Vous êtes sur la page 1sur 39

Oct.

1995

POINT GROUPS
A. Bossavit
lectricit de France, 1 Av. du Gal de Gaulle,
92141 Clamart

Abstract. An introduction to point groups, their generation and interpretation.

On notation
The order (number of elements) of a group G is denoted by |G|. The identity is always
denoted by 1. The abstract cyclic group of order n (one generator z and one relation, zn = 1) is
Zn . The symmetric group of order n (i.e., all substitutions in a set comprising n elements) is
S(n). The subgroup of even permutations in S(n) is A(n). The group of all isometries which
preserve a distinguished point (the origin, denoted by 0 in what follows), called the orthogonal
group, is labelled O3 . The subgroup of rotations among these is known as SO3 (the so-called
"special orthogonal" group).
Of course, the orthogonal group relative to another fixed point consists in a completely different
set of operations, so some mention of the origin should, in full rigor, be appended to the symbol O3 .
But we'll have no need for changes of origin, so we'll not maintain such fine distinctions.
When a function f maps some set X into a set Y, this is denoted by f X Y. By
"X Y", one thus means the set of all such maps. A construct like "f = x Exp(x)" should be
understood as "the map which assigns to x the result of evaluating the expression Exp for this
value of x".
What point groups are, and how they are generated
Point groups are closed subgroups of O3 . Closed here is taken in the sense of topology: the
limit of a sequence of isometries belonging to the subgroup also belongs to it. In practice, point
groups are finite, with the exceptions of C and D (to be presented below), and of course SO3
and O3 itself.
Isometries can be even or odd, depending on whether they preserve or change the orientation of
a figure (left shoes into right shoes, etc.). The function O3 IR such that o(g) = + 1 for
even and 1 for odd isometries is the first example of something we'll have a lot to say about, a
character, i.e., a real-valued function on a group G for which (gh) = (g) (h) and (1) = 1
(which implies (g1) = [(g)]1). That precise one is the orientation character, or parity character.

-1-

Isometries fixing 0 are either rotations (around some axis through 0), generically denoted by r,
which are even, or the central inversion (symmetry with respect to 0)1 , which is odd, or a
product of both (they commute: r = r, Fig. 0). Mirror reflections with respect to a plane are the
product of an inversion and a half-turn around the perpendicular axis, so they don't add anything
new. (They are, of course, odd.)

rx
P
0

r x
x

'

Figure 0. The interplay between the inversion , a -rotation r, and the mirror-reflection
with respect to plane P, orthogonal to the axis of rotation. Note that r = r = r' and
r = r = r', where r' is the rotation by the angle ' = .

If G is a point group, the set H G of all rotations in G, i.e., H = SO3 G, is clearly a


subgroup. Point groups are sometimes classified as being of the first kind when H = G, i.e., when
they contain only even operations, and of the second kind otherwise. (Cf., e.g., [Lyubar]. Another
common terminology is, respectively, proper and improper groups, cf., e.g., [LudFal].) A group of
the second kind thus contains some odd transform, say. Any odd transform g can now be
written as g = (1g), and 1g H, since (1g) = (1) (g) = () (g) = ( 1) ( 1) = 1.
One thus has
G H = { h : h H} H .
Since the above elements h are all different ( h1 = h2 implies h1 = h2 ), there are as many odd as
even isometries in a group of the second kind.
So G = H H, and as well (since one may take the identity g = (g1) as a starting point),
G = H H, for any odd transform G . [This shows that H is "of index 2", and "normal" in
G. For these notions, cf. Appendix 3.]
1

Beware "inversion" is ambiguous. We are not talking here of the well-known transformation on which inversive
geometry is based.
-2-

Exercise 1. Find the four isometries which preserve the shape of a banana. Spot the two odd ones
and the even ones (a rotation and, of course, the identity). Note that the inversion is not one
of them.
We said that point groups could be generated by rotations and the inversion about 0. A childish
blunder at this early stage would consist in arguing that, since a rotation and the inversion always
commute, any subgroup so generated will be Abelian. What makes this wrong is the possibility of
combining rotations around different axes, a non-commutative operation in general (the only
exception is for two half-turns around orthogonal axes). On the other hand, two or more different
undistinguished rotations will not generate a closed subgoup of SO3 : only when the axes meet at
some definite angles, and only for some values of the rotation angles, can this happen. Virtually all
treatises reproduce the famous proof by Klein [Klein] on which is based the enumeration of point
groups of the first kind, and we'll not give it again here. (See [Senechal] or [Ladd_76, Martin_82,
Yale].) Suffice to say that, except for the symmetry-groups of the Platonician solids, rotations that
belong to point groups are always by 2/n around some privileged axis or by around axes
perpendicular to the latter. This justifies the adoption of the setting of Fig. 0 as a convenient frame in
which to study point groups, because it allows the use of symbols in a uniform and consistent way.
So we assume an origin 0, lying in a horizontal plane P, and a vertical axis through 0. (This
axis will be 0z, if one imagines a Cartesian frame of reference superposed on this picture in the
usual way, with plane 0xy as P.) All mentioned axes are through 0. All mentioned polyhedra are
centered at 0. We call the vertical plane 0xz. Rotations are counted in radians as usual.
Rotations around the vertical axis are denoted by r(). If = 2/n, n integer 1, the short
notation rn may be used for r(). For n = 1, we have the identity, and for n = 2, a half-turn
around the vertical axis. Occasionally, we shall denote this by r, instead of r2 . Other reserved
symbols are: for the mirror-reflection with respect to P, s for the half-turn around the horizontal
axis 0x, and for the mirror-reflection with respect to . Note that = s = s.
Products like r() are called rotation-inversions, or rotatory inversions. Since r() =
r'( ) (Fig. 0), rotatory inversions are also rotatory reflections, i.e., the (commutative)
product of a mirror-reflection and a rotation around the vertical axis. The rotatory reflection rn
will be denoted by n , or just in the (infrequent) happy unambiguous contexts. (Note that 2 =
.) All the foregoing transforms can be used as generators for the point groups, and obviously they
are redundant. Depending on the group considered, however, some of them may be more
convenient, and careful choice is needed, which is one of the difficulties in this study.
To deal with the tetrahedron and the cube, we'll need the 2/3 rotation around the slanted axis
x = y = z. This will be denoted by .

-3-

A first, rapid description of all point groups


Cn : cyclic group generated by rn . (Isomorphic to Zn .) The symmetry-group of a steam
turbine.
Dn : generated by the previous rotation rn and the half-turn s around some horizontal axis.
Note that sr = rn1s = r1s. The symmetry-group of a (fully-reversible) n-bladed propeller.
T: all rotations that transform a tetrahedron into itself. (Isomorphic to A(4).) Generators: s,
r2 , and the 2/3 rotation around the diagonal axis x = y = z.
O: all rotations that transform a cube into itself. Generators: r4 and the 2/3-rotation
around the axis x = y = z.
Y: all rotations that transform a dodecahedron into itself.
C: all rotations around the vertical axis. (A funnel without its hook. A glass. A bottle.)
D: all rotations around the vertical and half-turns around horizontal axes. (A barrel. An
hourglass.)
S2 n: cyclic group of order 2n generated by the "/n rotation-inversion" 2 n. Note that 2 n2 =
rn , which generates Cn , thus a subgroup of S2 n.
Cn h: direct product of Cn and of {1, }. Abelian. Also the direct product of Cn and of
{ 1 , }. Generators: rn , , or rn , . The symmetry-group of a paddle wheel.
Cn v: generated by rn and the reflection . Isomorphic to Dn . The symmetry-group of an
n-sided pyramid.
Dn h: all isometries that transform a regular n-sided prism into itself. (To see the difference with
Dn : take a prism, twist it a little around the vertical axis: it loses the symmetry Dn h, to keep only the
Dn symmetry.) Generators: rn , , s.
Dn d: when n = 2, the symmetry-group of a tennis ball, or (after having removed irrelevant
accessories) of a horse saddle. Generators: 2n , s. For general n, take an n-sided pyramid, reflect
it through its base, and rotate the reflected part by /n (half the rotation angle around 0z). The
skewed object thus obtained has Dn d symmetry.
Th : Generated by T and by the central inversion
Td , Oh , Yh : all isometries (odd or even) that preserve, respectively, a tetrahedron, a cube2 , a
dodecahedron.
2

Note the inconsistency: since we called SO3 and O3, respectively, the maximal point groups of the first and the
second kind, we should perhaps use a consistent notation , like perhaps ST, SC, SI versus T, C, I, instead of the
present T, O, Y versus Th, Oh, Yh, for the Platonic groups. It seems wiser to stick to tradition, however.
-4-

Remark. Notation is somewhat abused here. We describe C3 , for instance, as the group of ternary
rotations about a privileged axis (the z-axis through 0), but a similar group relative to another
axis would, of course, be called C3 too. Symbols for point groups actually apply to whole
classes of different point groups, namely "conjugacy classes in the orthogonal group". See
Appendix 1 about this notion.
A visualization of point groups
Individual point groups are often described each as the group which leaves invariant some
familiar solid: a cube, a tetrahedron, some prism, etc. But this does not always convey a clear
image. Another descriptive style consists in giving the orbit of a general point under the action of the
group. We'll adopt a kind of synthesis between these two approaches, as follows.
Consider a -shaped token, with some width, black on one side, white on the other side (Fig.
1). Imagine this token lying horizontally, black face upwards, at some distance above the reference
horizontal plane P. Now, look at Fig. 2, which shows how the token is transformed by some of the
isometries mentioned so far. (We are looking from above, approximately along the vertical axis but
not quite, so that F can be seen under F. Black-looking objects are thus above P, and whitelooking ones underneath.)

