2. It can be constructed from a square, opposite
edges of which are glued together. The translations taking one edge to the other are
the face pairing isometries. The holonomy group of this space, generated by the face–
pairing translations, is isomorphic to the group of loops on the torus π1(T2) = Z×Z
(see figure 2).
In our case Σ = S3and its isometry group is the rotation group of the four–
dimensional space in which it sits, i.e. G=SO(4). In units of the curvature radius,
the volume of the spatial sections is given by
Vol(S3/Γ) = Vol(S3)
|Γ|=2π2
|Γ|,(11)

Topological Lensing in Spherical Spaces 6
where |Γ|is the order of the group Γ, i.e. the number of elements contained in Γ.
Since the elements gof Γ are isometries, they satisfy
∀x, y ∈Σ∀g∈Γ,dist[x, y] = dist[g(x), g(y)] (12)
An element gof Γ is a Clifford translation if it translates all points the same distance,
i.e. if
∀x, y ∈Σ dist[x, g(x)] = dist[y, g(y)],(13)
The significance of these particular holonomies, which are central to the detection of
the topology, will be explained in section 3.1. To perform computations we express
the isometries of the holonomy group Γ ⊂SO(4) as 4 ×4 matrices. To enumerate all
spherical manifolds we will need a classification of all finite, fixed-point free subgroups
of SO(4). This will be the purpose of sections 3 and 4.
Figure 2. Illustration of the general topological definitions in the case of a two–
dimensional torus (left). Its fundamental polyhedron is a square, opposite faces
of which are identified (middle). In its universal covering space (right) the two
translations taking the square to its nearest images generate the holonomy group.
2.3. Spherical spaces and cosmology
As early as 1917, de Sitter [25] distinguished the sphere S3from the projective space
P3(which he called elliptical space) in a cosmological context. Both spaceforms are
finite, with (comoving) volumes 2π2and π2, respectively. The pro jective space P3is
constructed from the sphere S3by identifying all pairs of antipodal points. The main
difference between them is that in a sphere all straight lines starting from a given
point reconverge at the antipodal point, whereas in projective space two straight lines
can have at most one point in common. Thus the sphere does not satisfy Euclid’s
first axiom, while projective space does. In S3the maximal distance between any two
points is π, and from any given point there is only one point, the antipodal one, at
that maximal distance. In P3the maximal distance is π/2 and the set of points lying
at maximal distance from a given point forms a two–dimensional projective plane P2.
Because each pair of antipodal point points in the 3–sphere projects to a single point in
projective space, if we adopt the 3–sphere as our space model we may be representing
the physical world in duplicate. For that reason, de Sitter claimed that P3was really
the simplest case, and that it was preferable to adopt this in a cosmological context.

Topological Lensing in Spherical Spaces 7
Einstein did not share this opinion, and argued that the simple connectivity of S3
was physically preferable [26]. Eddington [27], Friedmann [28] and Lemaˆıtre [29] also
referred to projective space as a more physical alternative to S3.
Narlikar and Seshadri [30] examined the conditions under which ghost images of
celestial objects may be visible in a Friedmann–Lemaˆıtre model with the topology
of projective space. De Sitter had correctly remarked that the most remote points
probably lie beyond the horizon, so that the antipodal point, if any, would remain
unobservable. Indeed, given the present data, it will be impossible to distinguish S3
from P3observationally if Ω0−1≪1, because our horizon radius is too small.
None of these authors mentioned other multi–connected spaces in a cosmological
context. This was first done by Ellis [31] and Gott [32]. More recently, lens spaces
have been investigated in the framework of quantum gravity models [33] and in terms
of their detectability [34]. Nevertheless, the literature on multi–connected spherical
cosmologies is underdeveloped; here we aim to fill the gap.
3. The Mathematics of Spherical Spaces I: Classification of S3subgroups
This section describes the classification of all spherical 3–manifolds and develops an
intuitive understanding of their topology and geometry. Of particular observational
relevance, we will see which spherical 3–manifolds have Clifford translations in their
holonomy groups and which do not. Threlfall and Seifert [35] gave the first complete
classification of these spherical 3–manifolds. Our approach borrows heavily from
theirs, while also making use of quaternions as in [36].
As emphasized in the previous section, technically what we will need in order
to perform any computation is the form of the holonomy transformations gas 4 ×4
matrices in SO(4), and an enumeration of all finite subgroups Γ ⊂S O(4). We will
enumerate the finite subgroups of SO(4) (see sections 3.1 and 4) in terms of the simpler
enumeration of finite subgroups of S O(3) (see section 3.2). The connection between
SO(4) and SO(3) will use quaternions (see Appendix A and Appendix B).
3.1. Generalities
Our starting point is the fact that for each isometry g∈O(n+ 1) there is a basis of
Rn+1 relative to which the matrix of ghas the form


















+1 0 ···
0...
+1
.
.
.−1
...
−1
cos α−sin α
sin αcos α
...


















(14)
If gis a holonomy transformation of an n–manifold, then ghas no fixed points and its
matrix has no +1 terms on the diagonal. Furthermore, each pair of −1 terms may be

Topological Lensing in Spherical Spaces 8
rewritten as a sine–cosine block with α=π. Thus when n= 3 the matrix takes the
form
M(θ, φ) = 



cos θ−sin θ0 0
sin θcos θ0 0
0 0 cos φ−sin φ
0 0 sin φcos φ




(15)
An immediate consequence of this decomposition is that every spherical 3–manifold
is orientable. Indeed all odd-dimensional spherical manifolds must be orientable for
this same reason. In even dimensions the only spherical manifolds are the n–sphere
Sn(which is orientable) and the n–dimensional pro jective space Pn(which is non
orientable).
If we replace the static matrix M(θ, φ) with the time–dependent matrix M(θt, φt),
we generate a flow on S3, i.e. to each point x∈S3we associate the flow line
x(t) = M(θt, φt)x. This flow is most beautiful in the special case θ=±φ. In this
special case all flow lines are geodesics (great circles), and the flow is homogeneous in
the sense that there is an isometry of S3taking any flow line to any other flow line.
The matrix M(θ, ±θ) defines a Clifford translation because it translates all points
the same distance (see equation 13). We further distinguish two families of Clifford
translations. When θ=φthe flow lines spiral clockwise around one another, and the
Clifford translation is considered right–handed whereas when θ=−φthe flow lines
spiral anticlockwise around one another, and the Clifford translation is considered
left–handed.
Every isometry M(θ, φ)∈SO(4) is the product of a right–handed Clifford
translation M(α, α) and a left–handed Clifford translation M(β, −β) as
M(θ, φ) = M(α, α)M(β, −β) = M(β , −β)M(α, α) (16)
where α≡(θ+φ)/2 and β≡(θ−φ)/2 and the order of the factors makes no
difference. This factorization is unique up to simultaneously multiplying both factors
by -1. Moreover every right–handed Clifford translation commutes with every left–
handed one, because there is always a coordinate system that simultaneously brings
both into their canonical form (15).
Just as the unit circle S1enjoys a group structure as the set S1of complex
numbers of unit length, the 3–sphere S3enjoys a group structure as the set S3of
quaternions of unit length (see Appendix A for details). Each right–handed Clifford
translation corresponds to left multiplication by a unit length quaternion (q→xq),
so the group of all right–handed Clifford translations is isomorphic to the group S3of
unit length quaternions, and similarly for the left–handed Clifford translations, which
correspond to right multiplication (q→qx). It follows that SO(4) is isomorphic
to S3× S3/{±(1,1)}, where 1is the identity quaternion, so the classification of all
subgroups of SO(4) can be deduced from the classification of all subgroups of S3.
Classifying all finite subgroups of S3seems difficult at first, but luckily it reduces
to a simpler problem. The key is to consider the action of the quaternions by
conjugation. That is, for each unit length quaternion x∈ S3, consider the isometry
pxthat sends each quaternion qto xqx−1
px:S3→ S3
q7−→ px(q) = xqx−1.(17)
The isometry pxfixes the identity quaternion 1, so in effect its action is confined to the
equatorial 2–sphere spanned by the remaining basis quaternions (i,j,k) [see Appendix

