Professional Documents
Culture Documents
Lectures On Groups of Transformations: J. L. Koszul
Lectures On Groups of Transformations: J. L. Koszul
Groups of Transformations
By
J. L. Koszul
J. L. Koszul
Notes by
R.R. Simha
and
R. Sridharan
Contents
1
0
1
2
3
4
. . . . . . . . . . . . . . . . . . . . . . . . . . .
Proper groups of transformations . . . . . . . . . .
Some properties of proper transformation... . . . .
A characterisation of proper transformation groups
Existence of invariant metrics . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
2
3
5
6
.
.
.
.
.
.
11
11
13
14
15
20
21
2
1
2
3
4
5
6
Slices . . . . . . . . . . . . . . . . . . . . . .
General Lemmas . . . . . . . . . . . . . . . .
Lie groups acting with compact isotropy groups
Proper differentiable action . . . . . . . . . . .
The discrete case . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
1
2
3
25
Finite presentations for discrete proper groups.... . . . . 25
Finite presentations for groups of automorphisms.... . . . 31
Groups generated by reflexions . . . . . . . . . . . . . . 37
1
2
3
43
Criterion for proper action for groups of isometries . . . 43
The rigidity of proper actions with compact quotients . . 46
Discrete subgroup of Lie group. Witts Theorem . . . . . 49
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Contents
ii
5
1
2
55
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
1
2
3
4
5
6
7
.
.
.
.
.
.
.
6
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
63
63
64
65
68
70
72
74
Chapter 1
This chapter collects some basic facts about proper actions of topolog- 1
ical groups on topological spaces; the existence of invariant metrics is
discussed in. 4 (Bourbaki [1], Palais [1]).
0
Let G ba a topological group, acting continuously on a topological space
X. We shall always suppose that the action is on the left, and if m :
G X X defines the action, we shall write, for s G and x
X, m(s, x) = sx.
Notation. For A, B X, we set
G(A/B) = s G sB A , .
1.
1.
G G/G(x)
Gx X
f is a closed continuous bijection, hence a homomorphism. In
other words, the orbits (with the topology induced from X) are
homogeneous spaces of G.
= (UK)(V K)1 ,
which is compact.
Using (iv), we see that every closed subgroup of G acts properly on
G/K.
1.
Proof of the lemma . W may be assumed open. Let V be a neighbourhood of x such that G(V|V) is relatively compact, and let A =
G(V|V) W. Then A G(x) = (note that G(x) W). Hence, for
there exist neighbourhoods Wt of t and Vt of x such that
every t A,
(Wt Vt ) Vt = . Since A is compact, we have a finite subset F of A such
T
S
Vt . Then clearly G(U|U) G(V|V)
Wt . Let U = V
that A
tF
tF
T
S
Wt = , hence G(U|U) W.
and G(U|U) A
G(Vt |Vt )
tF
tF
We now proceed with the proof of the theorem. Suppose that G acts
properly on X. Then for any (x, y) X X, f 1 ((x, y)) = G(x y) y is
compact. Hence we need only prove that f is closed.
Let F G X be closed. since f (G X) is the graph of the relation
f (G X).
defined by G, it is closed in X X (2, (ii)), so that f (F)
Let f (s, y) = (x, y) f (F). We must show that (x, y) f (F), i.e.,
f 1 ((x, y)) F , . Suppose this is false. since f 1 (x, y) = sG(y) y,
and G(y) is compact, we then have neighbourhoods W of G(y) and V of
y such that (sW V) F = (recall that F is closed). Now, by Lemma
1, there exists a neighbourhood U of y such that G(U|U) W; clearly
we may assume U V. We then have
f 1 (sU U) G(sU|U) U = sG(U U) U sW V.
Hence f 1 (sU U) F = . It follows that (sU U) f (F) = , which
is a contradiction since sU U is a neighbourhood of (x, y).
1.
u, v T X , s
f (sx)g(sT u, sT v) is a continuous function on G, whose
support is compact since f (sx) , 0 implies s G(B|{x}). Let ds be a
right-invariant Haar measure on G. If we set
Z
gX (uI , v) =
f (sx) g (sT u, sT v)ds,
G
with ds a right-invariant Haar measure on X. Then clearly D is a continuous G-invariant distance function on X. We shall now verify that it
G
Z
V
r
h(sx, sz)ds
2
ds.
