Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
OF
CLASSICAL MECHANICS
Princeton University
ERGODIC PROBLEMS
OF
CLASSICAL MECHANICS
V. I. ARNOLD
University of Moscow
and
A. AVEZ
University of Paris
PREFACE
vi
PREFACE
mann-Gibbs model, that is, a system of hard spheres with elastic collisions,
belongs also to this class; this proves the "ergodic conjecture."
This book is by no means a complete treatise on ergodic theory, and
references are not exhaustive.
The text presented here is based on lectures delivered during the spring
and fall of 1965 by one of the authors, who also wrote Chapter 4. The second
author is responsible for the proofs of Chapters 1,2, and 3.
We thank Professors Y. Choquet-Bruhat, H. Cabannes and P. Germain,
J. Kovalewsky, G. Reeb, L. Schwartz, R. Thorn, and M. Zerner, who welcomed the lecturer at their seminar. We also thank Professor S. Mandelbrojt,
who suggested that we write this book. The final manuscript was read by Y.
Sinai, who made a number of useful improvements for which we are sincerely
grateful.
The translator (A Avez) wishes to thank warmly Professors V. 1.
Arnold, S. Deser, and A S. Wightman, who prevented him from many
mistakes.
V. I. ARNOLD
A AVEz
CONTENTS
Preface
Chapter 1.
Chapter 2.
Dynamical Systems
1. Classical Systems
2. Abstract Dynamical Systems
3. Computations of Mean Values
4. Problems of Classification. Iwmorphism of
Abstract Dynamical Systems
5. Problems of Generic Cases
General References for Chapter 1
Ergodic Properties
6. Time Mean and Space Mean
7. Ergodicity
8. Mixing
9. Spectral InvaI:.iants
10. Lebesgue Spectrum
11. K -Systems
12. Entropy
General References for Chapter 2
vii
1
1
7
9
11
12
13
15
15
16
19
22
28
32
35
51
CONTENTS
viii
Chapter 3.
Chapter 4.
Appendixes.
Unstable Systems
13. C-Systems
14. Geodesic Flows on Compact Riemannian
Manifolds with Negative Curvature
15. The Two Foliations of a C-System
16. Structural Stability of C-Systems
17. Ergodic Properties of C-Systems
18. Boltzmann-Gibbs Conjecture
General References for Chapter 3
Stable Systems
19. The Swing and the Corresponding
Canonical Mapping
20. Fixed Points and Periodic Motions
21. Invariant Tori and Quasi-Periodic Motions
22. Perturbation Theory
23. Topological Instability and the Whiskered Tori
General References for Chapter 4
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
53
53
60
62
64
70
76
79
81
81
86
93
100
109
114
-tI5
117
119
120
121
123
125
127
129
131
132
134
138
143
145
147
153
CONTENTS
Negative Curvature
Proof of the Lobatchewsky-Hadamard Theorem
Proof of the Sinai Theorem
Smale Construction of C-Diffeomorphisms
Smale's Example
Proof of the Lemmas of the Anosov Theorem
Iritegrable Systems
Symplectic Linear Mappings of Plane
Stability of the Fixed Points
Parametric Resonances
The Averaging Method for Periodic Systems
Surfaces of Section
The Generating Functions of Canonical Mappings
Global Canonical Mappings
Proof of the Theorem on the Conservation of
Invariant Tori under Small Perturbations of
the Canonical Mapping
ix
158
163
168
178
191
194
196
201
210
215
219
221
227
230
235
243
249
Bibliography
271
Index
283
CHAPTER 1
DYNAMICAL SYSTEMS
This chapter contains examples of dynamical systems and related problems.
l. Classical Systems
DEFINITION
1.1
Let M be a smooth manifold, /l. a measure on M defined by a continuous positive density, 1>t: M .... M a one-parameter group of measure-preserving diffeomorphisms. The collection (M, /l., 1>t) is called a classical
dynamical system.
The parameter t is a real number or an integer. If t ( R, the group
1>t
1>t
feomorphism
1>
= 1, ... ,
= dimM.
1>
1> 1-
and
EXAMPLE
11-
is a translation group:
x=l,
y=a
(Z,
p/q rational:
>0
and p and q relatively prime. In the covering plane (x, y), the orbit with
~(x-xo)
q
I
I
-r-------I
I
I
I
-,--I
I
-T--------
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
Figure 1.3
DYNAMICM, SYSTEMS
EXAMPLE
=.
Tl Vx to V at x,
there is one, and only one, geodesic y passing through x with initial velocity vector ~. We denote by y (~, s) the point of y obtained from x in
time s when moving along y with velocity 1. The unit tangent vector to
y at y(~, s) is
(1.5)
Formula (1.5) defines a one-parameter group of diffeornorphisms of M
Tl V.
DEFINITION 1.6
(Liouville's theorem).
R2n , and
(1.9)
The equations
dq
aH
dp
dt
ap
dt
LIOUVILLE'S THEOREM
1.10
...
dP n . dq1 dqn
Proof:
...P-.
aq
(aH ) +
(_ aH) - 0
ap,
ap
. aq
-
(Q. E. D.)
1.11
dH
dt
aH aH + aH (_ aH) = 0
aq
ap
ap
aq
(Q. E. D.)
1.12
da
I grad HII
length,
EXAMPLE
p/ + q/,
12 =
pi + qi
DYNAMICAL SYSTEMS
The dynamical systems that are induced on these tori are isomorphic to
those of Example (1.2). Appendix 5 provides further examples.
REMARK 1.14. GLOBAL HAMILTONIAN FLOWS
More generally one may consider a symplectic
2n-dimensional mani-
x=
where /: T* Mx
for any ~
/w 1 '
M2n
TM x is defined by
Z.
!(x, y) mod 1\
automorphism is
(x,y) = (x+wl' y+( 2 )(modl),
w1
H.
i';
! (x, yY mod 1\
fined by:
- (1 1)
As Det
1 2
closed two-form
of rank n. Example;
dpl\dq
on R 2n .
. and then
)I
Figure 1.17
DYNAMICAL SYSTEMS
nA
bon of the plane. On M, this ribbon lies approximately in the neighborhood of an orbit of the system:
x=
According to Jacobi's
1,
the~rem
y=,\-l:
2.
DEFINITION
2.1 2
= Il(A)
If
1>/
1.
1>
1>1'
we
notation "mod 0." All of the preceding examples are abstract systems:
a compact Riemannian manifold M ~ith its canonical measure Il (Il(M)
= 1)
The space M.
Let Zn
m (M,
of the form
i ( Zn
i (Z,
Zn by setting:
i.
iI''''' a jk
p .... p . .
ikl
II
lk
where a;= a j _
..
m'= (".,a;,,,.),
To prove cp is measlire-
Hence:
p.[cp(Aj)
p.[A!H)
Pj
= p.(Aj).
J.
ith
EXAMPLE
2.3.
2' .
11
DYNAMICAL SYSTEMS
'J
1~
____________. -____
------~~
<p'A
Figure 2.4
'" '( x, Y) _
j (2x,
- 'I
<~
mod 1, if Y2 S x < 1 .
mod 1, if 0 S x
Y2y)
Of'
ox, twice as short in the direction oy. Then, cut off the right half of this
rectangle and move it, by translation, above the left half (see Figure 2.4).
As A C M and n converges to + "", 'n A is formed by a "very large
number" of segments which are parallel to ox.
Let us mention some problems on dynamical systems.
3.1
lim! Arg
/ ... "" t
l
k= 1
10
where the a k ,
(Uk
3.2
N~oo
EXAMPLE
3.3
= measureltlO~
t~ T,
y(t) (DI
lim - T
T~oo
These three problems are special cases of a more ,general one: Let f be a
complex-valued, /l-summable function defined on the space M of an abstract system (M, /l, t). Compute (if it exists)
lim
T~oo
1...
T
IT
!(tm)dt,
m ( M.
3.4
t ... + 00
(see Figures 1.17 and 2.4). Physical intuition makes it plausible that for
11
DYNAMICAL SYSTEMS
4. Probl('ms of ~Iassification
Isomorphism of Abstract Dynamical Systems
A natural way to classify the dynamical systems is to exhibit their invariants with respect to their corresponding group: ccnonical transformations for Hamiltonian flows, measure-preserving diffeomorphisms for general classical flows. Since abstract invariants are the deepest ones, we
give the definition:
DEFINITION
4.1
Two abstract dynamical systems (M, /1, ) and (M', /1" ') are isomorphic if there exists an isomorphism f: M .. MI (mod 0) of measurable
spaces making the following diagram commutative:
4.2
The Bernouilli schemes B (112, 1/8, 1/8, 1/8, 1/8) and B (1/4, 1/4, 1/4,
1/4) are isomorphic (see Meshalkin
EXAMPLE
4.3
The translations of the tori (1.1S) are not isomorphic to the automorphism of the torus (1.16) (see Chapter 2. 12.40).
EXAMPLE
4.4
On the torus T2
I(x, y) mod
11
'(x,y)
12
4.5
5.1
5.2
->
5.3
In the abstract frame, ergodic systems are generic in the weak topology (see Halmos [1]). By contrast, every Hamiltonian systemcIose enough
to the geodesic flow of the torus T2 (see Appendix 2) is nonergodic. See
also the three-body system (Chapter 4). Thus, ergodicity can be nongeneric in the classical frame.
DYNAMICAL SYSTEMS
13
CHAPTER 2
ERGODIC PROPERTIES
A series of concepts (ergodicity, mixing, spectrum, entropy, etc.) has
been introduced by the metric theory of dynamical systems to describe the
behavior of most of the orbits. This chapter is devoted to th'eir definition,
In Chapters 3 and 4 we shall make use of these concepts to describe classical systems.
6, Time Mf'an and Space Mean
DEFINITION
(6.2)
*
[(x)
M,
(6.2),
x ( M,
[(x)
1.)
15
f(x)dll
16
THEOREM
for every t,
(c)
A proof will be found in P. Halmos [1] for the discrete case, in NemytskiiStepanov [1] for the continuous case.
REMARK
6.5
*
I(x)
6.6
7.1
'mean a.e.:
(7.2)
1
I(x)
f(x), a.e.
try.
17
ERGODIC PROPERTIES
Thus, for an ergodic system, the time mean does not depend on the initial
point x.
EXAMPLE
7.3
M1,
M2 :
(2)
M2
CD
M,
M2
Figure 7.4
*
the time mean f(x)
REMARK
{1o
if x ( MI
if x ( M2 '
i(x) depends on x.
7.5
*
M1=lxlf(x)<a!,
M2
I x I f*(x)
::: a! .
18
COROLLARY
7.6
An abstract dynamical system is ergodic if, and only if, it is indecomposable, that is if every invariant measurable'set has measure 0 or 1.
The preceding argument proves more: a system is ergodic if, and only
if, any.invariant measurable function f ( Ll (M, p.) is constant a.e.
EXAMPLE
7.7
The rotation : x
-+
q> 0, p and
= ]
Xo
~ (1-
E)p.(I).
Thus, if n l
, ... , n k
nk[
are disjoint, we
get:
2 Small analytic perturbations of this metric' of
the geodesic flow (see Chapter 4).
r2
ERGODIC PROPERTIES
19
IL(A)?
;= 1
On the other hand, the orbit of a given endpoint of [ is dense (Jacobi theorem, Appendix 1). Since IL([) S E, there exist integers n 1 ., n k such that
the.sets
cp"I[,oo.,cp"k[
Consequently
and
ment proves that the systems of Examples (1.2) and (1.15) are ergodic as
soon as their orbits are everywhere dense (see Appendix 11). Appendixes
12 and 13 provide more examples.
8. \lixing
Let M be a shaker fun'of an ins,olJlpressible fluid, which consists of
20% rum and BO% Coca Cola (see Figure B.l). If A is the region originally
occupied by the rum, then, for any part B of the shaker, the percentage of
rum in B, after n repetitions of the act of stirring, is
In such a situation, physicists expect that, after the liquid has been
stirred sufficiently often
(Il
20
Figure 8.1
DEFINITION
8.2
lim
t->+
/L[tAnB]
/L(A) "/L(B)
00
0 or 1.
A; we get:
21
ERGODIC PROPEPTIES
The following example proves the converse is false: ergodicity does not
imply mixing. 3
EXAMPLE
8.5
An isometry
images
,p" A
,p
tion with another set B is sometimes empty, sometimes of positive measure. For instance, the ergodic translations of the tori (Examples 1.2 and
1.15) cannot be mixing.
EXAMPLE
8.(;
8.7
Mixing can be defined for endomorph isms (Appendix 6) which are not
automorphisms (see Appendix 14).
REMARK
8.8
lim
T->+oc
~ ~
1p.(,p"AnB)-p.(A)p.(B)1
n=O
3 1n contrast, if
p. (M) = 00, A. 8.
p., ,pt)
be
an ergodic system in which p.(M) = oc, A and B two measurable sets. For any
E
>
nB)
< E.
22
V. A. Rohlin [1] has proposed the concept of n-fold mixing as a new invariant of the dynamical systems (see Halmos [1]): A dynamical system
lim
inf
i '1i
9. Spectral Invariants
Let (M, /1, ) be an abstract dynamical system. Let us denote the Hilbert space of the complex-valued functions defined on M with /1-summable
square by L 2 (M, /1). If f, g ( L 2 (M, /1); we set
II f I
DEFINITION
Y<fTT>
9.1
We set
(9.2)
Uf(x) = ( (x)) ,
23
ERGODIC PROPERTIES
THEOREM
(b) V is a bijection: Any g ( L 2(M, /l) can be written as some Vf. Explicitly f (x) '"' g (cp -1 x).
(c) V is isometric: Since cp is /l-measure preserving, setting cpy '"' x,
we get
IlV f l1 2
'"'
If(CPy)12 d/l(Y)
(Q. E. D.)
REMARK 9.4
In the continuous case (M, /l, CPt) we obtain a continuous unitary oneparameter group V t .