Topological Lensing in Spherical Spaces 9
A for details and definitions concerning quaternions]. Thus, by restricting our
attention to the equatorial 2–sphere, we get an isometry
px:S2→S2(18)
In other words, each x∈ S3defines an element px∈SO(3), and the mapping
p:S3→SO(3)
x7−→ p(x) = px
(19)
is a homomorphism from S3to SO(3). To classify all subgroups of S3we must first
know the finite subgroups of SO(3).
3.2. Finite Subgroups of SO(3)
The finite subgroups of SO(3) are just the finite rotation groups of a 2–sphere, which
are known to be precisely the following:
•The cyclic groups Znof order n, generated by a rotation through an angle 2π/n
about some axis.
•The dihedral groups Dmof order 2m, generated by a rotation through an angle
2π/m about some axis as well as a half turn about some perpendicular axis.
•The tetrahedral group Tof order 12 consisting of all orientation–preserving
symmetries of a regular tetrahedron.
•The octahedral group Oof order 24 consisting of all orientation–preserving
symmetries of a regular octahedron.
•The icosahedral group Iof order 60 consisting of all orientation–preserving
symmetries of a regular icosahedron.
If the homomorphism p:S3→SO(3) were an isomorphism, the above list
would give the finite subgroups of S3directly. We are not quite that lucky, but
almost: the homomorphism pis two–to–one. It is easy to see that px=p−xbecause
xqx−1= (−x)q(−x)−1for all q. There are no other redundancies, so the kernel of p
is
Ker(p) = {±1}.(20)
Let Γ be a finite subgroup of S3and consider separately the cases that Γ does or
does not contain −1.
•Case 1: If Γ does not contain −1, then pmaps Γ one–to–one onto its image in
SO(3), and Γ is isomorphic to one of the groups in the above list. Moreover,
because we have excluded −1, and S3contains no other elements of order 2, we
know that Γ contains no elements of order 2. The only groups on the above list
without elements of order 2 are the cyclic groups Znof odd order. Thus only
cyclic groups of odd order may map isomorphically from S3into SO(3), and it is
easy to check that they all do, for example by choosing the generator of Znto be
the quaternion cos(2π/n)1+ sin(2π/n)i.
•Case 2: If Γ contains −1, then pmaps Γ two–to–one onto its image in S O(3),
and Γ is a two–fold cover of one of the groups in the above list. Conversely, every
group on the list lifts to a group Γ ⊂ S3. The construction of the two–fold cover is
trivially easy: just take the preimage of the group under the action of p. The result
is called the binary cyclic, binary dihedral, binary tetrahedral, binary octahedral,

Topological Lensing in Spherical Spaces 10
or binary icosahedral group. A “binary cyclic group” is just a cyclic group of twice
the order, so all even order cyclic groups occur in this fashion. The remaining
binary groups are not merely the product of the original polyhedral group with
aZ2factor, nor are they isomorphic to the so–called extended groups which
include the orientation–reversing as well as the orientation-preserving symmetries
of the given polyhedron, but are something completely new‡. Note that the
plain dihedral, tetrahedral, octahedral, and icosahedral groups do not occur as
subgroups as of S3– only their binary covers do.
3.3. Finite Subgroups of S3
Combining the results of the two previous cases, we get the complete classification of
finite subgroups of the group S3of unit length quaternions as follows:
•The cyclic groups Znof order n.
•The binary dihedral groups D∗
mof order 4m,m≥2.
•The binary tetrahedral group T∗of order 24.
•The binary octahedral group O∗of order 48.
•The binary icosahedral group I∗of order 120.
4. The Mathematics of Spherical Spaces II: Classification of spherical
spaceforms
There are three categories of spherical 3–manifolds. The single action manifolds are
those for which a subgroup Rof S3acts as pure right–handed Clifford translations.
The double action manifolds are those for which subgroups Rand Lof S3act
simultaneously as right– and left–handed Clifford translations, and every element of R
occurs with every element of L. The linked action manifolds are similar to the double
action manifolds, except that each element of Roccurs with only some of the elements
of L.
After introducing some definitions, we give the classifications of single action
manifolds (§4.1), double action manifolds (§4.2) and linked action manifolds (§4.3)
and present in §4.4 a summary of these classifications.
We define a lens space L(p, q) by identifying the lower surface of a lens-shaped
solid to the upper surface with a 2πq/p rotation (see figure 3), for relatively prime
integers pand qwith 0 < q < p. Furthermore, we may restrict our attention to
0< q ≤p/2 because for values of qin the range p/2< q < p the twist 2πq/p is the
same as −2π(p−q)/p, thus L(p, q) is the mirror image of L(p, p −q). When the lens is
drawn in Euclidean space its faces are convex, but when it is realized in the 3–sphere
its faces lie on great 2–spheres, filling a hemisphere of each. Exactly pcopies of the
lens tile the universal cover S3, just as the 2–dimensional surface of an orange may
be tiled with psections of orange peel, each of which is a bigon with straight sides
meeting at the poles. Two lens spaces L(p, q ) and L(p′, q′) are homeomorphic if and
only if p=p′and either q=±q′(mod p) or qq′=±1(mod p).
A cyclic group Znmay have several different realizations as holonomy groups
Γ⊂SO(4). For example, the lens spaces L(5,1) and L(5,2) are nonhomeomorphic
manifolds, even though their holonomy groups are both isomorphic to Z5. For
‡The binary dihedral group D∗
1is isomorphic to the plain cyclic group Z4.

Topological Lensing in Spherical Spaces 11
noncyclic groups, the realization as a holonomy group Γ ⊂S O(4) is unique up to
an orthonormal change of basis, and thus the resulting manifold is unique.
2 q/pp
Figure 3. Construction of a lens space L(p, q ).
4.1. Single Action Spherical 3-Manifolds
The finite subgroups of S3give the single action manifolds directly, which are thus
the simplest class of spherical 3–manifolds. They are all given as follows:
•Each cyclic group Zngives a lens space L(n, 1), whose fundamental domain is a
lens shaped solid, nof which tile the 3–sphere.
•Each binary dihedral group D∗
mgives a prism manifold, whose fundamental
domain is a 2m–sided prism, 4mof which tile the 3–sphere.
•The binary tetrahedral group T∗gives the octahedral space, whose fundamental
domain is a regular octahedron, 24 of which tile the 3–sphere in the pattern of a
regular 24–cell.
•The binary octahedral group O∗gives the truncated cube space, whose funda-
mental domain is a truncated cube, 48 of which tile the 3–sphere.
•The binary icosahedral group I∗gives the Poincar´e dodecahedral space, whose
fundamental domain is a regular dodecahedron, 120 of which tile the 3–sphere
in the pattern of a regular 120–cell. Poincar´e discovered this manifold in a
purely topological context, as the first example of a multiply connected homology
sphere [37]. A quarter century later Weber and Seifert glued opposite faces of
a dodecahedron and showed that the resulting manifold was homeomorphic to
Poincar´e’s homology sphere [38].
In figure 4, we depict the fundamental domains for the binary tetrahedral group
T∗, binary octahedral group O∗and binary icosahedral group I∗. The fundamental
polyhedron for the lens space L(n, 1) can be constructed by following the example
presented in figure 3. Finally, the fundamental domain of the prism manifold generated
by the binary dihedral group D∗
5is shown in figure 5.
To finish, let us emphasize that all single action manifolds are globally homoge-
neous, in the sense that there is an isometry hof the manifold taking any point xto
any other point y. If the manifold’s holonomy group is realized as left multiplication by
a group Γ = {gi}of quaternions, then the isometry his realized as right multiplication
by x−1y. To check that his well–defined on the quotient manifold S3/Γ, note that h
takes any point gixequivalent to xto a point gix(x−1y) = giyequivalent to y, thus
respecting equivalence classes of points.
4.2. Double action spherical 3-manifolds
The double action spherical 3–manifolds are obtained by letting one finite subgroup
R⊂ S3act as right–handed Clifford translations (equivalent to left multiplication