Figure 1. Showing the token and its reflection in a mirror.

For each group, we'll show all transforms of the token. To materialize the symmetry-elements
of the group, we shall use a thin line for the trace of a vertical mirror, a large caret on the right for the
horizontal mirror (cf. Fig. 4, right), a thick line for the horizontal axis of the half-turn s, a white
circle for the inversion , a black n-sided polygon (when n > 2) for rn and its powers. Rotatory
reflections should be conspicuous enough by themselves in each case not to need a special icon.

-5-

/2

x
s
r

Figure 2. Graphic convention, to show how some isometries act. The -shaped token,
black on its upper side, white on its underside, lies above the horizontal plane. (The
mirror-symetry /2 with respect to 0yz is dicussed in Appendix 5.)

For instance, Fig. 4 shows all the possible realizations, as point groups, of the abstract group
Z2 , with only two elements, the simplest non-trivial group. On the left, its realization as {1, r},
where r = r(), i.e., C2 . Next, as {1, }, i.e., as {1, } S2 (alias C1 ), then as {1, } C1 v
(alias D1 C1 ), last as {1, } C1 h. The unit group (the trivial group), that one may consistently call
C1 , is illustrated by Fig. 3 in this system.

Figure 3. Group C1 (the trivial group).

-6-

C2

S2

C1 h

C1 v

D1

Figure 4. Realizations of Z2.

An important remark already at this stage: whereas the five binary groups of Fig. 4 are
isomorphic (there is only one abstract group of order 2, namely Z2 ), there are obviously three
different geometric groups in Fig. 4, the top three. Groups C1 v and C1 h are the same, as are C2
and D1 . The precise meaning of "different" and "the same", however, is not that obvious in the
present context. The implied equivalence relation is conjugation in O3 : two point groups H and
H' are conjugate in this way iff there is some g O3 such that H' = gHg1 (cf. Appendix 1).
This amounts to saying that both groups have the same matrix expressions in two different
orthonormal coordinate frames.
The only point group of order 3 is C3 (Fig. 5). Same situation with C5 .

Figure 5. Groups C3 and C5.

-7-

There are two abstract groups of order 4: the cyclic group Z4 , and Klein's group V4 , which is
the direct product Z2 Z2 , with two generators. Again there are more than two different geometric
realizations (in the above sense) as point groups. They are presented on Fig. 6, from left to right.
First, C4 , cyclic, generated by r r(/2). Next, S4 , alias C4 C2 , generated by 4 , i.e., a quarterturn followed by a reflection in P, or by r (which happens to be 4 1). Then the three realizations
of V4 : as C2 v (D2 C2 ), generated by r and , as D2 , generated by r and s, and as C2 h (C2 ),
generated by r and . What about D1 h and D1 d? They are conjugate to C2 v and C2 h
respectively (Fig. 7).

C4

S4

C2 v

D2

C2 h

Figure 6. Five non-equivalent groups of order 4.

D1 d

D1 h

Figure 7. Other point-goups of order 4 in the list. They are equivalent (i.e., conjugate in
O3) to two of the groups of Fig. 6.

Let's continue, with groups of order 6. Algebraically, there are two different abstract groups of
order 6: Z6 and S(3). (Proof: Appendix 2.) Again, there are many realizations: C6 , S6 , D3 , C3 h,
C3 v (Fig. 8), all different. (C3 h is isomorphic to C6 . This is not supposed to be obvious.)

-8-

C6

D3

S6

C3 v

C3 h

Figure 8. Point groups of order 6.

Next group, in the direction of increasing order, is C7 , for which a picture is probably
unnecessary (it could not be done with straightedge and compass, anyway...). Let's go to order 8.
Again, there are five abstract groups (cf. [BauCha], pp. 150-52, [BouRic_74], pp. 156-58,
[Armstr_88, p. 70): Z8 , Z4 Z2 and Z2 Z2 Z2 , all Abelian, D4 , and the group of unitary
quaternions. Concrete realizations as point groups are: C8 , S8 , D4 , C4 h, C4 v, D2 h, D2 d, displayed
on Fig. 9 in this order. C8 and S8 realize Z8 , whereas D2 h and C4 h realize Z2 Z2 Z2 and
Z4 Z2 respectively. C4 v and D2 d are isomorphic to D4 . The quaternion group is thus the
smallest group not realizable as a point group.
At order 9, there is only C9 . The other abstract group, not realized in O3 , is Z3 Z3 .
The situation at order 10 is the same as at order 6: two abstract groups, Z1 0 and D5 .
Realizations are C1 0, S1 0, D5 , C5 h, C5 v. No picture is necessary. At order 11, only one group:
C1 1.
The next interesting case is thus at order 12. There are two abelian abstract groups of order 12
(cf. [BauCha], pp. 152-54, [BouRic_74], pp. 159-62, [Armstr_88], p. 116): Z1 2 (realized as C1 2
and S1 2), and Z2 Z2 Z3 , realized as C6 h. There are three non-abelian ones: D6 , A(4), and the
group generated by two elements a and b with relations, a3 = b4 = 1 and aba = b. This last
group (one of the "dicyclic" groups ([Armstr_88], p. 118) has no realization as a point group.
Realizations of D6 are C6 v and D3 d (Fig. 25), and of course the "concrete" D6 . This leaves us
with A(4), realized as T, the group of rotations associated with the regular tetrahedron (Fig. 10).
-9-

C8

S8

C4 v

D4

D2 h

C4 h

D2 d

Figure 9. Point groups of order 8. (As a rule, we describe each group in such diagrams by
using a minimal set of generators. Remark however that several choices of such sets may be
available.)

There is only one group of order 13, two of order 14 (same situation as with 6 and 10), one
only of order 15 ([BauCha], p. 131, [Moody], p. 80). Beyond that, the problem of listing all
groups of given order becomes increasingly difficult (see [Moody]3 , which solves it up to order
41), but this is of little concern to us because at this altitude, and as far as point groups are
concerned, some monotony settles: apart from the seven infinite sequences Cn , Cn h, Cn v, S2 n, Dn ,
Dn h, Dn d, the only accidents are at orders 24 (groups S(4), realized as O, or Td , and A(4) Z2 , as
Th ), 48 (group Oh , abstractly S(4) Z2 ), 60 and 120 (Y and Yh , that is, as abstract groups, A(5)
and A(5) Z2 ). We soon return to the detailed study of these, so we momentarily omit pictures, for
which the graphic convention of Fig. 2 is not ideal, anyway (a similar one, based on stereographic
projection, is sometimes used: cf. [Kaetal_72], p. 95).

Beware this author calls Dn (for even n) what is here called Dn/2.
- 10 -

Synoptic list of point groups, according to various authors


There are too many notational systems to describe them all. The most venerable is Schnflies',
to which we adhere here, up to minor variants (cf. Appendix 5). Crystallographers tend to prefer the
Hermann-Mauguin notation [Henlon_52], Hahn], also called international notation. Various
inconsistencies in these two systems, to say nothing of sheer conservatism, have been in the way of
unification till now (see a discussion of this issue in [Ladd_76]). More recent systems like Coxeter's
[Coxeter 69] rather insist on the way point groups are constructed from the basic (first kind) ones,
hence the use of the KH notation, as explained in Appendix 4.
The following table shows a sample of the systems used in some articles or books. The first
column corresponds to Schnflies notation, up to a few variants. Martin's notation (second column)
is close to Coxeter's. The international system is here labeled H.-M., for Hermann-Mauguin. (It
emphasizes rotation axes and their orders, and is desperately complicated: n refers to a rotation axis
of order n, i.e., to rn , the bar refers to rotatory inversions, with m used instead of 2, etc. Look
up [Opecho 86], p. 158, for an explanation.) Schnflies's notation and Weyl's are the same up to
minor details: Weyl used D' where we now put D, and the (presumably German?) initials T, W,
P where we now have T, O, Y.
Michel
Cn
Dn
T
O
Y
Cn h (even n)
Cn h (odd n)
Dn h (even n)
Dn h (odd n)
Th
Oh
Yh
S2 n (even n)
S2 n (odd n)
Cn v
Dn d (even n)
Dn d (odd n)
Td

Martin
I
Cn
C2 nCn
Dn
D2 nDn
T
O
I
C2 nCn
Cn
Dn Cn
D2 nDn
Dn
OT

Weyl
D'n
W
P
Cn
C2 nCn
D' n
D'2 nD'n
W
P
D'n Cn
WT

H.-M.
n
n 2 2 (or n 2)
23
432
5 3 2 (or 5 3)
n/m
2n
n/m m m
2n m 2 (or 2n 2 m)4
m3
m3m
53m
2n
n
n m m (or n m)
2n 2 m
n 2/m (or n m)
43m

Table 1. All point groups, in different notational systems.


