Sie sind auf Seite 1von 9

THE LYAPUNOV EXPONENTS OF GENERIC AREA

PRESERVING DIFFEOMORPHISMS
E

RICARDO MAN

Editors note. This article contains an outline of a proof of a result announced by


the author around 1980. The theorem and its generalization to higher dimensions are
reported in the authors address at the International Congress in 1983 [M]. The proofs
have not been published. For the 2-dimensional case, the closest to a written draft that
is available are some notes handed out by the author at a seminar in Paris. Except
for some very minor editing, this article is a direct transcription of those notes. We
stress that what is presented here is only the sketch of a proof, and that a substantial
amount of work is needed to fill in the details.

Let M be a compact connected boundaryless smooth 2-manifold and let be a


probability on its Borel -algebra generated by a smooth volume form. Let D (M )
be the space of -preserving C 1 diffeomorphisms endowed with the C 1 topology.
0.1. Theorem. There exixts a residual subset A D (M ) such that every f A is
either Anosov or
1
lim
log k(Dx f n )k = 0
n n
for a.e. x.
In these notes we shall sketch the proof of this result. This sketch will consist of the
statements of all the lemmas conducting to the proof of the theorem. With few exceptions these lemmas will be stated without proof. The sequence of lemmas is divided
in several steps. In the first step we introduce an entropy function h : D (M ) R
that is upper semicontinuous. Then its set A0 of points of continuity is residual. In
the second step we shall prove the existence of a residual set A1 D (M ) of diffeomorphisms such that all its hyperbolic sets have measure 0 or 1. The third, fourth
and fifth steps are devoted to show that diffeomorphisms in A0 do not exhibit certain
pathological behaviours, because these behaviours would produce discontinuities of h.
The perturbations required to show this discontinuity are constructed with the lemmas given in section III. Using this absence of pathological behaviour we shall show
that a diffeomorphism f A0 A1 satisfies one of the alternatives of the theorem.
1. First step: the entropy function.
If f D (M ) define (f ) as the set of points x M such that the limits:
1
log k(Dx f n )k
n |n|
lim

RICARDO MAN

exist and coincide. By Oseledecs theorem:


((f ))) = 1.
Define + : (f ) R by
1
log k(Dx f n )k
n+ n

+ (x) = lim
and

H (f ) = { x (f ) | + (x) 6= 0 } .
By Oseledecs theorem:
+ (x) 0
for all x (f ) and if + (x) 6= 0 there exists a splitting Tx M = E s (x) E u (x) such
that:
1
lim
log k(Dx f n )|E s (x)k = + (x)
n n
1
log k(Dx f n )|E u (x)k = + (x)
lim
n n
Define the entropy function h : D (M ) R by
Z
+ d.
h(f ) =
(f )

1.1. Lemma. h is upper semicontinuous.


Proof. Define
Z

log k(Dx f n )|E u (x)k d

cn =
(f )

then:
cn+m cn + cm
for all n > 0, m > 0. Hence
1
h(f ) = lim
n+ n

1
log k(Dx f n )|E u (x)k d = lim
cn
n+ n
(f )
Z
1
= inf cn = inf
log k(Dx f n )|E u (x)k d.
n
n n (f )


1.2. Definition. A0 = set of points of continuity of h.


1.3. Corollary. A0 is residual.

LYAPUNOV EXPONENTS

2. Second step: avoiding fat hyperbolic sets.


2.1. Proposition. There exists a residual subset A1 D (M ) such that if f A1
all its hyperbolic sets have measure 0 or 1.
Proof. Let B be a basis of neighborhoods of M and let B be the set of all finite subsets
of B such that BF B 6= M . If f D (M ) and F B define:

S(F, f ) = n f n BF B .
If > 0 and F B define D(, F ) as the set of diffeomorphisms f D (M ) such
that one of the following properties is satisfies:
(1) There exists a neighborhood U of f such that S(F, g) is not hyperbolic for all
g U.
(2) (S(F, f )) < .
Obviously D(, F ) is open. Moreover, it is dense because if f D (M ) is C 2 (that
is a dense property in D (M ) as proved by Zhender) then every hyperbolic set of f
has measure 0 or 1.