We proceed to find such a V. Let U be a compact symmetric neighbourhood of e in G such that for s U, h(x, sx) 2r . Then, since the
continuous function
(s, y)
1.
10
Chapter 2
1 Slices
Let G be a topological group, and H a subgroup acting on a apace Y. We
can then construct in a natural manner a topological space X on which
G acts. In fact, we let H operate on G Y (on the right) by setting
(s, y)t = (st, t1 y); s G, y Y, t H,
and take X = (G Y)/H. If q : G Y X is the natural mapping, then
the left action of G on X is defined by sq(r, y) = q(sr, y).
Note that in the above situation, if we set A = q(e Y), we have
(i) G(A|A)A = A, (ii) G(A|A) = H, (iii) the mapping (s, a)
sa of
G A into X is open. The property (iii) follows trivially from the fact
that the mapping y
q(e, y) of y onto A is a homeomorphism.
Conversely, let G be a transformation group of a space X, and A a
subset of X such that G(A|A)A = A. Then it is clear that H = G(A|A) is a
subgroup of G. By the above considerations, G acts on (G A)/G(A|A).
Let F : G A X be the map F(s, a) = sa, and q : G A G
11
2.
12
15
2. General Lemmas
13
2 General Lemmas
Lemma 1. Let G be a topological group, and H a subgroup of G acting
continuously on a space Y. Let X = (G Y)/H; we suppose that G acts
on X in the natural way. Let q : G Y X be the natural mapping.
Then we have;
(i) for any B Y, G(q(e B)|q(e B)) = H(B|B),
(ii) for any y Y, G(q(e y)) = H(y)
(iii) B Y is a slice in Y if and only if q(e B) is a slice in X.
(iv) B Y is a normal slice if and only if q(e B) is a normal slice.
(v) if y Y is regular, then q(e y) is regular;
(vi) if G is locally compact and H is closed, and if H acts properly on
Y, then G acts properly on X.
Proof. It is easy to verify (i), and (ii) is a special case. Also, once (iii)
is proved, (iv) and (v) follows from (i) and (ii). We shall prove (iii) and
(vi).
Proof of (iii). Let B Y be a slice for the action of H. We shall prove
that the natural mapping (G B)/G(B|B) X, which is clearly one-one
and commutes with the action of G, is actually an open mapping; since
B is a slice for (G B)/G(B|B), it will follow that q(e B) is a slice for
X.
To prove that the mapping (G B)/G(B|B) X is open, it is plainly
sufficient to prove that for any neighbourhood V of e in G, and any
neighbourhood W in B of any b B, the saturation by H of V W is a 16
neighbourhood of eb in GY. Now, if U is a symmetric neighbourhood
of e in G such that U 2 V, it is clear that (V W)H contains the
neighbourhood UX{(H U)W}.
The converse assertion in (iii) is easy to verify.
Proof of (vi). Suppose that H acts properly on Y. Let q(s, y), q(s , y )
X. Let V, V be neighbourhoods of y, y respectively in Y such that
2.
14
17
15
2.
16
(iii) dim N + dim Gx = dim X
Proof. By Theorem 2, Chapter 1, we can choose a G-invariant Riemannian metric on X. Let T (x) be the tangent bundle of X, and let be an
open neighbourhood of the zero section of T (x) on which the exponential mapping exp : X is defined (Nomizu [1]). Since G acts isometrically on X, we may assume that is stable for the induced action
of G on T (X); we denote this action by (s, u) { sT u, s G, u T (X).
We have the relations
s exp u = exp(sT u); s G, u T (X)
and
d(x, exp u) u; x X, u T x (X),
where d is the distance on X induced by the Riemannian metric, and ||u||
is the length of u.
Now let x X, and let T x (Gx) denote the subspace of T x (X) tangential to Gx. G(X) leaves T x (X) invariant, and clearly T x (Gx) is stable
under this action. Since G acts isometrically on X, the orthogonal complement N of T x (Gx) in T x (X) is also stable under G x :
N = {u T x (X) |< u, v > = 0 for all v T x (G x )}.
20
Clearly
this N has property (iii). Now for any r > 0, let Br = {u
N ||u|| < r}. Then Br is G(x)-stable, and is contained in if r is small.