4

This variant is one of Ladd's proposals [Ladd 76].


- 11 -

Lyubarskii [Lyubar] uses Schnflies' notation. He gives a set of two tables about the number
of conjugacy classes in each of the point groups (pp. 22 and 27). This is interesting information,
because the number of classes is also the number of (complex) irreps. His tables, unfortunately,
contain mistakes. The similar ones in [LudFal], a bit less complete, are accurate. (These authors
also use Schnflies' notation.)
Subgroup
Cn
Dn (even n)
Dn (odd n)
T
O
Y
Cn h
Dn h (even n)
Dn h (odd n)
Th
Oh
Yh
S2 n
Cn v (even n)
Cn v (odd n)
Dn d
Td

order
n
2n
2n
12
24
60
2n
4n
4n
24
48
120
2n
2n
2n
4n
24

Number of irreps
n
n/2 + 3
(n + 3)/2
4
5
5
2n
n+6
n+3
8
10
10
2n
n/2 + 3
(n + 3)/2
n+3
5

Table 2. Classes.

Groups related with the platonic solids


The tetrahedron and the cube (the "cubic system")
T is the group of rotations leaving a tetrahedron invariant. Besides the identity, there are two
such rotations ( /3) around the axes through the vertices, and one half-turn around each of the
three lines that join the middles of opposite edges, 12 in all (Fig. 10). To well understand the
situation, one should take two opposite edges as horizontal, as if the tetrahedron was inscribed in a
cube of center 0, the latter sitting normally on one of its faces, and the coordinate axes piercing it as
suggested by Fig. 10. The half-turns are then s, r2 (r, in the present context), and rs = sr. One of
the ternary rotations is . By inspection, one obtains the relations:
3 = 1, r2 = 1, (r)3 = 1
which fully determine the group (exercise), provided one sets s = r 2 .
- 12 -

The figure suggests that T is isomorphic to the group of even permutations of the four vertices,
A(4). To verify that, label a, b, c, d the vertices, and use the standard notation for substitutions
on this set of four elements: (ab) means exchange a and b, (abc) means the circular permutation
on a, b and c, (ab)(bc) is a product, etc. Now the correspondence r ~ (ab)(cd), s ~ (ac)(bd),
~ (bcd) establishes the isomorphism, since the above relations translate to the correct ones.

r4 , r2

z
y

0
x
a
0

d
c

Figure 10. Symmetries of the tetrahedron and of the cube.

The full symmetry group of the tetrahedron (including mirror symmetries) is Td , generated by
T and the reflection /4 with respect to the vertical plane x = y (simply denoted by ). Observe
that r = r, and = 2 . Again, the figure suggests that Td is isomorphic to the group of all
permutations of the four vertices, i.e., to S(4). Indeed, corresponds to the permutation (cd). It
is well known that S(4) can be generated by two permutations (like all groups S(n), cf.
[BouRic_74], pp. 180-81). So two geometrical transforms should be enough to generate Td .
Indeed, and foot the bill:
Exercise. Express r and s in terms of and .
Answer: The conjugate transforms 1 and 1 are also mirror reflections, in the planes y =
z and x = z. Their product is therefore s. Now, use r = s to get r = 1.
The rotation group of the cube, O, contains 24 elements: Besides the identity, there are 3
quaternary axes (9 rotations), 4 ternary axes (8 rotations), and 6 binary axes. (Eleven among
these rotations invary the inscribed tetrahedron of Fig. 10: the 8 ternary ones and the three half-turns
around binary axes. Therefore, T is a subgroup in O, of index 2.)

- 13 -

Figure 11. Solids with symmetry Th and how to build them. The one on the right comes
from [Yale], p. 98.

Generators for O are the above ternary rotation and r4 (here denoted by r), with the
relations:
r4 = 1, 3 = 1, (r)2 = 1.
O is isomorphic to S(4): this can be seen by considering the four ternary axes, which get permuted
in all possible ways by the rotations in O.
Exercise. Which rotation is r?
Answer: The half-turn around x = z.
Since T is of index 2 in O, one may form OT (cf. Appendix 4). It's easy to see that OT is
indeed Td : First enumerate all the direct symmetries (rotations) of the cube. Now consider one of
those not in T, a quarter-turn around the vertical axis, for instance, and let it be followed by an
inversion with respect to the origin. This (which is an element of OT, by definition of the latter)
leaves the tetrahedron unchanged. Since O and OT are isomorphic, as shown in Appendix 4, O
and Td are isomorphic too, which we already knew since both are isomorphic to S(4).
- 14 -

Taking the direct products of T and O by {1, }, one obtains Th and Oh . The latter is the
full symmetry group of the cube, and of its dual the octahedron. It is not so easy to find a solid
whose symmetry group is Th . Just taking the union of a tetrahedron and of its central reflection
won't do, because the stellated octahedron thus obtained (called "stella octangula", common in
Escher's graphical work) has symmetry Oh , a larger group than Th ). See Fig. 11 for solids with
symmetry Th (and also Fig. 13).
The icosahedron
Y is the group of rotations leaving a dodecahedron (or its dual, the icosahedron) invariant.
Besides the identity, there are four such rotations ( /5, 2/5) around the axes through the faces
(6 such axes, cf. Fig. 12), two rotations ( /3) around the axes through the vertices (10 such
axes), and one half-turn around each of the 15 lines which join the middles of opposite edges: 60
even symmetries in all.

z
r

y
x

Figure 12. Note that, in this placement of the solid with respect to a Cartesian frame, x =
y = z is one of the ternary axes, though not the one of the rotation labeled here.

A minimal system of generators is r2 (r in this Section) and (cf. Fig. 12, not the same as
the above , beware), with the relations:
r2 = 1, 3 = 1, (r)5 = 1.

- 15 -

If is the 2/5 rotation indicated on the picture, one has


r = 1
(exercise). Thus a more symmetrical presentation of the group, now with three generators r, ,
, is
r2 = 3 = 5 = r = 1.
(This was discovered by Hamilton [Hamil], quoted after [CoxMos 84], p. 67.)
Exercise (cf. [Montes], p. 198, or [Wolf], p. 85, or [CoxMos 84], p. 67). Propose similar
presentations with three generators for T and O.
Answer: For T, r2 = 3 = ' 3 = r' = 1, where ' is the ternary axis through c (Fig. 10). For
O, s2 = 3 = r4 = sr = 1, where s is the half-turn around the line x = z.
From Y, one obtains Yh , the direct product with {1, }, which is the full symmetry group of
the dodecahedron and of its dual. Just as with O, there is no need for notations like Yv , Yd , since
vertical mirror symmetries are already included in Yh .

Figure 13. One of the five cubes inscribed in the dodecahedron. (This shows that Y
contains O, and Yh, Oh.) Removing the matter contained in the cube from the solid
dodecahedron leaves an object whose symmetry group is Th.

Exercise. A(5), the group of even permutations on five objects, is dear to the heart of algebraists,
because of the relevance of its special structure (it is a "simple" group, and the smallest of
non-cyclic such groups) with the general non-solvability of algebraic equations of degree 5.
Virtually all texts dealing with this group signal that it is isomorphic to Y. But is that so
- 16 -

obvious? Where, in the dodecahedron, are the five objects that are permuted by elements of
Y?
Answer: Fig. 13. Cf. [Conway], or [BagGra_83], pp. 266-73. There is also a nice discussion in
http://math.ucr.edu/home/baez/six.html

One may refer to [Kappra 91], [WilliR79] (or, for French readers, to [Joly]) for a menagerie of
polyhedra illustrating the above groups, from which to devise exercises like the following one:
Exercise. What is the symmetry-group of the cuboctahedron of Fig. 14. (The cuboctahedron is
central to Buckminster Fuller's mystique, see [Edmons 87].)
Answer: Oh .

Figure 14. The cuboctahedron (B. Fuller's "vector equilibrium", [Edmons 87]).

"Temari balls" [Peders 83, Miya, Yamam] make a nice theme for exercises in point groups
recognition. (Charts, in the form of decision-trees, to help in such recognition appear in [Ladd_76].
For plane groups, such charts can be found in [Martin_82, Schatt_78].) There is also a simple way
to braid paper strips into various polyhedral shapes [Peders 83, Peder]. This provides in particular
for a nice visualization of Y (without the full Yh symmetry), cf. [Escher], p. 208, or [Peders 83],
p. 43. Other sources for such exercises include [SchWal, Wenninger].
Point groups and frieze patterns
Besides the seven "platonic" groups T, O, Y, Th , Td , Oh , Yh , there are seven infinite
"series" of space groups, namely Cn , Sn , Cn h, Dn , Cn v, Dn d, and Dn h. There are also seven
so-called "frieze-groups", i.e., symmetry-groups of linear patterns. Seven, seven, seven. Are such
equalities accidental? Most often, yes (there are also seven "syngonies", for instance, cf. Table 3).
- 17 -

The last of these equalities, however, is not a coincidence (Fig. 15). This rarely mentioned fact is
simply explained in [BagGra_83] (see also [Senech_75]): looking at a thin flat section about the
equator of a symmetrical solid, what one sees is a repetitive linear pattern, i.e., a frieze. (All of them
can be obtained this way: take a finite length off a frieze pattern, wrap it into a ring; the symmetry
group of what you have gotten is a point group. Cf. Fig. 16.)
Frieze-group (in the
A representative pattern
notation of [Martin_82])

Notation of
[BagGra_83]

Corresponding
point groups

F1

FFFFFFFFFFFFFFFFFFFFF

F1

Cn

F11

DDDDDDDDDDDDDDDDDD

F2

Cn h

F12

AAAAAAAAAAAAAAAA

F4

Cn v

F13

bpbpbpbpbpbpbpbpbpbp

F3

Dn d

F2

SSSSSSSSSSSSSSSSSS

F5

Dn

F21

HHHHHHHHHHHHHHHHHH

F6

Dn h

F22

M M M M M M M M

M M M M M M M

F7

Sn

Figure 15. The seven frieze groups, after [Martin_82]. (On the right, the symbol used in
[BagGra_83], and the corresponding series of point groups, according to the correspondence
suggested by Fig. 16.) Remark the wavering notation.