3. The fundamental perturbation lemma.
As we expalined in the introduction in the next section we shall have to produce
perturbations of diffeomorphisms exhibiting certain behaviours. The key tool in the
construction of these perturbations will be the next lemma.
3.1. Definition. If f : M - is a diffeomorphism and F Z+ we say that a sequence
of linear maps Lj : Tf j (x) M Tf j+1 (x) M , x M , j = 0, 1, . . . , m is F -adapted if
Lj = Df j (x) f for all j except for at most F values of j or if
(Ln Ln1 L0 )B1 (0) = (Dx f n )B1 (0)
for all 0 n m.
3.2. Fundamental Lemma. If f D (M ), F Z+ and U is a neighborhood of f
there exists > 0 such that if 1 M is a Borel set, 0 < 1 < and N : 1 Z+
is a measurable function with the property that for all x 1 and m > N (x) there
exists an F -adapted sequence Lj : Tf j (x) M Tf j+1 (x) M , 0 j m, satisfying


Lj (Df j (x) f )
for all 0 j m and
m
Y


Lj exp m1
j=0

then there exists g U such that


Z
h(g) h(f ) + 2 1

+ d .

n0 f n (1 )

This lemma is obtained from the following sequence of properties:

RICARDO MAN

3.3. Lemma. If f D (M ), 0 < k < 1 and U is a neighborhood of f , there exist


> 0 and r0 > 0 such that if x M , 0 < r < r0 , and L : Tx M Tf (x) M is a linear
map satisfying:
kL (Dx f )k <
then there exists g U such that:
g(y) = f (y)
when y
/ Br (x) and
Dy g = L
if y Bkr (x).
3.4. Lemma. Given f D (M ), F Z+ , 0 < k < 1, n > 0 and a neighborhood U
of f there exist > 0 and r0 > 0 such that for all 0 < r < r0 and every nonperiodic
point x M , if Lj : Tf j (x) M Tf j+1 (x) M , 0 j < n, is an F -adapted sequence
satisfying


Lj (Df j (x) f )
for all 0 j < n then for all 0 < r < r0 there exists g U and a compact set
K Br (x) with (K) > k (Br (x)) such that
f (y) = g(y)
if y
/ nm=0 f m (Br (x)) and
Dy g m = Lm1 Lm2 L0
if y K and 0 m < n.
3.5. Definition. Given f D (M ) we say that a family of Borel sets {U0 , . . . , Um }
is a tower if they are disjoint and Uj = f j (U ) for all 0 j m.
3.6. Lemma. Given f D (M ), 0 < k < 1, F Z+ and a neighborhood U of f
(n)
(n)
there exists > 0 such that if {U0 , . . . , Ujn }, n = 1, . . . , ` is a family of disjoint
(n)
towers and for all x `n=1 U0 we have an F -adapted sequence Lj (x) : Tf j (x) M
Tf j+1 (x) M , 0 j < jn , such that


Lj (x) (Df j (x) f )
(n)

for all 0 j < jn , then there exists a compact set K `n=1 U0

(1) g(x) = f (x) for x S,


(n) 
(2) (K) k `n=1 U0 ,
(n)
(3) If x K there exist 1 n ` and y U0 such that:
Dx g jn = Ljn 1 (y) L0 (y).
[Editors remark: S apparently not defined.]

and g U satisfying:

LYAPUNOV EXPONENTS

3.7. Lemma. If f D (M ), F Z+ and U is a neighborhood of f there exists


(n)
(n)
> 0 such that if 0 < 1 < and {U0 , . . . , Ujn }, n = 1, . . . , are disjoint towers
(n)
whose union is f -invariant and for all n 1 and x U0 there exists and F -adapted
sequence Lj (x) : Tf j (x) M Tf j+1 (x) M , 0 j < jn , such that


(1) Lj (Df j (x) f ) for all 0 j < jn ,
(2) kLjn 1 L0 k exp n1 .
[Editors remark: The author probably meant exp(jn 1 ).]

then there exists g U satisfying


Z
h(g) h(f ) + 2 1

+ d.

