DESCRIBING THE PLATYCOSMS

arXiv:math/0311476v1 [math.DG] 26 Nov 2003

J. H. CONWAY † AND J. P. ROSSETTI ‡

Abstract. We study in detail the closed flat Riemannian 3-manifolds.

Contents
1. Introduction 1
2. Dimension 2 2
3. Picturing the Platycosms - Space Groups 4
4. Parameters for Platycosms 11
5. Embedded Flat Surfaces 16
6. Infinite Platycosms 19
7. Fundamental Groups, Homology and Automorphisms 22
8. Double Covers 28
9. Diameters and Injectivity Radii 30
10. Some Formulae for Torocosms and Lattices 34
11. The Bravais-Voronoi Classes 35
Appendix I: Why there are just 10 platycosms 39
Appendix II: Conorms of Lattices 42
Appendix III: Dictionary of Names and Notations 44
References 46

1. Introduction
That there are just 10 closed flat 3-manifolds — we propose to call them
platycosms — has been known since around 1933. Since then, they have been
studied in many papers and several books, for example [Th], [We1], [Wo]. They
are of interest to speculative physicists as well as mathematicians, and indeed
there are some recent astronomical observations that suggest that the physical
universe might actually be a platycosm! (see for example [E], [HS], [NYT], [Spe],
[TOH], [URLW]).
We believe that these 10 manifolds should be more widely known. Accordingly,
our principal aim in this paper is to give a complete discussion of the ten compact
platycosms. In particular, we give them a uniform set of individual names, specify
2000 Mathematics Subject Classification. 20F34, 57S30, 20H15.
Key words and phrases. Platycosm, flat manifold, space group, automorphism, lattice, Bra-
vais, Voronoi, cover, diameter.
† NSF grant DMS-0072839. ‡ Supported by a Guggenheim fellowship.
1
2 J. H. CONWAY AND J. P. ROSSETTI

parameters in a systematic way, and give presentations for their fundamental

groups, which we use to derive their homology groups, automorphism groups and
list their double covers.
We then study their geometry, finding their Bravais types, diameters and in-
jectivity radii, and listing the compact embedded flat surfaces.
We have also included brief treatments of some related topics, namely the 2-
dimensional analogues of the platycosms, the 8 infinite platycosms and the 219
crystallographic groups.
Appendix I gives a brief proof that the list of 10 is complete, Appendix II
describes the “conorms” we use as parameters, while Appendix III is a dictionary
between our names and others.
An accompanying paper [RC] proves that, up to scale, there is a unique isospec-
tral pair of platycosms, the ‘DDT-example’ of [DR].
Acknowledgements: We have benefitted from conversations with Jeff Weeks, Bill
Thurston, and most particularly with Peter Doyle, who has also helped us in
many other ways. J. P. Rossetti is grateful to the Mathematics Department of
Princeton University for its hospitality during the writing of this paper.

2. Dimension 2
Although our main aim in this paper is to describe the platycosms or 3-
dimensional flat manifolds, particularly, the compact ones, we briefly discuss
their 2-dimensional analogues in this section.

The 2-dimensional flat manifolds. Everybody is familiar with the torus and
Klein bottle. Each of these can be given various locally Euclidean metrics — that
is to say, can be realized as a flat 2-manifold. We can see this by “rolling up”
the Euclidean plane in various ways: one obtains the typical flat torus TA B C by
dividing the plane by a 2-dimensional lattice of translations (Figure 1) — this is
mathematically more natural than identifying opposite sides of a parallelogram,
since each lattice can be defined by many different parallelograms. Beware: the
parameters A, B, C for the torus are the negatives of the inner products of w, x, y,
rather than their norms, which are A + B, A + C, B + C.
The most general flat Klein bottle KB A (Figure 1) can be obtained in a similar

way by dividing the plane by the group generated by the translation through a
vector x with norm N (x) = x · x = A together with a glide reflection based on
an orthogonal vector y of norm B.

Voronoi and Bravais types. The metric of a flat torus is determined by the
shape of its translation lattice. Voronoi classified lattices by the topological type
of their Voronoi cell, which is either hexagonal or rectangular, while Bravais
classified them by their symmetries, which are controlled by the shape of the
Delaunay cells. The lattice parameters A B C we recommend, called conorms
(see Figure 1 and Appendix II), easily yield both classifications (Table 1), since
the Voronoi cell is rectangular just if some conorm vanishes, while the shape of
the Delaunay cell is determined by this and which conorms are equal.
 DESCRIBING THE PLATYCOSMS 3

x A = −w · x y N (y) = B

C = −x · y w x
N (x) = A

y B = −w · y

A.
Figure 1. The torus TA B C and the Klein bottle KB

topological type shape of

conorms lattice shape
of Voronoi cell Delaunay cell
AB C scalene triangle generic lattice
AAB hexagonal isosceles triangle rhombic lattice
AAA equilateral triangle hexagonal lattice
AB 0 rectangle rectangular lattice
rectangular
AA0 square square lattice

Table 1. (it is understood that A, B, C are distinct and non-zero.)

If one lattice can be continuously deformed into another without changing

either its Bravais or its Voronoi class, we say they are in the same ‘BraVo’ class1
All these classifications can be applied to arbitrary flat manifolds and we dis-
cuss them for the platycosms in Section 11. All flat Klein bottles KB A lie in a

single BraVo class since A and B can be continuously and independently varied
through all positive values.
Infinite flat 2-manifolds. The torus and Klein bottle are finite 2-manifolds
— that is to say, they have finite volume, or (equivalently for flat manifolds)
are compact. There also exist just three types of infinite (or non-compact) flat
2-manifolds without boundary, namely the
Euclidean Plane R2 (∼ = T∞ ∞ = K∞ ∞ ),

= TA ∞ ∼
(infinite) Cylinder CA (∼ = K∞A ),
∼ ∞
Möbius Cylinder MB (= KB ),

1This is not the same as saying that they are in the same Bravais class and also the same
Voronoi class. The rhombic lattices ΛA A B split into two BraVo classes according as A > B or
A < B, between which we cannot pass without encountering a hexagonal lattice ΛA A A .
4 J. H. CONWAY AND J. P. ROSSETTI

illustrated in Figures 2 and 3 . Their 3-dimensional analogues are discussed in

Section 6.

(a) (b)

Figure 2. The (infinite) cylinder (a) and its covering plane (b).

(a) (b)

Figure 3. The Möbius cylinder (a) and its covering plane (b).

2-dimensional flat orbifolds. Dividing a Euclidean space by a discrete sub-

group of its symmetries yields a manifold only when the group is fixed-point-free
(i.e., only the identity element fixes any point of the space), or equivalently,
torsion-free (i.e., only the identity element has finite order). More generally, one
obtains a flat orbifold. There are just 17 types of compact flat 2-dimensional
orbifolds, corresponding to the 17 plane crystallographic groups, which are called
∗632, 632, ∗442, 4∗2, 442, ∗333, 3∗3, 333
∗2222, 2∗22, 22∗, 22×, 2222, ∗∗, ∗×, ××, ◦
in the orbifold notation [Co1], the last two being the Klein bottle and the torus.
The analogous results in 3 dimensions are discussed in Section 3.

3. Picturing the Platycosms - Space Groups

We shall use the term platycosm (“flat universe”) for a compact locally Eu-
clidean 3-manifold without boundary, since these are the simplest alternative
universes for us to think of living in [LSW]. (The program Curved Spaces of
Jeff Weeks — accompanying [We2] — allows you to ‘fly’ through platycosms and
other spaces.) If you lived in a small enough platycosm, you would appear to be
surrounded by images of yourself which can be arranged in one of ten essentially
different ways (Figures 12, 13, 4(i), 4(ii), 15, 16, 5, 6, 18(i), 18(ii)).
The images one sees of oneself lie in the manifold’s universal cover, which is of
course a Euclidean 3-space R3 . The symmetry operators that relate them form
 DESCRIBING THE PLATYCOSMS 5

a crystallographic space-group Γ, and geometrically the manifold is the quotient

space R3 /Γ.
If you hold something in one hand, your images will either all hold things with
the same hand (if the manifold is orientable, Figures 4, 12, 15, 16) or half with

(i) (ii)

Figure 4. (i) a tricosm, and (ii) a tetracosm.

their left hand and half with their right (if not; Figures 5, 6, 18). We therefore
call a platycosm chiral (“handed”) or amphichiral (“either handed”) according
as it is or is not orientable. Selecting the images of a particular handedness in
the picture for a non-orientable manifold X yields the picture for its orientable
double cover, Y , so we regard X as an ‘amphichiralized’ version of Y , and call it
‘an amphi-Y ’.
The chiral platycosms are the helicosms c1, c2, c3, c4, c6 (individually called
the torocosm, dicosm, tricosm, tetracosm, hexacosm) and the didicosm c22, also
known as the Hantzsche-Wendt manifold. In these notations, the letter “c” stands
for “chiral” while the digits indicate the point group, which is a cyclic group CN
of order N for cN , and C2 × C2 for c22.
A torocosm (usually called a 3-torus) is just the 3-dimensional analogue of
a torus. The space group for another helicosm cN is generated by a suitable
2-dimensional lattice of translations together with an orthogonal screw motion
of period2 N , while the space group for a didicosm is generated by orthogonal
period 2 screw motions whose axes bisect the faces of a ‘box’ as in Figure 7. The
space is tessellated into such boxes.
2We say a screw motion has period N if the lowest power of it that is a translation is the N th .
6 J. H. CONWAY AND J. P. ROSSETTI

Figure 5. A first amphicosm +a1.

Figure 6. A second amphicosm −a1.

Figure 7. The didicosm space group is generated by the half-

turn screw motions corresponding to the indicated lines.

The amphichiral platycosms are the first and second (or positive and negative)
amphicosms, ±a1, and amphidicosms, ±a2, whose orientable double covers are
c1 and c2 respectively (Figures 5, 6, 18). The figures become easier to follow if
 DESCRIBING THE PLATYCOSMS 7

we replace the human bodies (of Fig. 6, say) by boxes bearing the letters b,d,p,q
and insert some labelled o to indicate spacing (as in Fig. 8).

Figure 8. Another picture of a second amphicosm.

Such shorter figures for the amphis (Fig. 9) abbreviate nicely to

b o o b p o o p
b o d o b o o d b o q o b o o q
(1) d o d o d o d o
+a1 −a1 +a2 −a2

in which it is understood that the letters continue with period two in all directions.
Their space groups consist precisely of the operations that take the leading box
labelled ‘b’ to the other boxes labelled b, d, p or q.
We chose these standard forms because when x and y are orthogonal the first
amphicosm is the Cartesian product of a Klein bottle and circle (in later notation:
+a1D D 1
A:B = KA × SB ). However, they can be continuously deformed into the
variant forms:

d o o d q o o q
b o b o b o o b b o p o b o o p
(2) d o d o d o d o
+a1 −a1 +a2 −a2
in necessarily two different ways. For the amphicosms, one can use the ‘shear
motion’ that lifts each layer ‘one-half a unit’ more than the one in front of it (or,
in later notation, just change base from x, y, z to x, ±w, z). The amphidicosms
need only the ‘shift of motif’ that leaves the metric undisturbed, but moves each
8 J. H. CONWAY AND J. P. ROSSETTI

letter half way to the one above (in the b, d layers) or below it (in the p, q
ones). The variant forms are more useful in determining fundamental groups and
automorphisms.
A similar figure for the didicosm is

o p
b o q o
(3) c22 :
o d
in which the non-bold letters are those seen from behind, through the boxes
(Figure 10).