2/n

Figure 16. Finite part of a F2 frieze-pattern, wrapped round a ring, with a D8 symmetry.

- 18 -

On the prevalent suspicion against group theory among structural specialists


Glockner [Glockn73] once expressed surprise about the neglect of group theory in questions
where, to his views, it was obviously called for, like symmetry in structural mechanics. This paper
stirred a wake of discussion [LeiFra, Glockn 74]. Critics, one may fairly say, overreacted: "The
advantages of symmetry (...) are quite obvious. [This is why] no one has bothered for decades to
address a research paper to this topic. In fact, most automated structural analysis programs deal with
symmetry of structural systems in a routine fashion" [LeiFra]. To which Glockner replied [Glockn
74]: "From several statements made by these discussers, it appears that they missed some of the
points the paper was intended for. (...) Use and exploitation of symmetry is, in the opinion of this
writer, always challenging and exciting and very rarely, if ever, routine."
Glockner's case may have been a little weak, since he restricted his argument to mirror
symmetries and antisymmetries. But his critics probably did not themselves envision more complex
symmetries than cyclic. Otherwise, they would not have been so assertive about the state of the art at
the time: dihedral symmetry was then (and still is) a difficult challenge for "most automated
structural analysis programs" [Nastra].
It has also been this author's experience that careless mention of group theory in mixed
company (i.e., when applied mathematicians and engineers are simultaneously present) may lead one
into unexpected trouble. It may or may not be relevant, in this respect, to point that the whole
subject of symmetry seems to hatch crankish attitudes. From symmetry to polyhedra to the golden
ratio to the music of spheres to pyramidal power, the way seems to be all too easy. Cf. some
preposterous statements in [Ghyka] (pp. 14, 16, 66).
If this is any solace, group theory was also rejected firsthand by those very physicists who later
used it with such success in quantum mechanics. Says Wigner [Wigner]: "... in 1931, there was a
great reluctance among physicists toward accepting group theoretical arguments and the group
theoretical point of view". (See also [Dyson], where Dyson records how Jeans once expressed the
opinion that group theory would "never be of any use in physics".)
Symmetry in relation with finite elements is an underdeveloped subject. Nine books on finite
elements out of ten don't even mention symmetry. When they do so in a reasonably substantial way
(cf., e.g., [IroAhm], Chap. 19, pp. 317-34), they still avoid group theory at all, and don't go
beyond cyclic symmetry.
Miscellaneous notes
Around 1984-85, the European Spatial Agency has commissionned a work by C. Arduini, an
Italian scholar, on the dynamics of repetitive modular structures. Information may, or may not, be
obtained from ESTEC, Consulenze Generale Roma, Via Bruxelles, 55, Roma (Italy).

- 19 -

The (bilingual) Journal "Topologie Structurale" (Dpt. de Mathmatiques, U. du Qubec


Montral, CP 8888, succ. A, Montral, H3C 3P8, Canada), "an interdisciplinary journal on
geometry applied to problems of structure and morphology in design, architecture and engineering",
is a rare source of information, and possibly a forum for publication.
According to Evensen [Evensen], the cyclic symmetry features in NASTRAN were designed
after the landmark paper [Fortes] in which Fortescue introduced the idea of "symmetrical
components" (C3 -symmetry) in Electricity.

Irreps
A good source for irreps, although incomplete, is Serre's book [Serre]. For him, Cn and the
dihedral group Dn are algebraic objects (abstract groups). Cn is the abstract cyclic group denoted
by Zn by others. Dn has a presentation with two generators r and s and three relations: rn = 1,
s2 = 1, srs = r1. It can be realized as Dn or Cn v, in Schnflies' notation. And D2n can also be
realized as Dn d. In practice, it means that once found the irreps for Dn , one knows those of Cn v
and Dn d. For Dn h, take direct-product with Z2 . The group Sn being cyclic, the search for irreps is
almost over at this stage. What is left is T, O, Y and their by-products. The cases of T and OT
are treated by Serre, pp. 57-58. The case of Y is dealt with in [LudFal].
By the very nature of representations, two isomorphic groups have the same list of irreps. A
smart move is thus to class point groups according to isomorphism, hence the following table.
Isomorphism class

Point groups of this class

Zn

Cn

Z2 n

Cn h, S2 n

Dn

Dn , Cn v

D2 n

Dn h (odd n), Dn d (even n)

Dn Z2

Dn h (even n), Dn d (odd n)

A(4)

A(4) Z2

Th

S(4)

O, Td

S(4) Z2

Oh

A(5)

A(5) Z2

Yh
- 20 -

Irreps for Zn . Generator: r = r(2/n).


The complex irreps, all of degree 1, are k (r) = exp(2ik/n), k = 0, ..., n 1. The trivial one,
,
is
0 always real: 0 (r) = 1.
Irreps for Dn . Generators: r and s, with relations s2 = rn = (rs)2 = 1.
For odd n, there are 2 irreps of degree 1, which are the trivial one 0 (or, for consistency,
+) and , thus defined: (r) = 1, (s) = 1. The (n 1)/2 others, all of degree 2, are given,
for 0 < k < n/2, by
exp(2ik/n)

k (r) =

exp( 2ik/n)

, k (s) =
0

.
exp(2ik/n)

exp( 2ik/n)

So there are (n + 3)/2 irreps, corresponding to the same number of conjugacy classes (check against
table 2).
For even n, there are four irreps of degree 1, as follows:
+ +

r:

s:

and for 0 < k < n/2, the same degree-2 representations as in the odd case. The total number of
complex irreps is thus n/2 + 3.
Irreps for O
There are five, the trivial one 0 and i, i = 1 to 5, as follows: (...)
Irreps for T
There are four (...)

- 21 -

Nesting relationships of point groups


We'll endeavor here to investigate all the inclusion relations between point groups: is C2 v a
subgroup of C4 d? does Th contain D3 d, and how many times? does Oh ?and so on. (The reader
should now refer to Appendix 1, especially Exercise 2, for the precise meaning of "contain" and
"how many".)
It is a tedious quiz, and one might be tempted to dismiss such a problem as presenting no
practical interest, to say nothing of its purely mathematical interest.
The following example, however, will show the problem is not without motivation5 . Look
first at Fig. 17:

Figure 17. From left to right, deformation from Oh to Yh and back, via Th, of a
continuous family of polyhedra (first row) and of their duals (second row). All polyhedra are
centered at the origin, as indicated for the cube.

What is here depicted is a very interesting and a bit mysterious phenomenon of continuous
deformation of a structure, by which symmetry seems to be sometimes downgraded, sometimes
increased. Here, for instance, starting from the octahedron on the left (symmetry group Oh ), we
obtain a more complex and less symmetrical polyhedron (symmetry group Th ), which gradually
deforms until, suddenly and only for a moment, it becomes an icosahedron (with the relatively
"high" symmetry Yh ), and eventually a cuboctahedron (Oh ). The symmetry groups of the polyhedra
obtained through the process have been, from left to right,
Oh Th Yh Th Oh
5