(n)
n,j Uj

The fundamental lemma follows from the last one and the following well known
result about decompositions of invariant sets in unions of disjoint towers:
3.8. Lemma. If f D (M ) and M is a Borel set such that the measure of
the set of periodic points of f contained in is 0, then for every N > 0, can be
decomposed in a countable union of disjoint towers with its basis contained in 1 and
height larger than N .
[Editors remark: The statemnet of this lemma in probably not intended as is.]

4. Third step: bounding the angles of the Oseledec splitting.


If f D (M ) define 1 : H (f ) R by:

i (x) = inf E s (f n (x)), E u (f n (x))
n0

where (, ) is the angle between the subspaces between the brackets. Let 1 (f ) =
{ x H (f ) | 1 (f ) = 0 }. Then f (1 (f )) = 1 (f ).
4.1. Proposition. If f D (M ) and (1 (f )) =
6 0 then for every 0 > 0 and all
neighborhood U of f there exists g U with
Z
h(g) h(f ) + 0
+ d.
1 (f )

4.2. Corollary. f A0 = (1 (f )) = 0.
Sketch of the proof. Given f D (M ) with (1 (f )) > 0 and a neighborhood U of
f take > 0 given by the fundamental lemma. Fix any c > 0 and define:
c = {x 1 (f ) | + (x) > c}.
Then:
1
k(Dx f n ) | E s (x)k
log
2c
n n
k(Dx f n ) | E u (x)k
lim

RICARDO MAN

for all x c . [Editors remark: The author probably meant 2c.] Then there exists
a > 0 such that if x c then E s (f m (x)), E u (f m (x)) a for some m > 0. This
means that if we define:

a = {x c | E s (x), E u (x) > a}
we have:
c = n>0 f n (a ).
Take A > 0 and 0 < 1 < min(, a) and a measurable function N : a Z+ such
that if x a and n > N (x) then in every interval [j, j + n1 ] [0, n] there exists m
such that:

E s (f m (x), E u (f m (x)) 1


(Df m (x) f nm ) | E s (f m (x)) exp(n m)(+ (x) + A1 )
k(Dx f m )k exp m(+ (x) + A1 )


(Df m (x) f nm ) exp(n m)(+ (x) + A1 )
k(Dx f m ) | E s (x)k exp m(+ (x) + A1 ).
Choose m [(nn1 )/2, (n+n1 )/2] with all these properties and define a sequence
of linear mappings Lj : Tf j (x) M Tf j+1 (x) M putting Lj = (Df j (x) f ) when 0 j
m 2 or m j n 1 and
Lm1 = R(Df m1 (x) f )
where R is the rotation that sends E u (f m (x)) on E s (f m (x)). Then
Y
n1


Lj |E s (x) exp 2(+ (x) + A1 )1 n
j=0

n1
Y


Lj |E u (x) exp 2(+ (x) + A1 )1 n

j=0

Since (E s (x), E u (x)) a we can choose A = A(a) so small that these inequalities
imply:
Y
n1

Lj exp 3C1 n
j=0

where:
C = sup + (x).
x (f )

If 1 > 0 is small enough, the rotation R is so near to the identity that:




Lj (Df j (x) f ) <

LYAPUNOV EXPONENTS

for all 0 j n 1. Then the hypothesis of the fundamental lemma are satisfied
and there exists g U with
Z
h(g) h(f ) + 6 C1
+ d
Z
Zc
+
d +
+ d
= h(f ) + 6 C1
1 (f )
1 (f )c
Z

+ d.
h(f ) + 6 C1 + 6 1 (f ) c
1 (f )

Since 1 is arbitrarily small the proposition is proved.

5. Fourth step: the domination inequalities.


Define 2 : H (f ) R (i.e. R {+}) taking as 2 (x) the supremum over all
n > 0, m > 0 of



(Df n (x) f m ) | E s (f n (x)) (Df n+m (x) f m ) | E u (f n+m (x)) .
Define 2 (f ) as the set of points x H (f ) where 2 (x) = + then f (2 (f )) =
2 (f ).
5.1. Proposition. If f D (M ) and (2 (f ) 1 (f )c ) 6= 0 then for every neighborhood U of f and every > 0 there exists g U such that:
Z
+ d.
h(g) h(f ) +
2 (f )1 (f )c

5.2. Corollary. f A0 = (2 (f )) = 0.
This proposition also follows from the fundamental lemma. To apply it take K > 0,
n0 > 0, k > 0 and define as the set of points x 2 (f ) 1 (f )c such that


k(Dx f n0 ) | E s (x)k (Df n0 (x) f n0 ) | E u (f n0 (x)) > K,
1 (x) > a.
[Editors question: k = a?]