DEFINITION
9.S
It i~ clear that two dynamical systems (M, /l, cp) and (M', /l " cp '), which
are isomorphic under f (see Definition 4.1), induce operators V and V'
which are equivalent, that is, there exists an isomorphism F: L 2 (M, /l)
L 2 (M', /l ') such that V' '"' F V F- 1, according to the following diagram
-+
(see Figure 9.6). Thus, the invariants of V are certain invariants of the
dynamical system (M, /l, cp). They are called the spectral invariants. For
'instance, the spectrum of V is a spectral invariant. 4 A complete system
4
In
Ut ; or"again, the spectrum associated with the resolution of the identity E for
24
equivalent) does not imply isomorphism (see Chapter 2, 12, Entropy; Appendix 15, Anzai Skew-products).
cp'
Figure 9.6
9.7 (ERGODICITY)
(M, 11, ) is ergodic, if, and only if, 1 is a simple proper value 01 the
induced operator V.
Prool:
f. Now
is ergodic if, and only if, the invariant functions are all a.e. constant.
Since the functions that are a.e. constant are scalar multiples of one another, is ergodic if, and only if, the subspace' of-solutions of Vf = f
has dimension 1. In the continuous case, the ergodicity of t is equivalent to A = 0 having multiplicity one in the spectrum of V t'
ERGODIC PROPERTIES
25
The dynamical system (M, Il' t) is mixing if, and only if,
lim
(9.9)
T-+oo
< U/ I g >
Ut has properly continuous spectrum if its only proper functions are constants. It can be proved (see Halmos [1]) that a dynamical system has
properly continuous spectrum if, and only if, it is weakly mixing (see 8.9).
We turn to the case in which the spectrum of Ut is discrete.
EXAMPLE 9.10
= (). z,
()
e 2TTiW ,
tion zP, p ( Z:
=
W
sJ>
UzP
(Uz)P
Hence, the zP's are proper functions of U with corresponding proper values ()P. The set \zP I p.(
ZI.
A dynamical system
26
tem is ergodic if, and only if, the orbits are dense on M (see Example 7.8
and Appendix 1). Observe that ergodicity implies that all the proper values (}P are distinct and simple. The system is not mixing; take f = g =
zP
in (9.8); we get:
(Jpn
lim (JPn does not exist and (9.9) is not fulfilled. These results
n=oo
9.12
Izlz
(C,
Izl
1\;
Proof:
Since V is unitary, every proper value A has absolute value 1. It
follows that if f is a corresponding proper function
f(x) = Al(x) a.e. implies
If(x) I =
If(x)1 a.e.
I is con-
V(!!)
Vh
f
Vf
=
I'
27
ERGODIC PROPERTIES
VCt)
g~
= ::
A~-1 Ct)
This proves
(c).
Finally, if the system is mixing, take I = g equal to a proper function
with proper value in (9.9), we get:
lim < vn/l I>
n=oo
</11>'<11/>,
that is
lim
An
constant.
n=oo
9.13
(a) Two ergodic dynamical systems with discrete spectrum are isomorphic
ii, and only ii, the proper values 01 their induced operator coincide.
(b) Every countable subgroup 01 the circle group is the spectrum 01 an er-
28
10.1
We again consider Example (1.16): M is the torus l(x, y) (mod 1)\ with
its usual measure;
4)
is the automorphism:
.(x, y)
\e
p,q
(x, y) =
p, q (
Z\
is an orthonormal basis of L 2 (M, p.). The set D can be identified with the
lattice Z2
= \(p,
U induces an au-
tomorphism u on D:
Let us show that (0,0) is the unique finite orbit of u. Assume that (p, q)
( Z2 has a finite orbit. This orbit is a bounded subset of R2, invariant
under the linear operator of R2,
1>
has two proper values A1, A2, 0 < A2 < 1 < A1 Hence, 1> is "dilating" in the proper direction corresponding to A1 , and "contracting" in the
proper direction corresponding to A2 This implies that the only invariant
29
ERGODIC PROPERTIES
Let us go back to D
Ie p, q I p,
C.,
where f.l,n
Ii.l,n In ( Z\ ,
vectors of C i , then L 2 (M, /1) is the orthogonal sum of the H/s and ofthe
one-dimensional space of the constant functions. Each Hi is invariant
under V and has an orthonormal basis
Ii./,n In ( Z\
such that:
10.2
eM
eM
basis of L 2 (M, r11) formed by the function 1 and functions f.1,J. (i (1, j ( Z)
such that:
Vf 1,]
. .
f1,]+
. . l' for every i, j.
The flow is
10.3
00
-00
e 2TTit A dE (A)
.'
be the spectral resolution of Vt . It can be proved that (M, /1, q,t) has
30
continuous with respect to the Lebesgue measure. for every I ( L 2 (M. /L)
orthogonal to 1.
THEOREM 10.4
Lebes~ue
spectrum is
mixin~.
Prool:
From Theorem (9.8) we need to prove that:
lim
n"OO
for every I.
< Un I I ~ > = 0
are basis vectors. for the general case follows by continuity and linearity
If 1= I j i ~ = Ik r then
<I.J,n+.l/
J . k ,r >
which is null for n large enough.
COROLLARY 10.5
I (x. y)(mod
1) I
(Example 1.16) has Lebesgue spectrum (Example 10.1). Then. it is mixing and ergodic (Corollary 8.4).
EXAMPLE 10.6
Lebes~ue
spectrum.
In par-
\0.11.
31
ERGODIC PROPERTIES
Yn(x) =
if x = 0
if x = 1
1+ 1
nl
..... y
fIJc
Now. let
..
basis equivalent if some integer power of U carries one onto the other.
The function 1 constitutes its own equivalence class; the other basis
functions split into countably many equivalence classes. Each such equivalent class is in a one-to-one correspondence with Z: the action of U on
the class is to replace the element corresponding to n
Z by the element
corresponding to n+ 1.
To ~ummarize. there exists an orthonormal basis of L 2 (M. Il) consisting of the function 1 and of functions I.I,]. (i = 1. 2 ... ; j l Z) such that
UI 1,]
. . = l.I, J.+1
for every i. j. The number i is the number of the equivalence class. the
number j is the number of its element which corresponds, as described
above. to j
I 1, I;~j I in
11, I;,j I
..
10
L 2 (M 1.1l1) and
I~I, }+1
for every i, j. The isometry of L 2(M l' Ill) onto L 2(M 2 , 1l2) defined by
1 ... 1,
carries the spectral type of the first scheme into that of the other.
32
ll. K-Systems
In this section we define a class of abstract dynamical systems with
strongly stochastic properties.
DEFINITION
11.1
(a)
n
00
(b)
0,
nff =
n=-oo
where
0 is
(c)
i,
n (j' =
n=-oo
induced by .
n /i
(b ')
1.
t=-oo
00
(c ')
V /f.
t~-OQ
11.2
2.2)
I =~1ll, 0 = ;ll.
6 A. N. Kolmogorov [2) introduced this class under the name of quasi-regular sys
terns.
33
ERGODIC PROPERTIES
by the:
(Aj) = Aj+l
where is the shift. Hence ct is the algebra generated by the
k
At's,
1, and
ct c ct,
proving the property (a). On the other hand, every generator
q(A/)
Aj+q , i ~ 0, for q
A! of
1 is a
V nct
Let us now prove the property (c). Let
93
i.
93
for any B
there corresponds an N
f
-nct, n
~ N (exercise). Hence fl (A
and B (
fl (B) = fl (B
n B)
= [fl (8)]2 ,
n;; -net.
We conclude:
hence:
(Q. E. D.)
COROLLARY
n B)
n B)
11.3
34
Prool:
This system is isomorphic to B (V2, V2) (see Appendix 7).
1104
EXAMPLE
11.5
THEOREM
A K-system has a denumerably multiple Lebesgue spectrum. In particular, it is mixing and ergodic (Theorem lOA). This theorem is due to Kolmogorov [2] for K-automorphisms and to Sinai [6] for K-flows. We sketch a
proof for K-automorphisms, the complete proof will be found in Appendix 17.
H8UH
U-1 H8H
.........
.'
o. o.
0.00
000
000.00
:UH
U-1H
....................
x Uh 1
x hl
x U-1h 1
..........................
Hl)
Hd
x h2
000
00
oX
Uhj
x hj
II
Figure 11.6
n
~
U
n=-oo
n=-oo
= L 2 (M, f.L),
UnH
35
ERGODIC PROPERTIES
Ih.1
I
on the'orthocomplement H 9VH
Vh.,.... The H's are invariant under V, and their orthogonal sum is
I
Vjh.,
I
i(Z+,
j(Z,.
the e 1,1
. .'s and the function 1 constitute an orthononnal basis of L2(M,~)
such that:
~il1
be complete after it has been shown that the dimension of H 9 VH is infinite (see Appendix 17): V has a spectrum of infinite multiplicity.
12. Entropy
This section is devoted to the definition and the study of a non spectral invariant of dynamical systems introduced by A. N. Kolmogorov [4].
Throughout, z (t) denotes the function on [0, 1] defined by:
z(t) = {-tLOgt if O<t:;1
o if t = 0
where Log denotes the base-2 logarithm. We use the following properties
of z: z is nonnegative, continuous, strictly concave
< 0) ,
(z"(t) = - Lotge
(M -.U Ai)
1(1
if i f, j.
0,
II(A.nA.)=O
I
I
,..
be a finite
36
DEFINITION
12.17
EXAMPLE
IS:
~ Z(Il(A))
jll
Hence h({3) is nothing but the weighted logarithm of the number of the
elements of f3.
Observe that if two partitions are equivalent, that is, if their elements
coincide up to some sets of measure zero, their entropies are equal. In
particular, the elements of measure zero can be removed. Finally, h(a) =0
I'
e
Figure 12.3
I. A. Yaglom
[d.
37
ERGODIC PROPERTIES
means that a
RESPECT TO A PARTITION
{3
Let a = !Aili = 1, ... ,r\ and {3 = !Bjlj = 1, ... ,sl be two finite measurable partitions (in the future, for short: partitions). The entropy of a
with respect to (3 is defined by:
where
Il(A.nB.)
I}
Il(A./B. )
I}
1l(B.)
}
a induces on each B. a
finite measurable partition a B ' the elements of which are B. n AI' ... , B.
i
J }
n A.r After a suitable renormalization, B.} can be considered as a space
of measure 1 on which aBo has entropy:
]
h(aB ) = ~ z(Il(A./B.)).
i
I}
h(a/{3) = ~ Il(B.)h(a B )
}
Let a.
(12.6)
12.5
/3.
a~{3
(12.7)
(12.8)
(modO).
(12.9)
38
(12.10)
s h (aly) + h (f3/y) ,.
(12.11)
=v
in the above
h(a), we get:
(12.12)
a S f3 (mod 0)
h (a/ (3)
(12.13)
s h (a)
(12.14)
(12.15)
(a V (3)
(12.16)
h (a/ (3)
DEFINITION
a V f3 ,
h (aI m .
AN AUTOMORPHISM
Z+
LEMMA
12.18
39
ERGODIC PROPERTIES
Proof:
I sn I
is a nonincreasing sequence.
Proof:
According to (12.11):
(12.20)
sn
= h(cpnal a V"'V
cpn- l al .
Hence:
sn_l
h(cpn-lalaV"'Vcpn-2al;
THEOREM 12.21
Proof:
sn is a nonincreasing sequence of positive numbers: sn has a limit s.
n = oc 'n
s.
Theorem (12.21) follows at once from (12.20) and the very definition of
h (a, cp).
EXAMPLE 12.22. BERNOUILLI SCHEMES
40
A6
= Imlmo = il.
1, ... ,k.
1=
-~ Pi Log Pi .
Since n ..16
A~
are the:
n ... n Ain-I
n-l
'
'n
p .... p.
'0
'n-l
Log(p .... p.
'0
1.
'n-I
We get:
and by induction:
whence:
- ~ Pi Log Pi .
DEFINITION
bl.
,
[8].
'n-
) .
1
41
ERGODIC PROPERTIES
clear that h ()
O.
THEOREM 12.24
Let (M', /l', ') be a system isomorph.ic to (M, /1, ). (see Definition
-+ M', that is '= fe l . If
(12.16):
h((l,')
h(fa,fe l )
0=00
lim h[(aV"'Vo-la)]
n~oo
n:::oo
h(a,).
On the other hand, when a runs over all the partitions of M, fa runs over
all the partitions of M '. We deduce at OIlce:
sup h(a', ')
sup h(a, ) .
generated by a:
a is called a generator with respect to if:
00
n=-oo
,.
42
l:
h()
Pi Log Pi
i=1
Proof:
Let a' be the partition of 12.22, the elements of which are the
A~
Since
n A~
lmlmo
iI,
1.
1, ... ,k.
V..:ooo nm(a)
same entropy. Particular cases have been examined which indicate that
this may be correct. We mention a result due to Meshalkin. [1]: B(Pl' ... )
and B(ql' ... ) are isomorphic if they possess the same entropy and if the
P;'s, q;'s are negative integrC;lI powers olone and the same positive integer n. For instance, B('I" 'la, 'la, '10, '10) and B(';., ';., ';., ';.) are isomorphic
(here n
2). Blum and Hanson [1]. improved this result. In this direction
we mention a theorem due to Sinai [9]: . Two K-systems with the same en- .
43
ERGODIC PROPERTIES
If
rp
rp (x,
ad - bc
1,
where A1 is the proper value, whose modulus is greater than 1, of the matrix:
rp
rp
be an
h(rp)
I
Log
1\1
Ail>1
(See Genis [1] and Abramov [1] for a correction of the proof.)
COROLLARY 12.30 (SEE EXAMPLE 4.4)
; ) and (;
are nonisomorphic.
Those defined by:
(; D
and
(~
i)
possess the same spectral type and the same entropy. Whether they are
isomorphic is still an open problem. It is only known that they are weakly
isomorphic (Sinai [9])'.