We set f = exp Br . Clearly f has the properties (i) and (ii) of the
lemma. We shall now show that if r is small enough, (iv) is also valid
with B = Br .
We have as usual the commutative diagram
G BrO
OOO
OOO
q OOOO
''
//
r99 X
r
r
rrr
rr
r
r
r
(G Br )/G(x)
17
18
2.
22
23
19
is connected.
Now, q(e R ) is a dense set of regular points in the slice q(e
B),hence its in G \X is dense in G \R . On the other hand, since q(eR ) is
contained in the slice q(e B) at x, it is easy to verify that the mapping
R G \X obtained by composing the mappings R q(e R ) and
q(e RG \X , passes down to a mapping H \R G \X . Since G \R is connected, G \R thus contains a dense connected subset, hence is connected.
Proof of (iii). Use (ii), and (ii) of Lemma 3.
2.
20
3, there exists a neighbourhood V of y such that (z) < (y) for every
z V. Now R is dense in X, and , is constant on R, hence we have
(x) < (y) for every x R. The converse assertion (even without any
assumption on G \X) follows from Lemma 3, since we know that there
exists a slice at every x X.
25
Remark 1. We see from (iv) of Theorem 4 that for the proper differentiable action of a Lie group on a connected paracompact manifold,
the orbits of regular points are of maximal dimension. The converse is
not true, even if the Lie group is connected. For instance, consider the
group G = SO (3, R) of rotations of the two - sphere, acting on it self by
inner automorphisms. Then the regular points are the rotations of angle
, 0 or ; the isotropy group at such point is the one parameter group
through that point, consisting of rotations about the same axis, and the
orbit is a two sphere. For rotation of angle the isotropy group has two
connected components (the identity component being the one parameter
group through that point), and the orbit is a projective plane.
Remark 2. Let G be a connected Lie group. For any compact subgroup H of G, let [ H ] denote its conjugacy class. Now suppose
we are given two conjugacy classes T 1 , T 2 C (G). Then in the set
{[H1 H2 ], H1 T 1 , H2 T 2 }, there exists a ( unique ) minimal class
for the relation <. In fact G acts in the obvious manner on the connected
space G\H 1 G\H 2 , H1 T 1 , H2 T 2 , and the class we are looking for is
the conjugacy class of the isotropy groups at regular points. Thus, given
T 1 , T 2 C (G) we are able to associate with them an element T 1 T 2 of
C (G) characterised by the minimality property.
6.
21
26
6
Let G now be a compact lie group, action continuously on a completely
regular topological space. The following lemmas reduce the problem of
constructing a slice at a point of X to that of the differentiable case.
27
2.
22
Since the G(gi ) are compact Lie groups, we can find a finite subset J of
T
I such that H = G(gi ). For the E and u of the lemma, we can take
PiI
P
E = Ei , u = gi .
iJ
iJ
Lemma 6. Let G be a compact Lie group acting on a completely regular space X. Then for any xo X, there exists a finite dimensional
representation of G in a real vector-space E, and a mapping f : X E
commuting with the action of G, such that G( f (xo )) = G(xo ).
28
6.
23
Chapter 3
26
30
3.
27
sS
s S S sb
1
(t )W is a neigh-
3.
28
33
29
A.
Let now S = G(A|A). For every n, S n A satisfies conditions (1),
(2), (3) of the lemma. If n is large enough, S n A C, and therefore
satisfies condition (4). Hence there exists a finite presentation of G with 34
G(S n A|S n A) S 2n+1 as set of generators.
Remark 1. For a locally simply connected space X,the existence of a
Q
compact set C satisfying the condition of the theorem means that 1 (X)
is finitely generated.
Remark 2. Suppose that, in Theorem 1, we drop the assumption that A
is connected. We can still assert that G is finitely presentable, if X is
locally connected. We may assume that A satisfies conditions (1), (2)
and (4) of the lemma. The space Y constructed above need not now be
connected, so that we will have to enlarge K suitably.
We retain the notation of the proof of Theorem 1. Let B = interior
of S A, and let Bo be a connected component of B. We first prove that
X = G(Bo |B1 )G(B1 |B2 ) . . . G(Bn1 |Bn )Bn ,
(*)
where the union is over all finite sequences B1 , . . . , Bn connected components of B. In fact, since X is locally connected, the connected components of B are open, hence the right side X of () is open in X.