+a1 −a1

+a2 −a2

Figure 9. The amphicosms ±a1 and the amphidicosms ±a2.

Platycosms and space groups. Why are there only 10 platycosms? The real
reason is that just 10 of the crystallographic space groups are fixed-point-free,
analogously to the two plane crystallographic groups ◦ and ×× that yield the
torus and Klein bottle. In fact, Nowacki [No] found the 10 compact platycosms
in 1934 by finding which of the space groups of the International list were fixed-
point-free. A new and simple enumeration of the 219 space groups is given in
[CDHT], from which one can easily read off the numbers having various properties
(and so, in particular, pick out the platycosms). Of course one can obtain a much
shorter enumeration by restricting the argument to platycosms throughout, as
was done by Hantzsche and Wendt [HW] in 1934. We give a simpler proof of this
type in Appendix I.
 DESCRIBING THE PLATYCOSMS 9

Figure 10. A didicosm c22.

# parameters 1 2 3 4 6 #
# fibrations 0 1 3 ∞ ∞ -
total 350 1101 121 + 262 + 213 52 + 83 21 219
chiral 120 291 51 + 42 22 + 13 11 54
metachiral 1 10 0 0 0 11
∗432, 432 ∗22N , 22N
point groups
∗332, 3∗2 ∗NN , N ∗, NN ∗222, 222, ∗22 2∗, 22, ∗ ×, 1 -
in this case
332 2∗M , M ×
# pt. groups 5 19 3 3 2 32

# parameters 2 3 4 6 #
platycosms c3, c4, c6 c22, ±a2 c2, ±a1 c1 10
# fibrations 1 3 ∞ ∞ -
total 31 11 + 23 12 + 23 11 10
chiral 31 11 12 11 6
metachiral 3 0 0 0 3
point groups 33, 44, 66 222, ∗22 22, ∗ 1
# pt. groups 3 2 2 1 8

Table 2. The upper table arranges space groups according to

their point groups and numbers of parameters and Seifert fibra-
tions. The lower table restricts this information to platycosms.
The number M is 2 or 3, and N is 3 or 4 or 6.

We briefly describe the terms used here. A space-group is amphichiral (“either

handed”) or achiral (“not handed”)3 if it contains handedness-reversing opera-
tions; otherwise it is chiral (“handed”). It is metachiral if the group itself is
distinct from its mirror image.
3The terms “chiral” and “achiral” were introduced by Lord Kelvin before 1896, when he used
them in his Baltimore Lectures. Coxeter’s enlargement of the latter to amphichiral avoids using
the negating prefix for a positive property.
10 J. H. CONWAY AND J. P. ROSSETTI

The point group is the group obtained by identifying any two elements of the
space group that differ by a translation. We specify these finite crystallographic
groups in the orbifold notation [Co1]. The space groups with a given point group
constitute one of the 32 crystal classes.
The number of space-groups is often given as 230, because the metachiral ones
are counted twice, once for each of their two inequivalent orientations. From the
orbifold point of view this way of counting is incorrect — it would wrongly make
us say there were 13 platycosms! The numbers of types of oriented orbifolds and
platycosms are only 54+11=65 and 6+3 = 9 rather than 219 and 13. The 9 ori-
ented platycosm types arise as follows: the cases ±a1, ±a2, are non-orientable, so
yield no oriented types; each of c1, c2, c22 has an orientation-reversing symmetry,
so yields a single oriented type; finally c3, c4, c6 give two oriented types each,
since for them the defining screw motion may be either dextral (like a corkscrew)
or sinistral (like a reflected corkscrew).

Seifert fiber spaces. The orbifolds of many space groups can be realized as
Seifert fiber spaces, the one-dimensional fibers being the images in the orbifold of
a family of parallel lines that is fixed by the group. The number of such ‘Seifert
fibrations’, if not 0 or 1, is at least 3 (in fact usually 3, otherwise ∞), since if two
families of parallel lines are fixed so is the family of lines perpendicular to them4.
In Table 2, the numbers of cases with 3 or more fibrations appear with sub-
scripts 1, 2 or 3 according as the fibrations are of 1, 2 or 3 distinct types. Thus the
number with just three fibrations is given as 121 + 262 + 213 , meaning that there
are 12 in which the three fibrations are all of 1 type, 26 in which the fibrations
are of 2 distinct types, and 21 in which the fibrations are of 3 distinct types.
It happens that every platycosm has at least one realization as a Seifert fiber
space. Each such fibration determines a plane crystallographic group (by “looking
along the fibers”) and the “types” we now give (in Table 3) are the orbifold
notations for these groups:

torocosm c1 infinitely many fibrations, all of type ◦.

dicosm c2 one fibration of type 2222, infinity of type ××.
tricosm c3 one fibration of type 333.
tetracosm c4 one fibration of type 444.
hexacosm c6 one fibration of type 632.
didicosm c22 just three fibrations, all of type 22×.
first amphicosm +a1 one of type ◦, infinity of types ∗∗, ××.
second amphicosm −a1 one of type ◦, infinity of types ∗×, ××.
first amphidicosm +a2 just three fibrations, of types 22∗, ∗∗, ××.
second amphidicosm −a2 just three fibrations, of types 22×, ∗×, ××.

Table 3. The platycosms and their Seifert fibrations.

4and this third family will be rationally related to the lattice if the first two are.
 DESCRIBING THE PLATYCOSMS 11

4. Parameters for Platycosms

In the interest of consistency, the parameters we use are always the conorms
of what we call the Naming lattice N , generated by all the translation vectors,
screw vectors, glide vectors. Since our parameters are often squared lengths, we
introduce the convention that numbers inside a ‘square’ should be squared, for
instance c22 a b c means c22A B C , where A = a2 , B = b2 , C = c2 .
The Torocosm c1 = c1D EF
A B C . This is our proposed name for the 3-dimensional
torus. Geometrically it is the quotient R3 /T , where T is the normal sub-
group generated by translations — “the lattice of translations”. Our parameters
A, B, . . . , F are the conorms (see Appendix II) of T — in other words T has an
obtuse superbase x, y, z, t with y · z = −A, z · x = −B, x · y = −C, x · t = −D,
y · t = −E, z · t = −F . In the corresponding conorm diagram ((i) in Figure 11)
the Voronoi vector corresponding to any line is a translation vector.
(i) 0 (ii) 0 (iii) D

D F D 0 ∼
= 0 0
E 0 0

A B C A B C A B C

Figure 11. Conorms for ΛD EF D

A B C (i), and ΛA B C (ii) and (iii).

We omit conorms with value 0: for instance the lattice for c1D A B C has conorms
(ii), or equivalently (iii) in Figure 11. The lattice for c1A B C is spanned by three
orthogonal vectors of norms A, B, C as in Figure 12.
The torocosm is the helicosm for N = 1. The N -cosms for N = 2, 3, 4, 6
(the other helicosms) are generated by the translations of a 2-dimensional lattice
hw, x, yi where w + x + y = o, whose conorms are the lower parameters together
with a period N screw motion along a perpendicular vector z of norm D. The
particular cases are:
The Dicosm c2D A B C . The 2-dimensional lattice ΛA B C has an obtuse superbase
of three vectors w, x, y with w + x + y = o whose inner products w · x = −A,
w · y = −B, x · y = −C are non-positive. The half turn negating all these vectors
is an order 2 rotational symmetry. The space group for the dicosm c2D A B C is
generated by the translations of this lattice together with the period 2 screw
motion obtained by combining the above rotation with a translation through
a vector z of norm D perpendicular to ΛA B C , so the conorm function for the
naming lattice ΛDA B C is (ii) or (iii) in Figure 11.

The Tricosm c3D D D

A A A , Tetracosm c4A A , Hexacosm c6A A A . Only certain
special 2-dimensional lattices have higher order rotational symmetry. They are
ΛA A A which has order 3 and 6 rotations taking w → x → y → w and w →
12 J. H. CONWAY AND J. P. ROSSETTI

Figure 12. The orthogonal torocosm c1A B C .

Figure 13. The dicosm.

−y → x → −w respectively (Figure 14 left), and ΛA A (meaning ΛA A 0 ), which

has the order 4 rotation x → y → −x (Figure 14 right).
The space groups of the tricosm c3D D
A A A (Figure 4(i)), tetracosm c4A A (Fig-
ure 4(ii)), hexacosm c6D
A A A (Figure 15) are generated by the translations of the
appropriate lattice together with the period 3, 4 or 6 screw motion obtained by
combining the corresponding rotation with a translation through a perpendicular
vector of norm D.
Since a period N screw motion with N > 2 is either dextral or sinistral (see
Figure 15), the groups of these three manifolds have two enantiomorphic forms
each — that is to say, they are metachiral.
 DESCRIBING THE PLATYCOSMS 13

x −y y

−w w −x x
y −x
−y
ΛA A A ΛA A

Figure 14. The hexagonal lattice ΛA A A and square lattice ΛA A .

Figure 15. A sinistral hexacosm (left) and a dextral hexacosm (right).

The Didicosm c22A B C . For the didicosm (see Figures 10 and 16), the nam-
ing lattice is generated by the three defining screw vectors x, y, z, and so is a
rectangular lattice ΛA B C , whose conorms A, B, C are their squared lengths, say
a2 , b2 , c2 . The volume is 2abc (and so the determinant, or squared volume, is
4ABC), since a fundamental region consists of two a × b × c ‘boxes’, say the
leading box labelled b in Figure 10 and an adjacent one labelled o.
14 J. H. CONWAY AND J. P. ROSSETTI

Figure 16. The dizziness of the didicosm! (Figure 10 may be

more helpful.)

The Amphicosms ±a1D D

A:B C . The naming lattice for the amphicosms ±a1A:B C
is the lattice N = hw, x, y, zi (where w + x + y = 0) with Gram-matrix
w x y z
w A+B −A −B 0
x −A A+C −C 0 , but we must point out the ‘anomaly’ that z
y −B −C B+C 0
z 0 0 0 D
plays different rôles in the two cases — it is a translation vector for +a1, but
only half a translation vector for −a1: In consequence, the determinants are
D(AB + AC + BC) for +a1 and 4D(AB + AC + BC) for −a1. The translation
lattices T are h2x, y, zi and h2x, 2y, 2z, y + zi, and |N /T | = 2 or 4 respectively
(see Figure 17).

w w

x x

z (norm D) z (norm D)

Figure 17. Defining vectors for the amphicosms.

 DESCRIBING THE PLATYCOSMS 15

x y z
x A 0 0
The Amphidicosms ±a2D
A:B . For the amphidicosms the Gram-matrix ,
y 0 B 0
z 0 0 D
shows that the Naming lattice N = hx, y, zi is a rectangular 3-dimensional lattice
ΛD ∼ A B D . We have the same anomaly about the meaning of z, which leads
AB = Λ
to differing determinants: ABD for +a1D D
A:B and 4ABD for −a1A:B . The trans-
lation lattices T are h2x, 2y, zi and h2x, 2y, 2zi, and |N /T | = 4 or 8 respectively
(see Figure 19).

(i) (ii)

Figure 18. (i) a first amphidicosm +a2, and (ii) a second am-
phidicosm −a2.

y y

x x

z (norm D) z (norm D)

Figure 19. Defining vectors for the amphidicosms.