We also have problems in structural analysis where a part of the structure is "less symmetrical" than the whole.
Nested symmetry-groups are then clearly at work.
- 22 -

and the inclusion relations in this display clearly show what we meant by "increasing" or
"decreasing" symmetry. (The second row in Fig. 17 is the image of the first one via the geometrical
transfom known as "polar reciprocation"points to planes, etc., that turns a polyhedron into its
so-called dual.)
What happens here can be described in terms of strata of a specific set under a group action
[Michel]. The set, here, say P, is that of all convex polyhedra (in three dimensions), containing the
origin as an interior point. It can be provided with a distance (Hausdorff's, for instance), so that to
speak of a continuous trajectory in this space does make sense6 . A deformation of polyhedron p
is then, by definition, such a trajectory t p(t), going through p (which may be, conveniently,
p(0)).
Remark. There is no need to analytically describe the deformation, but this would be easy, as
follows: Start from the cube on the second row, assume its inscribed sphere is the unit sphere,
and its faces are parallel to the Cartesian coordinate planes; imagine a plane which contains the
axis x = 1, z = 0, and the transforms of this plane, itself included, under the action of Th .
These twelve planes define a unique convex polyhedron. By continuously varying the angle t
of this plane with 0yz, one obtains the polyhedra of the second row, and those of the first row
are their duals. The cube is p(0). (Exercise: for which values of t does one get the
icosahedron and the rhombohedron of Fig. 17?)
So much for the set. Now, the group is O3 (all isometries that fix the origin). It acts on
polyhedra the obvious way: isometry g transforms polyhedron p, whose vertices are v1 , v2 , etc.,
into polyhedron gp, whose vertices are gv1 , gv2 , etc. Subgroups of O3 (i.e., point groups) act on
P in the same way. The orbit of p under the action of subgroup G is the subset {gp : g G}.
For instance, if p is the octahedron of Fig. 17, its orbit under the action of SO3 consists of all
octahedra of same size and same center. On the other hand, its orbits under T, Th , O, Oh ,
respectively, all consist in a single point, p itself, and Oh is clearly the largest group in O3 with
this property. One says that Oh is the isotropy group, or little group, of p, with respect to the
O3 -action.
Exercise. Find, among subgroups conjugate to Y, one for which the orbit of the above p is as
small as possible. How many points are there in this orbit?
Now we may introduce strata: two polyhedra p and p' are in the same stratum (always with
respect to the action of O3 ), if their little groups G(p) and G(p') are conjugate. (One will easily
infer from this example the abstract definition.) By the very definition, a stratum is characterized by
a point group.
Thus, above, the octahedron p belonged to the Oh -stratum. There is more here than a fanciful
way to say that "the symmetry-group of p is Oh ", because of the geometrical grasp this viewpoint
gives on the concept of deformation. One should now envision the manifold P of convex
6

Or, even better, differentiability, since one may give a manifold structure to this set, with only minor technical
difficulties. (For instance, one may have to set an upper limit on the number of vertices, to keep the dimension of
such a manifold finite.)
- 23 -

polyhedra as stratified, according to the different point groups, and the trajectory t p(t) as
wandering across various strata, Th almost all the time, and occasionally Oh or Yh . So the study
of the topology of the system of strata is relevant to the study of deformation. More generally, to
quote from [Michel], "the first thing to do in studying a group action is to decompose it into strata
and orbits".
We'll stop at this point as regards deformations (see [Lalvan_86], [Wurtz_86], [Kappra 91],
Chaps. 8 & 9, and [Hofsta_85] for the unavoidable Escherian connection), and proceed with the
study of inclusion relations among point groups. The question to answer is thus, in general: is point
group H (realized as) a subgroup of (some group belonging to the conjugacy class of) point group
G? Is it normal in G? If not, how many times is it contained in G, i.e., how many subgroups of
G are conjugate to H, including H itself? Curien's contribution to the classic book [Kaetal_72] (p.
88) is rich in information about this (with a few mistakes, unfortunately). We shall address the
question in two successive stages: first about groups of the first kind (those containing only
rotations), then about groups of the second kind.
Point groups of the first kind are Cn , Dn , T, O and Y. (...)
Conclusion

References
[Armstr_88] M.A. Armstrong: Groups and Symmetry, Springer-Verlag (New York), 1988.
[BagGra_83] J.A. Baglivo, J.E. Graver: Incidence and symmetry in design and architecture, Cambridge
U.P. (Cambridge), 1983.
[BauCha] B. Baumslag, B. Chandler: Group Theory (Schaum's Outline Series), McGraw Hill (New York), 1968.
[Bieber I] L. Bieberbach: "ber die Bewegungsgruppen der Euklidischen Rame I", Math. Ann., 70 (1910), pp.
297-336.
[Bieber II] L. Bieberbach: "ber die Bewegungsgruppen der Euklidischen Rame II", Math. Ann., 72 (1912), pp.
400-12.
[BouRic_74] A. Bouvier, D. Richard: Groupes (Observations, Thorie, Pratique) (ASI 1383), Hermann
(Paris), 1974.
[BurGla_78] G. Burns, A.M. Glazer: Space Groups for Solid State Scientists, Ac. Press (New York),
1978.
[Conway] J.H. Conway: "Monsters and Moonshine", Math. Intelligencer, 2, 4 (1980), pp. 165-171.
[Coxeter 69] H.S.M. Coxeter: Introduction to Geometry, Wiley (New York), 1969.
[CoxMos 84] H.S.M. Coxeter, W.O.J. Moser: Generators and Relations for Discrete Groups, SpringerVerlag (Berlin), 1984.
[Dyson] F.J. Dyson: "Mathematics in the Physical Sciences", Scientific American, 24 (1964), pp. 129-46.

- 24 -

[Escher] H.S.M. Coxeter, M. Emmer, R. Penrose, M.L. Teuber (eds.): M.C. Escher, Art and Science (Proc.
Int. Congress on M.C. Escher, Rome, Italy, 26-28 March, 1985), North-Holland (Amsterdam), 1986.
[Edmonds 87] A.C. Edmondson: A Fuller Explanation (The Synergetic Geometry of R. Buckminster
Fuller), Birkhuser (Boston), 1987.
[Emmer_93] M. Emmer (ed.): The Visual Mind: Art and Mathematics, The MIT Press (Cambridge, Ma), 1993.
[Evensen] D.A. Evensen: "Vibration Analysis of Multi-Symmetric Structures", AIAA Journal, 14, 4 (1976), pp.
446-53.
[Farkas] D.R. Farkas: "Crystallographic groups and their mathematics", Rocky Mountain J. Math., 11, 4
(1981), pp. 511-51.
[Fortes] C.L. Fortescue: "Method of Symmetrical Co-ordinates Applied to the Solution of Polyphase Networks",
AIEE Trans., 37 (1918), pp. 1027-1140.
[FreSin_82] A.H. Frey, Jr., D. Singmaster: Handbook of Cubic Math, Enslow Publishers (Hillside, N.J.),
1982.
[Frob] G. Frobenius: "ber die unzerlegbaren diskreten Bewegungsgruppen", Sitzungsber. Preuss. Akad.
Wiss. Phys. Math. Kl. (1911), pp. 654-55.
[Ghyka] M. Ghyka: The Geometry of Art and Life, Sheed and Ward (New York), 1946 (Dover edition, New
York, 1977).
[Glockn73] P.G. Glockner: "Symmetry in Structural Mechanics", J. Struct. Div. ASCE, 98, ST1 (1973), pp.
71-89.
[Glockn 74] P.G. Glockner: "Symmetry in Structural Mechanics" (Discussion), J. Struct. Div. ASCE, 100,
ST8 (1974), pp. 1708-12.
[GoPrak] N.S. Goel, V. Prakash: "Tension icosahedron as a structure for use in terrestrial and outer space
environments", Comp. & Structures, 41, 2 (1991), pp. 189-196.
[Hahn] Th. Hahn (ed.): International Tables for Crystallography, Reidel (Dordrecht), 1983.
M. Hamermesh: Group Theory and its Application to Physical Problems, Addison-Wesley (Reading,
Ma.), 1962.
[Hamil] W.R. Hamilton: "Memorandum respecting a new system of roots of unity", Philos. Mag., 12, 4
(1856), p. 446.
[HenLon_52] N.F.M. Henry, K. Lonsdale (eds.): International Tables for X-Ray Crystallography, The
Kynoch Press (Birmingham, U.K.), 1952.
[Hiller] H. Hiller: "Crystallography and cohomology of groups", Amer. Math. Monthly (Dec. 1986), pp.
765-79.
[HilPed_88] P. Hilton, J. Pedersen: Build your own polyhedra, Addison-Wesley (Menlo Park, CA, USA),
1988.
[Hofsta_85] D.R. Hofstadter: Metamagical Themas: Questing for the Essence of Mind and Pattern, Basic Books
(New York), 1985.
[IroAhm] B. Irons, S. Ahmad: Techniques of Finite Elements, Ellis Horwood (Chichester), 1980.