Topological Lensing in Spherical Spaces 12
Figure 4. Fundamental domains for three single action 3–manifolds. From left to
right, the regular octahedron, the truncated cube and the regular dodecahedron
which respectively correspond to the spaces generated by the binary tetrahedral
group T∗, the binary octahedral group O∗and the binary icosahedral group I∗.
Figure 5. An example of fundamental domain for a prism manifold. This 10–
sided prism is the fundamental polyhedron of the space generated by the binary
dihedral group D∗
5, of order 20.
of quaternions) while a different finite subgroup L⊂ S3simultaneously acts as left–
handed Clifford translations (equivalent to right multiplication of quaternions). A
priori any two subgroups of S3could be used. However, if an element M(θ, θ) of R
and an element M(φ, −φ) of Lshare the same translation distance |θ|=|φ|, then
their composition, which equals M(2θ, 0) or M(0,2θ), will have fixed points unless
θ=φ=π. Allowing fixed points would take us into the realm of orbifolds, which
most cosmologists consider unphysical to describe our universe, so we do not consider
them here. In practice this means that the groups Rand Lcannot contain elements of
the same order, with the possible exception of ±1. The binary dihedral, tetrahedral,
octahedral, and icosahedral groups all contain elements of order 4, and so cannot be
paired with one another.
Thus either Ror Lmust be cyclic. Without loss of generality we may assume
that Lis cyclic. If Ris also cyclic, then Land Rcannot both contain elements of
order 4, and so we may assume that it is Lthat has no order 4 elements. If Ris
not cyclic, then Ris a binary polyhedral group, and again Lcan have no elements of
order 4. Thus either L=Znor L=Z2n, with nodd. Recalling that the group Zn
consists of the powers {qi}0≤i<n of a quaternion qof order n, it’s convenient to think
of the group Z2nas {qi}0≤i<n ∪{−qi}0≤i<n. If Rcontains −1, then nothing is gained
by including the {−qi}in L, because each possible element (r)(−l) already occurs
as (−r)(l). In other words, L=Znand L=Z2nproduce the same result, the only
difference being that with L=Z2neach element of the resulting group is generated
twice, once as (r)(l) and once as (−r)(−l). If Rdoes not contain −1, then it is cyclic
of odd order and we swap the roles of Rand L. Either way, we may assume Lis cyclic
of odd order. The double action spherical 3–manifolds are therefore the following:
•R=Zmand L=Zn, with mand nrelatively prime, always yields a lens space

Topological Lensing in Spherical Spaces 13
L(mn, q). However, not all lens spaces arise in this way.
•R=D∗
mand L=Zn, with gcd(4m, n) = 1, yields an n–fold quotient of a prism
manifold that is simultaneously a 4m–fold quotient of the lens space L(n, 1).
•R=T∗and L=Zn, with gcd(24, n) = 1, yields an n–fold quotient of the
octahedral space that is simultaneously a 24–fold quotient of the lens space
L(n, 1).
•R=O∗and L=Zn, with gcd(48, n) = 1, yields an n–fold quotient of the
truncated cube space that is simultaneously a 48–fold quotient of the lens space
L(n, 1).
•R=I∗and L=Zn, with gcd(120, n) = 1, yields an n–fold quotient of the
Poincar´e dodecahedral space that is simultaneously a 120–fold quotient of the
lens space L(n, 1).
In figure 6, we present the fundamental domain of the double action manifold
generated by the binary octahedral group R=O∗and the cyclic group L=Z5, of
orders 48 and 5 respectively.
Figure 6. A fundamental domain for the double action manifold of order 240
generated by the binary octahedral group R=O∗and the cyclic group L=Z5.
4.3. Linked Action Spherical 3–Manifolds
The third and final way to construct spherical 3–manifolds is to choose groups Rand
Las before, but allow each element r∈Rto pair with a restricted class of elements
l∈L, being careful to exclude combinations of rand lthat would create fixed points.
The following linked action manifolds arise in this way.
•Rand Lare the cyclic groups respectively generated by
r=Mp+q+ 1
2p2π, p+q+ 1
2p2π
and
l=Mp−q+ 1
2p2π, −p−q+ 1
2p2π
with 0 < q < p and gcd(p, q) = 1. The generator ris linked to the generator l, and
their powers are linked accordingly. This yields the lens space L(p, q ), generated
by rl =M(2π/p, 2πq/p). Note that when p+qis even, each element of L(p, q)
is produced twice, once as rklkand once as rp+klp+k=rprklplk= (−rk)(−lk).
•R=T∗and L=Z9nwith nodd. The plain (not binary) tetrahedral group T
contains a normal subgroup H≃D2consisting of the three half turns plus the
identity. Each element rin the binary tetrahedral group T∗is assigned an index
0, 1, or 2 according to the coset of Hin which its projection pr∈SO(3) lies.
Each element lin Z9nis assigned an index 0, 1, or 2 equal to its residue modulo 3.

Topological Lensing in Spherical Spaces 14
An element ris linked to an element lif and only if their indices are equal. This
yields a holonomy group with only a third as many elements as the full double
action group would have, and avoids elements with fixed points.
•R=D∗
mand L=Z8nwith gcd(m, 8n) = 1. Each element rin D∗
mis assigned
an index 0 or 1 according to whether its projection pr∈SO(3) lies in the cyclic
part of the plain Dmor not. Each element lin Z8nis assigned an index 0 or 1
equal to its residue modulo 2. An element ris linked to an element lif and only
if their indices are equal. This yields a holonomy group with only half as many
elements as the full double action group would have, and avoids elements with
fixed points. Note that because Rand Lboth contain −1, each element in the
group is produced twice, once as rl and once as (−r)(−l).
As an example, figure 7 shows the fundamental polyhedron of the linked action
manifold generated by the binary tetrahedral group R=T∗and the cyclic group
L=Z9, of orders 24 and 9 respectively.
Figure 7. Fundamental domain of the linked action manifold of order 72
generated by the binary tetrahedral group R=T∗and the cyclic group L=Z9.
4.4. Summary of classification
The preceding sections constructed all three–dimensional spherical manifolds and clas-
sified them in three families. To sum up, figure 8 organizes these manifolds by de-
creasing volume. Since none homeomorphic lens spaces can have isomorphic holonomy
group, figure 9 charts this special case of lens spaces.
On Figure 10, we show the number of distinct spherical 3–manifolds of a given
order. The vast majority of these manifolds are lens spaces. This does not necessarily
mean that a spherical universe is “more likely” to be a lens space. It does, however,
reflect the fact that there is a free parameter governing the amount of twist in the
construction of a lens space of order n. The remaining groups of order nhave no
free parameters in their construction. The difference arises because each lens space
is generated by a single element and is therefore subject to minimal consistency
constraints; the remaining manifolds, which each requires at least two generators,
are subject to stronger consistency conditions.
In the construction of a lens space of order n, there are roughly nchoices for
the amount of twist (the exact number of choices varies with n, because the twist
parameter must be relatively prime to n), so the number of distinct spherical manifolds
of order exactly ngrows linearly with n(see Figure 10, left plot). Thus the total
number of spherical 3–manifolds of order nor less is the sum of an arithmetic series,
and therefore grows quadratically with n(see Figure 10, right plot).