Which is which? It can be quite difficult to recognize which kind of amphicosm

or amphidicosm one finds oneself in! Thus it took 20 years before it was noticed
that the four figures of amphichiral platycosms in the celebrated book The Shape
of Space [We1], only define two inequivalent manifolds!5
There are several ways to answer this question. One is that it is an amphicosm
or amphidicosm according as its point group has size 2 or 4, and first or second
5Figures 7.7 and 7.10 in that book are two first amphicosms with different metrics, while Fig-
ures 7.11 and 7.12 are actually isometric first amphidicosms! To understand these equivalences,
see the remark on ‘variant forms’ (2) in Section 3.
16 J. H. CONWAY AND J. P. ROSSETTI

according as |N /T | is 1 or 2 times this. However, the safest rule we have found

is
An ‘amphi’ is of first or second type if and only if it has a glide mirror of first
or second return, respectively.
The reflecting plane of a glide reflection (a glide mirror) is of first or second
return if and only if the geodesic normal to its image at an arbitrary point closes
exactly at its first or second return to the surface respectively.
So to find which kind of amphichiral platycosm you are in, starting from a
general point of some glide mirror, walk perpendicularly to it until you first
hit that mirror again. If this is always at the same point, you are in a first (or
positive) amphicosm or amphidicosm — if always at a different point, in a second
(or negative) one. But beware: in an amphidicosm, there are also ‘ambiguous’,
glide mirrors that meet most normals twice, but some only once. For more details,
see the next section.

5. Embedded Flat Surfaces

In this section, “embedded surface” always means “embedded compact flat
2-manifold”. It will be either a torus or a Klein bottle, and either type may be
embedded either 1-sidedly or 2-sidedly6. Any such surface is the image of a plane
in the universal cover with the property that modulo Γ it is compact.
Then the planes parallel to π will have the same properties and so yield other
embedded surfaces. The configuration space of any such parallel family may
be a circle, which we denote by (2K) or (2T ) according as its elements (which
are necessarily 2-sided) are Klein bottles or tori. Alternatively, it may be an
interval, in which case only the extreme members will be 1-sided; we will write
[1K (2T ) 1T ] (say) for such an interval whose end surfaces are a 1-sided Klein
bottle and a 1-sided torus, and whose interior surfaces are 2-sided tori.
It is useful to classify 1-sided surfaces more closely, which we do by inserting
signs and the letters g and s. We write +1 for a first return (or ‘positive’) surface,
meaning that the normal at any point p intersects it only at p, and −1 for a second
return (or ‘negative’) one, when the normal at any p intersects it in exactly two
distinct points (p and the ‘negative’ of p), and finally ∓1 for an ‘ambiguous’ one,
which intersects most normals twice but some only once.
Again if a surface is 1-sided then there is either a glide reflection or a screw
motion (or both) preserving it. We call it respectively a glide surface or screw
surface (or both), and indicate this by the letters g or s (or gs). So +1gT refers
to a positive (1-sided) glide torus, ∓1sK to an ambiguous screw Klein bottle.
Since embedded flat 2-manifolds are the images of planes in the universal cover,
we must look for such planes whose image modulo Γ is a torus or Klein bottle
(the only compact possibilities).

6For surfaces embedded in Euclidean space, 2-sidedness is equivalent to orientability, but in

more general 3-manifolds this is no longer true.
 DESCRIBING THE PLATYCOSMS 17

Some considerations limit the search. The plane π must not intersect any of
its images under Γ other than itself, which entails that all its images are parallel.
Equivalently, Γ preserves the family of parallel lines perpendicular to the plane.
There is a standard argument that usually proves π is basal, that is, parallel
to hx, yi, or perpendal, that is, perpendicular to hx, yi. For if not, a unit vector
v orthogonal to π would resolve into basal and perpendal parts as v = v1 + v2
with v1 6= 0, v2 6= 0. Since the typical operation γ ∈ Γ takes this to v = v1′ + v2′
where v2′ = ±v2 we must have v1′ = ±v1 with the same sign. However for every
case other than c1 we can find a γ for which this does not happen.
For the hexacosm c6, tetracosm c4 and tricosm c3 the argument actually
proves more, for then the defining screw motion takes v1 to a vector v1′ which is
not equal to ±v1 , unless v1 = 0, which forces v = v2 and π to be basal. The
basal planes yield a circular family (2T ) of 2-sidedly embedded tori.
For the dicosm c2 the same is true of the basal planes, but now there are also
an infinite number of families of perpendal planes, such as the family parallel to
hx, zi. These yield an interval [1sK (2T ) 1sK] of embedded surfaces as suggested
in Figure 20. We obtain a similar interval of perpendal planes for each primitive
vector x′ (counted up to sign) of the lattice hx, yi, since any such vector appears
in some basis x′ , y′ . On the other hand, the standard argument shows that

y
1
2
y
1sK
2T
1sK
x

Figure 20. The extremes (through 0 and 12 y) are 1-sided Klein

bottles while all others are 2-sided tori.

every embedded surface must be basal or perpendal, since the defining screw
motion takes v1 to −v1 but v2 to +v2 . This shows that the infinity of intervals
[1sK (2T ) 1sK] of the preceding paragraph do indeed comprise all the non-basal
embedded surfaces, agreeing with the entry (2T )1 , [1sK (2T ) 1sK]∞ in Table 4.
For the torocosm c1 the standard argument does not apply, but the answer
is easy — there is a circular family (2T ) of 2-sided tori corresponding to each
2-dimensional section7 of T = hx, y, zi, or equivalently to each pair of primitive
vectors ±v∗ of the dual lattice T ∗ , yielding the entry (2T )∞ in Table 4.
For the didicosm c22 the standard argument can be applied in different di-
rections, showing that π must be parallel or perpendicular to hy, zi as well as
to hx, yi, which forces it to be parallel to one of the three coordinate planes
hx, yi, hx, zi, hy, zi. We obtain three intervals of type [∓1sK (2T ) ∓1sK] (Fig-
ure 21).
7a section of a lattice is its full intersection with some subspace.
18 J. H. CONWAY AND J. P. ROSSETTI

α β γ

Figure 21. The plane α is a Klein bottle, which is embedded

1-sidedly as shown by the screw motion taking b to q, and the
same is true for γ as shown by that taking b to d. In view of these
two screw motions, we need only consider planes like β between
α and γ — these are 2-sided tori.

For the amphicosms ±a1 the standard argument applies since the glide re-
flection takes v1 to +v1 but v2 to −v2 . We first handle the perpendal planes.
We can see that the planes parallel to hx, zi or hy, zi yield circles of 2-sided
Klein bottles and 2-sided tori respectively, since x is a glide vector and y is a
translation vector.8 But any primitive vector of hx, yi belongs to a superbase
w′ , x′ , y′ whose vectors are respectively congruent modulo 2 to w, x, y, which
makes w′ , x′ be glide vectors and y′ a translation vector, and any such superbase
is equivalent to w, x, y by some isotopy of the lattice. So the perpendal planes
yield infinitely many circles of each type (2K) or (2T ). The basal planes form an
interval of type [+1gT (2T ) +1gT ] for +a1, [−1gT (2T ) −1gT ] for −a1.
Finally, for the amphidicosms ±a2 the standard argument applies in both
hx, yi and hx, zi directions, showing that π must be parallel to a coordinate plane.
The perpendal planes parallel to hx, zi form a circle (2K) and those parallel to
hy, zi an interval [∓1sK (2T ) ∓1gK] while the basal planes (parallel to hx, yi)
form an interval of type [+1gsK (2K) +1gsK] for +a2 and [−1gT (2T ) −1sK]
for −a2.
We summarize the results in Table 4.

8In view of the symmetry interchanging w and x, the planes parallel to hw, zi behave like
those parallel to hx, zi.
 DESCRIBING THE PLATYCOSMS 19

platycosm families of surfaces

c1 (2T )∞
c2 (2T ) ; [1sK (2T ) 1sK]∞
1

c3 (2T )1
c4 (2T )1
c6 (2T )1
c22 [∓1sK (2T ) ∓1sK]3
+a1 [+1gT (2T ) +1gT ]1 ; (2K)∞ , (2T )∞
−a1 [−1gT (2T ) −1gT ]1 ; (2K)∞ , (2T )∞
+a2 [+1gsK (2K) +1gsK]1 ; (2K)1 , [∓1sK (2T ) ∓1gT ]1
−a2 [−1gT (2T ) −1sK]1 ; (2K)1 , [∓1sK (2T ) ∓1gT ]1
Table 4. The parallel families of embedded surfaces, with the
type symbol defined in the text and the number of families of this
type indicated by the superscript. Those before a semicolon are
images of basal planes, those after of perpendal ones.

6. Infinite Platycosms
Although elsewhere in this paper ‘platycosm’ means ‘finite platycosm’, in this
section we briefly discuss and name the infinite ones. An infinite platycosm is a
boundaryless flat manifold that has infinite volume, or equivalently is not com-
pact.
One kind of infinite platycosm, the ‘Product Space’ or Prospace, is topologi-
cally the cartesian product of a compact flat manifold of dimension 0, 1, 2 by a
Euclidean space of the complementary dimension 3, 2, 1. The cases are

compact complementing notation and

name
factor (fiber) space parameters
Point R3 Euclidean Space EU C
Circle R2 Circular Prospace CP SA (θ)
Torus R1 Toroidal Prospace T P SA B C
Klein Bottle R1 Kleinian Prospace KP SBA

The others are related to these in the way a Möbius strip is to an annulus, so
we call them Möbius spaces, or ‘Mospaces’. Technically, they are fibrations whose
base is a compact flat submanifold of dimension 1 or 2 and whose fiber is the
‘complementary’ Euclidean space of dimension 2 or 1, which is reflected when we
traverse at least one closed path in the base. This path must be homotopically
non-trivial and may be orientation preserving (+) or orientation-reversing (−).
The cases are
20 J. H. CONWAY AND J. P. ROSSETTI

compact compl. path notation and

name
base fiber type parameters
Circle R2 + Circular Mospace CM SA
Torus R1 + Toroidal Mospace T M SA B:C
Klein Bottle R1 − chiral Kleinian Mospace +KM SB A

Klein Bottle R1 + achiral Kleinian Mospace −KM SB A

Figures 22, 23 and 24 picture these manifolds (strictly, their universal covers).
√ group for CP SA (θ) is generated by a screw motion of angle θ and length
The
A. In the untwisted case when θ = 0 this is the direct product of a circle and
a plane even in the metrical sense, in the twisted cases (when θ is not a multiple
of 2π) it is only topologically so.

Figure 22. One’s images in the simplest infinite platycosms.

√
In CP SA (θ) (= CP S a (θ)) there is always a closed geodesic of length a = A:
θ
this is the only primitive closed geodesic unless 2π is rational, when the associated
screw motion has a finite period N , and all geodesics parallel to it are closed, of
length aN . For CM SA (= CM S a ) the group is generated by a glide reflection
√
of length a = A. The manifolds described so far are illustrated in Figure 22.
For T P SA B C the group is a 2-dimensional lattice of translations for which
some superbase v0 , v1 , v2 has conorms A, B, C. In T M SA B:C the translations
through v0 and v1 (so not v2 ) are replaced by the glide reflections obtained
by combining them with the reflection in the base plane hv0 , v1 , v2 i. These
manifolds appear in Figure 23.
The group for KP SB A (= KP S a ) is generated by a translation along v of
b 1
√ √
length a = A and a glide reflection along a vector v2 of length b = B per-
pendicular to v1 whose reflecting plane is orthogonal to the base plane hv1 , v2 i.
For +KM SB A (= +KM S a ) we compose the latter with the reflection in the
b
base plane, so that it becomes a screw motion. For −KM SB A (= −KM S a ) we
b
instead compose the former with that reflection, so making it a glide reflection.
Figure 24 shows these three Kleinian manifolds.
 DESCRIBING THE PLATYCOSMS 21

Figure 23. The toroidal Product and Möbius spaces.