THEOREM 12.31 10 ENTROPY OF K-SYSTEMS
44
Proof:
According to Definition (11.1) there exists a subalgebra d' of
such
that:
6=
1.
nd'
-00
(12.32)
k 93)
k$.n
93 c
k P,
n, n' (
Z.
k$.n'
93 c
k<n'
93.:
I
satisfying:
for every i. As
, we obtain:
93
(see I\ppen-
45
ERGODIC PROPERTIES
n=oo
V k
k<-1
(3) .
Now, it is sufficient to show thath ({31 Vk <-1 k(3) > O. Assume that
h({3!Vk <_1 k(3)
{3 ~
k{3 (mod 0) .
k~-1
Consequently:
that is
(Q. E. D.)
12.33
12.34
46
KOUCHNIRENKO'S THEOREM
sup
a classical
(independent of a)
satisfying:
S(a)
(12.36)
S(a)
< A
Hence, we obtain:
11
See Kouchnirenko
[11.
[2].
47
ERGODIC PROPERTIES
i = 1
i= 1
If we denote by S (a) the sum of the areas of the boundaries of the elements of a (each boundary is counted twice), clearly:
n
~ [/L(A i )](N-1)IN
:S
i= 1
Log l
i= 1
/L(A.)Log/L(A.)
1
< N. Log ,\ +
h (a,
S N Log ,\ ,
(12:38)'
constan~
48
(Q. E. D)
COROLLARY 12.39
22n-n:-1 hmes
COROLLARY 12.40
In fact the system is isomorphic to a compact, finite dimensional abelian group on which a translation
h (1))
1>
O.
A PROBLEM 12.41
1>
is an open question.
REMARK 12.42
='
C (1)),
= 1..(1))
REMARK 12.43
ERGODIC PROPERTIES
49
<h <
A-entropy is 0
< 2h <
00.
~ h,
Since 2h
Its
the horocyclic flow; however, they have both zero usual entropy and
countable Lebesgue spectrum.
REMARK
12.44
13
(a)
Sn~n = ~n;
(b)
(S)n
n
E (identity), (S )k ~ E for k
n
< n.
i = 1
Katok and Stepin have proved some important theorems which connect the
l3
Moscow, 1966.
50
transf~rmations
= O.
>
~l
~3
~3
~l
dt
dy
F(x, y)
dt
F (x, y)
1967.
Il>YHKl.\HOHaJlbHblH aHaJIH3
ERGODIC PROPERTIES
51
J.
CUAPTER
lI\STABLE SYSTE\IS
This chapter contains the study of classical systems with strongly stochastic properties, the so-called C-systems.
The orbits of a C-system are highly unstable: two orbits with close
initial data are exponentially divergent. This property turns out to imply
the asymptotic independence of past and future: C-automorphisms are ergodic, mixing, have Lebesgue spectrum, have positive entropy, and, in general, are K-automorphisms. The set of the C-systems defined on a prescribed manifold M is an open set in the space of the classical systems
defined. on M. Consequently, ev,ery classical system, close enough to a Csystem, is
a C-system.
13.1
: M ... M:
The length
IIXII
of a tangent vector
X is
Our terminology was introduced by Anosov; C is used because these systems satisfy a "condition C", in Russian J.J YCAOBHC y. 'H
.
53
54
(~ ~)
Hence,
has two real proper values Al and A2 , (0 < A2 < 1 < AI)' with
Y. Then:
II*(II
A1
'1I(1I
if (( X m ( \ > 1) ;
II"(II ::;
A2
'11(11
if (( Ym , (0 < A2 < 1)
Figure 13.2
55
UNSTABLE SYSTEMS
DEFINITION 13.3
fi~/ds
of tangent
110.
II(-n)*tll
11(n)*tll
:S b e-.\nlltll,
:S b e-.\nlltll, if
if
t ( X m,
t ( Ym
depend 2
but a and
contracting space.
Example (13.1) is a C-diffeomorphism:
a = b =
1,
.\.
= "1'
-.\
= "2
This definition extends to the continuous case (t (R): Let t be a oneparameter group of C 2 -differentiable diffeomorphisms of a compact, connected, smooth manifold M.
d~
(0)
(1)
k .;, 0,
dim Ym
I ,;, 0 i
And hence for every metric: Let gl and g2 be two Riemannian metrics on M.
Due to the
compactness
t(
f3
such that:
for every
TM . . Thus, if the inequalities 2 hold for
~ I with constants a and b, they still hold for ~2 with corresponding constants
(al (3) a
and
metric g.
(f3/ a) b.
56
(2) for any positive real number t and for some Riemannian metric g:
for t and
t,
but
and b depend on the metric g. Condition (D) means that the system has
no equilibrium position. Condition (2) describes the behavior of the orbits. A C-diffeomorphism or a C-flow will be called, for short, a C-system
REMARK
13.4
tively, the "most dilating" and the "most contracting" subs paces of
TM m );
M)~
13.5 (SMALE)
The Space M.
Let T2
Iu I 0
fy T 2 x
101
x, y), 1) ;:
where
1>
X+ y,
x+ 2y), 0) (mod 1) ,
1> : ( ; ) ->
nD(;)
(mod 1) .
57
UNSTABLE SYSTEMS
il
Figure 13.6
= ! u (mod
Hence, M is a fibre
bundle 3
1) l, p (x, y, u)
=u
The flow /.
We define a flow / by its infinitesimal generator:
x = 0,
(13.7)
= 0,
u=
1 .
<
A2
(~ ~).
We define a Riemannian metric on T2 x [0, 1] by:
3
compact, connected, smooth (n+ q) -dimensional manifold M; (ii) a smooth, n-dimensional manifold
whose rank is n everywhere and called the projection. The p-l(b) 's,
p,i-> B
b ( B, are
called the fibres. They are q-dimensional manifolds diffeomorphic one to another.
58
(13.8)
c)::-,vt" )
It is readily proved that this metric is invariant under the substitution;
x
->
x + y,
x + 2y,
->
->
u-1 .
I
111.
(M, t) is a C-f1ow.
x I ul
of :
: T2 x lui
(~)
(b
->
->
T2
x lui ,
n(~)
(mod 1)
R.
Hence, II(;)~II
(13.3):
a
= 1,
59
UNSTABLE SYSTEMS
->-00
(or +
00;
CPt m
Figure 13.9
13.10
where v ( V,
4
By a foliated manifold
over
60
~.
is a Riemannian submanifold of V.
~
at
The geodesic flow t of a Riemannian manifold V describes the movement of a material point on the frictionless manifold II without external
forces [(see (1. 7)], If V has negative curvature the geodesics are very un-
stable: if v, Vo ( Tl V , the distance ltV, tVol increases exponentially with t. To be precise, we. have the following result:
LOBA TCHEWSK Y - HADAMARD THEOREM 7 14.3
Let V be a compact, connected Riemannian manifold of negative curvature. Then, the geodesic flow on the unitary tangent bundle M
Tl V
is a C-flow.
~ppendix
V=
universal coyering of
V.
This theorem is "due to Lobatchewsky for surfac:.es of constant negative curvature. Hadamard
[1]
61
UNSTABLE SYSTEMS'
y'
positive
asymptotes
negative
asymptotes
Figure 14.4
Let y (u, t)
= y (t) = y
(14.5)
5. b e->'"
t ,
t ::::
0,
where the constants b and A are positive and independent for y, y', t.
62
we have:
integral manifolds are the horospheres S+ and S-. Both of these foliations are invariant under the geodesic flow, for the horospheres are orthogonal trajectories of (n-l) -parameter families of geodesics of V .
. We turn to prove that general C-systems admit two invariant foliations.
15.1
X and
and that are respectively tangent to the dilating field Xm and the contracting field Ym. Hence, these fields are always integrable.
(2) Every diffeomorphism ': M
-+
M, C 2 -cIose enough to , is a
X, and
~' of '
their discussion was concerned with the particular case of small perturba-
63
UNSTABLE SYSTEMS
15.2
cp*p(cp- 1 m).
(15.3)
for
IX -P 1 1 < 0, IX -P 2 1 < 0,
and
still holds for any diffeomorphism cp'. C 2 -c1ose enough to cp, since cp'*
is C 1-close to cp *. We deduce from the contracting mapping theorem that
a mapping verifying (15.3) admits a fixed point. The fixed point of the
mapping cp is X, but for cp' it is another field p'. Clearly:
p'
(cp ,**)n X ,
lim
n=~
cp'*p'(m)
p'(cp'm) ,
and the field p' is dilating for cp'. A similar study of (M, cp -1) leads to
the contracting field of cp' that is close to Y. Thus cp' is a C.diffeomorphism.
THE INVARIANT FOLIATIONS
15:4
and
p'
y,
64
But ( ,**)n X is the field of the i<Lplanes tangent to the foliation ,n:x: .
Thus, the limit field is. integrable, that is tangent to a certain foliation
:X:'.
This completes the proof of the part (2) of Theorem (15.1). As a matter of
fact, the previous argument proves that there exists a foliation
:X:.
Let us
(n
->
dim M).
Consider a foliation
:X:?1
!/J.(
C.);
which consists in planes parailel to !/J~I (X ml ). If the local charts
I
I
:X:?I
at m
C.I
is close enough to the dilating plane Xm ' for all' m. Obviously there are
se~eral planes.
X?(m)
passing through m if m belongs to several C I..
I
:X:7
M and (15.3) implies that their fields of tangent planes con verge to X as
->
+ 00. We deduce readily that there exists a limit foliation :x: tangent to
15.5
Each sheet of :x: is C1-differentiable. But the foliation :x: is not required to be smooth: the normal derivative to t!le sheets may not exist. If
k = I = 1, the field X is C1-differentiable (see Arnold and Sinai [6]). It
seems probable that :x: is not smooth in the general case. Anosov concocted an example where :x: and 'Yare not C 2 -differentiable.
REMARK
15.6
16.1'.
STRUCTURAL STABILITY
(A) Diffeomorphisms
-+
M a CT-differenti_
6S
UNSTABLE SYSTEMS
t/I (
that is: k.
t/I. k. In
of It/ln I n
I" In ( ZI
ZI.
\
d.,l., m I
at
'f't
t=
m (M.
16.2
y = -x-Ky,
Two mappings (or two vector fields) are C'-close if their derivatives of order
inferior to r+ 1 are close.
9
[1].
66
Consequently, two systems with distinct values of K would not be topologically conjugate and the focus (0,0) would not be structurally stable.
y
focus
Figure 16.3
UNSTABLE SYSTEMS
67
Figure 16.4
[1].
1 the dimension is greater than two, the situation is involved. For instance, the system (13.1) is structurally stable but very complicated 11 (ergodic, the cycles are everywhere dense, and so on).
On the other hand, Smale [2] constructed an example (M, ) such that:
dim M ~ 3, every diffeomorphism close enough to is not structurally stable (see Appendix 24). This invalidates the genericity of structurally stable systems.
ANOSOV'S THEOREM 12
16.5
11
12
See Anos ov
[11.
68
'==k..r 1
W()
< 1 .
mM
Let '
Z.
From the very definition of C-systems we see that there exists at most
one point m' satisfying (16.6). In fact, for any
00
~.f,
0, we have:
f3
X', close to m,
69
UNSTABLE SYSTEMS
for the precise meaning). It can be proved (Lemma B, Appendix 25) that
there exists, among these sheets, a unique sheet (j(m) such that its images
<O.
The same device proves that there exists a unique contracting sheet
8(m) (
'Y'
whose images cp '"8(m) are close to cp"(m) for all n (' Z. The
'Y'
unique point m' in the neighborhood of m. It is eas y to see that cp '(m ')
=
(cp(m)) , and that m' varies continuously with m; thus the mapping k:
C-FLOWS 16.7
Figure 16.8
A similar construction (Figure 16.8) gives two sheets, (3(m) and 8(m).
They are formed with orbits of cp; (respectively for t
which are
a~ymptotiC
->
70
: ( ; ) -THEOREM
G~ )(;)
(mod 1) .
17.1
(1'. 17.2
The matrix:
(~ ~)
y
x
Figure 17.3
f.
71
UNSTABLE SYSTEMS
(3) and four triangles (1, 2, 1 " 2 '), with sides parallel to X and Y (see
Figure 17.3). By pairwise identification of the opposite sides of the unit
square, 1 and 1', 2 and 2', and 3 give rise to three disjoint parallelograms
P1,P2, P3 on the torus M. Again split P1, P2 , and P3 into parallelograms
the sides of which are parallel toX and Y and whose lengths are inferior
V (p-n{3.
n~O
The subalgebra
ct
Appendix 18).
17.4
LEMMA
ct
ing direction Y.
(2) Let I be a given positive number. The measure (in the sense
of dxdy) of the union of those atoms whose length is inferior to I isbound-
ct.. is a
Proof:
B 1, B2
'f
{3.
Ai- 1 . A
(x, y) ~
M of
T ~ (1,0),
72
it
c (X, y) =
are six segments, the lengths of which are inferior to A and whose mutual
distances are greater than K - A, where:
K
as Ik 1 + 1 k'i
I<
'
e is the angle
1 1 A, A)
::; A.
-+
part of Lemma (17.4), Al > 1. Let L be the sum of the lengths of the
sides of the elements of {3 that are parallel to X. The analogous sum for
P{3, p ( Z, is
At L.
above by.
def
I-A -1
C.
73
UNSTABLE SYSTEMS
tracting sheets (see 15). This foliation is called ergodic if any union of
sheets, with positive measure, coincides with M up to a set of measure
zero.
LEMMA
17.6
Assume
A; .f.
A;' f
tradicts the obvious fact that the length of any sheet has to be greater
than 1.
V nU'
1.
n~O
V na
n~O
74
where the A;'s are segments parallel to Y and with length inferior to A.