Topological Lensing in Spherical Spaces 15
Order Single action Double action Linked action
1Z1
2Z2
3Z3
4Z4
5Z5Z5
6Z6
7Z7Z7
8Z8D∗
2Z8
...
12 Z12 D∗
3Z3×Z4
...
72 Z72 D∗
18 Z8×Z9D∗
2×Z9Z72 T∗×Z9
D∗
9×Z8
...
120 Z120 D∗
30 Z40 ×Z3Z24 ×Z5Z8×Z15 Z120 D∗
3×Z40
I∗D∗
10 ×Z3D∗
6×Z5D∗
2×Z15 D∗
5×Z24 D∗
15 ×Z8
T∗×Z5
...
216 Z216 D∗
54 Z8×Z27 D∗
2×Z27 Z216 T∗×Z27
D∗
27 ×Z8
...
240 Z240 D∗
60 Z80 ×Z3Z48 ×Z5Z16 ×Z15 D∗
20 ×Z3Z240
D∗
3×Z80 D∗
12 ×Z5D∗
4×Z15 D∗
5×Z48 D∗
15 ×Z16
O∗×Z5
Figure 8. Classification of groups generating single, double and linked action
spherical 3–manifolds. The first column gives the order |Γ|of the holonomy group,
i.e. along each row the volume of space is 2π2/|Γ|. The simply connected 3–sphere
is S3/Z1, and the projective space P3is S3/Z2. Since some cyclic groups Znhave
different realizations as single action, double action, and linked action manifolds
they may appear more than once on a given line. For example, the lens spaces
L(5,1) and L(5,2) are nonhomeomorphic manifolds, even though their holonomy
groups are both isomorphic to Z5. The double action groups Zm×Zn≃Zmn
all yield lens spaces, as do the single action groups Zmn. For instance L(12,1)
is the single action manifold generated by Z12 while L(12,5) is the double action
manifold generated by Z3×Z4≃Z12. For noncyclic groups, the realization as a
holonomy group is unique, and thus the resulting manifold is unique, named after
its holonomy group.
5. Crystallographic Simulations
As we emphasized, the PSH method applies as long as the holonomy group has at
least one Clifford translation. Thus we use the PSH method for most of the spherical
manifolds but we will need the CCP method in certain exceptional cases.
Section 5.1 determines the radius χmax of the observable portion of the covering 3–
sphere in units of the curvature radius, as a function of the cosmological parameters Ωm
and ΩΛand a redshift cutoff zc. For plausible values of χmax, Section 5.2 determines
which topologies are likely to be detectable. Section 5.3 explains the expected form of
the Pair Separation Histogram first in the 3–sphere, and then predicts the location and

Topological Lensing in Spherical Spaces 16
Order Single action Double action Linked action
2L(2,1)
3L(3,1)
4L(4,1)
5L(5,1) L(5,2)
6L(6,1)
7L(7,1) L(7,2)
8L(8,1) L(8,3)
...
12 L(12,1) L(12,5)
...
72 L(72,1) L(72,17) L(72,5)
+ 5 more
...
120 L(120,1) L(120,31) L(120,7)
L(120,41) + 7 more
L(120,49)
...
216 L(216,1) L(216,55) L(216,5)
+ 17 more
...
240 L(240,1) L(240,31) L(240,7)
L(240,41) + 15 more
L(240,49)
Figure 9. Classification of lens spaces. In this chart each lens space occurs only
once in the first valid column (e.g. if the lens space is a single action manifold
we ignore the trivial expression of it as a double or linked action manifold). This
also takes into account the various equivalences so, for example L(7,2) appears
while L(7,3) does not because L(7,3) = L(7,2).
height of the spikes in a multiply connected spherical universe. Computer simulations
(§5.4) confirm these predictions. In §5.5, we briefly recall the applicability of the
CCP method.
5.1. Observational prospects
Observations indicate that |Ωm0+ ΩΛ0−1|is at most about 0.1, and this value fixes
the physical curvature scale Rphys
C0from (8). To detect topology one requires that
the size of the manifold be smaller than the diameter of the observable universe
in at least one direction. Equation (10) gives the distance χin radians from the
observer to a source at redshift z. Figure 11 uses equation (10) to plot the maximal
radial distance χmax that is accessible in a catalog of sources extending to redshift
zc, as a function of the cosmological parameters Ωm0and ΩΛ0. This maximal radial
distance χmax may also be used to compute the volume of the observable universe
vol(χmax) = π(2χmax −sin 2χmax )≃4
3πχ3
max in curvature radius units and compare
it to the total volume 2π2of the 3–sphere.
In practice one requires that shortest distance between topological images be less

Topological Lensing in Spherical Spaces 17
0
10
20
30
40
50
60
70
100 150 200 250
Number of spaces
Order of the group
,
0
1000
2000
3000
4000
5000
050 100 150 200 250
Cumulative number of spaces
Order of the group
Polynomial fit of order 2
N (< Γ) = -9.5066 + 1.4864 Γ + 0.0774 Γ2
Correlation = 0.99999
Figure 10. Left: The number of distinct spherical 3–manifolds of order exactly
|Γ|. The vast majority of these manifolds are lens spaces for which there are
roughly |Γ|choices for the amount of twist in their construction. So, the number
of distinct spherical manifolds of order exactly |Γ|grows linearly with |Γ|. Right:
The total number of spherical 3–manifolds of order |Γ|or less is the sum of an
arithmetic series, and therefore grows quadratically with |Γ|.
than the effective cutoff radius χmax. For example, a cyclic group Znwill satisfy this
criterion if and only if
2π
n< χmax.(21)
Figure 12 shows the implications of this equation. In principle one could detect
topology even if the shortest translations were almost as large as the diameter of
the observable universe, i.e. 2χmax , but when using statistical methods the signal
would become too weak.
5.2. Geometrical expectations
In this section we determine which topologies are likely to be detectable for plausible
values of χmax (see also [34] for detectability of lens spaces).
In a single action manifold every holonomy is a Clifford transformation, so in
principle the PSH detects the entire holonomy group. In practice we see only a small
portion of the covering 3–sphere. For example, with Ωm0= 0.35, ΩΛ0= 0.75, and
a redshift cutoff of z= 3, we see a ball of radius χmax ≃1/2 (see figure 11). With
this horizon size the binary tetrahedral, binary octahedral, and binary icosahedral
groups would be hard to detect because their shortest translations distances are 2π/6,
2π/8, and 2π/10, respectively. If, however, we extend the redshift cutoff to z= 1000,
then the horizon radius expands to χ≃1, putting the binary tetrahedral, binary
octahedral, and binary icosahedral groups within the range of CMB methods.
The cyclic groups Znand the binary dihedral groups D∗
mare much more amenable
to detection, because the shortest translation distances, 2π/n and 2π/2mrespectively,
can be arbitrarily small for sufficiently large nand m. For the example given above
with χmax ≃1/2, a cyclic group Znwill be detectable if n > 2π/(1/2) ≃12, and
similarly a binary dihedral group will be detectable if 2m > 12. On the other hand,

Topological Lensing in Spherical Spaces 18
Figure 11. The maximum radial distance χmax accessible in a catalog of depth
zc= 5 as a function of the cosmological parameters.
the lack of obvious periodicity on a scale of 1/20 the horizon radius§implies that
nand 2mmay not exceed 2π/(1/20) ≃120. In conclusion, the orders of the cyclic
groups Znand the binary dihedral groups D∗
mthat are likely to be detectable must
lie in the ranges
12 <
∼n<
∼120 and 12 <
∼2m<
∼120.
The holonomy group of a double action manifold (recall §4.2) consists of the products
rl, as rranges over a group Rof right–handed Clifford translations and lranges over
a group Lof left–handed Clifford translations. Because only pure Clifford translations
generate spikes, the PSH detects a product rl if and only if r=±1or l=±1. Thus
the set of spikes in the PSH for the double action manifold defined by Rand Lwill be
the union of the spikes for the two single action manifolds defined by Rand Lalone,
or possibly R× {±1}or L× {±1}. For example, if R=D∗
2and L=Z3, then we
get the union of the spikes for D∗
2with the spikes for Z3× {±1}=Z6(see figure 14).
The conditions for observability thus reduce to the conditions described for a single
action manifold. That is, a double action manifold is most easily detectable if one of
its factors (possibly extended by −1) is a cyclic group Znor a binary dihedral group
§Note that this upper bound is very approximate and that a detailed study will be required.