Figure 24. Kleinian Product and Möbius spaces.

Beware: not every 3-dimensional product of flat manifolds is a ‘Prospace’ —

we use that term only for the product of a compact manifold and a Euclidean
space. The following little table shows that no fewer than 8 of the 18 platycosm
types include non-trivial direct products (Table 5).

product of line circle

plane EU C CP S ∗
cylinder CP S ∗ T P S ∗
Möbius cylinder CM S T M S ∗
torus TPS c1 ∗
Klein bottle KP S +a1 ∗

Table 5. An asterisk indicates that the metrical direct products

of this kind do not yield all parameter sets in this case.
22 J. H. CONWAY AND J. P. ROSSETTI

7. Fundamental Groups, Homology and Automorphisms

The fundamental group of a platycosm is just the space group Γ we see in its
universal cover. We explain how to find a presentation, using the amphicosms
and amphidicosms as our examples. Let us mark any letter b, d, p, q in our
diagrams with the element of Γ by which it can be obtained from the initial b —
this sets up a 1-1 correspondence, since these letters have no symmetry.9 Then a
presentation for Γ can be computed as follows. First find generators for Γ, and
then generators for T in terms of these. Then a set of relations will be sufficient
if and only if they imply the correct structures for
i) the subgroup T ;
ii) the action of each generator on T ;
iii) the point group G = Γ/T .
Since the two amphicosms ±a1 are the hardest cases, we discuss them in detail.
It is easy to see that the operations W, X, Z defined by Figures 17 and 25 are
generators. The vectors associated to these are w, x and z′ = z or 2z, and we see
that the translation lattice is generated in each case by 2w, 2x, w + x, z′ , showing
that the subgroup T is generated by W 2 , X 2 , W X, Z.
+a1: −a1: +a2: −a2:
d o o d q o o q
W b o W o b W p o W o p
b o b o b o b o
X X X X
d o d o d o d o
Z Z Z Z

Figure 25. Generators for the ‘amphis’. In these, X is always

a glide reflection with vector x that takes the leading b to the d
below it, while Z is the translation with vector z or 2z that takes
it to the next b to its right. For +a1, W is a glide reflection with
vector w. For −a1, it is a glide reflection with vector w + z while
the W ’s for ±a2 are (different) screw motions, with vectors y.

We obtain the correct structure for T by demanding that these commute, and
that (W X)2 = W 2 X 2 or W 2 X 2 Z in the two cases. Let us suppose in general that
we have found some generators G1 , G2 . . . for Γ, and certain products w1 , w2 , . . .
of them that generate T , and relations R1 , R2 , . . . that define its structure. Our
next step is to define the correct action on T , for which it suffices to express each
G
wi j = G−1
j wi Gj as a function of the wi . For the first amphicosm, we find that
W and X 2 are fixed by all three of W, X, Z, while Z is fixed by itself (obviously),
2

but inverted by each of W and X, so we add these assertions as relations.

9Not even front-to-back!
 DESCRIBING THE PLATYCOSMS 23

Finally, we must ensure that our relations imply the correct structure for the
point group G = Γ/T . However, for the amphicosms they already do, since
modulo T we have W 2 ≡ X 2 ≡ W X ≡ Z ≡ 1, which show that in the point
group, W and X map to the same element of order two, while Z maps to 1.
The relations one obtains in this way can almost always be simplified. The
best way to do this is to adjoin the shortest relations one can find and then delete
any redundant ones. For the first or second amphicosms, we find three simple
relations, namely
Z W = Z X = Z −1 and W X = W or ZW
in the two cases.
The fact that W, X invert Z implies that W 2 , X 2 , W X commute with Z. But
also, these three relations imply that W 2 commutes with X 2 — obviously for
the first amphicosm, while for the negative one ZW 2 Z W = ZW 2 Z −1 = W 2 .
Moreover, they imply that
XW XW = X 2 W X W = X 2 (1 or Z)W 2 ,
which was the further relation needed for the structure of T .
Finally, our three simple relations specify the action of W, X on Z and each
other, so in particular on W 2 , X 2 , W X, Z, the generators of T . Rewriting the
last relation as
[X, W ] = 1 or Z,
we see that:
The fundamental groups of the amphicosms have presentations:
for + a1 hW, X, Z : Z X = Z W = Z −1 , [X, W ] = 1i
for − a1 hW, X, Z : Z X = Z W = Z −1 , [X, W ] = Zi.
Of course, their first homology groups are obtained by abelianizing these. They
are
for + a1 hW, X, Z| abelian, Z 2 = 1i ∼ = C2 × C∞ × C∞
for − a1 hW, X, Z| abelian, Z = 1i ∼ = C∞ × C∞ ,
since in the +a1 case Z was conjugate to Z −1 , so Z 2 maps to 1, while for −a1
Z was a commutator, so itself maps to 1. In Table 6, we abbreviate these group
structures to 2 · ∞2 and ∞2 .
Helicosms. For the helicosms we have the presentations of the table, in which Z
is the defining screw motion, while X and Y generate the 2-dimensional lattice
perpendicular to it.
Didicosm. For the didicosm c22 we find in this way that the translation subgroup
is generated by the squares of the generating screw motions X, Y, Z, which invert
the squares of each other, and satisfy XY Z = 1. It turns out that this presenta-
tion reduces to hX, Y | X = Y 2 XY 2 , Y = X 2 Y X 2 i when we define Z = (XY )−1 .
The abelianization is C4 × C4 , which we denote by 42 . This proves the well
known fact that the Hantzsche-Wendt didicosm c22 is the only platycosm with
finite homology, or equivalently, with zero first Betti number.
24 J. H. CONWAY AND J. P. ROSSETTI

presentation H1 transl. lattice

c1 X ⇋Y ⇋Z⇋X ∞3 X, Y, Z

X ⇋Y
c2 22 · ∞ X, Y, Z 2
Z : X → X −1 , Y → Y −1

X ⇋Y
c3 3·∞ X, Y, Z 3
Z : X → Y → (XY )−1

X ⇋Y
c4 2·∞ X, Y, Z 4
Z : X → Y → X −1

X ⇋Y
c6 ∞ X, Y, Z 6
Z : X → XY → Y

c22 hX, Y |X = Y 2 XY 2 , Y = X 2 Y X 2 i 42 X 2 , Y 2 , Z 2 = (XY )−2

+a1 W, X : Z → Z −1 , [X, W ] = 1 2 · ∞2 W 2 , X 2 , W X, Z

−a1 W, X : Z → Z −1 , [X, W ] = Z ∞2 W 2 , X 2 , W X, Z

+a2 W, X : Z → Z −1 , W : X → X −1 22 · ∞ W 2, X 2 , Z

−a2 W, X : Z → Z −1 , W : X → X −1 Z 4·∞ W 2, X 2 , Z
Table 6.

We remark that the fundamental groups of c1, c2, +a1, +a2 need three gener-
ators; each of the others can be generated by two elements.
Automorphisms. The largest group of ‘symmetries’ of Γ is its affine normal-
izer10, which consists of the affine automorphisms of R3 that take Γ into itself.
We discuss the four ‘parts’ of this group11 illustrated in Fig. 26:
Part I is the connected part (the component of the identity), which by continu-
ity must fix Γ pointwise since the identity does. In other words, it is central. For
the platycosms, it consists only of translations and its dimension is the first Betti
number β1 . It is a subtle theorem (cf. [ChV]) that factoring out part I yields
precisely the automorphisms of Γ.12
Part II, the inner part, is the space group Γ, which acts trivially on M , but
(of course) induces inner automorphisms on Γ. Since these induce the identity
on M , they must be factored out to obtain the group of affinities of M .
The intersection of parts I and II is the translation subgroup of Γ. If we
factor out both, we obtain the outer automorphism group of Γ, which is equally
10so called because it is the normalizer of Γ in the group of affine transformations of R3 .
11Beware: these ‘parts’ are not always subgroups and not always disjoint.
12Our argument, although it refers to the way an automorphism acts on Γ, does not in fact
use this theorem.
 DESCRIBING THE PLATYCOSMS 25

deformable:
Bravais
subgroups B1 B2 IV
rigidly
isometric: III
Au

Γ
tom

of
ms
orp

connected: inner:

his
his

I II

orp
ms

tom
T
of

Au
M

Figure 26.

the discontinous part of the (affine) symmetry-group of M . The elements of

the connected and inner parts are isometric, but not all elements of the outer
automorphism group need be so. We therefore separate it into:
Part III, the rigidly isometric part, the normal subgroup that contains precisely
those elements that are isometric for all the values of the parameters. Factoring
these out, we obtain:
Part IV, the deformable part. For example, the variant form (2) for +a1 shows
that it has an automorphism interchanging W and X (see Fig. 17), which is not
always an isometry. It will however became an isometry if B = C. In general there
are certain finite subgroups, the Bravais subgroups B1 , B2 , . . . , of the deformable
part that just consist of the symmetries that can become isometric for various
choices of the parameters. They are completely enumerated in Section 11.
Combining the results of that section and this, our paper describes all isome-
tries of all platycosms.
These groups are tabulated in Table 7. We shall discuss only the two hardest
cases, c22 and −a1.

The second amphicosm. For −a1 it is easily checked that the most general inner
automorphism (m, n, σ), say, takes W → W Z m , X → XZ n , Z → Z σ , where
σ = ±1, m − n = σ−1 2 , the ones induced by W, X, Z being (0, +1, −), (−1, 0, −),
(2, 2, +), respectively.
Now the glide planes of −a1 from a parallel series, say
π−1 π0 π1 π2
... ... ... ...

and the inner automorphisms can move the W and X planes (say π0 and π1 ) any
even number of steps, while the further automorphism that takes (W, X, Y ) to
(W Z, XZ, Z −1 ) moves them just one step.
26 J. H. CONWAY AND J. P. ROSSETTI

Modulo these we may suppose π0 and π1 are fixed or interchanged. The most
general automorphism that does this is
W → W aX b , X → W cX d , Z → Zǫ (ǫ = ±1),
where
   
a b 1 0
(4) if ǫ = +1, ≡ (mod 2), while
c d 0 1
   
a b 0 1
(5) if ǫ = −1, ≡ (mod 2).
c d 1 0
The group defined by (4) is often called Γ(2). This explains the entry in Table 7.