Now let x be any point of X. Since GB = X, we have x tB ,
for t G and some connected component B of B. Now suppose the
neighbourhood tB of x meets X , say
Then
hence
3.
30
@@ g
@@
@@
~~
~~
~
~~~~ f
36
31
3.
32
easily seen that any path in Y can be lifted to a path in X with any given
initial point.
If X is connected, and f : X Y is a covering such that every lift
of any loop in Y is a loop in X, then f is bijective. In fact, it is sufficient
to assume that for a point yo Y and an xo f 1 (yo ), the lift through xo
of any loop at yo in Y of length breadth Y is a loop.
38
39
33
// G
// X
3.
34
It is easy to see that this defines a graph structure on G which is leftinvariant, connected, and finite at each point. We shall now prove that
the breadth of this graph is finite.
Let L(S ) be the free group on S ; for s S , we denote be sL the
corresponding generator of L(S ). Let K be the kernel of the natural
mapping L(S ) G. Since any loop at e in G can be written in the
form (e, s1 , s1 s2 , . . . sn = e), we see that K is naturally isomorphic to the
group of homotopy classes of loops at e.
Now, since G is finitely presentable, we have by a theorem of Schreier a finite subset F of K such that K is the normal closure of F in L(G).
It follows easily that, if for each element of F we choose a representative
loop at e, and 1 is an upper bound for the lengths of these loops, our
graph structure on G has breadth 1.
Remark 2. Let G be a group which has a finitely presentable normal
subgroup N such that G/N is finitely presentable. Then G is finitely
presentable. In fact, by the above remark, there exists a G-invariant
graph structure on G/N which is connected, finite at each point, and of
finite breadth. Since G acts transitively on G/N with isotropy group N
which is finitely presentable, it follows from Theorem 2 that G is finitely
presentable.
As an application of Theorem 2, we shall prove the following
Theorem 3 (Behr [2]). For any finite set P of primes, the group
GL(nZ[P1 ]) is finitely presentable.
35
+1 if ri > 0,
(i) =
0
if ri = 0
1 if ri < 0.
Then the lattice C with the ci as basis obviously satisfies (i).
43
Now let D R, and let r = sup(d(D, A), d(D, B)). Then clearly
pr D A B C. On the other hand, for each i, pr ai and pr+ri i D,
hence pr+(i) i D; this means that pr C D. This proves (ii).
It follows that R is connected: the above proposition shows that,
given A, B R, there exists a path of length d(A, B) joining A and B.
3.
36
We shall now prove that R has breadth 8. We shall show that any
loop (Ao , . . . , An ) in R with n > 8 is homotopic to a product of loops of
length < n, and this will prove our assertion.
Case 1. n even. Let n = 2m. If d(Ao , Am ) < m, there exists a path of
length < m from Ao to Am , and it is obvious that the given loop is homotopic to the product of two loops of length < 2m. Let then d(Ao , Am ) =
m. We choose C R such that d(Ao , C) = 1, and d(C, Am ) = m 1. By
the proposition above, we have then d(Am2 , C) m 2. Since there
exists a path from C to Am (resp. Am2 ) of length m 1(resp. m 2).
It follows easily that the given loop is homotopic to the product of four
loops, each of length < 2m(see the figure below).
2
2
1
44
(In the figure, the lengths of the paths are less than or equal to the
numbers marked along them.)
Case 2. n odd. Let n = 2m + 1. Then d(Ao , Am+1 ) m. If d(Ao , Am+1 ) <
m, there is a path of length < m from Ao to Am+1 , and we are through. If
d(Ao , Am+1 ) = m, we proceed as in Case 1. See the figure below:
2
1
37
3.
38
has exactly two connected components which are carried each into the
other by r, and that M(r) is a (not necessarily connected) submanifold
of M of codimension one.
46
Theorem 4. Let G be a discrete proper group of differentiable automorphisms of a simply connected differentiable manifold M, generated by
reflexions. Then G has a presentation of the form {r ; (r r ) p = e},
where the r are reflexions.
Proof. Let R be the set of all reflexions belonging to G. Since M(r))rR
S
is a locally finite family, M M(r) is an open set; we denote the set of
rR
o , .