Topological Lensing in Spherical Spaces 19
0,1
1
10
40 80 120 160 200
Redshift of the first image
Order of the cyclic group
Figure 12. The redshift of the first topological image as a function of the order
of the cyclic group Znwhen Ω0= 0.35 and ΩΛ0= 0.75.
D∗
m, with
12 <
∼n<
∼120 and 12 <
∼2m<
∼120.
Happily the shortest translations are all Clifford translations.
In a linked action manifold (see §4.3) not every element of Roccurs with every
element of L. Let R′be the subgroup consisting of those elements r∈Rthat occur
with ±1of L, and let L′be the subgroup consisting of those elements l∈Lthat occur
with ±1of R. With this notation, the Clifford translations are the union of R′and
L′, possibly extended by −1.
Linked action manifolds are the most difficult to detect observationally, because
the groups R′and L′may be much smaller than Rand L. Indeed for some linked
action manifolds the groups R′and L′are trivial. For example, the lens space L(5,2)
is obtained as follows. Let R=Z5and L=Z5, and let the preferred generator rof
Rbe a right-handed Clifford translation through an angle 2π/5 while the preferred
generator lof Lis a left-handed Clifford translation through an angle 4π/5. Link the
generator rto the generator l, and link their powers accordingly. The identity element
of R(resp. L) occurs only with the identity element of L(resp. R), so both R′and L′
are trivial. In this case the PSH contains no spikes and we must use the CCP method
or CMB methods instead.
We emphasize that the detectability of a given topology depends not on the or-
der of the group, but on the order of the cyclic factor(s). Given that we can see to
a distance of about 2π/12 in comoving coordinates (see discussion above), the lens
space L(17,1), a single action manifold of order 17, would be detectable. However,
the lens space L(99,10), a double action manifold of order 99 generated by Z9×Z11,
would probably not be detectable, because its shortest Clifford translations have dis-
tance 2π/11. The only way we might detect L(99,10) would be if the Milky Way,
just by chance, happened to lie on a screw axis, where the flow lines of the right–
and left–handed factors coincide. On the other hand, a double action manifold like
S3/(Z97 ×Z98) probably is detectable. It has Clifford translations of order 97 and
98, so crystallographic methods should in principle work, and it does not violate the
observed lack of periodicity at 1/20 the horizon scale because its screw axis (along
which the minimum translation distance is 2π/(97 ∗98) = 2π/9506) probably comes

Topological Lensing in Spherical Spaces 20
nowhere near the Milky Way.
Putting the results all together and using the complete classification of spherical
spaces, we estimate that a few thousands of potentially observable topologies are
consistent with current data. This number is reasonably low in view of search strategies
for the detection of the shape of space through crystallogaphic or CMB methods. We
note that similar lower bounds were recently obtained in [34] in the particular case of
lens spaces, and this analysis does not give upper bounds and is less general than the
one presented here.
5.3. PSH spectra
Let us now discuss the form of ideal PSH spectra. We start by considering the expected
PSH spectrum in a simply–connected spherical space. As shown in [39], it is obtained
by computing the probablility P(χa, χl)dχlthat two points in a ball of radius χaare
separated by a distance between χland χl+ dχl
P(χa, χl) = 8 sin2χl
[2χa−sin 2χa]2{[2χa−sin 2χa−π] +
Θ(2π−2χa−χl)×[ sin2χa+π−χa−χl/2
−cosχasec(χl/2) sin(χa−χl/2)]},(22)
valid for all χa∈[0, π] and χl∈[0,min(2χa, π)] and in which Θ stands for the
Heavyside function.
As seen in the sample PSH spectra of figure 14, the PSH of the simply–connected
space S3provides the background contribution of the Poisson distribution over which
the spikes of the multi–connected space will appear.
Each Clifford translation gof the holonomy group will generate a spike in the
PSH spectrum. A Clifford translation takes the form M(θ, θ) or M(θ, −θ) (see §3.1)
and so its corresonding spike occurs at χ=θ. The number of group elements sharing
the same value of |θ|, with θnormalized to the inverval [−π, π], is the multiplicity
mult(θ) of the translation distance θ.
To compute the amplitude of the spike, we first consider the case that there is
only one source in the fundamental polyhedron. In that case, the number of images in
the covering 3–sphere equals the order of the holonomy group |Γ|. Each single image
sees mult(θ) neighbours lying at distance θfrom it, so that this distance appears
|Γ|mult(θ)/2 times (the division by two compensates for the fact that we counted the
distance from image A to image B as well as the distance from image B to image A).
We conclude that if we have Nsources in the fundamental polyhedron, the amplitude
of the spike located at χ=θis
amplitude(θ) = 1
2N|Γ|mult(θ).
As an example, let us consider the case of the cyclic group Z6consisting of six
Clifford translations through distances (−2π/3,−π/3,0, π/3,2π/3, π). These Clifford
translations yield spikes of multiplicity 1 at χ= 0 and χ=π, and spikes of multiplicity
2 at χ=π/3 and χ= 2π/3, as shown in table 2. Applying the above formula to the
spike at χ=π/3, in the case of N= 300 distinct sources in the fundamental domain,
we expect the spike to reach a height 1
2N|Γ|mult(θ) = 1
2(100)(6)(2) = 1800 above
the background distribution, in agreement with computer simulations (see figure 13).

Topological Lensing in Spherical Spaces 21
χ/π 0 1/2 1
multiplicity 1 6 1
Table 1. Position and the multiplicity of the spikes for the binary dihedral group
D∗
2. Compare with figure 14.
χ/π 0 1/3 2/3 1
multiplicity 1 2 2 1
Table 2. Position and the multiplicity of the spikes for the binary cyclic group
Z6. Compare with figure 14.
χ/π 0 1/3 1/2 2/3 1
multiplicity 1 2 6 2 1
Table 3. Position and the multiplicity of the spikes for D∗
2×Z3. As expected,
we find that it can be obtained from the data for the binary dihedral group D∗
2
(table 1) and Z3× {±1}=Z6(table 2). Compare with figure 14.
χ/π 0 1/5 1/3 2/5 1/2 3/5 2/3 4/5 1
multiplicity 1 12 20 12 30 12 20 12 1
Table 4. Position and the multiplicity of the spikes for the binary icosahedral
group I∗. Compare with figure 15.
In tables 1 to 4, we give the translation distances |θ|and the multiplicities mult(θ)
for the binary dihedral group D∗
2, the cyclic group Z6, the product D∗
2×Z3, and the
binary icosahedral group I∗, respectively. They correspond to the PSH spectra shown
in figures 14 and 16. Note that the spectrum for D∗
2×Z3(table 3) is obtained from
those of the binary dihedral group D∗
2(table 1) and Z3× {±1}=Z6(table 2).
5.4. PSH numerical Simulations
The previous section presented the theoretical expectations for PSH in multiply
connected spherical spaces. Here we carry out numerical simulations confirming those
expectations. Later in this section we will also take into account the approximate
flatness of the observable universe, which implies that we are seeing only a small
part of the covering space, and therefore can observe spikes only at small comoving
distances χ.
In figure 13 we draw the histogram for the lens space L(6,1) and we check that
our geometrical expectations about the positions and the heights of the spikes are
satisfied.
In figures 14 to 16 we present various simulations. We have chosen to draw
the normalised PSH in the covering space, i.e. the PSH divided by the number of
pairs and the width of the bin, and to show in grey regions the part of the PSH
that is observable if the redshift cut–off is respectively zc= 1,3,1000, which roughly
corresponds to catalogs of galaxies, quasars and the CMB.
We start by showing in figure 14 the “additivity” of the spectrum by considering
the group D∗
2×Z3for which the positions of the spikes in its PSH can be found

Topological Lensing in Spherical Spaces 22
from the PSH of the binary dihedral group D∗
2and of Z3× {±1}=Z6. Indeed, the
amplitude of the spikes is smaller since the same number of pairs has to contribute to
more spikes.
We then show in figures 15 and 16 different groups of the same order, chosen to be
120. The aim here is first to show that the number of spikes is not directly related to
the order of the group and that the order of the cyclic factor, if any, is more important.
While increasing this order, the number of spikes grows but their amplitudes diminish.
To get a physical understanding of this effect, we show in figure 17 the view for an
observer inside the manifold for the binary octahedral group O∗, for the cyclic group
Z17, and then for their product.
Finally, we applied our calculations by applying the PSH to real data, namely
a catalog of about 900 Abell and ACO clusters with published redshifts. The depth
of the catalog is zcut = 0.26, corresponding to 730h−1Mpc in a spherical universe
(see [14] for a more detailed description of this catalog). The PSH exhibits no spike;
this gives a constraint on the minimal size of spherical space, which corresponds to a
maximum order of the cyclic factor of the holonomy group of about 200.
0
1000
2000
3000
4000
5000
00,2 0,4 0,6 0,8 1
Z6
Pair separation histogram
Distance/π
Figure 13. The PSH for the lens space L(6,1) generated by the cyclic group
Z6with N= 300 sources in the fundamental domain. The height of the peaks
at χ=π/3 and χ= 2π/3 is of order 4200 while the background distribution is
of order 2400, giving the peaks a height above the background of about 1800, in
agreement with our theoretical analysis.
5.5. Robustness of PSH
In [20] we addressed the question of the stability of crystallograpic methods, i.e. to
what extent the topological signal is robust for less than perfect data. We listed the
various sources of observational uncertainties in catalogs of cosmic objects as:
(A) the errors in the positions of observed objects, namely
(A1) the uncertainty in the determination of the redshifts due to spectroscopic
imprecision,
(A2) the uncertainty in the position due to peculiar velocities of objects,
(A3) the uncertainty in the cosmological parameters, which induces an error in
the determination of the radial distance,