Let
(K, n0 , k) = n>0 f n ().
If K and n0 are large and k is small, the measure of 2 (f ) 1 (f )c (K, n0 , k) is
small. Now we take 1 > 0 and a measurable function N : (K, n0 , k) Z+ such
that if x (K, n0 , k) and n > N (x) in every interval [j, j + n1 ] contains some m
such that f m (x) . We define the sequence of linear maps Lj as follows: choose
m [(n n1 )/2, (n + n1 )/2] with f m (x) . This property implies that there
exists a subspace S Tf m (x) M with the properties described in the picture:

RICARDO MAN

Define Lm1 as the composition of (Df m1 (x) f ) with a linear map near to I and sending E s (f m (x)) on S and leaving E u (f m (x)) invariant. Define Lm+n0 1 as (Df m+n0 1 (x) f )
composed with a rotation R mapping (Df m (x) f n0 )(S) on E u (f m+n0 (x)). Finally define
Ln1 as the composition of Df n1 (x) f with a linear map L near to the identity that
maps (Df m+n0 (x) f nm )RE u (f m+n0 (x)) on E s (f n (x)) and leaves E u (f n (x)) invariant.
Qn1
Lj maps E s (x) on
The other linear maps Lj are defined as (Df j (x) f ). Then j=0
E u (f n (x)) and E u (x) on E s (f n (x)) and the norm can be bounded by exp 3Cn1 as
in the previous section. In fact this estimate as well as the possibility of taking L
near to the identity require some more precautions in the choice of m that we didnt
explain, hoping that this technical oversimplification may help to put in evidence the
underlying idea of the proof.
6. Fifth step: more domination inequalities.
Define 3 : H (f ) R and 3 : H (f ) R by
3 (x) = sup{n > 0 | k(Dx f m )|E s (x)k k(Df m (x) f m )|E u (f m (x))k > 1/2
for all 0 m n}
3 (x) = sup 3 (f n (x)).
n>0

Let 3 (f ) = {x | 3 (x) = +}.


6.1. Proposition. If f D (M ) and (3 (f ) 2 (f )c 1 (f )c ) > 0 [then] for all
> 0 and every neighborhood U of f there exists g U satisfying:
Z
h(g) h(f ) +
+ d.
3 (f )2 (f )c 1 (f )c

6.2. Corollary. f A0 = (3 (f )) = 0.
6.3. Lemma. If f D (M ) then for all c > 0 the closure of the set
c (f ) = {x | 3 (x) < c}
is hyperbolic if c (f ) 6= .

LYAPUNOV EXPONENTS

Proof. The definition of c (f ) implies the existence of constants K > 0, 0 < < 1
such that:


k(Dx f n )|E s (x)k (Df n (x) f n )|E u (f n (x)) Kn
for all n 0, x c (f ). This condition implies
inf (E s (x), E u (x)) > 0.
xc (f )

Using the fact that f is area preserving and dim M = 2 it follows that there exist
< 1 such that:
> 0, 0 <
K
k(Dx f n )|E s (x)k Kn


(Df n (x) f n )|E u (x) Kn
for all x c (f ), n 0. From the uniformity of these estimates, the hyperbolicity of
c (f ) follows easily.

Proof of the Theorem. Suppose that f A0 A1 and (H (f )) 6= 0. We shall prove
that f is Anosov. By the corollary in section 1, the hypothesis f A0 implies:
(3 (f )) = 0.
Hence (H (f ) 3 (f )) > 0. But
H (f ) 3 (f ) = c>0 c (f ).
Therefore (c (f )) > 0 for some c > 0. By the lemma the set c (f ) is hyperbolic
and since f A1 its measure must be 1. Then f is Anosov.

References
[M] R. Ma
ne, Oseledecs theorem from the generic viewpoint, Proceedings of the International Congress of Mathematicians, August 1983, Warzawa, Volume 2, p. 12591276 (Theorem C and its
Corollary in particular).

Das könnte Ihnen auch gefallen