The Didicosm. To fix our ideas we take A = B = C = 1. Then the generators

X, Y are two screw motions of the minimal length 1 at the minimal distance 1/2.
The same must be true of their images, since if either length or distance were to be
increased the resulting elements would no longer generate. Now X, Y determine
Z via XY Z = 1. The lines of these three generators are edges of a unique 12 × 21 × 21
cube (or ‘cubelet’) that constitutes one eighth of our defining ‘box’. The general
automorphism will take it to a similar cubelet, constituting an eighth of some
other box of the tessellation.
We first show that there are enough automorphisms to take this standard
cubelet to any of the eight cubelets in any box. It may be taken to the correspond-
ing one in any other box by a translation through a typical vector αx + βy + γz
of the naming lattice N = hx, y, zi, which achieves
(X, Y, Z) 7−→ (XY 2β Z 2γ , Y X 2α Z 2γ , ZX 2α Y 2β ).
(This is inner just if α, β, γ are even, since then the associated translation belongs
to T = h2x, 2y, 2zi.)
The automorphisms we have just found effectively allow us to suppose that
there is only one box, which contains just eight cubelets. The inner automor-
phisms by X, Y, Z take the standard cubelet to four of these. (We have now found
all inner automorphisms, since these elements represent the four cosets of T in Γ.)
So we need only find a further automorphism that takes the standard cubelet
to one of the missing four. This is
(X, Y, Z) 7−→ (Y Z −1 , ZX −1 , XY −1 ).
(the latter three elements are half turn screw motions about lines bisecting the
other three faces of the standard box.)
All the above automorphisms fix the three coordinate directions, so will be
isometric even in the general case (A, B, C distinct). Modulo them, we can
suppose X, Y, Z go to some permutation of X ±1 , Y ±1 , Z ±1 , since these are the
only length 1 screw motions associated with the initial cubelet. The condition
XY Z = 1 restricts us to the cyclic permutations of X, Y, Z or Z −1 , Y −1 , X −1 .
The identity is the only one of these that is isometric in the general case.
 DESCRIBING THE PLATYCOSMS 27

I II III IV V
β1 inner automorphisms outer automorphisms # of
rigidly isometric deformable Bi ’s
c1 3 trivial X −1 , Y −1 , Z −1 2 PGL3 (Z) on X, Y, Z 14
X, Y, XZ
X , Y ǫ, X aY b Z
ǫ
c2 1 X, Y, Y Z 23 PGL2 (Z) on X, Y 5
: a, b ∈ 2Z
X, Y, Z −1

X, Y, X a Y b Z
X, Y, XZ
Y, W, X a Y b Z
c3 1 X −1 , Y −1 , Z 2 × S3 1 1
W, X, X a Y b Z
Y, X, Z −1
: a + b ∈ 3Z
X, Y, X a Y b Z
Y, X −1 , X a Y b Z
X, Y, XZ
c4 1 X −1 , Y −1 , X a Y b Z 22 1 1
Y, X, Z −1
Y −1 , X, X a Y b Z
: a + b ∈ 2Z
X σ , Y σ , X aY bZ
Y σ , W σ , X aY bZ
c6 1 Y, X, Z −1 2 1 1
W σ , X σ , X aY bZ
: σ = ±1, a + b ∈ Z
X, Y X 2 , ZX 2
T b c T a c T a b XY 2 , Y, ZY 2
X Y Z ,Y X Z ,Z X Y
c22 0 24 S3 3
: a, b, c ∈ 4Z, T = 1, X, Y, Z XZ 2 , Y Z 2 , Z
Y Z −1 , ZX −1 , XY −1
W Z m , XZ m , Z σ Γ(2) on W, X
+a1 2 W Z, XZ, Z −1 2 5
: σ = ±1, m ∈ 2Z X, W, Z −1
W Z m , XZ n , Z σ Γ(2) on W, X
−a1 2 W Z, XZ, Z −1 2 5
: σ = ±1, m − n = σ−1 2 X, W, Z −1
W Z, XZ, Z
W X 2a Z 2b , X δ Z 2b , Z ǫ
+a2 1 W −1 , X, Z 23 1 1
: δ, ǫ = ±1, 2a ≡ δ − ǫ mod 4
W, X −1 , Z
W X aZ b, X δ Z c, Z ǫ W Z, XZ, Z
−a2 1 : δ, ǫ = ±1, 2c + δ ≡ 1 mod 4 W −1 , X, Z 23 1 1
a ≡ δ − ǫ mod 4, 2b + 1 = 2c + ǫ W, X −1 , Z

Table 7. Columns I–IV concern the appropriate parts. Thus I

gives the dimension of the connected component, II the generic
inner automorphisms, III generators and structure (modulo inner
automorphisms) of the rigidly isometric part, IV the deformable
part. Column V gives the number of distinct types of Bravais
subgroups Bi (see Section 11). Automorphisms are specified by
giving images of X, Y, Z for the chiral platycosms, W, X, Z for the
achiral ones.

The conclusion is that the general inner automorphism takes (X, Y, Z) to

(X T Y 2β Z 2γ , Y T X 2α Z 2γ , Z T X 2α Y 2β ), (α, β, γ even , T = 1, X, Y, Z);
28 J. H. CONWAY AND J. P. ROSSETTI

while the full automorphism group is obtained by letting α, β, γ be arbitrary

integers, and adjoining the maps taking (X, Y, Z) to any even permutation of
(X, Y, Z) or (Y Z −1 , ZX −1 , XY −1 ) or any odd permutation of the inverses of
those. Note that the argument proves c22A B C to be invariant under all permu-
tations of A, B, C (not just the more obvious even ones).
Let θ1 , θ2 , θ3 , φ be the automorphisms that take (X, Y, Z) to
(X, Y X 2 , ZX 2 ), (XY 2 , Y, ZY 2 ), (XZ 2 , Y Z 2 , Z), (Y Z −1 , ZX −1 , XY −1 ).
Then it is easily checked that modulo inner automorphisms, θ1 , θ2 , θ3 , φ are
multiplicatively commuting elements of order two. The elements π that take
(X, Y, Z) to even permutations of (X, Y, Z) or odd permutations of (X −1 , Y −1 , Z −1 )
form an S3 that acts in the obvious way on θ1 , θ2 , θ3 and takes φ to itself or
θ1 θ2 θ3 φ, according as π is even or odd. This shows that the outer automorphism
group is a split extension 24 :S3 . The structure of this group was found (in a
different form) by Zimmermann [Zi], who disproved an earlier assertion ([ChV],
[Ch]) that the group was a split extension of form 23 :(2 × S3 ).
The outer automorphism groups of flat manifolds have often been studied
(e.g., [Ch], [MS], [Hi]), the last of which used them to enumerate compact flat
4-manifolds. Our treatment avoids uses of subtle theorems, and clearly distin-
guishes the isometric part.

8. Double Covers
A double cover of a manifold M determines and is determined by a homomor-
phism from its fundamental group π1 (M ) onto {±1}, the image of a path being
−1 just if it interchanges the two sheets. To enumerate such homomorphisms,
we can obviously abelianize π1 (M ) (to H1 (M )) and then make H1 (M ) have ex-
ponent 2, turning it into an elementary abelian 2-group. If there are r generators
after this, the number of double covers will be 2r − 1. We discuss the cases.

Torocosm. Since H1 (M ) = ∞3 , a torocosm has 7 double covers, themselves all

torocosms.

Dicosm. Now H1 (M ) = 22 · ∞, so the dicosm also has 7 double covers, of which

just one is a torocosm. Each of the other 6 is a dicosm, since some period 2 screw
motion survives in the kernel of its defining homomorphism.

Tricosm. Since W, X, Y are conjugate in π1 (M ), they must map to the same

sign, which must be +, since W XY = 1. There is therefore just one double cover
(given by W 7→ +, X 7→ +, Y 7→ +, Z 7→ −) which is another tricosm c3.

Hexacosm. A similar argument shows that the hexacosm c6 also has a single
double cover, a tricosm c3.

Tetracosm. X and Y are again conjugate, but now can both map to + or both
to −. There are therefore three double covers: (X, Y, Z) 7→ + + − (type c2) or
− − + or − − − (both of type c4).
 DESCRIBING THE PLATYCOSMS 29

Didicosm. The relation XY Z = 1 shows that two of X, Y, Z must map to −

and the third to +. So we have three double covers, all dicosms c2.

Amphicosms and Amphidicosms. The abelianized groups show that the pos-
itive cases each have seven double covers and the negative ones three. The types
are
for +a1: 1 of type c1, 4 of type +a1, 2 of type −a1;
for −a1: 1 of type c1, 2 of type +a1;
for +a2: 1 of type c2, 2 of type +a1, 2 of type +a2, 2 of type −a2;
for −a2: 1 of type c2, 2 of type +a1.

platycosm generators double covering # total #

WXY Z
c1 c1 7 7
c2D
ABC (+ + +) − c14D
ABC 1
(+ − −) ± D
c2[A] B C , etc. 2 of each 7
c3D
AAA (+ + +) − c34D
AAA 1 1
c4D
AA (++) − c24D
AA 1
(−−) ± c4D2A 2A 1 of each 3
c6D
AAA (+ + +) − c34D
AAA 1 1
c22A B C (+ − −) c2A4B 4C , etc. 1 of each 3

X Y Z +a1D A:B C −a1D A:B C +a2D

A:B −a2DA:B
− + + c1D
[A] B C c1 see (6) +a14A
B:D +a14A
[0]:B D
+ − + +a1D
[B]:A C +a14D
[B]:A C +a1D
A:4B +a14D
A:4B
− − + +a1D 4D c2B c2B
[C]:A B +a1[C]:A B 4A D 4A 4D
± + − +a14D
A:B C — +a24D
A:B —
± − − −a1D
A:B C — −a2D
A:B —
total # 7 3 7 3

Table 8. Parameters for the double covers.

Parameters. Tables 8 give the types and parameters of the double covers cor-
responding to all homomorphisms : π1 (M ) → {±1}. The parameters for the
orientable double covers (c1) of −a1 take one of the four forms
c12D 2D
2C 2C+4A B−C−D c12D 2D
2B 2B+4A C−B−D c12B 2B 4A
2C 2C D−B−C
(if B ≥ C + D) (if C ≥ B + D) (if D ≥ B + C)
(6)
c1C+D−B B+D−C B+C−D
C+D−B B+D−C B+C−D+4A
(otherwise)
30 J. H. CONWAY AND J. P. ROSSETTI

and in several other cases, there are two forms, which we abbreviate using the
notation

w − v : 2v 2v + 4u if v ≤ w;
(7) [u]:v w =
v − w : 2w 2w + 4u if v ≥ w,
and the similar notation without a colon.

9. Diameters and Injectivity Radii

The covering radius of a lattice T is the minimal radius for which the closed
balls of that radius centered at points of T cover the space. It coincides with
the diameter of the torus Rn /T , where the diameter of a Riemannian manifold
is defined in general to be the maximal distance between any pair of points in it.
The packing radius of T is the maximal radius for which the open balls centered
at points of T are disjoint. In the torus Rn /T , it becomes the injectivity radius,
which is defined for every Riemannian manifold to be the maximal radius for
which balls of that radius centered at any point of the manifold will embed in
the manifold. These are easier to determine:

Injectivity Radii.
Proposition. The injectivity radius of a platycosm is bounded below by the pack-
ing radius of its naming lattice N . Moreover, these two concepts coincide for 8
of the 10 platycosms.
Proof. The injectivity radius equals one half of the length of the shortest closed ge-
odesic in the manifold. Every closed geodesic corresponds to a translation, screw
motion or glide reflection and its length is exactly the length of the translation
vector, screw vector or glide vector, respectively, and all these are by definition
in the naming lattice N . This proves the first assertion.
On the other hand, one can check that the shortest possible vectors of N do
correspond to such ‘geodesic’ vectors in 8 out of the 10 cases, the exceptions being
the two ‘negatives’ −a1 and −a2. 
We now discuss these two exceptions.

Injectivity radius of −a1. Recall that N = hw, x, zi, whose conorms and vonorms
we display:

D B +C A+C A+B

conorms: vonorms: A+C +D

0 0 A+B +D
0 B +C +D

A B C D

Figure 27.
 DESCRIBING THE PLATYCOSMS 31

We now classify the geodesic vectors:

kind of vector vectors norm of minimal vector

glide vector odd · w + even · x A + B = N (w)
even · w + odd · x A + C = N (x)
translation vec. odd · w + odd · x + odd · z B + C + D = N (w + x + z)
even · w + even · x + even · z 4 · min(all vonorms)
seeing that they lie in four cosets of N in 2N . But by vonorm theory (see
Appendix II), the minimal norms in these cosets are the numbers in the right
column.
So the squared injectivity radius is the minimum of the ten numbers

A + B, A + C, A + B + D, 4D, 4(B + C)
4(A + B), 4(A + C), 4(A + B + D), 4(A + C + D), 4(B + C + D),

of which those in the second line are dominated. Any of the five numbers in the
first line can be the minimum!