Topological Lensing in Spherical Spaces 23
0
1
2
3
4
5
00,2 0,4 0,6 0,8 1
D*2
Normalised pair separation histogram
Distance/π
0
0,5
1
1,5
2
2,5
3
00,2 0,4 0,6 0,8 1
Z6
Normalised pair separation histogram
Distance/π
0
0,5
1
1,5
2
2,5
3
00,2 0,4 0,6 0,8 1
D*2xZ
3
Normalised pair separation histogram
Distance/π
Figure 14. The PSH spectra for (upper left) the binary dihedral group D∗
2,
(upper right) the cyclic group Z6and (bottom) the group D∗
2×Z3. One can
trace the spikes from each subgroup and the black line represents the analytic
distribution in a simply–connected universe.
(A4) the angular displacement due to gravitational lensing by large scale structure,
(B) the incompleteness of the catalog, namely
(B1) selection effects,
(B2) the partial coverage of the celestial sphere.
We showed that the PSH method was robust to observational imprecisions, but
able to detect only topologies whose holonomy groups contain Clifford translations,
and we performed numerical calculations in Euclidean topologies. Fortunately the
shortest translations in spherical universes are typically Clifford translations (even
though more distant translations might not be), so the PSH is well suited to detecting
spherical topologies. In the present work we check the robustness of PSH in spherical
topologies.
Let us consider various spherical spaces of the same order 120 – namely the
lens space L(120,1), the binary dihedral space D∗
30 and the Poincar´e space I∗– and
calculate the corresponding PSH’s. The density parameters are fixed to Ωm0= 0.35

Topological Lensing in Spherical Spaces 24
0
1
2
3
4
5
6
7
8
00,2 0,4 0,6 0,8 1
D*30
Normalised pair separation histogram
Distance/π
z = 1
z = 3
z = 1000
0
1
2
3
4
5
00,2 0,4 0,6 0,8 1
I*
Normalised pair separation histogram
Distance/π
z = 1
z = 3
z = 1000
Figure 15. The PSH spectra for two spherical spaces of order 120 but with
no cyclic factor: (left) the binary dihedral group D∗
30 and (right) the binary
icosahedral group I∗
Table 5. Values of the limit percentage plof rejection above which the PSH
spikes disappear, as a function of the redshift cut–off for the Poincar´e space I∗.
zcut 1 3 ≥5
plNo signal 80% 90%
and ΩΛ0= 0.75. In the runs, the number of objects in the catalog is kept constant.
We examine separately the effects of errors in position due to redshift uncertainty ∆z,
and the effects of catalog incompleteness. Each of these effects will contribute to spoil
the sharpness of the topological signal. For a given depth of the catalog, namely a
redshift cut-off zcut, we perform the runs to look for the critical value of the error at
which the topological signal fades out.
Figure 18 gives the critical redshift error ∆zlabove which the topological spikes
disappear for the spherical space Z120 with Ω0= 0.35, ΩΛ0 = 0.75.
Next, we simulate an incomplete catalog where we randomly throw out p% of the
objects from the ideal catalog. For the cyclic group Z120 and the binary dihedral group
D∗
30, the topological signal is destroyed only for a very large rejection percentage (above
90%). For the less favorable case of the Poincar´e group I∗, the results are summarized
in table 5. Thus our calculations confirm that the PSH method is perfectly robust for
all spherical topologies containing Clifford translations.
As we have seen the PSH method applies for most of the spherical manifolds.
Nevertheless when the order of the group or of one of its cyclic subgroups is too
high then the spikes are numerous and have a small amplitude. Thus they may be
very difficult to detect individually. This the case for instance for some L(p, q) as
well as for linked action manifolds. In that case the CCP method, which gathers the
topological signal into a single index, is more suitable. Again the topological signal is
rather insensitive to reasonable observational errors as soon as the underlying geometry
contains enough Clifford translations.

Topological Lensing in Spherical Spaces 25
0
0,5
1
1,5
2
2,5
3
00,2 0,4 0,6 0,8 1
D*6xZ5
Normalised pair separation histogram
Distance/π
z = 1
z = 3
z = 1000
0
0,5
1
1,5
2
2,5
00,2 0,4 0,6 0,8 1
D*5xZ24
subgroup
Normalised pair separation histogram
Distance/π
z = 1
z = 3
z = 1000
0
0,5
1
1,5
2
2,5
00,2 0,4 0,6 0,8 1
Normalised pair separation histogram
Distance/π
z = 1
z = 3
z = 1000
Z120
Figure 16. The PSH spectra for three spaces of order 120 with different cyclic
factors: (upper left) D∗
6×Z∗
5, (upper right) D∗
5×Z∗
12 and (bottom) Z120. As
explained in the text, the order of the group being fixed, the higher the order of
the cyclic subgroup the larger the number of spikes. However the amplitude of
these spikes is smaller.
6. Conclusion and Perspectives
In this article, we have investigated the possible topologies of a locally spherical
universe in the framework of Friedmann–Lemaˆıtre spacetimes. We have given the
first primer of the classification of three-dimensional spherical spaceforms, including
the constructions of these spaces.
We have determined the topologies which are likely to be detectable in three–
dimensional catalogs of cosmic objects using crystallographic methods, as a function
of the cosmological paramaters and the depth of the survey. The expected form of the
Pair Separation Histogram is predicted, including both the background distribution
and the location and height of the spikes. We have performed computer simulations of
PSH in various spherical spaces to check our predictions. The stability of the method
with respect to observational uncertainties in real data was also proved.
Such a complete and exhaustive investigation of the geometrical properties of

Topological Lensing in Spherical Spaces 26
Figure 17. To have an intuitive understanding of which subgroups give rise to
spikes and how, it is fruitful to have a look from the inside. In the upper left panel,
we see the view inside the quotient of the 3–sphere S3by the binary octahedral
group O∗. The fundamental domain is a truncated cube, 48 copies of which tile
S3. The tiling’s 1–dimensional edges are shown in the figure. In the upper left
panel, we see the view in L(17,1). All 17 translates of the Earth align along a
Clifford parallel. When mixing both groups to get O∗×Z17 in the bottom panel,
we see the simultaneous effects of the cyclic factor Z17, which generates lines of
images, and the binary octahedral factor, which translates one line of images to
another. This illustrates how important the order of the cyclic group is, because
it alone determines the distance to our nearest translate.
Note: More distant images of the Earth are always dimmer than closer images
(because of artificial “fog”), but the apparent size of an image decreases only until
the images reaches a distance of π/2, after which more distant images appear
larger because the light from them follows geodesics that reconverge in S3. As an
image approaches the antipodal point, at a distance of π, it fills the whole sky.
spherical spaces will be useful for further studies dealing with spherical topologies,
including two–dimensional methods using CMB data. We plan to investigate the
applicability of the circle-matching method [40] for spherical topologies.