Since A is small enough, then
Ao n Al n ... n n An
is a connected
-+
+ 00, we have A2 A
-+
-00
H is the union of atoms of (j', the union of atoms of -1(j', -2(j', and so
<
E p.(H),
where C is the constant of Lemma (17.4). Up to a set E of measure inferior to E p.(H), the atoms of (j' have a side parallel to Y whose length
is greater than I. Since -1 is measure-preserving and dilating in the
direction Y with ratio A;l, one concludes that, up to a set -nE of measure inferior to . p.(H), the atoms of -n(j' have a side parallel to Y
whose length is greater than
~-n
. 1. As n
-+
1. Whence:
00
(Q. E. D.)
f3.
A theorem due to
75
UNSTABLE SYSTEMS
17.7
If for any such pair of neighboring manifolds the map f has a continuous generalized Jacobian, and if for smaIl deformations of II' this Jacobian varies continuously, then the foliation is caIled absolutely continuous.
For both of the two foliations of a C-system (see 15), Anosov proved:
THEOREM
17.8
We cannot give here the proof of Theorem (17.8), which is too long. We
only announce the final results obtained with its help. The statement of
Theorem (17.1) extends and leads to the following theorems (Anosov [2]).
THEOREM
17.9
17.10,
(SEE SINAI
[11])
have Lebesgue spectrum (see 1O) and cannot be a K-flow (Theorem 11.5).
The following theorem proves that, in some sense, Example (15.5) is the
only one exception.
THEOREM
Let
1>t
17. 11
be a C-fIow on a compact, n-dimensional manifold M, then:
(1) either
(2)
1>t
1>t
is a K-system; or
In case (2), such a proper function is continuous and there exists a compact, (n -1) -dimensional submanifold V of M and a C-diffeomorphism
76
1>:
->
V such that
1>/
1>
t (t ...
C t, C = constant).
COROLLARY 17.12
The geodesic flow on the unitary tangent bundle of a Riemannian manifold 'V of negative curvature is a K-system.
Proof:
According to Theorem (14.3), the geodesic flow
from Theorem (17.9),
1>/
1>/
is a C-flow. Hence,
is ergodic.
T2 since the Gauss-Bonnet formula implies that the E;uler-Poincare characteristic is negative. Consequently, according to Corollary (A 16.10) of
Appendix 16, the only continuous proper functions of
1>/
are constants.
The second case of Theorem (17.11) cannot occur and this theorem proves
that
1>/
is a K-flow. 13
Thus,
1>/
14 This result is due to Gelfand and Fomin [2] for arb~trary dimension and constant negative curvature.
15 This result
1S
due to E. Hopf
77
UNSTABLE SYSTEMS
and wall, are supposed to be perfectly elastic. Sinai has proved ergodicity17 of this system on each manifold T
constant ~ O.
two~dimensional
we consider one of the particles as fixed. The second particle (which can
now be regarded as a point) moves on a "torus billiard table" (Figure 18.1),
being reflected from
t~e
f3."
Let us, at the same time, consider an elliptic billiard table (Figure
I
I
I
I
I
I
-;------------I
Figure 18.1
17
This result was conjectured as the "quasi-ergodic hypothesis." At the beginning of Ergodic Theory this conjecture occasioned sharp discussions. But the
problem was, perhaps, overvalued: statistical mechanics deals with asymptotic
behavior as N -> +00 (N = number of parhcles) and not as I'" +00 for fixed N.
78
point moves along a geodesic, passing at each reflection from one side to
the other. In the same way a torus billiard table can be regarded as a twosided torus with a hole on which the point moves along a geodesic. But if
the two-sided ellipse is a'n oblate ellipsoid, the two-sided torus with a hole
will be an oblate surface of genus 2. Thus, motion on our tor~s billiard
table is a limiting case of geodesic flow on a surface of genus 2.
Figure 18.2
We now turn to our billiard tables and consider the curvature to which
Figures 18.1 and 18.2 correspond. The ellipsoid has a positive curvature
whose integral is equal to 411 (Gauss-Bonnet formula). On flattening the
ellipsoid, all the curvature is accumuiated along the boundary of an ellipse.
For a surface of genus 2 the integral of the t:urvature is equal to -411. Thus,
a two-sided torus billiard table can be regarded as an oblate surface with
negative curvature everywhere: on flattening, all the curvature is accumulated along the circumference.
The preceding arguments (Arnold [4] p. 184) are not, of course, a proof
of ergodicity, not even in the simplest cases we considered. Nevertheless,
using methods and notions dealing with C-systems (asymptotic orbits, transverse foliations), Sinai [4], [5] succeeded in proving that the BoltzmannGibbs model is ergodic on each submanifold T
more, is a K-system.
constant
f.
0 and, even
UNSTABLE SYSTEMS
79
304.
Sinai, Va, Dynamical Systems with Countably Multiple Lebesgue Spectra
CHAPTER 4
STABLE SYSTEMS
Many dynamical systems are known, the orbits of which, with remarkable stability, fail to fill up the "energy level"
H =, Ct -ergodically
and remain (to the end) in their particular corner of phase-space. The systems that are close to an "integrable" s'ystem, and the systems to which
the Theory of Perturbations of Celestial Mechanics applies, belong to this
class. Let us mention, for instance, the three-body problem, th: fast rotations of a heavy-rigid body, the motion of a free point in a geodesic on a
convex surface, and the adiabatic invariants.
Only in recent times, "beginning with the work of C. L. Siegel [1] (1942)
and A. N. Kolmogorov [5] (1954), has some progress been made in studying
these systems. This chapter reports on the present status of this problem.
We refer to V. Arnold [4], [5] for further details.
Let us begin with an example.
19. The SWing and the Corresponding
Canonical Mapping
EQUATIONS OF THE MOTION 19.1
q = p,
2
P = -w Sin q,
where w is the "proper frequency" that depends on the length of the penduillm.
A swing is a pendulum whose length I periodically varies (Figure 19.3).
Its equations can be written as
81
82
Figure 19.3
(19.4)
q = p,
_w 2 (t) sin q,
where w(t+ r) '" ,w(t). Appendix 5 studies (19.2) by using the phase-plan.e.
Equations (19.4) contain the time variable t explicitly.
Thus, we deal
Figure 19.5
83
STABLE SYSTEMS
THE MAPPING T 19.6
q(t). The time periodicity of Equations (19.4) allows one to identify the
surfaces t
=0
and t
the space R1 x Sl x Sl
T of the surface
~ (t= 0)
into itself:
Clearly:
p (nr), q (nr) = Tn(po' qo) ,
and the study of p (t), q (t) as t ... + oc reduces to the study of the iterates Tn, n ( Z. The mapping T is canonical, for the Equations (19.4)
are canonical. Therefore, T preserves the area dp 1\ dq.
The equilibrium positions p
= 0,
Let us
be~in
"inte~rable
with the
case" (,)
ed with the mapping T. In this case, the system is conservative; thus, the
energy is invariant. In other words, on the surface ~,the curves (Figure
AS.l, Appendix S):
r:
'hp2 -
(,)2
cos q
< (,)2).
r[
r[.
The coordi-
r[
84
=0
CUT
as 11
A=
belongs to a
= (p, q)
a fixed point of Tn, and the orbit of x consists in a finite number of points
(Figure 19.8). If the angle A (I) that corresponds to 11 is
incommen~urate
Figure 19.8
= q =
IXol = jp~ + q~
IT" xol remains small for any
n ( Z.
1): It
= p,
fJ
= _CU 2 q
0 is stable: if
is a rotation with
frequency CU (see Appendix 5). This gives rise to a rotation with angle II)T after
time
T.
85
STABLE SYSTEMS
Suppose that
o<
The mapping TE , which corresponds to
T. This mapping
WE'
1,
v = 27T/r.
rl
-+
+ 00 and E
(1) E
I,
T; as
Inl
(2) E I,
<n <
< E- k
00
(Theory of Stability);
(Asymptotic Theory of the k-th
approximation).
If E is small enough,
curves
t~an
wg
cos q
:S
wg),
of T. Besides, for E
fill up the domain interior to the separaup to a set whose Lebesgue measure is
r I , whose angles
27T, do
not collapse but are only slightly deformed. The images TEnx of x
are all contained in
t ~
~.
the space p, q, t (see Figure 19.11). Therefore a trajectory of (19.4) peginning in a gap between two invariant tori cannot pass out of this gap.
Thus, Theorem (19.10) allows us to reach conclusions regarding the stability of motion.
2 "That is a conditionally periodic motion of the swing.
86
Figure 19.11
understand better the structure of the gaps between two invariant curves
r;.
The
STABLE SYSTEMS
STABILITY OF THE LOWEST POSITION OF EQUILIBRIUM 20.2
If E
=T
= q =
= 21T(CU/V).
2cu
2cu 2cu
= -1, 2' 3 ' ....
= q =
swing alters (and the swing starts oscillating) if one deflects during an integral number of half-periods of the proper oscillations. This result!s wellknown empirically. 3
2w
20)
,Figure 20.4
3 Besides, observe that the amplitude of the oscillations of system (19.4) remains
small if e is small enough. For, according to Theorem (19.10), the orbit remains
between two curves ~. This follows from the nonlinearity of our system: the frequency A(1) depends on the amphtude; condition (20.3) is no longer fulfilled since
the oscillations do not have tIme enough to amphfy.
88
dA /
dT r O.
217 E!.. ,
n
r.
T and close to r: namely, the curves r+ and r-, with angles of rota-
tion A+ > A> A- (Figure 20.6). Thus, the mapping T" rotates r+ posi-
r-
tivelyand
negatively.
Figure 20.6
This property still holds for Ten if E is small enough. Thus, on each
radius
1'" n
r.
:s 277)
form a
89
STABLE SYSTEMS
TenRe cannot be surrounded by Re and, conversely, Re cannot be surrounded by TenRe' Thus, Re and ~nRe intersect (Figure 20.7).
The
points of intersection are fixed points of Ten, for Ten moves each point of
in the neighborliood of
r.
Figure 20.7
FIXED POINTS OF THE ITERATES OF
What is the type of these fixed points? Are they elliptic or hyperbolic?
If = 0, then they are all parabolic with proper value '\12 = 1. Consequently, for small enough,. '\12 "" 1 and the hyperbolic case with reflection is impossible. On the other hand, consider the "radial displacement":
90
~n.
= d~/d
In the "generic
is even.
Let us make correspond to any point x the vector whose extremity is
Te n x (see Figure 20.7). It is readily seen that the index (Appendix 27) of
this vector field at a fixed point is:
Ind
sign of ( dA d~ \
dl
d)
Thus, half of the fixed points have index + 1 and the others have index -1This means that half of these points are elliptic and half of hyperbolic
type (an elliptie point has index + 1, an hyperbolic point has index -1).
Elliptic and hyperbolic points are illustrated in Figure 20.7.
Now, consider an elliptic~ fixed point: Ten x
x. The orbit of x is
All points of the orbit of x are fixed points of Ten and are elliptic since
they l1ave the same proper value. Hence, the set of the elliptic fixed points
splits into orbits consisting of
bits, then there are
.0
91
STABLE SYSTEMS
tween these curves, and so on. Between these islands and curves
, re-
main zones around the hyperbolic points. In fact,4 the separatrices of hyperbolic fixed points of the Ten,s intersecting each other create an intricate network, as depicted in Figure 20.10. On discovering this, Poincart:
wrote ([2], V.3, Chap. 33, p. 389):
"One is struck by the complexity of thIs figure that I am not even
attempting to draw. N othmg can give us a better idea of the complexity of the three-body problem and of all problems of dynamICS
where there is no holomorphlc integral and Bohlin's series diverge."
Figure 20.10
92
REMARK
20.11
The above argument does not prove there exist infinitely many elliptic
islands for a given 1. Poincare's last geometrical theoremS proves
that there exist infinitely many fixed points of ~n (n
->
0.1
o
-0.1
0 0.1
Figure 20.12
Yr-~-.-r-.--~~~-r-.--r-~-r~-.r-~
.5
.4
"-,3
..
"
...: "
-.5-.4-.3-.2-.1
,/
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Y
Figure 20.13.
93
STABLE SYSTEMS
(Theorem
19.10). Perhaps some of these points are not elliptic but hyperbolic with
reflection. Numerical computations 6 seem to support this conclusion.
y
0.8
0.6
0.2
o
-0.2
-0.04
-0.6
-0.8
-0.8 -0.6 -0.04-0.2
Flgure 20.14
The example we considered in 19 and 20 is a particular case of a situation which occurs for each system close enough to an "integrable" system.
6
b);
[3];
94
If one takes a look at the "integrable" problems of Classical Mechanics,7 one finds that, for all of these problems, bounded orbits are either periodic or quasi-periodic. In other words, the phase-space is stratified
i~to
E'XAMPLE 21.2
Assume the phase-space Q is the product of a bounded domain Bn of
(P l '
, .. , Pn )
be coordinates on Bn and q
(ql' "" qn) (mod 277) coordinates on Tn. The Hamiltonian equations, with
f 0,
"
7 For instance, the motion of a free point along a geodesic on the surface of a triaxial ellipsoid or a torus (see (1.7) and Appendix 2), a heavy solid body (Euler,
Lagrange, and Kovalewskaia cases),
95
STABLE SYSTEMS
(21.5)
oH I
oq
---,
For most initial data, A. N. Kolmogorov [6] proved that the motion remains
quasi-periodic (see Theorem 21. 7). Consequently, (21.5) is not ergodic on
the "energy surface" H = Ct and, among the ergodic components, there
are components with discrete spectrum, the complement of which has small
Lebesgue measure as H I is small.
Assume that the function H (p) is anal ytic in a c0cnplex domain [0] of
the phase-space:
O(~p,
I~ pi
'R q, (0,
< p,
I~ ql
< p) .
(21.6)
(U
I~
0
lo2H
Det __0_
op
op2
oJ- O.
the invariant torus To(w*) of the unperturbed system (21.3) are p = p*,
where wo(P*) = w*. Thus, the system (21.3) has frequencies w* on
To(w*).