Injectivity radius of −a2. The vectors corresponding to a letter ‘d’ in the figure
are of the form odd · x + even · w. In this way we obtain

letter vector minimal vector norm

d odd · x + even · y x A
q odd · y y B
p odd · y + odd · z y+z B+D
b even · x + even · y + even · z (in 2N ) 4 · min(vonorms)
Hence the answer is min(A, B, 4D), since the other numbers are dominated by
A or B.
For c1 the answer is the minimal vonorm — see the end of this section.
For cNADB C (N > 1), it is generically min(B + C, C + A, A + B, D). The
particular cases are listed in Table 9.

Diameters. We illustrate our method for finding diameters by considering that

of the Klein bottle. The solid rectangles in Figure 28 are the Voronoi cells (see
Section 11) of the images of p0 under Γ. This is an orbit which is also a lattice,
so that the rectangles form the Voronoi tiling of an orbit lattice.
Moreover, this tiling is translatable in the sense that it remains a tiling if we move
its tiles (say as indicated by the dashed lines) so that their centers form the orbit
of an arbitrary other point p1 (although the tiles might no longer be Voronoi
cells).
It follows that the diameter of the Klein bottle is exactly the circumradius
of these rectangles, since the corners of the original rectangles are exactly this
distance from p0 , and clearly no point can be further than this distance from the
generic point p1 . The argument proves in general
32 J. H. CONWAY AND J. P. ROSSETTI

p1
p0

Figure 28. Translating the Voronoi cells.

Proposition. The diameter of a platycosm P is bounded below by the covering

radius of any orbit lattice. If P has an orbit lattice whose Voronoi tiling is
translatable then its diameter equals the covering radius of that orbit lattice. 

The orbit lattice bound R0 for the diameter of a platycosm P is the maximum
of the covering radii of all orbit lattices for P , and we say P has the orbit lattice
property if its diameter equals this lower bound. We believe

(The orbit lattice conjecture for platycosms). All 10 platycosms have the
orbit lattice property.

The proposition establishes this for 7 of the 10 since they have orbit lattices
whose Voronoi tilings are translatable. The authors have worked out the orbit
lattice bound for the three exceptions, and now summarize the results.

Didicosm. The didicosm c22A B C has 8 orbit lattices, of just 4 distinct shapes,
for which the values of 4R2 are:
(B + C)2 (C + A)2 (A + B)2
α =A+ , β=B+ , γ=C+
max(B, C) max(C, A) max(A, B)

and
 
2 1 1 1 2
δ = A + B + C + m · max + + − , 0 , where m = min(A, B, C).
A B C m

Hence the orbit lattice bound is given by 4R02 = max(α, β, γ), since it can be
shown that δ is dominated by any of α, β, γ.

Second amphidicosm. The second amphidicosm −a2C A:B has infinitely many orbit
lattices, of just three different shapes, for which the values of 4R2 are the above
β, γ, δ. The orbit lattice bound is therefore given by 4R02 = max(β, γ).
 DESCRIBING THE PLATYCOSMS 33

platycosm squared injectivity radius squared diameter

c1D EF
A B C the minimal vonorm — see below see Section 10
(B+C)(C+A)(A+B) D
c2D
ABC min(B + C, C + A, A + B, D) 4(BC+CA+AB) + 4
2 D
c3D
A A A min(2A, D) 3A + 4
1 D
c4D
A A min(A, D) 2A + 4
2 D
c6D
A A A min(2A, D) 3A + 4
c22A B C min(A, B, C) ≥ 14 max(α, β, γ)
(B+C)(C+A)(A+B) D
+a1D
A:B C min(A + B, B + C, A + C, D) 4(BC+CA+AB) + 4
−a1D
A:B C min(A + B, A + C, B + C + D, 4D, 4(B + C)) see text
1
+a2C
A:B min(A, B, C) 4 (A + B + C)
−a2C
A:B min(A, B, 4C) ≥ 41 max(β, γ)

Table 9. Injectivity radii and diameters. Here

(B+C)2 (C+A)2 (A+B)2
α = A + max(B,C) , β = B + max(C,A) , γ = C + max(A,B) .

Second amphicosm. The hardest case is −a1D A:B C , which has just 2 distinct shapes
of orbit lattices. For the first of these,


 I if C ≥ D,
max(II, V ) if C ≤ D ≤ A + C, B + C,

2
(8) 4R =

 max(III, V) if B + C ≤ D, B ≤ A,
max(IV, V ) if B + C ≤ D, B ≥ A,


where
I = A + B + 4D + (A + B)(D − C)2 /Ω
II = A + B + 4C + (A + B + 4C)(D − C)2 /Ω
III = 2B + 2C + D + (B + C)2 /D + (B + C)(B − A)2 /Ω
IV = 2A + 2C + D + (A + C)2 /D + (A + C)(B − A)2 /Ω
V = A + B + 2C + C 2 /D
where Ω = BC + CA + AB. Equivalently, it can be shown to be the minimum
of the four expressions in (8), which may be rewritten as
 
max min(I, II, III, IV ), min(I, V ) .

For the second orbit lattice, 4R2 is given by the similar expression
 
max min(I ′ , II ′ , III ′ , IV ′ ), min(I ′ , V ′ ) .
34 J. H. CONWAY AND J. P. ROSSETTI

found by interchanging B and C.

Therefore the orbit lattice bound is given by
 
4R02 = max min(I, II, III, IV ), min(I, V ), min(I ′ , II ′ , III ′ , IV ′ ), min(I ′ , V ′ ) .

10. Some Formulae for Torocosms and Lattices

Many functions of a 3-dimensional lattice (or its associated torocosm) can be
expressed as symmetrical functions of all 7 conorms. For instance, the squared
packing or injectivity radius is the minimal vonorm, and of course the typical
conorm is just the sum of conorms not on a given line, so the lattice ΛD E F G has
ABC
the conorms and vonorms of Figure 29. The G conorm is not necessarily the zero

G B +E +C +F C +F +A+D A+D+B +E

A+E +C +G
conorms: D F has vonorms: A+B +F +G
E D+B +C +G

A B C D+E +F +G

Figure 29.

one. If it is, the squared injectivity radius (of c1D EF

A B C ) simplifies to


min B+E+C+F, C+F +A+D, A+D+B+E, D+B+C, A+E+C, A+B+F, D+E+F .
The determinant of this lattice (which is the squared volume of the corre-
sponding torocosm) is the sum of the conorm products over all 28 triangles in the
conorm plane. When G is zero, 12 of those products vanish, and the determinant
simplifies to
∆ = (A+D)(BE+CF )+(B+E)(CF +AD)+(C+F )(AD+BE)+DBC+AEC+ABF +DEF.
The actual length of the edge in Figure 30 marked with a given conorm x is
r
x ∂∆
(9)
∆ ∂x
where ∂∆
∂x denotes the formal derivative of the above expression for ∆ with respect
to the variable x. It is easy to check that (9) can vanish only when x does.
Another useful expression is ∆′ , the sum of conorm products over the comple-
ments of all triangles; when G = 0 this simplifies to
∆′ = AD(B + E)(C + F ) + BE(C + F )(A + D) + CF (A + D)(B + E),
since now 16 of the products vanish. It can be shown that the covering radius is
given (when G = 0) by
∆′ + 4 · min(BECF, CF AD, ADBE)
(10) 4R2 = A + B + C + D + E + F −
∆
 DESCRIBING THE PLATYCOSMS 35

Finally, we remark that the lattice Λda eb cf g dual to ΛD EF

A B C has conorms given by
a∆ = AD − m, d∆ = AE + AF + EF + m,
b∆ = BE − m, e∆ = BD + BF + DF + m,
c∆ = CF − m, f ∆ = CD + CE + DE + m,
g∆ = AB + AC + BC + m,
where m = min(AD, BE, CF ).

11. The Bravais-Voronoi Classes

Voronoi classification of lattices. The points of space that are closer to one
particular lattice point than to any other constitute the Voronoi cell of that point.
In [Co2] conorms were used to give a simple derivation of Voronoi’s classification
of lattices by the topological type of their Voronoi cells.

β
c a 0
β c
a
α γ
α γ a c
truncated a c b
Octahedron b b
γ α α γ
γ β α β
b β b
a c

β=0
0
c a
hexarhombic a b c
Dodecahedron α γ

b α 0 γ

b=0 α=0 hexagonal

rhombic c a Prism 0
Dodecahedron c a
0 α γ γ a c
b

a c
α=0 b=0 b
0 0 0 γ
c a
α 0 γ 0
γ
rectangular
Cuboid a 0 c

0 0 γ

Figure 30. The shape of the Voronoi cell.

36 J. H. CONWAY AND J. P. ROSSETTI

The generic Voronoi cell is a truncated octahedron as in Figure 30, adapted

from [Co2]. The six families of parallel edges correspond to its six non-zero
conorms. When any conorm vanishes, the corresponding edges shrink to points,
changing the Voronoi cell into one of the four other shapes: a hexarhombic do-
decahedron, a rhombic dodecahedron, a hexagonal prism, or a rectangular cuboid
(see Figure 30).
Each of these figures has antipodal symmetry. Any other topological symme-
try will become a metrical one provided certain conorms are equal. Since the
possible topological symmetries depend on which conorms vanish, we have found
it convenient to adjust the lettering to display this more clearly. The cases are:

lattice (abbreviating) effect of symmetries on the conorms

Make two of the interchanges A ↔ D, B ↔ E, C ↔ F .
ΛD EF
ABC Bodily permute the three columns.
BO Make either of the interchanges A ↔ B, C ↔ D.
ΛA B
C D:E ΛA
CDE Bodily permute the these two couples.
ΛA B ΛA BO Permute A, B, C, D in any way.
CD CDO

ΛD ΛD OO Permute A, B, C in any way.

AB C AB C

ΛA B C ΛA BC
OOO Permute A, B, C in any way.