- 25 -

[EDM] S. Iyanaga, Y. Kawada (eds.): Encyclopedic Dictionary of Mathematics, The MIT Press (Cambridge,
Ma.), 1980.
[Joly] L. Joly: Les Polydres, Blanchard (Paris), 1979.
[Kaetal_72] T. Kahan et al.: Thorie des groupes en physique classique et quantique, t. 2, Dunod
(Paris), 1971.
[Kappra 91] J. Kappraff: Connections: the geometric bridge between art and science, McGraw-Hill (New York),
1991.
[Klein] F. Klein: Lectures on the Icosahedron, Dover (New York), 1987 (first published in 1884).
[KocTar_93] A.S. Koch, T. Tarnai: "The Aesthetics of Viruses", in [Emmer_93], pp. 223-8.
[Ladd_76] M.F.C. Ladd: "Computer-assisted methods in the study of point-group symmetry", Int. J. Math.
Educ. Sci. Technol., 7, 4 (1976), pp. 395-400.
[Lalvan_86] H. Lalvani: "Metamorphosis and Cycle in Curved Space Structures", in Ref. [Escher], pp. 187-93.
[LeiFra] K.R. Leimbach, D. Franz: "Symmetry in Structural Mechanics" (Discussion), J. Struct. Div. ASCE,
99, ST8 (1973), pp 1792-94.
[Lomont] J.S. Lomont: Application of Finite Groups, Academic Press (New York, London), 1959.
[LudFal] W. Ludwig, C. Falter: Symmetries in Physics (Group Theory Applied to Physical
Problems), Springer-Verlag (Berlin), 1988.
[Lyubar] G.Ya. Lyubarskii: The Application of Group Theory in Physics, Pergamon Press (New York),
1960.
[Martin_82] G.E. Martin: Transformation Geometry (An Introduction to Symmetry), Springer-Verlag
(New York), 1982.
[Nastra] R.H. Mac Neal, R.L. Harder: "Nastran Cyclic Symmetry Capability", in Nastran User's Experience
3rd Colloquium (NASA Tech. Memo. TM X-2893, Langley Research Center, Hampton, Va, USA), 1973,
pp. 395-421.
[Michel] L. Michel: "Symmetry Defects and Broken Symmetry. Configurations. Hidden Symmetry.", Rev. Mod.
Phys. 52, 3 (1980), pp. 617-651.
[Miya] K. Miyazaki: An Adventure in Multidimensional Space, Wiley (Chichester), 1986.
[Montes] J.M. Montesinos: Classical Tessellations and Three-Manifolds, Springer-Verlag (Berlin), 1987.
[Moody] J.A. Moody: Groups for Undergraduates, World Scientific (Singapore), 1994.
[Opecho 86] W. Opechowski: Crystallographic and Metacrystallographic groups, North-Holland
(Amsterdam), 1986.
[Peders 83] J. Pedersen: "Geometry: The Unity of Theory and Practice", Math. Intelligencer, 5, 4 (1983), pp.
37-49.
[Peder] J. Pedersen: "Braiding Escher models", in Ref. [Escher], pp. 203-10.
[Penros 79] R. Penrose: "Pentaplexity: A Class of Non-Periodic Tilings of the Plane", Math. Intelligencer, 2,
1 (1979), pp. 33-43.
- 26 -

[Schatt_78] D. Schattschneider: "The plane symmetry groups: their recognition and notation", Amer. Math.
Monthly, 85, 6 (1978), pp. 439-50.
[SchWal] D. Schattschneider, W. Walker: M.C. Escher Kaleidocycles, Tarquin Publications (Stradbroke),
1978.
[ScBGC] D. Schechtman, I. Blech, D. Gratias, J.W. Cahn: "Metallic Phase with Long-Range Orientational Order
and No Translational Symmetry", Phys. Rev. Letters, 53, 20 (1984), pp. 1951-53.
[Schonf 91] A.M. Schnflies: Kristallsysteme und Kristallstruktur, Teubner (1891).
[Senech_75] M. Senechal: "Point groups and color symmetry", Zeitschrift fr Kristallographie, 142 (1975),
pp. 1-23.
[Senechal] M. Senechal: "Finding the Finite Groups of Symmetries of the Sphere", Amer. Math. Monthly, 9 7,
4 (1990), pp. 329-35.
[Serre] J.P. Serre: Reprsentations linaires des groupes finis, Hermann (Paris), 1978.
[Watson] J.D. Watson: Molecular Biology of the Gene, Benjamin (Menlo Park, Cal.), 1976.
[Wenninger] M.J. Wenninger: Polyhedron Models, Cambridge U.P. (Cambridge, U.K.), 1971.
[Weyl] H. Weyl: Symmetry, Princeton U. P. (Princeton, N.J.), 1952.
[Wigner] E.P. Wigner: Gruppen Theorie und Ihre Anwendung auf die Quantenmechanik der
Atomspektren, Friedr. Vieweg (1931), (Eng. transl.: Ac. Press, 1959).
[WilliR79] R. Williams: The Geometrical Foundation of Natural Structure, A Source Book of Design,
Eudmon Press (Moorpark, Cal.), 1972 (Dover edition, New York, 1979).
[Wolf] J.A. Wolf: Spaces of constant curvature, Publish or Perish (Wilmington, Del.), 1984.
[Wurtz_86] M. Wurtz: "Structural and Dynamic Transformations in Molecular Biology", in Ref. [Escher], pp.
305-12.
[Yale] P.B. Yale: Geometry and Symmetry, Dover (New York), 1988 (first published in 1968, Holden-Day).
[Yamam] K. Yamamoto: "Temari Workshop", in Ref. [Escher], pp. 237-38 and 398.

Appendix 1
Conjugation is one of the very primitive concepts of group theory. Two elements g and g'
are conjugate in a group G iff there exists h G such that g'h = hg, i.e., g' = hgh1.
Conjugation is an equivalence relation, and conjugacy classes are the equivalence classes with
respect to this relation. The notion of conjugation applies to subgroups: two subgroups H and H'
of a group G are conjugate if any h H has a conjugate h' H', and vice versa.

- 27 -

The concrete meaning of these definitions may be immediately grasped by some, but some help
in building for oneself an appropriate mental image of conjugation may also be useful.
Suppose the Lefthanded Persons Liberation Front wants to demonstrate by having the Statue of
Liberty raise the left hand. They have good connections, some money, and a theoretically perfect
photocopier at their clandestine headquarters. What do they do?
Dismantle the statue, have blueprints of all pieces made, bring them to the machine, take copies,
reduce them, mount them all into a single A4 frame, on a transparent sheet. End of Phase One.
Now flip the sheet (Phase Two). Phase Three, the last one, will consist in undoing all the operations
of Phase One, in reverse order: dissect the inverted blueprint, make blow-ups, have pieces forged,
moulded, or whatever, from the inverted blueprints, and have the new sinister thing erected,
according to instructions, by sympathizers of the LPLF (who, of course, can read inverted script,
like Leonardo; observe that some care should be exerted when drawing the blueprints, in order to
make sure the flip will really swap right and left, not top and bottom, or back and frontan aside
puzzle, interesting in its own left).
In this story, we have three groups: H = {do_nothing, change_arms}, of order two, H' =
H, and a gigantic group G generated by all the operations
involved in the whole process. The operations h = change_arms and h' = flip are conjugate in G.
Note that groups H and H', both belonging to G, are conjugate in G.
{no_flip, flip}, of course isomorphic to

More seriously, think about what scholar A means when he says to his guest, scholar B:
"Hey, we happen to have the same kind of wall paper, only in my office, it's an op-art pattern
instead of your Escher's lizards." On which he draws from his wallet a photograph (he has it next to
his wife's), put it at arm's length at the right distance of the wall and at the right angle, and says:
"Look; see what I mean?" What A has done is to find an affine transformation that maps the plane
of the photograph to the plane of the wall (or, rather, the best approximation to such a transform he
has been able to manage, despite distortions due to parallax, etc.), in order to bring the generating
patterns of both paper designs into coincidence. A and B can then satisfy themselves that their
wall-paper groups are "the same" (all copies of the patterns, properly flipped, rotated and translated,
do match), whereas, actually, they are conjugate in the affine group of IR2 .
The two wallpaper groups, being conjugate, are also isomorphic. Remarkably, the converse is
true, according to a theorem due to Bieberbach [Bieber II]: conjugate subgroups in the affine group
are isomorphic (in all dimensions). Hence the classification one may find in many publications, of
wallpaper groups according to their classes of isomorphism (there are 17 such classes; in
dimension 3, i.e., for crystallographic groups, there are 219 of them). What is intended, in fact,
as demonstrated by the above anecdote, is a classification in conjugacy classes (in the relevant affine
group).
Another striking illustration of conjugation is given by the craze of the early 80's, Rubik's cube
[FreSin_82]. "Cubers" soon learned that some simple manipulation, say T, had a definite effect, for
instance to swap two pairs of small cubes ("cubies"), located at specific positions ("cubicles"). One
was then able to swap some differently located cubies: first bring them into those privileged
- 28 -

positions by some transform S, then perform T, then carefully undo all unwanted changes by
performing S1. To master a transform T thus meant that one also had all its conjugates STS1 at
one's command. Children often seemed to discover this spontaneously.
We have been using this notion in the main text practically all the time, explicitly or not. For
instance, groups of (2/n)-rotations around two different axes, say d and d', are different, yet we
called them all Cn , instead of, say, Cn (d) and Cn (d'). It is easy to see that Cn (d) and Cn (d') are
in fact conjugate in a larger group (the group of isometries of the Euclidean three-dimensional space,
if d and d' do not intersect; if they do, the group of rotations about the common point is enough).
So the symbol Cn really refers to conjugacy classes of isometry groups.
A concrete example where such fine distinctions may be called for is when we consider
subgroups of a concrete group. A cube, for instance, has three quaternary axes, so its rotation group
O (all rotations that preserve it) contains three different, conjugate two by two, "instances" of C4
(note they are conjugate not only in the affine group, but in O as well). We'll say, rather loosely,
that O "contains C4 three times".
Exercise 2. Show that D3 contains C2 three times, and C3 only once.
Answer. This is actually an abstract phenomenon, independent of the realization of D3 as a point
group. Let D3 = {1, r, r2 , s, sr, sr2 }, with r3 = s2 = (rs)2 = 1. (Note that these relations do
determine the group.) The three instances of C2 are C2 ' = {1, s}, C2 '' = {1, sr}, C2 ''' =
{1, sr2 }. They are conjugate in D3 : for instance,
r C'2 r1 r {1, s} r2 = {1, rsr2 } = {1, sr} = C2 ' ' ,
etc. This proof may as well be carried over in concrete geometrical terms: the three subgroups
correspond to three instances of C2 , relative to three different coplanar axes of rotation through
the origin, at angles /3 to each other. The subgroup {1, r, r2 } is the only realization of C3
in D3 , because r and r2 are the only elements of order three in D3 .
About the previous exercise, note that C3 is normal or invariant, as a subgroup, in D3 (cf.
Appendix 3): its conjugates g{1, r, r2 }g1 all coincide with it (try g = r, g = s, etc.). Therefore,
C3 is alone in its conjugacy class in D3 . One might be tempted to infer from this that "when a group
H is normal in a larger group G, it is 'contained only once' in G". Indeed, in that case, H is
alone in its conjugacy class, if one refers to conjugacy in G. But there may be other subgroups of
G which are conjugate to H in a group larger than G itself. Better to explain that with an
example, as provided by the following:
Exercise. How many instances of C2 are there in D2 ? In D4 ?
Answer. 3. 5. (Count the axes of possible half-turns.)