(ri )L (r j )L
|ri , r j R , M(ri ) M(ri ) W
We denote by : G = L (R )/K G the natural homomorphism
induced by rL
r. Also, for any r R , we denote rL mod K by r. We
shall prove by induction on the dimension of X that
(1) : G G is a bijection
(2) G acts freely transitively on the set (Wi )iI .
47
39
q(q, W))
, . Here q : G W
(ii) f is locally injective . It is sufficient to know that f is injective 48
o . Now G (V x W
o)
in a neighbourhood of any q(e, x), x W
x
o )) is a neighis saturated with respect to q, hence q(G x (V x W
bourhood of q(e, x). Using the inductive assertions a) and b), we
o ).
see that f is injective on q(G x (V x W
o = M. Now, (G)W
o is
(iii) f is surjective. We must show that (G)W
obviously closed in M, being a locally finite union of closed sets.
o , (G )W
o = GxW
o is a
But it is also open, since for any x W
x
neighbourhood of x by the inductive assertion b).
(iv) f has local sections. Since f commutes with the action of G, and
o = M, it is sufficient to consider points of W
o.
(G)W
Now let N be the subgroup of G defined by
o = W
o }.
N = {n G|(n)W
Clearly N ker . For n N and r R , it is clear that rn (=
o , , we
r(n) ) R . Further if r, r inR and M(r) M(r ) W
3.
40
49
41
Chapter 4
This chapter contains results related with the following kind of problem: 51
given a discrete group of continuous transformations, use information
on the behaviour of a set of generators to prove that the action of the
group is proper. The solution of such a problem is based here on a
Lemma (Lemma 2) related to the methods of Chapter 3 as well as to a
Theorem of Weil on discrete subgroup of Lie groups (A. Weil [1], [2]).
4.
44
G AF
FF
FF
FF
FF
""
52
// Y
f
53
45
2. f has a local section. For any xo X, let B = x X|d(xo , x) <
/2 , and let N = t GtA B . For each t N, we have a
section t : B Y of f defined by t (z) = q(t, t1 z), z B. Note
that, for t, t in N, we have either t = t or t (B) t (B) = .
Let q(r, a) f 1 (B), i.e. ra B. Let s S be such that sA x
X d(x, a) < . Then rsA x X d(x, ra) < B, which means
4.
46
47
Proof of a) and b). With the assumptions of the theorem, we shall prove
that there exists a compact connected subset A of X containing K, a relatively compact open neighbourhood S of e in G, and a neighbourhood
W of mo in M I such that, for every mW, A and S m = S Gm (A|A)
satisfy the conditions of Theorem 1. Then W will satisfy a) and b).
Let C be a compact subset of X such that mo (G, C) = X. Since X
is locally connected, locally compact and connected, there exists a connected compact neighbourhood A of C containing K. Let B be an open
relatively compact set in X, containing A. We set S = Gmo (B|B). Clearly
S is a symmetric open relatively compact neighbourhood of e in G. For
mM, we set S m = S Gm (A|A). Clearly S m is also a neighbourhood of
e.
(i) There exists a neighbourhood W1 of mo in M such, that, for
any mW1 , and any s, s S m , A m(s, A) m(s , A) , implies
s1 s S m .
In fact, L = S 2 S is compact, and mo (L, A) A = . Hence
there exists a neighbourhood W1 of mo in M such that, for any 56
m W, m(L, A) A = . It is easily verified that W1 has the
required property.
(ii) There exists a neighbourhood W2 of mo in M such that, for any
m W2 , S m generates G.
tT
4.
48
57
58
49
1, t a
a (t) =
0, t > a + 2
a (t) 1.
4.
50
(ii) ho () is discrete;
(iii) ho ()G is compact.
for every x X, s G, t W.
Proof. In view of Theorem 2, we can find a neighbourhood W1 of 0 in
R, and a compact set A in X, such that the action of G on W1 X defined
by s(t, x) = (t, mt (s, x)) is proper, and such that mt (G, A) = X for every
t W1 . Then there exists a G-invariant Riemannian metric on W1 X
(Theorem 2, Chapter 1). Let p : W1 X W1 be the natural projection,
and let H be the vector-field on W1 X orthogonal to the fibres of p such
d
that pT H = . It is easily seen that H is G-invariant.
dt
51
(*)
d
, we have
dt
(0, x) = (, a (x))
where a : X X is a diffeomorphism. Using the fact that (0, m0
(s, x)) = s (o, x) (which is () with t = 0), we see that the at , |t| <, 61
satisfy the conditions of the theorem.