THEOR.EM
21.7
If HI is small enough, then for almost 9 all w*, there exists an invari-
ant torus T(w*) of the perturbed system (21.5) and T(w*) is close to
To (w*). To be precise:
That is, (w, k) ,;, 0 for all integers k.
All, except for a set of Lebesgue measure zero.
96
For all K> 0 there exist E> 0 and a mapping p = p(Q), q = q(Q) of
an abstract torus T =
I Q (mod
21T)} into
n such that,
according to (21.5)
Q = w*,
and:
Ip(Q)-p*1 <K,
Iq(Q)-QI <K,
if:
holds in [n].
Moreover, the tori T (w*) form a set of positive measure whose complement set has a measure that tends to zero as
IH 11
->
O. Proof of Theorem
The do-
mains of the complement set are toric annuli betwee~ these invariant tori
(see Figure 19.11).
do not belong to the tori T (w*) can travel very far along H
h (see 23).
three-body problem. 10
10 A. N. Kolmogorov [7].
11 V. I. Arnold [5].
97
STABLE SYSTEMS
- O.
the Lagrange periodic solutions for the reduced problem of the three-body
(plane and circular).
V. Arnold [8], [9], [10], studied the generation ~f new frequencies from
the perturbation of degenerate systems. As a corollary, one obtains the
For in-
stance, for systems with two degrees of freedom, it is sufficient (hat 333
derivatives exist!
(D) INVARIANT TORI OF CANONICAL MAPPINGS 21.9
Theorem (21.7) can be reformulated by using the construction of the
98
Consider a sub-
0,
f.
-->
~2n-2 is well
ql = O.
Since
then P2' ... , Pn; q2' ... , qn (mod 211) are "action-angle" coordinates in this
neighborhood. The mapping A is canonical (see Appendix 31).
Now, consider the unperturbed system (H 1
(21.10)
-->
invariant tori of
d'
ping
d':
tori of (21.5) and there is a theorem, similar to Theorem (21.7), for mappings (see Theorem 21.11).
Let
Assume that B: p, q
-->
that is:
Pdq=~Pdq,
Y
By
99
STABLE SYSTEMS
[a] of a:
Let A: p, q
-+
ic function w(p) in [a], and To(w*) the torus p = p*, w(p*) = w* that
is invariant under A..
THEOREM
21.11
12
all w *, there
-+
A, such that:
O(Q+w*)
8A0(Q),
01-1
Q - Q+w*,
and:
Ip(Q)
-p*1 < K,
Iq(Q)-QI<K,
provided that:
holds in [01. Moreover the tori T (w*) form a set of positive measure
whose complement set has measure tending to zero as
Ip' - pi
+ Iq' -
ql
-+
O.
Theorem (19.10) is a direct corollary of Theorem (21.11): n = 1. Theorem (21.11) has been known since 1954, though its proof was never published.
J.
Moser [1] gave a proof for mappings of the plane (n = 1). This
proof makes use of the topology H2. Appendix 34 gives a proof for arbitrary n: the topological part is reducl'd to the technique of generating
functions of global canonical transformations (see Appendix 33).
12 All, except for a set of Lebesgue measure zero.
100
I ~; I
Det
ji 0 ,
I aw I
ji 0
ap
(21.13)
Det
a2H 0
ap2
ji 0,
Det
a2H0
aHo
ap2
ap
aHo
ap
.;, 0
These conditions (21.13) are clearly independent. Each of them is sufficient to emmre that there exist invariant tori. The second one ensures,
additionally, that there exists such tori on each "energy surface"; this
implies the stability (see Figure 19.11) for systems with two degrees of
freedom (n = 2 for Theorem 21.7, n = 1 for Theorem 21.11).
In most applications, both of the conditions (21.13) are simultaneously
fulfilled or invalid.
101
STABLE SYSTEMS
II
)1
= ([1' .. ,1 1
(22.2)
It is, obviously, a generalization of system (21.3): each torus I = con-
{~
I
w([)+ Et(I,)
EF(I, )
where {
1> + 277)
:=
F([, ),
E 1.
1. Notable ef-
fects, of order 1, of the evolution appear only after a long enough time:
t '" 1/ E .
Perturbation Theory proceeds to study the perturbed system as follows.
Let
fi(n
be the mean:
j = Efi(]).
(22.4)
For E
(22.5)
Ho).
13 Thls method goes back to Lagrange, Laplace, and Gauss, who used It in Celestia I Mechanics.
102
I (t)
I[(t) 71(t)1
be~mes
(U
~ 0 then:
< 1.E
2 = 12 ,
11 = E,
i2
= Eacos(1-2)
11 = E,
12 = 0
0,
1
2 = arcos a
1 (t)
-----
--....,.
~
~
~
--....,.
--....,.
---+
---+
~
(t)
-+
~
11 , J 1
Figure 22.7
STABLE SYSTEMS
103
Then:
1.
Thus,
I[(l/s) - ](l/s)1
In other words, after the interval of time
lis,
1.
the averaged motion loses
any relation with the real motion which remains locked in by the resonance
wI = w 2
For instance, there exist attracting tori of (22.3) which correspond to the
[2], [5], and 1._ Kupka [1] proved that attracting tori still exist for perturbed
systems.
This approach does not apply to Hamiltonian systems because attracting points do not exist according to the Liouville theorem (see 1.10).'
(2) One can study the relations between I (t) and] (t) for most (in
the sense of measure theory) initial data, neglecting points that correspond to resonance. For instance, Anosov [3] and Kasuga [1] proved
theorems of the following type:
Let R ( s, p) be the subset of
I[(t) -](t)1
for certain 0
>P
-+
This approach allows one to obtain similar results for systems much
more general than (22.3); whence its weakness: estimates of the measure
104
of R (f;., p) are not realistic and one has no information concerning the motion in R(E,p).
(3) One can study passages through states of resonance,
(.4) One restricts oneself to Hamiltonian systems to obtain more information.
(D) PASSAGE THROUGH STATES OF RESONANCE 22.9
1>2 = 12,
il
= E,
i2
= E cos(1 -2)
j2=0.
J 1 =E,
-- -- ---- -
-~
,~
----.
--
-----
Figure
--
2(O)
22.10
I (t)
= 11 (0)
12 (0)-1
= w2:
O.
1/2(t}-11
fiEjTcosx2.dX,
o
T=..jE/2t.
l?
105
STABLE SYSTEMS
For t
liE, obviously,
I/( t) - J( t) I =
C YE.
cp(O). The scattering of 12 after going through the resonance is of the order of
Vc (see
quencies (k
If the quantity:
(22.11)
< t < !.
E
changes sign at 11
A
t.
t.
0 is violated:
tem.
The idea used in proving (22.11) is that the scattering produced by
each resonance is of the order C YEand that, among the infinitely many
resonances
w/ w2
< In !. )
E
106
Passage through resonance!': for systems with more than two frequencies (k > 2) has not been studied.
(E) EVOLUTION OF HAMILTONIAN SYSTEMS 22.12
= p, = q, k
= n:
aH 1
al
w(I) +' E - -
~, i
- E--
aH 1
a
where
w
The averaged system is ] = 0, for
o.
In other words, there is no evo}ution for nondegenerate Hamiltonian systerns: ] = constant.
Theorem (21.7) of conservation of quasi-periodic motions rigorously
establishes this conclusion. In fact, Theorem (21. 7) implies that:
2, and
a2 H o
aP
E<
Eo(K)
107
STABLE SYSTEMS
and for most initial data in the general case). This illuminates the part
that conditions of conservation play in Theorem (21.7): they prevent evo1ution. ls In the same way, evolution is prohibited in Theorem (21.11), for
the mapping is globally canonical.
On the other hand, one also understands the part that the condition of
nondegeneracy plays. In fact, in case of degeneracy, the generic orbit of
the unperturbed system is ergodic on k-dimensiona1 tori (k strictly inferior
to n). In such a case, the algorithm of perturbation theory allows one to
predict the averaging on the k-dimensiona1 torus. Hence evolution becomes
possible, even for canonical systems.
IS This particular feature of canonical systems already comes out from simple examples. Let us consider the following perturbations of a center (Figure 22.13):
{t
-x - Ey
x.
Figure 22.13
The first perturbation, which is canonical, moves the orbits along a direction
which is orthogonal to the perturbation, and does not give rise to evolution. The
second perturbation, which is not canonical, gives rise to an evolution to zero.
Volume-preserving examples with evolution can b~ constructed in the four-dimens~onal
space x, y, z, u:
x = y,
-X-E Y,
.i
u,
ti
-Z+EU.
108
EXAMPLE
22.14
<P ( Tn
a2H o
-f, O.
a1 2
This system has the form (22.3), with k
< n,
system is:
J0
= (jk+1' ... ,
J)
where
W k) ,....,
"slow" frequencies (w k+ l' ... , Wn ) ,...., E that arise from the averaging system.
In the general case, when the averaged system is not integrable, the
relation between the solutions of the perturbed and the averaged systems
is still unknown even for 0 < t
approaches 2 and 3 (22.8). Moreover, observe that, even for nondegenerate systems, we need a study of the motion in the zones of instabilify
16 V. Arnold.[101. [4].
109
STABLE SYSTEMS
(complement set of the invariant tori) for n> 2, at least for t '" 1/E: (or
t '" liE: m).
invariant tori of "elliptic" or "hyperbolic" type that generalize, in arbitrary dimension, periodic motions of 20. If n > 2, recall that n~dimen
sional invariant tori do not divide the (2n -1) -dimensional energy level
H
The
-00
<t <
00.
According to
Theorems (21.7) and (21.11), this system is stable for most initial data
(the corresponding motions are quasi-periodic). The secular changes of
I (t) have the velocity exp (-1/ {E.) and consequently cannot be dealt
23.1
= M+
n M-,
where
17 One can find motivation in V. Arnold [I4]. Since this was written, the proof
was given independently by V. K. Melnikov [2],
J.
[1].
18 Example (23.10) is rather artificial, but we believe that the mechanism of
"transition chains" which guarantees that instability in our example is also
applicable to the generic case (for instance, to the three-body problem).
110
Ix{t) - TI
-+
-+
-00:
0 for x(O) ( M+
-00
lim Ix{t) - TI
+ 00
t ......
x=
(23.2)
A x,
in the system:
z =
= - f.L y,
if> =
0,
(U
(A, f.L > 0, cp(mod 211) ( Tk, (U incommensurate) defined in the space R/+
x R/- x RIo x Tk has a (/+ + k)-dimensional whisker
(y = z = 0) and
n.
spiral
For instance, 19 a
Chapter 3). If the whisker~ torus T has the property that the images of
an arbitrary neighborhood V of an arbitrary point ~ of one of its arriving
whiskers M- obstruct the departing ~hisker M + at an arbitrary point TJ of
M+, then the torus will be saip to be a transition torus (see Figure 23.4).
LEMMA 23.5
The torus x
= y =
Proof:
We set ~ = (0, Yo, 0, cpo),
TJ = (Xl' 0, 0, CPl).
CPo + (Uti
-+
19 Articles by Sitnikov
[1]
and A. Leontovich
[11 are
U
00
V(t)
111
STABLE SYSTEMS
we denote the set of all points of all the orbits emanating from U. Then
contains the set of all the images gtlV, where gt is the transformations
I-------....,..T)
Figure 23.4
group' defined by (23.2). For t; large enough, these images gt, V intersect
the neighborhood of TJ (because ,\ > 0). The intersections have equations:
Y = Y,
J
Thus
e -Ilt; Yo ... 0
Assume that the dynamical system has transition tori T1, T2, ... , Ts'
These tori will be said to form a transition chain if the departing whisker
112
Figure 23.7):
Mi nM;
M; M;_l
Figure 23.7
LEMMA 23.8
Let T1 , T2 ,
.. ,
(0)
V,
(t)
M; :
V for a certain t.
Proof:
Consider the future U ~
torus, then
Uobstructs Mi
Ut > 0 V(t)
at ~1 ~
of V. Since Tl is a transiti~n
Mi nM;.
Thus, M; intersects
M;
at TJ.
(Q. E. D.)
20
113
STABLE SYSTEMS
where J.L E
THEOREM
1.
23.11
< A < B. For every E> 0 there exists a J.Lo = J.Lo(A, B, E,)
> 0 such that for 0 < J.L < J.Lo the system (23.10) has a solution satisfying:
12 (0) < A, 12 (t) > B for a certair) t.
Assume 0
Ts'
LEMMA
23.12
(23.10).
In fact:
(1) Tw is clearly an invariant torus of (23.10);
11
(3) for J.L
t-
=+
-
2"jEsin .p1 ,
2
0 and small enough, the whiskers still exist and can be found
[tl.
114
LEMMA 23.13
Assume A < w < B.
Iw - w'l < K,
-1/ Vs).
Lemmas (23.12) and (23.13) imply that the whiskered tori T , ... , T
Wi
(w. irrational,
I
Iw. 1
(J.).
1+
11
< K,
w1 < A, ws > B)
Ws
APPENDIX]
W f
! x (mod
is irrational.
Proof: Assume w rational:
Let w = pi q where p, q .f Z, g > O. Here q is the identity transformation. Since every point of our circle is left fixed by q, every orbit
is closed and consists of a finite set of points.
Assume w irrational:
Let x be an arbitrary point of 51. The n x's are distinct for
Setting
In - ml
115
116
Let Tn
x
->
Rn /Zn
<P
be the translation:
APPENDIX 2
circle of radius r about a line Oz in the plane of the circle at the distance
1 (> r) from the center of the circle. Equations of V in geographical coordinates are:
x ~ (1 + r cos
t/J)
cos
if>
y ~ (1 + r cos
t/J)
sin
if>
z
where
t/J,
r sin
-",
\
\
\
\
Flgure A2.1
117
118
Figure A2.2
h = constant.
under rotations
-+
Y2"
APPENDIX 3
The rigid body is a system with three degrees of freedom and a six-dimensional phase space. There exist four independent single-valued first integrals: the energy T and three components of the angular momentum vector m. These four functions of position in the phase space do not change
their values for a given motion. The points of the six-dimensional phase
space for which the four functions have given values form, in general, a
two-dimensional manifold M. These manifolds M(T, m) are tori. In fact,
M is invariant, and so the phase-velocity vector at each point of M is tangent to M; consequently M admits a vector field without singular point.