- 29 -

Appendix 2
Proposition. There are 2 (classes of isomorphic) groups of order 6.
Proof. Let G be a group of order 6 and a one of its elements of maximal order (the order is the
smallest integer n such that an = 1). This order n, being a divisor of |G|, must be 6, 3 or 2. In
the first case, G = Z6 . If n = 3, G contains the subgroup {1, a, a2 } and three other elements. Let
b be one of them. Then the group elements b, ba, ba2 are all different, and different from the first
three, so we have all the group. What about ab? Not being equal to 1 (because a1 = a2 , not b),
or to a (since this would imply b = 1), or a2 (since b a), or b (since a 1), ab is either ba
or ba2 . If ab = ba, then (ba2 )2 = a and (ba2 )3 = b, so here we are with an element of order larger
than 3, a contradiction. Therefore ab = ba2 , and by completing the multiplication table we see that
G = S(3). Now suppose n = 2, i.e., all non-trivial elements of G have order 2. Let a, b be two
of them. Then abab = 1, hence bab = a, hence ab = ba (premultiply by a, then b, and use a2 = 1,
b2 = 1). So G would contain the subgroup {1, a, b, ab ba}, of order 4, which is impossible
since 4 does not divide 6. So, with Z6 and S(3), we have exhausted all possibilities.

Appendix 3
The notion of index
We shall recall the notion of index of a subgroup H in a group G.
To keep the discussion concrete, we'll bear in mind the case in which G is some point group
(for instance, C3 v), and H the subgroup of all rotations in G (in that case, C3 ). Assume that H is
strictly contained in G, i.e., there are operations in G that are not rotations. Observe that such an
operation, let's say, is odd (i.e., reverses orientation). Observe also that 2 thus has to be a
rotation, since it is even. In our example, may be the reflection , for instance.
Let us start from the well-known notion of coset: Left cosets of a subgroup H are the sets gH
{gh : h H} where g spans G. There are three important things about them. First, for x and
y to be in the same coset constitutes an equivalence relation (an easy exercise), and therefore cosets
are equivalence classes for this relation, which means that either gH = g'H, or gH and g'H are
disjoint.
The number of such disjoint classes, when finite, is called the index of H in G. (Right cosets,
Hg, of course form a partition too. The index is also the number of right cosets.) Observe that H
always coincides with one of the cosets, the one that contains 1.
In our example, there are two cosets: one is {1, r, r2 }, i.e., H itself, the other is
{, r , r2 } .
Second, the number of elements in gH is always |H|, since gh = gh' implies h = h'. Since
the union of all cosets is obviously G, both observations together yield this: when G is finite, |H|
- 30 -

divides |G| (this is the Lagrange theorem) and the number of cosets, the index, is the integer ratio
|G|/|H|.
And the third important thing? It's that, when left and right cosets coincide, i.e., gH = Hg for
all ga condition under which H is said to be normal in Gone may confer a group structure on
the set of cosets, then denoted7 by G/H, the quotient of G by H. The product of cosets g1 H
and g2 H is the coset g1 g2 H, the inverse of gH is g1H .
Exercise. Show that "being normal" and "being a union of conjugacy classes" are equivalent
properties for a subgroup.
In our example, C3 is indeed normal in C3 v. This is no accident:
Proposition.

Subgroups of index 2 are always normal.

Proof. If g H, then gH includes gg1 = 1, so gH = H, and symmetrically, Hg = H. If g


H, gH = G H, since it is the other coset than H, and Hg = G H by the same argument. Hence
gH = Hg for all g.
Note the above argument does not assume G finite.

Appendix 4
The notation KH.
We explain how groups of the second kind can be enumerated, once all those of the first kind
are known.
Let G be a point group of the second kind. Consider the subgroup H of even elements in G.
As we know, G H = { h : h H} H H, where is any element in G H. Therefore,
G = H H = H H, as a set. Either G contains , the central symmetry, or it doesn't.
If G does contain , G = H H : for transforms in G are rotations (those in H) and
products of these by . Since such products commute, G is the direct product of H by the
subgroup {1, }. (Often enough, one denotes this direct product by H.)
Now, suppose G does not contain . (We know by Exer. 1 that such groups exist. Think
also about the full tetrahedron group.) Let K be the set union H (G H). (Note that all
elements in K are even, so K cannot be G.)
Proposition 1. K is a group (isomorphic to G).

Actually, both the notation G/H and the word "quotient" are often used to refer to the set of cosets, even when H
is not normal. This may be a bit confusing, and we avoid it.
- 31 -

Proof. Let G H. As we know, G = H H = H H, therefore K = H H = H H,


since commutes with everything. Products of two elements of K are thus of one of the forms
h1 h2 , (h1 )h2 , h1 (h2 ) (all clearly in K), or (h1 )(h2 ) = h1 h2 , an element of G which is also
in H, since it is even, as a product of two odd transforms. Thus K is closed with respect to group
multiplication. Inverses of h or h, i.e., h1 and h11, are also in K, which therefore is a
group. The isomorphism is achieved via the mapping g if g H t h e n g e l s e g.
So H is a subgroup, both of the original G and of an associated group K, of the first kind.
Note that |K| = |G| = 2 |H|, and observe the symmetrical relations
K H = (G H),

G H = (K H),

depicted (in a perhaps superfluous way) on Fig. 18. One uses the notation
G = KH
as a shorthand to specify that G and H are in such a relationship. (Remark that K = GH as well.)

H
Figure 18. Relationships between G = KH, K, and H = G K. (Observe that K =
GH.) A deceptive notation, isn't it?

The practical implications, as regards the classification of point groups, are as follows: those of
the second kind are of two sorts, 1- direct products of groups of the first kind by {1, } , 2groups KH, where K and H both are of the first kind, H a subgroup of K, with |K| = 2 |H|.
The list of such pairs is easy to establish, hence the classification of all point groups, as it appears in
Table 1.
For practice, let us consider the groups of lowest order that can be obtained in this fashion. The
first is C2 C1 : Here, K = C2 , and H C1 is reduced to the identity. There is only one element in
K H C2 C1 , namely the half-turn r2 around z. Since C2 C1 (K H) H, its only nontrivial element is r2 , i.e., the plane reflection . So C2 C1 is C1 h. Same reasoning about D1 C1 :
single out the non-trivial element in D1 , which is the half-turn s around the x axis, compose with
- 32 -

, hence, since s = , D1 C1 = {1, } C1 v. The game is more interesting if H is a little bigger,


say H = C2 . Let's take K = C4 . Then KH = {1, r4 , r2 , r4 3 }, i.e., the cyclic group generated by
r4 , that is, S4 (cf. Fig. 6). Still with H = C2 , another choice for K is D2 = {1, s, r2 , sr2 }: then
KH = {1, s, r2 , sr2 }, that is, C2 v (Fig. 6).
Exercise. Work out D2 D1 , D3 C3 , D4 D2 . Find a familiar object with D4 D2 symmetry.