For other applications, we need the following modification of Theorem 1.
Theorem 5. Let G be a discrete group acting isometrically on a connected locally connected, simply connected metric space X. Let C be
a connected compact subset of X, and S a finite subset of G(C|C) such
that
(i) e S ,
(ii) for any s, s S , C sC sC , implies s1 s S ,
(iii) S C is a neighbourhood of C,
(iv) S generates G.
This S = G(C|C), the action of G on X is proper, and GC = X.
4.
52
62
Theorem 6 (E. Witt [1]). Let G be the group generated by the set
{r1 , . . . , rn } with the relations (ri r j ) pi j = e, 1 i, j n, where the pi j
are integers satisfying
Pii = 1, pi j = p ji > 1 if j , i, 1 i, j n.
Then G is finite if and only if the matrix cos
Q!
pi j
is positive defi-
nite.
Proof. Let (ei )1in denote the canonical basis of Rn , and
Q B the symn
metric bilinear form on R defined by B(ei , e j ) = cos
. We define
pi j
the standard representation of G in Rn by setting
ri e j = e j 2B(ei , e j )ei .
Clearly, B is invariant under G.
n
We first prove that B is non-degenerate. Let N = x Rn |B(x, y) =
0 for every y Rn . Since N is G-stable and G is finite, there exists a
53
Then prove the positive-definiteness, we consider any non-trivial irreducible G-subspace L of Rn . We see as above that there exists an
ei L.On the other hand, there exists on L a G-invariant positive
definite bilinear form, say Bo , and (Since L is irreducible) a R
such that B|L = Bo . Since B(ei , ei ) = 1, we must have > 0. Hence
B|L is positive definite. Since B is non-degenerate, it follows that B 63
is positive definite.
b) B positive definite G is finite. (The following proof is based on
Buisson [1]). Let C Rn be defined by
C = x Rn B(x, ei ) 0 for every i .
We shall prove the following statements by induction on n.
1) G is finite
2) GC = Rn
3) If s G and c C are such that sc C, then sc = c, and in fact s
belongs to the subgroup of G generated by the ri belonging to G(c).
If n = 2, B is automatically positive definite, and the above statements are easily verified. Thus let n 3, and let us assume that 1), 2)
and 3) are truefor n 1.
P
P
Let
= x Rn B(x, x) = 1 , and let A = C. For each i,
4.
54
65
Ca = y L|B(y, ei ) 0 for all ei L ,
Chapter 5
For proper action a discrete group on a space with compact orbit space 66
/X, there are rather strong connections between the topological properties of X and the properties of . The theory of ends, due to Freundenthal
[1] and Hopf [1] is the most conspicuous example of such a connection.
1
Let X be a connected topological space. We denote by L the set of all
sequences (ai ) of connected sets in X such that
(i) for every i, ai , ,
(ii) ai ai+1 for all i,
(iii) each ai has compact boundary,
(iv) for every compact set K in X , there exists an i such that ai K = .
For (ai ), (bi ) L, we write (ai ) (bi ) if for every i there exists a j
such that ai b j . The relation is an equivalence relation L. Indeed
we need only check that it is symmetric. Let (ai ) (bi ). For any i, there
exists a j such that a j bi = . Since a j is connected and a j 1 X bi ,
it follows that a j bi .
An equivalence class of L with respect to the relation is called an
end of X. The set of all ends of X is denoted by E (X) .
55
5.
56
67
Remark . For (ai ), (bi ) L with (ai )(bi ), there exists a k such that
ak bk = . In fact, there exists an i such that, for every j, ai 2 b j . On
the other hand, for a sufficiently large j, we have ai b j = . Then for
k > i, j we have ak bk = .
Let a E (X), and let (ai ) L represent a. By a neighbourhood
of a we mean any subset of X which contains ai for some i. If V is a
neighbourhood of a, it it clear that, for any (ai ) representing a, V ai
for some i.