Topological Lensing in Spherical Spaces 27
0
50
100
150
200
250
300
1 1,5 2 2,5 3 3,5 4 4,5 5
∆zl × 103
zcut
Figure 18. Plot of ∆zlas a function of the depth of the catalog, for the spherical
space Z120 with Ω0= 0.35, ΩΛ0 = 0.75, using the PSH method.
Appendix A. Quaternions
The quaternions are a four–dimensional generalization of the familiar complex
numbers. While the complex numbers have a single imaginary quantity isatisfying
i2=−1, the quaternions have three imaginary quantities i,j, and ksatisfying
i2=j2=k2=−1 (A.1)
which anti–commute
{i,j}= 0,{j,k}= 0,{k,i}= 0 (A.2)
and are subject to the multiplication rules
[i,j] = 2k,[j,k] = 2i,[k,i] = 2j,(A.3)
and
[i,1] = 0,[j,1] = 0,[k,1] = 0,(A.4)
where [] and {} are the usual commutation and anti–commutation symbols.
Geometrically, the set of all quaternions
q=a1+bi+cj+dk,(a, b, c, d)∈R4(A.5)
defines four–dimensional Euclidean space, and the set of all unit length quaternions,
that is, all quaternions a1+bi+cj+dksatisfying a2+b2+c2+d2= 1, defines the
3–sphere. The quaternions are associative, (qr)s=q(rs), even though they are not
commutative. Introducing the conjugate quaternion ¯
qof qby
¯
q≡a1−bi−cj−dk(A.6)
and the modulus of qby
|q| ≡ √q¯
q=pa2+b2+c2+d2(A.7)
a unit quaternion satisfies
|q|= 1 ⇐⇒ ¯
q=q−1.(A.8)

Topological Lensing in Spherical Spaces 28
The identity quaternion 1is fundamentally different from the purely imaginary
quaternions i,jand k, but among the unit length purely imaginary quaternions
bi+cj+dkthere is nothing special about the basis quaternions i,jand k. Any
other orthonormal basis of purely imaginary quaternions would serve equally well.
Lemma A1:Quaternion change of basis
Let Mbe a 3 ×3 orthogonal matrix, and define
i′=M11i+M12 j+M13k
j′=M21i+M22 j+M23k
k′=M31i+M32 j+M33k(A.9)
then i′,j′and k′satisfy the quaternion relations
i′2=j′2=k′2=−1
i′j′=k′=−j′i′
j′k′=i′=−k′j′
k′i′=j′=−i′k′(A.10)
Lemma A1 says that an arbitrary purely imaginary quaternion bi+cj+dkmay, by
change of basis, be written as b′i′. If the purely imaginary quaternion bi+cj+dkhas
unit length, it may be written even more simply as i′. A not–necessarily–imaginary
quaternion a1+bi+cj+dkmay be transformed to a′1+b′i′. If it has unit length it
may be written as cos θ1+ sin θi′for some θ.
The unit length quaternions, which we continue to visualize as the 3–sphere, may
act on themselves by conjugation or by left or right multiplication.
Proposition A2:Conjugation by quaternions
Let qbe a unit length quaternion. According to the preceding discussion, we may
choose a basis {1,i′,j′,k′}such that q= cos θ1+ sin θi′for some θ. It is easy to
compute how qacts by conjugation on the basis {1,i′,j′,k′}:
(cos θ1+ sin θi′)1(cos θ1−sin θi′) = 1(A.11)
(cos θ1+ sin θi′)i′(cos θ1−sin θi′) = i′(A.12)
(cos θ1+ sin θi′)j′(cos θ1−sin θi′) = cos 2θj′+ sin 2θk′(A.13)
(cos θ1+ sin θi′)k′(cos θ1−sin θi′) = −sin 2θj′+ cos 2θk′(A.14)
Conjugation by any quaternion fixes 1(“the north pole”), so the action is always
confined to the “equatorial 2–sphere” spanned by {i′,j′,k′}. Within the equatorial
2–sphere, conjugation by the particular quaternion cos θ1+ sin θi′rotates about the
i′axis through an angle 2θ. Unlike the preceding action by conjugation, which always
has fixed points, when the quaternions act by left or right multiplication they never
have fixed points.
Proposition A3:Left multiplication by quaternions
Let qbe a unit length quaternion. Choose a basis {1,i′,j′,k′}such that q=
cos θ1+ sin θi′for some θ. It is easy to compute how qacts by left multiplication on
the basis {1,i′,j′,k′}.
(cos θ1+ sin θi′)1= cos θ1+ sin θi′(A.15)
(cos θ1+ sin θi′)i′=−sin θ1+ cos θi′(A.16)
(cos θ1+ sin θi′)j′= cos θj′+ sin θk′(A.17)
(cos θ1+ sin θi′)k′=−sin θj′+ cos θk′.(A.18)

Topological Lensing in Spherical Spaces 29
We see that left multiplication rotates 1towards i′while simultaneously rotating j′
towards k′. The result is a screw motion along so–called Clifford parallels. The Clifford
parallels are geodesics, and they are homogeneous in the sense that there is a parallel–
preserving isometry of S3taking any one of them to any other (see section 3.1). Action
by right multiplication is similar, but yields left–handed Clifford parallels instead of
right–handed ones.
Appendix B. Matrices
This appendix defines each finite subgroup of SO(3) as an explicit set of rotations.
Lifting a subgroup of SO(3) to the corresponding binary subgroup of S3is easy.
Proposition A.2, together with the change of basis principle in Proposition A.1, implies
that a rotation through an angle θabout an axis (x, y , z) is realized by both the unit
length quaternion
q= cos θ
21+1
px2+y2+z2sin θ
2(xi+yj+zk) (B.1)
and by its negative. This one-to-two mapping yields a binary subgroup of S3that is
exactly twice as big as the original subgroup of SO(3).
Converting from a quaternion in S3to a matrix in SO(4) is also easy. When a
quaternion q=a1+bi+cj+dkacts on the left its effect on the basis vectors 1,i,j,
and kis, respectively,
(a1+bi+cj+dk)1=a1+bi+cj+dk
(a1+bi+cj+dk)i=−b1+ai+dj−ck
(a1+bi+cj+dk)j=−c1−di+aj+bk
(a1+bi+cj+dk)k=−d1+ci−bj+ak(B.2)
Thus, relative to the orthonormal basis {1,i,j,k}, the matrix for this action is
Mleft(q) = 



a−b−c−d
b a −d c
c d a −b
d−c b a




.(B.3)
Similarly, when the same quaternion q=a1+bi+cj+dkacts on the right, the
resulting isometry has matrix
Mright(q) = 



a−b−c−d
b a d −c
c−d a b
d c −b a




.(B.4)
In the case of double action and linked action manifolds (§4.2 and §4.3), one
quaternion racts on the left (giving a right–handed Clifford translation) while a
different quaternion lacts on the right (giving a left–handed Clifford translation).
Their combined action is given by the product of their respective matrices. The order
in which we multiply the two matrices doesn’t matter because right– and left–handed
Clifford translations always commute.
The following realizations of the finite subgroups of SO(3) are unique up to an
orthonormal change of coordinates.

Topological Lensing in Spherical Spaces 30
Appendix B.1. Cyclic Groups Zn
Each cyclic group Znconsists of rotations about the axis (0,0,1) through an angle
2πk/n, for 0 ≤k < n. This defines ndistinct rotations.
Appendix B.2. Dihedral Groups Dm
The dihedral group Dmcontains the mrotations of the cyclic group Zmabout the
axis (0,0,1) as well as mhalf turns about the axes (cos kπ
m,sin kπ
m,0) for 0 ≤k < m.
This defines 2mdistinct rotations.
Appendix B.3. The Tetrahedral Group T
The tetrahedral group (Figure B1) consists of
•the identity,
•order 2 rotations about the midpoints of the tetrahedron’s edges, realized as π
rotations about the three axes (1,0,0), (0,1,0), and (0,0,1), and
•order 3 rotations about the tetrahedron’s vertices and face centers, realized as
counterclockwise 2π/3 rotations about the eight axes (±1,±1,±1).
This defines a tota l of 1 + 3 + 8 = 12 = |T|distinct rotations. Applying equation (B.1)
yields the 24 quaternions of the binary tetrahedral group T∗.
Figure B1. Visualisation of the symmetry group Tof the tetrahedron.
Appendix B.4. The Octahedral Group O
The octahedral group consists of
•the identity,
•order 2 rotations about the midpoints of the octahedron’s edges, realized as π
rotations about the six axes (±1,1,0), (0,±1,1), and (1,0,±1), and
•order 2 rotations about the octahedron’s vertices, realized as πrotations about
the three axes (1,0,0), (0,1,0), and (0,0,1), and
•order 3 rotations about the centers of the octahedron’s faces, realized as
counterclockwise 2π/3 rotations about the eight axes (±1,±1,±1).