Appendix 5 (References: [EDM], entry 94 B, [LudFal], p. 24)


The Schnflies notation [Schonf 91], while quite redundant (it happens that different symbols
refer to the same group), is still the most intuitive of all. Beware however that precisely because of
this redundancy, many small variations are perceptible in the practice of authors who claim to use
this notation.
It uses two series of symbols: the capitals C, D, S, T, O, Y, and the subscripts n, v, h, i,
s, d. (There was also a capital V, for Klein's "Viergrup" V4 , which we discard here: V was a
substitute for D2 .) Capitals C and D stand for "cyclic" and "dihedral", and the subscript n refers
to the order of the rotation r(2/n) around the z axis. S is used for cyclic groups generated by a
rotatory reflection. Hence the basic symbols Cn , Sn and Dn . Symbols T and O refer to the
groups of the first kind that leave invariant a tetrahedron and an octahedron, respectively. As for Y
(or I, for many...), this is a modern addition: Y, the group of the icosahedron, was not actually
considered by Schnflies, who was only interested in point groups of crystalline structures, for
which it was known (from an easily established mathematical result) that rotational symmetries of
order five could not happen. (The realization that such symmetries are not incompatible with
crystal-like arrangements of atoms is recent [ScBGC]. The corresponding mathematical objects,
called aperiodic pavings, had been known since the mid-Seventies [Penros 79]. On the other hand,
not so few biological structures have icosahedral symmetry: radiolaria ([Ghyka], pp. 91 & 100,
[WilliR79], p. 140), the Aids and Herpes viruses and other adenoviruses ([Watson], [Miya], p. 19,
[KocTar_93]), etc. This and the contemporary craze about "fullerenes", those artificial carbon
molecules with Y-symmetry, and wonderful properties, forbid us to ignore this group.)
Small-case symbols v, h and i, subscripted, stand for "vertical", "horizontal" and "invertive",
respectively. They refer to symmetry operations which can be combined with all those in one of the
basic groups to yield a larger group (twice as large, actually). For instance, Cn v is obtained by
adding to the list of elements of Cn , and by considering the group thus generated. This is what
"to combine" is taken to mean. (Recall that is the reflection with respect to the vertical plane
0xzbut any vertical plane through 0z would do, in order to yield Cn v from Cn .)
Thus, Cn {1, r, r2 , ..., rn1} combined with the horizontal plane reflection yields Cn h.
Since and r commute, the combination simply consists in adjoining to Cn all products of the
form rk , k = 0, ..., n 1. The group Cn h thus obtained is what is called the direct product of Cn
and of the two-element group {1, }. The combination with also yields direct products, since
commutes with all other transforms, hence the group Cni, denoted by Cn in some other systems.

- 33 -

With , which does not commute with elements of Cn other than 1, the "combination" is a less
trivial operation than a plain direct product of groups.
At this stage, we begin to see the redundancy at work: C2i is equal to C2 h (strictly equal, not
merely conjugate in O3 ), C4i is C4 h (Fig. 19), etc.: Cni = Cn h for even n.

C2

C2i

C4h

C4i

Figure 19. For even n, Cnh = Cni.

What about odd values of n? Working out the case n = 3 will be enough:
Exercise. Compare C3i and S6 .
Answer: Fig. 20.

r
1

Figure 20. Note that = r.

Clearly, Cni = S2 n for odd n, just as above in the case n = 3, since r(2/n) is the inverse of
the rotatory reflection r(/n) 2 n, which generates S2 n. Faced with this redundancy, Schnflies
used Cni for odd n (with Ci standing for C1i), rather than S2 n, and Cn h rather than Cni for even
n (with again a charming variation of his, Cs standing for C1 h, the only appearance of subscript s,
meaning "reflection"). The logic of his system still called for S2 n for even values of n, but groups
S8 , S1 2, etc., could not be associated with crystals, so he had little use for S (cast only once, in
S4 ). Note also that there is no use for the notation Sn when n is odd: the group generated by
(2/n) is then Cn h.

- 34 -

Combining Dn with or yields Dn h and Dni (alias Dn ), again direct products, again
redundant notation when n is even:
Exercise. Compare D2i and D2 h.
Answer: Fig. 21.

D2

D2i

D2h

Figure 21. From D2 to D2i (alias D2h).

How can one combine with Dn ? Now, being with respect to 0xz is important because,
contrary to what happened with Cn v, not any vertical reflection can be combined with Dn . To
understand this, we need to investigate (Fig. 22) how to combine the half-turn s around 0x and the
mirror reflection through a vertical plane of equation y = x tan (/2 < /2, the
equation being x = 0 in the latter case): Their composition s is the product of three reflections,
through P, and successively, and the last two amount to a rotation by 2 (twice the angle
of the two vertical mirrors), according to a well known result. So s = (2), with obvious
notation. We may assume a rational /, since otherwise (2) would not be of finite order.
How does this combine with rn ?

s
Figure 22. The product s is equal to the rotatory reflection r(2).
- 35 -

Working out the exercise, one finds that the only meaningful cases are when is a multiple of
/2n. For even multiples, contains a horizontal rotation axis, so the enlarged group is the same
as the one obtained when = 0, for which Dn v would be a logical name. But since s = , the
would-be Dn v is actually Dn h, and this case brings in no novelty (Fig. 23), whatever the parity of
n.

Figure 23. From D3 (left) to "D3v" (middle), which turns out to be D3h (right).

In the case of odd multiples, the plane bisects the angle between two horizontal rotation
axes. The enlargement of Dn thus obtained is what Schnflies called Dn d (Figs. 7, 9), d for
"diagonal" (mirror), the last subscript in the above list.
Exercise. Compare D1i and D1 d, D3i and D3 d.
Answer: Figs. 24 and 25.

s
D1

s
D1 d

Figure 24. From D1 to D1d (alias D1i)

- 36 -

D1i

Figure 25. From D3 (left) to D3d (middle), the same as D3i (right).

We have now exhausted the combinatorial possibilities (as far as symbols of the principal series,
C, D, S, are concernedfor T, O, Y, cf. the relevant sections in the main text). There is no
meaning to Cn d, which is just Cn v. As for combining Sn with , , , nothing new either: for
odd n, Sni = C(2n)h, Sn h = Sn = Cn h, and for even n, Sni = Cn h, and Sn v = D(n/2)d. This makes for
numerous and a bit tedious exercises, except perhaps for the latter case.
Exercise. Compare S4 v and D2 d.
Answer: Fig. 26.

Figure 26. From S4 (left) to S4v (middle), which is actually D2d (right).

Not only did Schnflies eliminate redundant symbols when two or more of them referred to the
same group, but he also did that with symbols referring to equivalent groups (in the above sense,
conjugation in O3 ). Thus were retained, for groups of order 2, the symbols C2 , Ci (here replaced
by S2 ), Cs (our C1 h), and neither C1 v nor D1 or D1 d (for which Cs and C2 can stand).
Similarly, D1 d could be discarded, the group it denotes being conjugate to C2 h.
What remained was a list of 32 groups (the crystallographic point groups) corresponding to the
32 "crystal classes" described by Hessel in 1831:

- 37 -

Point groups of the 32 crystal classes


Crystalline system
(the so-called "syngony")

Point groups

Triclinic

C1 , Ci

Monoclinic

Cs, C2 , C2 h

Rhombic8

C2 v, D2 , D2 h

Hexagonal

C3 h, C6 , C6 h, D3 h, C6 v, D6 , D6 h

Tetragonal

S4 , C4 , C4 h, D2 d, C4 v, D4 , D4 h

Rhombohedral9

C3 , C3i, C3 v, D3 , D3 d

Cubic

T, Th , Td , O, Oh

Table 3. Syngonies.

The 230 crystallographic space groups (groups of isometries containing three independent
translations) distribute among these 32 classes. See, e.g., [BurGla_78, Kaetal_72, LudFal] for their
description. They were first enumerated by Fedorov (1885, or perhaps 1891; there is an ongoing
priority squabble around this). An isometry g can always and uniquely be expressed as the product
r t, where r fixes the origin and t is a translation. When g spans the space group G, r thus
spans a point group H: one says that G belongs to the class H. (Beware that G may not include
H as a subgroup.) As shown by Table 3, the 32 classes distribute among 7 broader classes, called
"syngonies", characterized by the symmetry of the system of three generating vectors of the lattice
(cf. [BurGla_78], Chap. 2). There is an even more refined classification among 11 "Laue classes",
useful to X-ray crystallographers only, cf. [Kaetal_72], p. 89, [BurGla_78], p. 76.
Remark. It's sometimes quite puzzling to read in some places about 219 space groups, instead of
230. This is a delicate matter of definition, which has to do with what exactly we mean when
we say that two groups of isometries are "the same". It can be any of the following three
things:
(a) The two groups are isomorphic (same multiplication table, up to labelling of the elements),
(b) The two groups are conjugate in the group of all affine transformations1 0,
8

Also "orthorhombic".
Also "trigonal".
10
meaning that, if a given object is to undergo any transformation belonging to the first group, this can be performed
in three phases, as follows: first, translating, rotating, stretching, reflecting, etc. (all imaginable affine
tranformations), the object in an appropriate way, then, applying some transform which belongs to the second group,
9

- 38 -

(c) The two groups are conjugate in the group of all affine, orientation preserving
transformations.
If we mean (c) (the strictest rule, hence the most discriminating one), there are 230 "different"
space groups. If, like Frobenius [Frob], we mean (b), only 219: eleven pairs in the previous
list are enantiomorphic, i.e., refer to crystal structures which are mirror twins. If, like
Bieberbach [Bieber I], we mean (a), there are 219 classes again, because criteria (a) and (b)
are actually equivalent (in all space-dimensions): this is Bieberbach's theorem [Bieber II]. In
dimension 2, it happens that criteria (b) and (c) also give the same classification: there are thus
17 classes of "equivalent" wall-paper groups, whatever we mean by that. For more
information, cf. [Hiller] (a first-class paper, for the mathematically minded) or, for the
physicist's approach, [Opecho 86], pp. 263-68 (not a very satisfactory reference, but one out
of a precious few).

and finally, undoing all the affine transformations of the first phase, in reverse order. This is the concrete interpretation
of "conjugation" (cf. the Appendix on conjugation).
- 39 -

Vous aimerez peut-être aussi