We also need the notion of ends of graphs. Let X be a connected
graph (see Chapter 3, 2). For any A X, we define the boundary of A,
denoted by A, as the set
x X (x) A , , (x) (X A) , .
It is easily seen that if C X is connected and C A = , then
either C A or C X A. Now let L be the set all sequences (ai ) of
connected subsets of X such that
(i) ai , for every i,
(ii) ai ai+1 for every i,
(iii) ai is finite for every i,
T
ai = .
(iv)
i
1.
57
(1) for every y Y, f 1 (y) is finite.
Then there exists a unique map f ; E (X) E (Y) such that, for
any a E (X) and any neighbourhood V of f (a), f 1 (V) is a
neighbourhood of a. If we further suppose that
58
70
71
5.
1.
59
60
73
5.
2
Lemma 1. Let X be a connected graph, and let A, B, H be connected
subsets such that A H and B A H. Then either B A or
A B = X.
74
2.
61
(3)
n
n
n
n
since
a(3)
i tS S = , it follows that tS ai S . Hence
( j)
n
Lemma 1 can be applied, with A = a(3)
i , H = S , and B = tai . Since
( j)
( j)
the tai , j = 1, 2, 3 are mutually disjoint, we must have tai a(3)
i for
at least two the j s. This proves the theorem.
62
5.
3) The free product of the cyclic group of order 2 with the cyclic group
of order 3 (which is isomorphic to the classical modular group) has
infinitely many ends, none of which is invariant. This example shows
incidentally that in Theorem 3, the assumption that G \X is compact
cannot be dropped.
Chapter 6
Discrete linear groups acting properly on convex open cones in real vec- 76
tor spaces are of special interest for the applications. In that case, the
existence of a stable lattice or, more generally, of certain stable discrete
subsets gives rise to special methods of constructing fundamental domains. The material here is due to Koecher [1] and Siegel [1].
1
Let E be a real vector space, of dimension n 2. A subset of E is
called a cone if t for every real t > 0. The cone in the dual
E of E, defined by
= X E < x , x >> 0 for all x {0}
P
is called the dual cone of . is always open in E ; in fact, if
denotes
the unit sphere in E (with respect to some
norm on E), we have
P
.
= X E < x , x >> 0 for all x
Assume E to be a Euclidean vector space with scalar product h, i. If,
under the canonical identification of E with E, we have = , we
say that is a self- dual cone
(or a positivity domain). Clearly,
is self
{0} .
- dual if and only if = x E hx, yi > 0 for all y
63
6.
64
Examples.
and
77
hA, Bi = T r (AB) = T r ( A A B B)
= T r ( B A A B)
= T r (( A B) ( A B)) > 0.
2
We now state elementary properties of cones and their duals.
(i) For any cone , is convex.
(ii) If the cone in E contains a basis of E the is a non-degenerate
convex cone (A convex set non- degenerate if it does not contain
any straight line). In fact, let x , y E , and suppose x +ty
{0}, we have 0 hx +
for every t R. Then, for every z
ty , zi = hx , zi + thy , zi, for every t R.
78
{0}, hence y = 0.
Hence hy , zi = 0 for every z
3.
65
Using (i) and (ii) we have (iii) If contains a basis of E, then
x , x imply x = 0.
on K x(
) is continuous and > 0. We can take (K) to be the infimum
of this function.
Remark. If is open, the statement analogous to that of Lemma 1, with
the roles of and interchanged, is also true; the proof is the same.
3
Let E be a non-degenerate cone; we then have , . Let D be a
{0}. For any x , we define
discrete subset of E contained in
(x ) = inf hx , di.
dD
6.
66
that
a
2
a
for d D M(x ). (We may suppose that <.) Thus, there exists a
2
neighbourhood Vi of x such that y Vi implies
hx , di > (x ) +
a
hy , di > (x ) + , d D M(x ).
2
Finally there exists neighbourhood V2 of x such that y V2
implies
a
|hy , d > (x ) < ; d M(x ).
2
Clearly, U = K V1 V2 satisfies conditions of the lemma.
A point x of is called perfect it M(x ) contains a basis of E.
Since, for any > 0, M(x ) = M(x ), we shall assume that , for a
perfect point x , (x ) = 1.
80
(b) (y + to x ) = (y )
(c) M M(y + to x )
3.
(y
67
+
x )
<
(y ).