It is evident that M is orientable and compact. The only compact two-
motion is, in general, quasi-periodic and the orbits are dense on M. Its
two periodic components are called, respectively, precession and rotation
(Euler [1]).
119
APPENDIX 4
GEODESIC FLOWS OF LIE GROUPS
(See Example 1. 7)
The geodesic flow of a Lie group carrying a left-invariant (or rightinvariant) metric has important applications:
The geodesic flow of the connected component SO(3} of the group of
the rotations of the three-dimensional Euclidean space corresponds to the
rotations of a heavy solid moving around a fixed point. Each orbit corresponds to a motion.
The homotheties of positive ratio and the translations of the n-dimensional space Rn form a group that generates the geodesic flow of the
(n+ l}-dimensional space of constant negative curvature.
< V,
>
Ii) v2 dx
cosy,!... Vy
O.
120
APPENDIX 5
THE PENDULUM
(See 1.13)
The equation of the pendulum is:
q+ k
sin q
0, where k is a posi-
q ~ p
p=
-k sin q.
The Hamiltonian function is H = (p2/2) -k cos q and Figure (AS. 1) depicts the orbits. The system is invariant under the symmetry p'" - p and
the translations:
(q, p) ... (q + 2K7T, p),
Z.
Figure AS.!
121
122
The points (kl7,O) are critical points: points (217k, 0) are centers (stable
equilibrium) and points ((2k + 1)17,0) are saddle points (unstablE! equilibrium). The orbits split into three types: the orbits 1 (small oscillations);
the separatrices 2 joining two saddle points; and the orbits 3 (complete
rotation around the hanging point).
The natural phase space is no longer the plane (p, q) but it is rather
the cylinder (q mod 17, pl. This is an example of global Hamiltonian flow.
APPENDIX 6
MEASURE SPACE
(See Chapter 1, Section 2)
An a-algebra
93
93, 11)
93
'B
is called
measurable subsets of M and 11 is the measure. (See Halmos [2] and Rohlin
[3j for these concepts.)
In fact, the object of interest is not (M,
93, 11)
(M,I1) of measure spaces. Let us make this point clearer. Let A and B
be elements of
93,
n B)
~ O. This
;.
93
93
un-
imply
M - A1
M - A2 (mod 0) .
93
(mod 0).
123
93, 11)
by the study of
124
(M, ~ (mod 0), p.), which we still denote unambiguously by (M, p.), because
the function p. defines completely ~ (mod 0) (but not ~). In other words,
we identify (M,~, p.) and (M, ~ " p.') if their measurable a-algebras ~
(mod 0) and ~' (mod 0) coincide.
Let (M, p.) and (M', p.') be two measure spaces. A mapping : M
-+
M'
:B ' (mod
0)
-+
:B
(mod 0).
If (M, p.) = (M', p.') then is called an endomorphism (mod 0). If both
of the mappings and -l are homomorphisms (mod 0), then is called
an isomorphism (mod 0); if, additionally, (M, p.) and (M', p.') coincide,
then is called an automorphism (mod 0).
APPENDIX 7
De/inition 01 I. Let m
f(m)
a_ 1 , aO' a 1 ,
...
(A7.1)
= ... ,
x=
i= 0
a -,.
00
y=
2i+1
~
i= 1
f(Aj)
126
consists in 21il rectangles, the sides of which are 1 and 1/2I i l+1. Thus,
we get:
where
ai
al _ 1,
00
k=O
that is to say:
f (2x, 'hY)
if a o
0, i.e., 0 ~ x
Consequently: I 1- 1 = '.
< 'h
~ x < 1.
(Q. E.. D)
APPENDIX 8
is:
M=
of
YH.
{(x,
The matux
<
The line
y = ("2 - 1)s,
s (
rii
->
M.
This
we have:
and 0
<
127
128
N-l
1N ~
f(rpnm) =
~ ~
n=O
e 2TTixA,f.
n=O
0, there-
n=oo
fore:
N-l
lim
f(m)
N-++oo N
f(rpn m)
1 .
n= 0
Thus, whatever the point m of the dense subset y be, we have: [*(m)
though f is analytic and
rp
is classical.
f T,
APPENDIX 9
I z ( C, \z\
I\, (z)
8 z, 8 =
e2lTiW ,
w is irrational
Prool.
lsi ca~e: i(z) = zP, P (
N-I
~ ~
N-I
~ ~
I(n z ) ,=
n=O
z.
We get:
(8 n z)P =
1
1
N-
n=O
2nd case:
'" I(z)
zP
eN P - l
8p -1
if p =
if p ~ 0
{I
if p = 0
p (Z,
129
Z (
M,
ifp~O.
so we get:
130
in which
*
i(z)
BO
r.
and
j(Pe+(z)-Pe-(Z))d ll
<
E.
N-l
~ lim sup 1
N ~
N~oo
f(n z)
~ + dll
*
i(z)
= f
(Q. E. D.)
APPENDIX 10
tl-~
n-1
are nonnegative (tl is the Laplacian _Va Va' R is the scalar curvature).
In particular (L. W. Green):
131
APPENDIX 11
and e 277ix means (e277ixl, ... , e277ixn). The measure of M is the usual
THEOREM.
Z and k ( Zn imply k = O.
Prool:
Let ( be a measurable invariant function. Its Fourier coefficients are:
ak =
e- 277ik
x I(x)d/l
M-
Z.
132
133
APPENDIX 11
If the
w;'s
e 21Tik x is a non-
t:
e 21Tix
->
e 21Ti (X + tw) ,
we have a similar result: (M, /L, t) is ergodic, if, and only if, k ( Zn
and kw = 0 imply k = 0 (or if, and only if, the orbits are everywhere
dense; see Jacobi's theorem).
APPENDIX 12
A12.1
, An abstract dynamical systeJll (M, /l, /) is ergodic if, and only if, the
sojourn time reT) in an arbitrary measurable set A of an orbit
I /x I 0 $. t s; TI
is asymptotically proportional to the measure of A:
(A12.2)
dT)
T->oo
11m - -
It (A),
*
Assume (M, /l, /) is ergodic and A is measurable. We have f(x)
for every f ( Ll (M, It) and for almost every x ( M (see 7.1). Take f
T
XA
XA
Theorem (A12 .1) clearly holds in the discrete case (M, Il. ).
134
135
APPENDIX 12
I e 27Tix I x
If k ( Zn artd k (U
Z imply k
dix 11). Thus, relation (AI2.2) holds for almost every initial point. This
can be rephrased as follows: denote by r(N, A) the number of elements of
the sequence
11m
- - - = Il
(AI2.4)
N~oo
= ~A'
<
(k +
P . lOT
136
that is to say:
r + Log10k S n Log 10 2 ::; r + Log10(k+ 1) .
Set a = Log 10 2 and (n a) = na - [n a], where [ ] means the integer
part. The above inequality may be written:
Log10k S (na)
< Log10(k+
1)
->
le 21Tix I x
RI.
r(N, A)
-N-
N->oo
/l(A)
But r(N, A) is nothing but the number of elements of the sequence 1, 2, ... ,
12n In
expects from an inspection of the first terms: 1,2,4,8, 1,3, 6, 1, 2,5, ...
This is due to the fact that a = 0, 30103 '"
REMARK
A12.6
137
APPENDIX 12
lim !l
T-+oo
where
~ x ITT(X)
T
TT(x) =
measure
- !l(A)
< ;;
yT
I t I tX
A, 0
$nl
217
:s. t :s. Tl
Ca e- u2/2 du
-00
and C is a constant.
APPENDIX 13
1 Arg
iwkt ,
~ ak . e
t-++oo
k= 1
= lim
(A13.3)
are
k = 0 and k ( Zn imply k = O.
(A13.4)
The Pk's depend on the lakl only. Ifl p(ak ; al' ... , ~k' ... , an) is the
probability that an (n-l) -linkage with prescribed sides al' ... , ~k' ... , an
spans a distance inferior to ak (see formula A13.12), then:
1
means cancellation.
138
139
APPENDIX 13
o =AO
Figure A13.1.
(A13.S)
In particular, for n '" 3, if there exists a triangle the sides of which
&re
formula):
(A13.6)
n '"
_A_l_w..;.1_+_A...;:2:....w.....;2'--.+_A_3:....w_3;:....
17
1 provides
140
A13.7
Tn
is the n-dimensional torus, p. is the usual measure, and >t is the translation group:
The phase space of the n-linkage is M, and >t depicts the movement. Let
us define a function a on M by:
n
(A13.B)
a(z) = Arg
lakl
Zk'
k=l
This function is discontinuous over the slit l:
= I z I a(z) = 01,
and is not
(A13.9)
is analytic outside of S. The limit (k13.1), if it exists, is nothing but the
time mean ~ of {3:
(A13.10)
where
A13.11
(Uk's
141
APPENDIX 13
f1*
n ~ i3
f1
is Lebes-gue-integrable. Thus,
f1
f1.
i3.
Relation (A13.9)
f1
depends lin-
217,
(U2
(Un
We have:
1 -
217
where
Relation (AI3.8) allows one to carry out the integration over (Jl:
where
(Un
2"
142
A13.13
Hence, formulas (A13.4) and (A13.s) are proved for almost every initial
phase Arg a k . To prove them for all initial phases we use a special device inaugurated by Bohl [1] for n
n
(3(z.)
J
N(z) is bounded.
N=O
Fii:Ure
~13.1S
m:
(~13.14)
N*
continuo~s
*=
N exists every-
(Q. E. D.)
APPENDIX 14
= l(x, y) mod 11
I'
~ ~
@ ~
Figure A14.l
~--------------------~
Which is called "multiplication of loaves" since Figure (A14.1) shows the solution of a well-known historical problem.
143
144
(x, y)
~ (2x,2y-l)
(2x-1,2y)
if 0 $ x,
y<\6
O$y<Y2
, (2x-1,2y-l) if \6 $ x, y < 1.
The application is not one-to-one, in fact it is everywhere four-to-one.
is measure-preserving. We
4- N p.(A),
->
k ( Z+
APPENDIX I))
SKEW-PRODUCTS
(See Definition 9.5)
Let (M, p.) and (M', p.')
b~.two
1M , A' ( ~;
1M , A' ( 1M "
I Mx M'
of Mx M' is gen-
M an automorphism .pm: M' .. M' such that (m, m') ... .pm(m')
IMxM '
whose characteristic
X F [( x I.pD(m, m')]d(p.xp.')
MxM'
[,x
F(m,
rp mm ')dp.
[
145
dp.
dp.
146
(Il X Il')F
A 15.1
EXAMPLE
If t/J m
then the skew-product is the direct product of (M, Il' ) and (M', Il', ').
EXAMPLE
A 15.2
Tflke M
51
->
51
an ergod'ic translation:
x = x + w (mod 1),
w irrational. Now, let n be an integer.
->
1)1
51 a
(lX
(mod 1)
nO.
Inl p Ipl,
phic although they have the same spectral type and the same vanishing entropy.
APPENDIX 16
A 16.1
u,t
of an ergodic
S. b l
2 This fact is general for systems with continuous proper functions (Avez [2]).
147
148
THEOREM A 16.2
C :s
b1
16.3
group H 1(M, R) has finite integral base: Y1"'" Yb l ' Each Yk is a closed
curve which can be assumed differentiable. Let a
:s T I
Itx\ x ( M,
O:s t
Fieure A 16.3'
Let (w k ) be the dual base of '(Yk) in the first cohomology group H1(M, Z),
that is the closed one-forms satisfying:
if i = k
ifi,tk.
149
APPENDIX 16
We obtain:
(A 16.4)
(y, w k ) is the value of the one-form w k for the infinitesimal generator y of t at point tx. Since length {3 < diameter M, we have:
where
(A 16.5)
!. (Wi<
lim
1{3
T-+oo
(A 16.7)
nk(T)
hm - T
T-+oo
f.
(y, wk)dJL
de!
JLk
= e2TTil-'lc,
the circle ~roup; ~ is called the ~roup of the windin~ numbers. It is read-
ily seen that ~ does not depend on the base (Yk). Thus, t defines a
r:al homology class:
the winding numbers JLk define the "homological position" of a generic orbit. In other words, they define how an "average" orbit w-anders around M.
'This concept was first introduced by H. Poincare [1] for flows on the
torus T2. Further investi~ations of systems:
x=
F(x, y),
G(x, y)
150
From its very construction, the rank of the winding numbers group is
bounded from above by b 1 , Therefore, Theorem (A 16.2) is a corollary of
the following lemma.
LEMMA A
16.8
(y,
(A 16.9)
for t
df(x))
217iAf(x)
constant
a.e.,
that is to say
li(x)1 = constant';' 0 everywhere,
where 1jI: M
-+
(A 16.10)
Therefore dljl defines a closed current of degree 1 in De Rham's sense. 4
3 See also I. M. Gelfand and Shapiro-Piatetski
[1].
S. Schwartzman
[1].
151
APPENDIX 16
[dy,] + dh .
Assume that M is metrized with a Riemannian metric whose volume element is dfL. Since "if> t is fL-measure-preserving, the infinitesimal generator
yof
; is co-closed (oy = 0)
f
u
[dy,]
dy,
= Y, [u (1)] -
y, [u (0)] ( Z .
(Q. E. D.)
COROLLAR-V' A 16.11 (Arnold [2], [3]).
Let V be a compact orientable Riemannian manifo: { which is not. a torus and the dimension of which is greater than !'ne. If the geodesic flow on
the unitary tangent bundle M
152
From
we deduce:
f (1
V
S"- 1
(y, w) a)1J =
Jrv
(1
S"-1
ya, w)1J
o
(0. E. D.)