Topological Lensing in Spherical Spaces 31
•order 4 rotations about the octahedron’s vertices, realized as counterclockwise
π/2 rotations about the six axes (±1,0,0), (0,±1,0), and (0,0,±1).
This defines a total of 1 + 6 + 3 + 8 + 6 = 24 = |O|distinct rotations. Applying
equation (B.1) yields the 48 quaternions of the binary octahedral group O∗. Note
that the octahedral group contains the tetrahedral group within it.
Appendix B.5. Icosahedral Group I
The icosahedron’s twelve vertices being at (±φ, ±1,0), (0,±φ, ±1), and (±1,0,±φ),
where φ= (√5−1)/2 is the golden ratio, the icosahedral group consists of
•the identity,
•order 2 rotations about the midpoints of the icosahedron’s edges, realized as
πrotations about the fifteen axes (1,0,0), (0,1,0), (0,0,1), (±φ, ±(φ+ 1),1),
(1,±φ, ±(φ+ 1)), and (±(φ+ 1),1,±φ), and
•order 3 rotations about the centers of the icosahedron’s faces, realized as
counterclockwise 2π/3 rotations about the twenty axes (±1,±1,±1), (0,±(φ+
2),±1), (±1,0,±(φ+ 2)), (±(φ+ 2),±1,0), and
•order 5 rotations about the icosahedron’s vertices, realized as both counterclock-
wise 2π/5 rotations and counterclockwise 4π/5 rotations about the twelve axes
(±φ, ±1,0), (0,±φ, ±1), and (±1,0,±φ).
This defines a total of 1 + 15 + 20 + 24 = 60 = |I|distinct rotations. Applying
equation (B.1) yields the 120 quaternions of the binary icosahedral group I∗. Note
that the icosahedral group also contains the tetrahedral group within it.
Acknowledgments
JRW thanks the MacArthur Foundation for its support, and thanks the remaining
authors for their warm hospitality. E.G. thanks FAPESP–Brazil (Proc. 00/04911-8)
for financial support.
[1] A.G. Riess et al., Astron. J. 116 (1998) 1009; S.Perlmutter et al., Nature (London) 391 (1998)
51.
[2] P. de Bernardis et al., Nature (London) 404 (2000) 955; ibid, [astro-ph/0105296].
[3] Y. Mellier, Ann. Rev. Astron. Astrophys. 37 (1999) 127.
[4] R. Juszkiewicz et al., Science 287 (2000) 109.
[5] B. Roukema and G. Mamon, Astron. Astrophys. 358 (2000) 395.
[6] C. Wetterich, Nucl. Phys. B302 (1988) 668; R.R. Caldwell, R. Dave, and P.J. Steinhardt, Phys.
Rev. Lett. 80 (1998) 1582; I. Zlatev, L. Wang, and P.J. Steinhardt, Phys. Rev. Lett. 82
(1999) 896; P.G. Ferreira and M. Joyce, Phys, Rev. D58 (1998) 023503; J.–P. Uzan, Phys.
Rev. D59 (1999) 123510.
[7] M. White, D. Scott, and E. Pierpaoli, [astro-ph/0004385].
[8] A. H. Jaffe et al., Phys. Rev. Lett. 86 (2000) 3475.
[9] G.D. Mostow, Ann. Math. Studies 78 (1973), Princeton University Press, Princeton, New Jersey.
[10] M. Lachi`eze-Rey and J.-P. Luminet, Phys. Rep. 254 (1995) 135.
[11] J.–P. Uzan, Int. J. Theor. Physics 36 (1997) 2439.
[12] J-P. Luminet and B.F. Roukema, in Theoretical and Observational Cosmology, M. Lachi`eze-Rey
(Ed.), Kluwer Ac. Pub., pp.117-157, [astro-ph/9901364].
[13] J-P. Uzan, R. Lehoucq, and J-P. Luminet, Proc. of the XIXth Texas meeting, Paris 14–18
December 1998, Eds. E. Aubourg, T. Montmerle, J. Paul and P. Peter, article no04/25,
[gr-qc/0005128].
[14] R. Lehoucq, M. Lachi`eze-Rey, and J-P. Luminet, Astron. Astrophys. 313 (1996) 339.
[15] R. Lehoucq, J-P. Luminet, and J-P. Uzan, Astron. Astrophys. 344 (1999) 735.

Topological Lensing in Spherical Spaces 32
[16] G.I. Gomero, A.F.F. Texeira, M.J. Rebou¸cas, and A. Bernui, [gr-qc/9811038].
[17] H.V. Fagundes and E. Gausmann, [astro-ph/9811368].
[18] H.V. Fagundes and E. Gausmann, Phys. Lett. A261 (1999) 235.
[19] G.I. Gomero, M.J. Rebou¸cas, and A.F.F. Teixeira, [astro-ph/9909078]; ibid.,
[astro-ph/9911049].
[20] R. Lehoucq, J-P. Uzan, and J-P. Luminet, Astron. Astrophys. 363 (2000) 1.
[21] J-P. Uzan, R. Lehoucq, and J-P. Luminet, Astron. Astrophys. 351 (1999) 766.
[22] J.A. Wolf, Spaces of constant curvature, fifth edition, Publish or Perish Inc., Wilmington USA
(1967).
[23] A.F. Beardon, The geometry of discrete groups, New York, Springer (1983).
[24] M. Nakahara, Geometry, Topology and Physics, Adam Hilger, Bristol, New-York (1990).
[25] W. de Sitter, Month. Not. R. Astron. Soc 78 (1917) 3 ; Proceedings of the Royal Academy of
Amsterdam 20 (1917) 229.
[26] J.-P. Luminet, L’invention du big bang, introduction to Essais de Cosmologie de A. Friedmann
et G. Lemaˆıtre, Seuil, Paris, 1997, p. 78.
[27] A.S. Eddington, The mathematical theory of relativity, chap. 5, (Cambridge University Press,
Cambridge, 1923).
[28] A. Friedmann, Zeitschr. f¨ur Phys. 2(1924) 326.
[29] G. Lemaˆıtre, Rev. Quest. Sci. (1929) 189 (English translation in The Primeval atom, Van
Nostrand, New York, 1950).
[30] J.V. Narlikar and T.R. Seshadri, Astrophys. J. 288 (1985) 43.
[31] G.F. R. Ellis, Gen. Rel. Grav. 2(1971) 7.
[32] J. R. Gott, Month. Not. R. Astron. Soc. 193 (1980) 153.
[33] R. Ionicioiu and R. Williams, Class. Quant. Grav. 15 (1998) 3469.
[34] G.I. Gomero, M.J. Rebou¸cas, and R. Tavakol, [gr-qc/0105002].
[35] W. Threlfall and H. Seifert, “Topologische Untersuchung der Diskontinuit¨atsbereiche endlicher
Bewegungsgruppen des dreidimensionalen sph¨arischen Raumes”, Math. Annalen 104 (1930)
1–70 and 107 (1932) 543–586.
[36] W.P. Thurston, Three-dimensional geometry and topology (1997) Princeton Mathematical series
35, Ed. S. Levy, (Princeton University Press, Princeton, USA).
[37] H. Poincar´e, “Cinqui`eme compl´ement `a l’analysis situs”, Rend. Circ. Mat. Palermo 18 (1904)
45-110.
[38] C. Weber and H. Seifert, “Die beiden Dodekaederr¨aume”, Math. Zeitschrift 37 (1933) 237-253.
[39] A. Bernui and A.F.F. Teixeira, [astro-ph/9904180].
[40] N. Cornish, D. Spergel, and G. Starkman, Class. Quant. Grav. 15 (1998) 2657.