(
)
D
(I)
6.
68
hx , zi =
ai hx , xi i = (x )
b j hx , y j i (x )
ai
bj
P
P
P
Since (x ) , 0, we have ai b j , hence, by symmetry ai =
b j . It follows that hx , y j i = (x ), i.e. y j M(x ) for every j.
Similarly, xi M(y ) for every i, i.e. z PM(x ) PM(y). The first
assertion of the lemma is therefore proved. The second is then clear.
4
{0}) is said to
Definition . The discrete set D in E (contained in
83
4.
69
(I)
70
6.
86
5.
71
we have
(a) A = ,
(b) (A|A) is finite,
6.
72
(c) (A|A)A is a neighbourhood of A in ,
(d) is finitely presentable.
Proof. Using Lemma 6 and the fact that L = P (and since sM(x ) =
M(sx ), x P, s ), it is easy to see that A = . The proof of
(b) is similar to that of the fact that the action of on is proper; we
have only to use the remark following
Lemma 7. Since A = , (c)
follows. (Note that sA|s is a locally finite family of closed sets
88
Remark. The condition in the above theorem that \P be finite is satisfied if we assume the following: there exists a finite subset B of D such
that, for every y P, there exists s such that convex envelope of
M(sy ) B meets . In fact, let y P, and let s be as above. Then
P
1
a , where r =number of elements of M(sy ) B.
b=
r aM(sy )B
Now,
1X
hsy , bi =
hsy , ai = 1.
r
1
Hence |sy | , where b = (K) of Lemma 1 with K = {b}.
b
The number of points b is finite. Since P is discrete, our assertion
follows.
6
We now apply the preceding results to the case of GL(n, Z) acting on
the space of symmetric positive definite matrices. Thus let E be the
vector space of all real n n symmetric matrices with the scalar product
hA, Bi = T r (AB), and let be the (self-dual) cone of positive definite
matrices in E. = GL(n, Z) acts on : (S , A)
S AS = S [A].
Let D = UU |U Zn , U , . We have seen that D satisfies the
6.
73
Lemma 8, M(A)
contains n independent elements U1 . . . , Un .
Now there exists S such that, for 1 i n,
1
ai
.
..
i
a
j
S Ui = i , O ai < aii for 1 j < i
O
.
..
O
The proof of this fact is analogous to the Elementary Divisor Theorem (see van der Waerden [1]). Clearly the S Ui are independent. Since
6.
74
j
90
Lemma 11. There exists a finite subset B of D such that for every A P.
there exists S such that the convex envelope of B M(S [A]) meets
.
Proof. Let B = UU |U L , where L is as in Lemma 10. Let A P.
7
Let be an open non-degenerate convex cone in a real vector space E.
Let G be a subgroup of G(), and let : G R+ be a homomorphism
(R+ denotes the group of real numbers > O).
Definition. A norm on (with respect to : G R+ ) is a continuous
map : R+ such that
(i) (sx) = (s)(x) for s G, x ,
(ii) (x) as x
(iii) for every x and r O,
x |(x) r is compact.
(x + )
Examples. i) If is the cone of positive definite real n n matrices
and G = GL(n, R), then (A) = det A, A , is a norm on for
: G R+ defined by (S ) = (det S )2 .
7.
75
ehx,y i dy
(x) =
91
L a + .
Proof. We first remark that for any compact set K , there exists
such that K + . Now, for any y P, PM(y) is a finite
union of pyramids PMi , where the Mi M(y ) consists of precisely
n independent elements. It is sufficient to find each i precisely n independent elements. It is sufficient to find for each i an ai such that
for, by the remark above, there exists a such
PMi L ai + ;
that ai a + for each i, and clearly a will satisfy the condition of the
lemma.
Let M = a1 , . . . , an be any one of the Mi . We have M L; let
6.
76
iI
Examples.
0 1 0
2 0 3
1 0 0 and 0 1 0
0 0 1
1 0 2
93
7.
77
= (t1 , . . . , tn ) Rn |ti > O for all i by setting
(t1 , . . . , tn ) = (1 ()t1 , . . . , n ()tn ),
Let D = (1, . . . , 1 . It is a classical result that for any i, 1 i n,
Bibliography
80
BIBLIOGRAPHY
BIBLIOGRAPHY
81