APPENDIX 17
SPECTRA OF K-SYSTEMS
(See Theorem 11.5)
A. Subalgebras of Measurable Sets
Let (M, /L) be a measure space. We denote by 1 the algebra of all the
measurable sets and by
DEFINITION
6 the
A 17.1
which is closed
under the formation of complement and the denumerable union, and which
contains M.
INCLUSION A
17.2
1,
a subalgebra of (tl' that is, that every element of (to is an element of (fl.
The relation C is a reflexive partial ordering on the family of sub algebras
of 1.
INTERSECTION A 17.3
Let
(1'); ([
(f
;(/
A 17.4
Likewise, we denote by
153
154
V G';
; ( I
which contains
every (J;.
THE SPACE
L 2 (G'). A17.5
G'.
G' c 93
L2(
n G';)
;(1
L 2 (6)
".
n L 2(G';)
;(1
B. Spectra of K-Systems
A 17.6
G'
of
such that:
(A 17.7)
6 '=
V nG'
1.
n=-oo
17.8
155
APPENDIX 17
n""
HO =
UnH
cc UH
Let us denote by H
C H CC
n=-oo
0=-00
e H0
be written again:
101
""
n
UnH'C"'CUH'CH'CU-1H'C'"
0==-00
""
U
e Ho
0=-00
Proof:
(1, and
if (x)
=
if (x) ( A ,
that is,
{
U~A(x)
Thus,
uXA
0 if x
I -lA
1 if x ( -lA .
UL 2(f) = L2(-1(f)
A17.9
dim (H
U H) .
Proof:
Let
the closure of the subspace spanned by hi' Uh i ,.... From their very construction, the Uih I.'s, and so the J{.'s
are orthogonal to each other. From
I
Lemma (A 17.8) we have
plete system of H ':
n""
n=-~
UnH' =
101.
therefore lUih.!
is a comI
156
(A 17.10t
On the other hand, the relation:
can be rewritten:
oJ
v-nH'
L~.
n= 0
EB
~
i
V-n}{i
n=O
Let us set:
(A 17.11)
H,
U V- n }{.,
,
n=O
(A 17.12)
EB
~ Hi
i
According to (A 17.11), from the basis Ih i' Vh i' ... 1 of }{ i we obtain a complete orthonormal basis of Hi' namely:
Furthermore:
Ve ,,]
..
for every i and j.
Together with (A 17.12), this proves that V has Lebesgue spectrum,
the multiplicity of which is equal to the cardinality of IH ii, that is
dim (H' e VH') = dim (H
uR)
157
APPENDIX 17
LEMMA
A 17.13
dim(H
uR)
00
Proal:
Lemma (A 17.8). implies:
$. dim UH $. dim H $. dim U-1H $.
This proves dim H
dimUnH
dimU n+1H
Similarly, dim UH
function I ( H
00
00.
F =
1m I m (
Ig:tFlg
Lo
i(
01
00.
F)
the space L1
=
M,/(m) ,;,
(HI
= O.
Thus Lo C H
e UH,
00.
:t ~2'
we find that the Ih 1, Ih2' ... ( L are linearly independent. But they belong to Lo because h ( UH implies:
(h k h ( UH is ortJlOgonal to
n.
(Q. E. D.)
The preceding lemmas prove Theorem (A 17.6).
APPENDIX 18
A. Measurable Partitions
DEFINITION
A 18.1
O.
lB.J LJ l J
ther Aj C B k , Aj
rt.
B k , or Aj
rt.
B k , Aj C B k
From the very definition, it follows that we can remove the elements of
measure zero from a partition.
Mo~e
159
APPENDIX 18
o.
DEFINITION A 18.4
Let
la;\iEI
ai
iff
n AIJ A.J
a ={
a. for all
jff
The operation
i}.
DEFINITION A 18.5
= (3 <
a :S. (3
'l]j (
>m(a)
= m({3)
< >m(a)
C ')J{(f3}
ai )
iEI
m(a)
iEI
8. Entropy of a Given {3
Let a
IAi I i
= 1, ... , d and {3 =
IB j 1 i
160
DEFINITION
A 18.6
z(t)
- t log t if 0 < t
:s 1
ift=O.
f3
is:
Il(A./B.)
1
Il(A. nB.)
1
Il(B.)
]
Let a
12.5
f3 = IBjl,
1Ai\'
=-ICkl
Then:
h (al f3) ~ 0 with equality if, and only if a
(12.6)
:s f3;
(12.7)
:s f3
(12.8)
(12.9)
f3:s
(12.10)
> Maly)
:s h (f3/y) ;
> h(a/y)
~ h(alf3};
h(aVf3/y)
:s h(aly) + h(f3/y)
Proof:
Proof of (12.6) is left to the reader as an easy exercIse.
The elements of aV f3 and aVy are, respectively, of the form: Ai
B j and Ai
n Ck
Therefore
h (aV f3/y)
But we have
161
APPENDIX 18
jl(A,nBjnC k )
jl(Aj~Ck)
jl(AjnBjnC k )
jl(C k )
jl(A,nC k )
jl(C k )
= jl(A/CkhdB/ A( n'C k )
= -
i,j,k
= -
jl(A j
n C k ) Log
jl(A/C k ) -
i,k
h(aly) + h({3/aVy)
(3 and relations
We deduce:
162
z(p.(A/Ck))p.(CkIB j ) $. z[p.(A/B;)].
and
h(aV(3/y)
The preceding definitions and properties extend to denumerable measurable partitions (see Rohlin and Sinai [5]).
APPENDIX 19
ENTROPY OF AN AUTOMORPHISM
(See Theorem 12.26)
The purpose of this appendix is to prove the following theorem dl.e to
Kolmogorov.
THEOREM
A 19.1
The proof breaks into several lemmas. Denote by F the set d all finite measurable partitions. Given a, {3
F we write
Ia, {31
= h (a
I (3) +
h ({31 a).
LEMMA
19.2
Ia, {31
duce:
Ia, {31
> h (a I (3)
= h ({3
la, (31
I a)
='
>a $. {3 and {3 $. a
>a = {3.
(12.9) we have:
h(a/y) = h(aVy) -h(y) $. h(aV{3Vy) -h({3Vy) + h(j3vy) -h(y)
= h(aI{3vy)+h({3/y) ~ h(a/{3)+h({3!y)
163
164
and symmetrically:
h(y/ a)
5 h({3/a) + h(y/{3)
Addition yields:
19.3
Proof:
Given a, {3 ( F, we set:
Since h (
I) ~
0, we deduce:
Symmetrically:
Addition yields:
513
fT ,
'we have:
165
APPENDIX 19 _
00,
LEMMA A 19.4
00
V
n~l
m(an )
1 .
~ an
{3
s.
an'
is dense in
( m(an)
1,
Since
for every 0' > 0 there eXist an n and subsets AI',, A~_l
such that:
i~1 .... ,m-1.
It is clear that {3
s.
la. (31 ~
r hi (3) + h ({31 a)
...
Bm defined by:
166
IL(A)
!.
Il(B k nA j )
Il(B k )
- !
k
Il(A j n B k )
Il(A) !
- !
Log [ Il (A I. n B k)
Il (A j)
Log
[1l(B k nA.h
I
Il(B k )
Il (B k)
i,k
Al
la, ~I
depends continuously on
A l , ... , A~_l
F and q =
0,1, ... :
We have:
~ l
F such
167
APPENDIX 19
n + m -1
m
h(>.V V 2 q -2>..)
2q-1
2q-1
Thus, passing to
~he
00
h (a, ) .
sup h ({3, )
B'
sup h({3, )
F
~ h() ,
\PPENDIX 20
I tit
( R I. An ele-
g:
-+
yt +
x, y (R,
x,
y > 0,
we obtain:
= (x', y'),
g'(g(t))
y'(yt+x)+x'
y'yt+y'x+x'.
-+
IS
(_xy-l, y-l).
01.
(0, 1) is:
dx 2 + di
y2
168
169
APPENDIX 20
Proof:
To any element X = (x, y) of G corresponds the left translation LX
L x(U)
x 1. U,
where U
(u, v)
G .
We have:
!:) ,
( -u-x
y-' y
the tangent mapping of which is:
(A 20.2)
Define a metric over the Lie algebra TG e by setting:
(~1)2 + (~2)2
DEFINITION A 20.,+
The upper half-plane G endowed with the metric (A 20.3) is called the
Lobatchewsky-Poincare plane.
It can be useful to represent a point (x, y) of G by the complex number
z = x + iy.
THEOREM A 20.S.
THE ISOMETRIES OF G
-+
170
(A20.6)
az + b
cz + d ;
z ... z
a, b, c, d (
R. ad-bc
Proof:
Proof is purely computational and easy if one observes that:
ds 2
-4dzdz
h
- - - - , were z
(z_z)2
X-ly.
Proof:
i's proportional to dx 2 + di .
constant, y > 0
and the upper half-circles centered on ox. In particular, there exists one,
and only one, geodesic passing through two given distinct points.
Proof:
Let ab be a segment of x
b we have:
ds 2
ab
171
APPENDIX 20
THEOREM
A20.9.
CURVATURE
=0
A+ B+ C ':"
ABC gives:
1T
ff
K da
1T
+ K area L\
t"
~
=0
O. As the
1T
-1.
----O~=-B---------/------~~==~C~--~x
r
Fi~re
A20.10
172
as t1
->
}J!J.- has a
limit position
-+
(resp. y(-oo)) of y with ox (see Figure A20.12). Thus our geodesic has a
limit position, namely the upper half-circle centered on ox and passing
through g and y (+ 00) (resp. y (-00)). Consequently, this limit position is
a geodesic.
y
'Y(-m)
Figure A 20.12
173
APPENDIX 20
=0
the straight lines y = C > 0 are horocycles. They are positive horocycl/
of the axis oy (y
-+
00).
Proof:
Theorem (A 20.14) follows at once from the elementary properties of conjuii!,ate pencils of circles.
The points y(+oo) and y(-oo), which do not belong to G, have to be
removed.
THEOREM A20.1S. RIEMANNIAN CIRCLES
nothi~g
but the upper part of the pencil of circles passing through m and m'.
(Q. E. D.)
In particular, the power of any point d of ox with respect to one of these
Riemannian circles centered at m is:
174
THEOREM
A 20.16
Horocycles are Riemannian circles the radii of which are infinite and
the centers of which are at infinity (on y
0). '/
Proof:
Consider the Riemannian circle passing through a fixed point n of a
geodesic y and centered at m ( y (see Figure A 20.17). If m moves to
infinity along y, that is, if m converges to ox, then mm' ... O. Therefore,
the power of any point of ox with respect to our circle tends to zero. Thus
our circle has a limit position which is the circle tangent to ox at y (+ 00)
and which passes through n. Theorem (A 20.16) shows that this limit position is an horocycle. Conversely, any horocycle is obtained from the
(Q. E. D.)
above construction.
'Y(+OO )
m'
Fi\t:ure A20.17
175
APPENDIX 20
THEOREM
A 20.18
Let y(u, t) and y'(u', t) be two geodesics which are positively (to
fix the idea) asymptotic one to the other. We denote their arc length counted from their origins nand n' by t. Then, after a suitable selection of
nand n " we have:
d (y (t), y '(t)) $. nn 'e- t ,
t 2. 0,
where d means the Rierrftmnian distance, and nn' is the arc-length of the
horocycIe.
Proof:
Origins nand n' are selected on the same horocycle 1 (Figure A 20.19).
Denote by m and m' the intersections of y and y' with another horocycle
2. Arcs
nm
and
nm =
mm'
t.
tion:
x
r sin u, y
= r + r cos
u.
Symmetrically, on horocycle 1:
nn
un'
tg -
- tg ~
om =
n'm'
Log 1tg
Log tg
~n I -
Log I tg u;
Uri -
Log tg
I'
U; 'I .
176
Figure A 20.19
Consequently:
tg
et
u
_n_
2
u
tg .2!..
2
tg
2--u
tg .2!..
2
mm
tg ~ - tg
2
tg
- tg
I-t
nn e
:s.
mm' .
-!!.
u
.2!!_
nn
mm
177
APPENDIX 20
GENERALIZATION A 20.20
(x )2
n
constant and
o.
APPENDIX 21
PROOF OF TilE
LOBATCHEWSKY-IIADAMARD THEOREM
(See 14.3, Chapter 3)
A. Manifolds of Neglltive Curvature
Fa:>,
of negative
'me.
THEOREM A 21.1
178
179
APPENDIX 21
II,
t)
= y (t) = y
with initial velocity-vector u and arc length t. The point of y corresponding to t is denoted also by y (t). The Riemannian distance of two points
a and b is denoted by
la, bl.
21.3
->
+00 (resp. t
->
-(0).
v'
Y(v',u',t)
----------------~------------~----~--~Y
Fi~re
A21.4
The points v' and y (t 1 ) define one, and only one, geodesic y (v', u l' t ).
We set
s1 =
(A21.1) to the geodesic triangle v.', y(t 1), y(t 2 ); With obvious notations
we have:
t1 whence:
lv, v'l
~ s1 ~ t1 +
180
Similarly:
We deduce:
1,
that is to say:
->
00.,
The geodesic y (v', u ',t) is the limit position of y (v', u l' t), for the
exponential mapping Expv' is continuous.
->
-(0).
REMARK
A 21.5
given point of the positive asymptote y (v', u', t) is nothing but y (geometrically). Therefore, we may speak of a positive asymptote to y without referring to a definite point v'. Furthermore, the set of the positive
asymptotes to y is a (dim V -1)- parameter family of geodesics.
C. The Horospheres 1 of V
11,
bitrary point of V.
LEMMA
A21.6
See A. Grant
[1].