Deducing the Bravai classification. The arrows in the following figures show
how certain equalities between conorms permit additional metrical symmetries
(the arrows being downward or rightward according as the symmetries move
numbers within their columns or across columns). They show that 3-dimensional
lattices belong to just 24 ‘BraVo’ (Bravais-Voronoi) classes, where we say that two
lattices are in the same BraVo class just if each can be continuously deformed into
the other while keeping all its symmetries and without changing the topological
shape of its Voronoi cell.
Below each conorm array is a naming letter A,B,. . . ,X, and the orbifold nota-
tion for the corresponding point group.
Truncated Octahedron (tO).

a b c a a b a a a
−→ −→
α β γ α α β α α α
A  × D  2∗ G 2∗3
y y |
|
a b c a a b
−→ ——————–|
a b γ a a β |
B  2∗ E  ∗222
|
y y y

a b c a a b a a a
−→ −→
a b c a a b a a a
C ∗222 F ∗422 H ∗432
 DESCRIBING THE PLATYCOSMS 37

Hexarhombic Dodecahedron (hD).

a b a a a a
−→ −→
α β :γ α β :γ α α :γ
I  × K 2∗ L  ∗222
y | y
|
a b | a a
———————————−→
——
a b :γ a a :γ
J 2∗ M ∗422
Rhombic Dodecahedron (rD).

a b a a a a
−→ −→
α β α β α α
N × O  2∗ Q  ∗222
y y
a a a a
−→
a α a a
P 2∗3 R ∗432
Hexagonal Prism (hP).

d d d
−→ −→
α β γ α α β α α α
S 2∗ T ∗222 U ∗622
Rectangular Cuboid (rC).

a b c a a b a a a
−→ −→
V ∗222 W ∗422 X ∗432
The easiest proof that there are just 14 Bravais classes is to see how certain
continuous variations link the 24 BraVo classes. For example, take the lattice
Λaα aαaα (case G) and vary α until it becomes a small negative number −ǫ, forcing
us to renormalize the conorms as in Figure 31.
Then the new lattice has shape Λaa ab (case P), showing that cases G and P are
in the same Bravais class. To obtain the exact partition into Bravais classes one
merely has to vary the parameters like this in all possible ways for which the new
lattice has no more symmetries than the old one13. Similarly, making b negative
in case F leads to case M. All other identifications can be achieved just by making
some conorms vanish, as in Figure 32.
This gives Table 10, whose last column gives the “symmetry factors” by which
the sizes of the isometry groups have been increased.
To prove this list is complete, it suffices to show that any other cases with
the same group are in distinct Bravais classes, which follows from the fact that
each of E,H,R,S,T,U,V,W,X is in a Bravais class of its own. Why is this? For
13although at isolated intermediate times the symmetry might have been larger, as it was in
this case when α = 0
38 J. H. CONWAY AND J. P. ROSSETTI

0 −ǫ ǫ

a−ǫ
a
a
a a− 3ǫ a−3ǫ a−3ǫ a− 3ǫ a−3ǫ

−ǫ −ǫ −ǫ 2ǫ 0 0 0 0 0
ǫ 0

a+ǫ a−ǫ a−2ǫ a− 2ǫ a−4ǫ

a− ǫ

0 −2ǫ −2ǫ ǫ ǫ −ǫ

Figure 31.

A
K B D C L c=0
a=0 c=0
γ=0 or or β=0 c=0 γ=0
b=0 γ=0 I
γ=0

O J Q
N

Figure 32.

H,R,V,W,X no parameter can pass through 0 without the lattice degenerating.

The same is true for the parameter  a in E and d in 
S,T,U. For E this leaves

a a ǫ a−ǫ a−ǫ ǫ
essentially only the normalization of to ,
a a β a−ǫ a−ǫ β+ǫ
which is again in case E. The argument for S,T,U is even easier, and completes
the proof of the Bravais classification.

Bravais types for the other platycosms. For the remaining helicosms cNADB C
the Bravais classification reduces to that of the 2-dimensional lattice ΛA B C , since
D can be varied independently of A, B, C. Thus c2 has five Bravais types:
c2D D D
A B C −→ c2A A B −→ c2A A A , c2D D
A B −→ c2A A

where c2D D
A B means c2A B 0 , and c3, c4, c6 just one each

c3D D D
A A A , c4A A , c6A A A .

For the didicosm c22, the classification is just that of the rectangular 3-dimensional
lattice ΛA B C :
c22A B C −→ c22A A B −→ c22A A A .
 DESCRIBING THE PLATYCOSMS 39

Voronoi classes Bravais classes

tO hD rD hP rC crystallographic name symmetry factor
A I N Triclinic 1
BD JK O base-centered Monoclinic 2
C L Q body-centered Orthorhombic 4
E face-centered Orthorhombic 4
F M body-centered Tetragonal 8
G P Rhombohedral (or Trigonal) 6
H body-centered Cubic (bcc) 24
R face-centered Cubic (fcc) 24
S primitive Monocline 2
T base-centered Orthorhombic 4
U Hexagonal 12
V Orthorhombic 4
W primitive Tetragonal 8
X primitive Cubic 24
Table 10. How the 24 BraVo classes correspond to the 5 Voronoi
and 14 Bravais classes.

For the two amphicosms ±a1D A:B C , the Bravais classification reduces to that of
the 2-dimensional lattice ΛA B C , but taking into account the distinguished rôle
of A:
+a1D D
A:B C −→ +a1A:B B , +a1D D
:B C −→ +a1:B B , +a1DA:B ,

−a1D D
A:B C −→ −a1A:B B , −a1D D
:B C −→ −a1:B B , −a1D
A:B ,
where, of course, ±a1D D D D
A:B and ±a1:B C abbreviate ±a1A:B 0 and ±a10:B C respec-
tively.
Finally, each of the amphidicosms ±a2DA:B forms a single Bravais type since all
three parameters are distinguishable. We conclude:
The numbers of Bravais types of platycosms are
14 5 1 1 1 3 5 5 1 1 (total 37)
for c1 c2 c3 c4 c6 c22 +a1 −a1 +a2 −a2

Appendix I: Why there are just 10 platycosms

Why are there exactly 10 compact platycosms? We sketch a proof of this to
make our paper complete. The proof is outlined in Figure 33 and the accompa-
nying explanations (i)–(vi). We quote Bieberbach’s theorem that the subgroup
of translations has finite index.
(i) When there are screw motions there is a
Splitting Lemma. If v is the smallest screw vector in a given direction, say
vertical, then the translation lattice T decomposes as hN vi ⊕ T ′ , where N is the
period of the corresponding screw motion σ.
40 J. H. CONWAY AND J. P. ROSSETTI

Proof. It suffices to prove that there is no translation τ whose vertical part cv is

strictly between 0 and N v. But then στ −1 if 0 < c < 1, or τ σ −m if m < c <
m + 1 (m = 1, . . . , N − 1) would be a vertical screw motion with shorter vector
than v. 
(ii) If all screw vectors are parallel we have a cyclic point group whose order N
is the least common multiple of their periods and which acts on the 2-dimensional
lattice T ′ . Obviously Γ is generated by the shortest screw motion of this period
together with the translations of T ′ . The identification with c2, c3, c4 or c6 follows
from the well known lemma:
Barlow’s Lemma. If a rotation of order N fixes a 2-dimensional lattice T ′ ,
then N = 1, 2, 3, 4, 6.
Proof. Apply the rotation to a minimal non-zero vector of the lattice. We obtain
a ‘star’ of N vectors, the difference of adjacent members of which will be a shorter

vector if N ≥ 7. A rotation of order 5 would combine with negation to produce

one of order 10. 
(iii) Two non parallel screw motions σ1 and σ2 would make T decompose in
two ways, say hv1 i ⊕ T1 ∼ = hv2 i ⊕ T2 , showing immediately that v1 and v2 are
orthogonal. If σi had a period other than 2, then σj and σjσi would not be
orthogonal. Finally, if σ1 and σ2 are orthogonal and of period 2, then σ1 σ2 is a
screw motion in the direction of v3 orthogonal to v1 and v2 . These together with
the translations must generate Γ, and can be identified with the X, Y, Z of c22.
(iv) Any glide reflection maps to a reflection through the origin when ignore
translations. The product of two of them is therefore a translation or a screw
motion since the product of two reflections through o is the identity or a non-
trivial rotation according as their planes are parallel or not.
(v) If all glide reflections are in parallel planes — call them basal — the point
group has order 2, so Γ is generated by a single glide reflection, together with its
translations. We resolve any translation vector into basal and perpendal parts
v1 + v2 , and see that since τ takes this to v1 − v2 , 2v2 is also a translation
vector. If v2 itself (and so v1 ) is a translation vector, then the translation lattice
T decomposes into its basal and perpendal parts, leading to +a1. Otherwise,
adjoining the translation through v2 embeds our platycosm in a copy of +a1,
and easily identifies the original platycosm with −a1.
 DESCRIBING THE PLATYCOSMS 41

Start Here

no Are there no Are there

c1 screw- glide-
motions? reflections?

yes yes
(i) (iv)
c2
then the lattice then their product
decomposes! is a translation or
a screw motion.
c3

c4 yes are they

(ii)
are all their
(v)
yes
+a1
all glide mirrors
parallel? parallel? −a1
c6 no no
(iii) (vi)
then they are then they are in vertical +a2
c22 orthogonal and planes, the glide vector
of one is horizontal,
of period 2. and their glide mirrors
are orthogonal. −a2

Figure 33. Guide to the proof.

(vi) If two glide reflections g, g′ have non-parallel planes intersecting in a line

we will call vertical, then their product will be a screw motion s whose screw
vector v will be vertical. We choose things so that v is as short as possible. Then
we can multiply by a power of s to reduce g until the vertical component of its
glide vector λv has |λ| ≤ 21 . But then λ must be 0 since g2 is a translation and
by the Splitting Lemma its vertical component must be a multiple of N v, where
N ≥ 2 is the period of s.
The glide vector of the new g′ (defined by gg′ = s) will have the same vertical
component v as s. But now the translation vector (g′ )2 has vertical component
2v, showing (by the Splitting Lemma) that s must have period 2, from which it
follows that the planes of g and g′ are orthogonal.
We can now see that the translation lattice is generated by 3 orthogonal trans-
lations, say (x, y, z) 7→ (x + a, y, z), (x, y + b, z), (x, y, z + c). We can also see
that the point group has order 4, i.e. that g, g′ , together with translations, must
generate Γ. For if not, there would be a glide mirror not parallel to either of
those of g, g′ , and so perpendicular to both, by the previous paragraph. But then
42 J. H. CONWAY AND J. P. ROSSETTI

the product of the corresponding glide reflection with s would have the form

(x, y, z) 7→ (constant − x, constant − y, constant − z),

which fixes a point. In these coordinates

1
s : (x, y, z) → (−x, −y, z + c)
2
1
g : (x, y, z) → (x + a, λb − y, z)
2
where we may take 0 ≤ λ < 1 by compounding g with translations. From these
we find
1 1
g′ = g−1 s : (x, y, z) → ( a − x, y − λb, z + c)
2 2
′ 2
(g ) : (x, y, z) → (x, y − 2λb, z + c)

showing that 2λ must be an integer, so either λ = 0, which gives +a2; or λ = 21 ,

which gives −a2. In the former case, our screw vector (0, 0, 12 c) is also a glide
vector (of g′ ) — in the latter case, no screw vector is a glide vector.

Appendix II: Conorms of Lattices

Introduction. The conorms of a lattice L are certain numbers that are deter-
mined by L (up to equivalence) and return the compliment by determining L,
at least in low dimensions. More precisely L has a conorm function defined
on conorm space, which is a finite set that has the structure of a projective
(n − 1)-dimensional space over the field of order two. At least for n ≤ 4, two n-
dimensional lattices L and L′ are isometric if and only if there is an isomorphism
between their conorm spaces that takes one conorm function to the other.
The theory in low dimensions is greatly simplified by the observation that for
n ≤ 3 every lattice has an obtuse superbase, which implies in particular that the
conorms are ≥ 0. A superbase for an n-dimensional lattice L is an (n + 1)-tuple
{v0 , v1 , . . . , vn } of vectors that generate L and sum to zero. It is obtuse if all inner
products vi · vj of distinct vectors are non-positive (and strictly obtuse if they are
strictly negative). For a lattice with an obtuse superbase, it can be shown that
the conorms are the negatives of the inner products of pairs of distinct superbase
vectors, supplemented by zeros. (If n ≥ 4 other things can happen; a lattice may
not have an obtuse superbase, and some conorms may be strictly negative.)
For n = 0, conorm space is empty, so there are no conorms.
For n = 1, conorm space is a single point, and there is one strictly positive
A
conorm A. We represent this by the picture •; it means that
 the lattice has an
A −A
obtuse suberbase {v, −v} with Gram-matrix ; equivalently a base
−A A
{v} with Gram-matrix [A].
 DESCRIBING THE PLATYCOSMS 43

For n = 2, conorm space is a 3-point projective line, and the 3 conorms A, B, C

A B C
are non-negative. We represent this by the picture •−− −•−−−• ; it means that the lat-

B+C −C −B
tice has an obtuse superbase {v0 , v1 , v2 } with matrix  −C C +A −A ,
−B −A A+B
 
B+C −C
or equivalently a base {v0 , v1 } with matrix .
−C C+A
For n = 3, conorm space is the 7-point “Fano plane”, represented by Figure 34.
The superbase has Gram-matrix
 
p0|123 −p12 −p13 −p01
 −p12 p1|023 −p23 −p02 
 −p13 −p23 p2|013 −p03  ,
 

−p01 −p02 −p03 p3|012

where pij = pji ≥ 0, pi|jkl := pij + pik + pil , and the minimal conorm is 0.

p02

p12 p23
0

p01 p13 p03

Figure 34. A Fano plane with p01 , p02 , p03 , p12 , p13 , p23 , 0 in the
usual arrangement.

The sum of the four conorms not on a given line in Fig. 34 is a vonorm, which
can be defined either as the norm of a Voronoi vector, or as the minimal norm of
any vector in a non-trivial coset of L in 2L.

Putative conorms and the reduction algorithm. A 3-dimensional lattice

has many systems of “putative conorms” in addition to its unique system of actual
conorms, for instance the numbers obtained by arranging 0 and the negatives
pij := −vi · vj of the inner products of distinct members of any superbase on
a Fano plane in the manner of Figure 34. If the putative conorms are all non-
negative, they will be the actual conorms. Otherwise, the following algorithm
quoted from [Co2] and [CoS] will produce the latter.
Select a ‘working line’ that contains both a 0 conorm and a negative one, say
−ǫ. Then we transform to an improved system of putative conorms by adding ǫ
to the 3 conorms on the working line, and subtracting ǫ from the 4 conorms off
this line. If the improved system still has a negative conorm, we can define a new
working line and repeat the procedure. A finite number of repetitions will suffice
to produce the actual conorms.
44 J. H. CONWAY AND J. P. ROSSETTI

As an example Figure 35 finds the conorms  for the 

lattice whose Gram-matrix
2 1 1
with respect to a suitable base v1 , v2 , v3 is  1 3 1  The Gram-matrix for the
1 1 4
 
2 1 1 −4
 1 3 1 −5 
superbase v1 , v2 , v3 , −v1 − v2 − v3 is   (found by making
 1 1 4 −6 
−4 −5 −6 15
each row and column sum to 0), which leads to the putative conorms of Figure
35(a). Transforming this using the working line indicated on it, we obtain Figure

5 4 2
(a) (c) (e)

−1 −1 0 0 0 0
0 −1 1

4 −1 6 1 2 3 1 0 3

6 3

−2 −2 −1 1
1 0

(b) (d)

3 0 5 2 1 2

Figure 35.

35(b), and we proceed from this in a similar way to Figures 35(c), 35(d), 35(e),
the last of which gives the actual conorms. The reader might like to verify that
another choice of working line yields the same conorm function.
For a 2-dimensional lattice with putative conorms A, B, C there is a similar
algorithm. We select a ‘working point’ at which there is a negative conorm, say
−ǫ, and transform by adding 2ǫ to this conorm and subtracting 2ǫ from the other
−3 5 10 3 −1 4 1 1 2
two conorms. Thus •−−−•−−−• transforms through •−−−•−−−• to •−−−•−−−• .

Appendix III: Dictionary of Names and Notations

The purpose of this appendix is to give to readers who have met with a platy-
cosm in some other notation a way to recognize it quickly. Some of the notations
are for the corresponding space groups. We take the translations of these in the
form
(x, y, z) → (x + l, y + m, z + n) (l, m, n ∈ Z)
and use the additional generators given in the last column (in which the coordi-
nates are not necessarily orthonormal). These are taken from CARAT (cf. [CiS]),
which has such information up to dimension 6.
 DESCRIBING THE PLATYCOSMS 45

our name symbol other names Wolf

torocosm c1 3-torus G1
dicosm c2 half turn space G2
tricosm c3 one-third turn space G3
tetracosm c4 quarter turn space G4
hexacosm c6 one-sixth turn space G5
didicosm c22 Hantzsche-Wendt space G6
first amphicosm +a1 Klein bottle times circle B1
second amphicosm −a1 B2
first amphidicosm +a2 B3
second amphidicosm −a2 B4

Table 11. Names and notations for platycosms.

internatl. non-translation
symbol [CDHT]
no. name generators
c1 (◦) 1. P 1 —
c2 ¯ ×)
(21 21 21 21 ) = (× ¯ 4. P 21 (−x, −y, z + 1/2)
144. P 31
c3 (31 31 31 ) (−x − y, x, z + 1/3)
145. P 32
76. P 41
c4 (41 41 21 ) (−y, x, z + 1/4)
78. P 43
169. P 61
c6 (61 31 21 ) (x + y, −x, z + 1/6)
170. P 65
¯ (−x, y + 1/2, −z + 1/2)
c22 (21 21 ×) 19. P 21 21 21
(x + 1/2, −y, −z)
+a1 ◦0 ) = (∗:∗:) = (××0 )
(¯ 7. P c (x + 1/2, y, −z)
−a1 ◦1 ) = (∗:×) = (××1 )
(¯ 9. Cc (x + 1/2, z, y)
¯ 0 ) 29. P ca21 (−x, y, z + 1/2)
+a2 (21 21 ∗:) = (¯
∗:¯
∗:) = (××
(x + 1/2, −y, z)
¯ = (××
¯ 1 ) 33. P a21 (−x, y + 1/2, z + 1/2)
−a2 (21 21 ×) = (∗:×)
(x + 1/2, −y, z)

Table 12. Notations for space groups.

The two international numbers and names given in the three metachiral cases
correspond to the two orientations.
The infinite — or non-compact — platycosms were systematically treated prob-
ably for the first time in [Wo] p.123, with the symbols indicated in the correspond-
ing column of the following table:
46 J. H. CONWAY AND J. P. ROSSETTI

our name symbol Wolf

Euclidean Space EU C E
Circular Product space CP SA (θ) S1θ
Circular Möbius space CM SA S2
Toroidal Product space T P SA B C T1
Toroidal Möbius space T M SA B:C T2
Kleinian Product space KP SBA K2
chiral Kleinian Möbius space +KM SB A K1
achiral Kleinian Möbius space −KM SB A K3

Table 13. Infinite platycosms.

References
[Ch] Charlap, L.S., Bieberbach groups and flat manifolds. Universitext. Springer-Verlag,
New York, 1986.
[ChV] Charlap, L. S.; Vasquez, A. T., Compact flat riemannian manifolds. III. The group
of affinities. Amer. J. Math. 95 (1973), 471–494.
[CiS] Cid, C.; Schulz, T., Computation of Five and Six dimensional Bieberbach groups.
Experiment Math. 10 (2001), 109–115.
[Co1] Conway, J.H., The orbifold notation for surface groups. In Groups, combinatorics
and geometry (Durham, 1990), 438–447, London Math. Soc. Lecture Note Ser., 165,
Cambridge Univ. Press, Cambridge, 1992.
[Co2] Conway, J.H. (with the assistance of F.Y.C. Fung), The sensual (quadratic) form,
Carus Mathematical Monographs, 26 Mathematical Association of America, Washing-
ton, DC (1997).
[CDHT] Conway, J.H.; Delgado Friedrichs, O.; Huson, D.H.; Thurston, W.P., On
three-dimensional space groups. Beiträge Algebra Geom. 42 (2001), no. 2, 475–507.
[CoS] Conway, J.H.; Sloane, N.J.A., Low-dimensional lattices. VI. Voronoı̆ reduction of
three-dimensional lattices, Proc. Roy. Soc. London Ser. A 436 (1992), no. 1896, 55–68.
[DR] Doyle, P.G.; Rossetti, J.P., Tetra and Didi, the cosmic spectral twins. Preprint.
[E] Efstathiou, G., Is the Low CMB Quadrupole a Signature of Spatial Curvature?
Submitted to MNRAS, arXiv:astro-ph/0303127.
[HS] Hajian, A.; Souradeep, T., Statistical Isotropy of CMB and Cosmic Topology. Sub-
mitted to Phys. Rev. Lett., arXiv: astro-ph/0301590.
[HW] Hantzsche W.; Wendt H., Dreidimensionale euklidische Raumformen. Math. An-
nalen. 110 (1934–35), 593–611.
[Hi] Hillman, J.A., Flat 4-manifold groups. New Zealand J. Math. 24 (1995), no. 1, 29–40.
[LSW] Luminet, J.P.; Starkman, G.D.; Weeks, J.R., Is Space Finite? Scientific American
(1999), no. April, 90–97.
[MS] Malfait, W.; Szczepański, A., The structure of the (outer) automorphism group of
a Bieberbach group. Compositio Math. 136 (2003), no. 1, 89–101.
[No] Nowacki, W., Die euklidischen, dreidimensionalen, geschlossenen und offenen Raum-
formen. Comm. Math. Helv. 7 (1934), 81–93.
[NYT] Dennis Overbye, Universe as Doughnut: New Data, New Debate. New York Times
March 11, 2003.
[RC] Rossetti, J.P.; Conway, J.H., Hearing the Platycosms. Preprint.
[Spe] Spergel, D.N.; et al., First Year Wilkinson Microwave Anisotropy Probe (WMAP)
Observations: Determination of Cosmological Parameters. To appear in The Astro-
physical Journal, astro-ph/0302209.
 DESCRIBING THE PLATYCOSMS 47

[TOH] Tegmark, M.; de Oliveira-Costa, A.; Hamilton, A., A high resolution foreground
cleaned CMB map from WMAP. arXiv:astro-ph/0302496.
[Th] Thurston, W.P., Three-dimensional geometry and topology. Vol. 1. Edited by Silvio
Levy. Princeton Mathematical Series, 35. Princeton University Press, Princeton, NJ,
1997.
[URLW] Uzan, J.-P.; Riazuelo, A.; Lehoucq, R; Weeks, J., Cosmic microwave back-
ground constraints on multi-connected spherical spaces. Submitted to Phys.Rev.Lett.,
arXiv:astro-ph/0303580.
[We1] Weeks, J.R. The shape of space. Second edition. Monographs and Textbooks in Pure
and Applied Mathematics, 249. Marcel Dekker, Inc., New York, 2002.
[We2] Weeks, J.R. Real-Time Rendering in Curved Spaces. IEEE Computer Graphics and
Applications 22 (2002), no 6, 90–99.
[Wo] Wolf, J.A., Spaces of constant curvature, Fifth edition. Publish or Perish, Inc., Hous-
ton, TX, 1984.
[Zi] Zimmermann, B., On the Hantzsche-Wendt manifold. Monatsh. Math. 110 (1990),
no. 3-4, 321–327.

Department of Mathematics, Princeton University, Fine Hall, Princeton, NJ

08544, USA.
E-mail address: [email protected]

FaMAF(Ciem), Universidad Nacional de Córdoba, Ciudad Universitaria, 5000-

Córdoba, Argentina. (Visiting Princeton University).
E-mail address: [email protected], [email protected]