->
181
APPENDIX 21
Proof:
Take t2 > t l' The triangular inequality applied to v, Y (t l)' Y(t 2 )
gives:
(t2)
lv', y(t2)1 -Iv, y(t2)1 ::: lv', y(tl)1 + Iy(tl)' y(t2)1 -Iv, y(t2)1
=
(tl) .
lim
t ..... +
00
that is
Obviously:
L(v'; y, v) - L(v'; y, Vl) = vVl
(A 21. 7)
where
vVl
vVl
'
DEFINITION A 21.8
The locus of the points x for which L (x; y,O) = 0 is called the
pos~
L(x; y,vt )
OVl
Now we obtain the horospheres as spheres with center at infinity and radius
182
A21.9
as t -+ +00.
Proof:
-+
0 as t
-+
+00.
-----+--------~--------+-------~y
Figure A21.10
We have:
lx, b(t)1 = Ix,y(t)I-ly(t), b(t)1
Ix,y(t)I-IO,y(t)1
-+
0 as t
-+
+00.
This means that every point of H+(y, 0) is a limit point of the spheres
L(y(t),O) as t
-+
lim b(t).
t -+ +00
Ilx, y(t)1 -10, y(t)11 < I lx, y(t)1 -Ib(t), y(t)11 + Ilb(t), y(t)1 -10, y(t) II
Ix, b( t) I
-+
0 as
-+
+ 00
183
APPENDIX 21
Therefore: L (x; y, 0)
0, that is,
x
H+(y,O).
COROLLARY A21.11
IS
which goes to infinity along y, and these balls are convex (see A 21.2).
LEMMA
A21.12
~ (y (t),
~ir(t),
0) such that:
lim a(t) ~ a,
t ... + 00
0') such that:
lim
t ... + 00
a'(t)
a'.
< 10,0'1.
To fix the ideas assume that the point 0' lies between the points 0 and
y (t). We obtain the following contradiction:
laCt), y(t)1 ::; laCt), a'(t)1 + la'(t), y(t)1
~ 10,y(t)1
LEMMA
< 10,0'1 +
la(t),y(t)I
la'(t), y(t)1
(Q. E. D.)
A21.13
Two positive horospheres H+(y, 0) and H+(y, 0 ') cut off an arc of
length 10,0' I on every positive asymptote to y.
184
Proof:
Y(a',Uo,t)
-----+--------~------------~-----------Y
0
FIgure A 21.14
lim
-L(a'; y,O).
We deduce:
I la,y(t)I-IO,y(t)1 I :; Ila,a(t)1 + la(t),y(t)I-IO,y(t)11
=
"
y (t) II
->
0 as t ... + 00
21.15
->
-00).
185
APPENDIX 21
D. The Horospheres of Tl V
-+
= y (u, t)
H -(u) passing through pu. The set of the unitary vectors orthogonal to
H+(u) (resp. H-(u)) along H+(u) (resp. H-(u)) and oriented like u is a
A 21.16
(1) The y(u, t)'s and the }{+(u)'s, }{-(u)'s are the sheets of three
foliations of Tl v.
(2) At each point u ( Tl V these foliations are transverse, that is:
T(T1 V)u
where
X:
(resp.
X;;,
X;
E9
X;; Zu
y (u, t) ) at u.
Proof:
(1) Follows from the very construction of the sheets.
(2) Follows from the strict convexity of H + (resp. H -, see A 21.11).
(3) Follows from Theorem (A 21.15).
The invariance of the foliations reduces the study of the differential
Iv
Now,
of a compact
186
LEMMA
A21.17
Let rs(t) be a one-parameter family (s> 0) of numerical, C"-differentiable functions. Assume that:
2
r. s >
- k r s
(k = constant
> 0)
<
r (t)
S
r (0)
s
sinh [k (s - t)]
,forOStS4.
cosh (ks)
Proof:
The function:
r s(O)
rs(t) -
cosh[k(s-t)]
cosh (ks)
verifies:
i'
.s
(t)
>
k21 (t),
s
0, s, conse-
quently I s(t) S 0 for 0 S t S s. This proves the first part. This proves
also that is increases between 0 and s; thus: i s(t) S is(s) for 0 S t
in particular,
rs(s)
... 0 as s ... + 00
187
APPENDIX 21
THEOREM
A 21.18
Let >t be the geodesic flow of Tl V. Then, for any positive number
1I>;(il S..b'e-ktII(II,
II>; (;1
a'ektll(ll,
11>~lll S
a and
b'e-ktll(11
are independent of
if ( l X~
t and
~,
and
!I
denotes the length of a vector of Tl V equipped with its natural Riemannian metric.
Proof:
We p.rove the first inequality, the others can be proved in the same way.
Let y (0, u, t)
y (t)
Tl V.
Let r s(t) be the Riemannian distance of their projections y(t) and y s(t)
on V. To compute r s(t) 'we consider a Jacobi field 2 1/1 (t) -along y, 'that is
orthogonal to y and vanishes for t
s. By definition:
2 See
J.
Milnor
[1].
_k 2 , consequently:
188
V <VI/I,I/I>
YlV 2 \11/I112
Yl d2211 1/1112 ,
dt
that is
rs
>
k 2 I s , and I s (0) > 0,
-
ls(s)
0 .
Lemma (A 21.17) and the classical possibility to select the Jacobi field
(A21.19)
cosh[k(s-t)]
, for 0 ::;, t ::;, s .
cosh (ks)
->
+ 00. Thus,
i)s)
->
0 as s
->
~o
plies again:
I; s(t)1 <
(A 21.20)
As s
->
sinh (ks)
ys(t)
tance of y (t) to y '(t), then the inequalities (A 21.19) and (A 21.20) imply (s
->
+(0):
for t > 0 .
189
APPENDIX 21
Tl V verifies:
Due to this theorem, the sheets J{+(u) (resp. J{ -(u) ) are called the
~
A 21.21
Let W be a compact, connected Riemannian manifold of negative curvature, then the geodesic flow on Tl W is a C-flow.
Proof:
Let V =
IV
IV . . W. V
IV
IV,
for
of Tl IV: if u',
J{ .(u
U" f
Tl
IV
REMARK
A 21.22
The horospheres of a compact, n-dimensional manifold Ware diffeomorphic to Rn-l. In fact, let us consider the horosphere J{ +.
It is a
See J. Hadamard [l]. Proofs of SectlOns Band C are mainly due to H. Busemann: Metric Methods in Finsler Spaces and Geometry. Ann. Math. Study. No.8.
190
1> t J{ +
1>; 1D i~
to Rn -
1,
~oes
Consider the
FIgure A 21.23
1 is an
APPENnIX 22
PROOF OF THE SINAI THEOREM
(See Section 15, Chapter 3)
Let (M, ) be a C-diffeomorphism and X m {resp. Ym ) the k-dimensional dilating space at m Crespo the I-dimensional contracting space). A Rie-
mannian metric is definitively selected on M. Thus, Xm and Ym are Euclidean subspaces of TMm'
THE METRIC SPACE OF THE FIELDS OF TANGENT k-PLANES
A22.1
The tangent space TMm is the direct sum Xm
III
Ym . Therefore, the
IVm -V'[
m
IIP(U m )-P(U')II
m
sup
X(Xm
.lx[<l
IP(v)x-P(U')xl
m
IV-V'I
= sup
m(M
IV m -
V'
I.
V, V' ( K.
191
192
LEMMA A22.2
Let RI and R2 be two n-dimensional Euclidean spaces (n
Assume that R j (i
Rj =
Let A: RI
->
dim M).
Xj
III
lj,
dim Xj
k,
dim Y.I
I:
AX I
Y2 ,
Ixl
for x ( Xl
for y ( YI
IIAxl
(A 22.3)
AYI =
~ fL
Ci'
spond to the k-planes of RI the k-planes of R 2 . If V and V' are transverse to YI , then:
Proof:
By definition:
lCi'v - Ci'v'l
Ixl < I
X(X 2
sup
Ixl
IA[P(U)A-Ix] - A'[P(U')A-Ix]1
<1
X(X 2
:s.
fL- l . Therefore:
:s.
sup
Ixl<l
sup
Izl<rl
X(X 2
X(X I
and
:Ci'v-Ci'v'l:s.
fL- l
sup
!z < 1
Z(
XI
IA[P(U)-P(U')]z!
193
APPENDIX 22
I s:.
~ a'
sup I[P(U)-P(U')]zl
Izl < 1
a'IU-U'1 .
Z(X I
Finally we have:
IAU-AU'I ~ /I-laIU-U'1
(Q. E. D.)
INEQUALITY
for
AYI = Y2 are verified. Inequalities (A 22.3) are consequences of the axioms of C-systems:
e=
I
In fact
**
**
to the k-planes.
(Q. E. D.)
APPENDIX 23
C: ;J
f1
Ip + qy'3l p, q (
of G, the elements
of which satisfy
x, y, z ( Q( v'3) ,
X
It is readily proved that
x,
y,
=Z
and, con-
M.
a(G
mapping : G G by:
194
APPENDIX 23
1> (x,
195
AX, ii Y, i7 Z)
where:
1>
1>
2 + V3,
v = (2-V3)2,
P. = Av = 2-y3,
f, and
(M, ) is a C-Diffeomorphismo
C:iO;g)
The metric
of TG e defines a right mvariant metric on G and, consequently, a Riemannian metric on M = Golf. The Lie algebra TG e splits into the sum X + Y,
where the elements of X (respo Y) are of the form:
(respo)
IS
g of G:
TG~ = X~ + ~
Thus, the tangent space TMm at m ( M splits into:
m
= Xm+ Ym
IS
dilating on X m
APPENDIX 24
A24.1
1/1',
C 1-c1ose to
1/1,
1/1:
is structurally stable.
1/1.
>.
1
(p . . q P (P
(mod 1)
z ... ~z
Let By, be the ball (Figure A 24.2) of T2 x R with center (0, 0, 2) and ra-
dius
1/2:
x2
We defioe a diffeomo",hism
+;+ (z_2)2
~;
>1:
{Of
:v: i~~
~ ~ .
T2 x Iz 1
y ... ~y
z ... 2z-2 .
196
~ z ~
3\ by setting
APPENDIX 24
197
x: dilating direction
Y: contracting direction
t
Figure A 24.2
198
LEMMA
A24.3
i ,-
. A 24.4
(~)
(A 24.5)
. . 0 0(;)
(mod 1).
Let us recall some properties of this diffeomorphism: There exist two foliations
X and
X (or
are dense in T2 x 10\. This fact can be proved by observing that every rational point (pi q, p 'I q) ( T2 is periodic.
Now we pass to the diffeomorphism : T3 ...
that the periodic points of
cp,
in T2 x I z
I-
1 S; z
r.
s:.
It is easy to see
1 \, coincide with
I-
1 S; z
s:. z S; 11,
where
Y is some sheet of ~.
B. The Diffeomorphism ifJ
d2 .
s:.
T2x
10\
such that
N';
199
APPENDIX 24
nG =0 .
(x, y, z)
r/J by setting:
= (x', y', z ')
outside G
>
is
" z)\
Now the curve tjJI(O,.O,
s.
l' This region is foliated into contracting planes (see A 24. 4). Let
~xla) + (ylb) =
(x, y, z). Among these planes intersecting the bump B, we select the plane
~, for which a is maximum (see Figure A 24.2). Either j= contains a peri-
odic point of tjJ, or it does not. In the first case, the bump is called periodic and in the second case, nonperiodic.
LEMMA A24.7
tjJ is not structurally stable.
->
b~tween
h=
h tjJ. In fact,
bumps, contracting
200
LEMMA
A24.8
Every diffeomorphism
!/J',
Ciclose to
!/J,
1,
and so on:
APPENDIX 25
Let (M, ) be a C-diffeomorphism. We select definitively a Riemannian metric on M. Since M is compact, there exists a number d > 0 such
that, whatever be the ball B (p ; d) C TM
with radius d and center p ( M,
.
p
the restriction
Exppl
B(p; d)
= B(nm, d)
C TM,J..n
o..p
m.
and the restriction IjJ I-B is Exp ,J..n' . Let us denote by X the dilating
n
"fJ m
n
k-space X (n m) of TMnm' and by Yn the contracting I-space y(nm),
Tke invariant dilating and contracting foliations rand
new foliations on B:
201
of M induce
202
Bn
maps B
into B
n+
tions concerning the range of 1' Assume d small enough, then the sheets
of the foliations
X1
= y(O)
= x(O)
+ gn(Y' x(O))
where x ( X n , Y ( Yn and where in' gn' and their first derivatives can bl:
made arbitrarily small by a suitable choice of d.
a n of '!:J n which passes through the center 0 of
The mapping:
e:
whose restriction
eIY
n:
I Yn I n
Yn
->
( ZI
an
->
I an In (
is defined for
ZI
y ( Yn n B n
by
Figure A2S.1
203
APPENDIX 25
In
and
ASSERTION A 25.2
It follows from the very definition of C-systems that the restriction
is contracting:
(A 25.3)
where
e is a constant.
REMARK A 25.4
To be precise, (A25.3) holds for a certain iteration ~ of 2: we
must "kill" the constant b in the definition of C-systems. For simplicity,
we assume that (A 25.3) already holds for v = 1. Now let ' be a diffeo
morphism C 2.close to ' Then ' is a C-diffeomorphism (Sinai theorem,
Section 15) and the foliations :t~ = 1j1- 1:t.' 'lJ; =rljl-1'lJ' and the mapping
; induced by ':
;: B ... B.
; = 1j1-1'1jI,
IBn:
Bn ... B n+ 1 ,
:t~ are close to those of :t1 and transverse to the sheet an' Therefore,
there exists a projection IT:
e- 1 IT;e,
204
cp'
Figure A 25.5
ASSERTION A25.6
;. is C-c1ose to 2: to any
(A 25.7)
where
I II c 2
If
II'- I c 2 < 0
for any y ( Yn
implies:
Bn ,
is the C 2 -norm.
Proof:
{ is close to 1' II: ; an -+ an+ 1 is small for { an '" 1 an =
a n + 1 , and the sheets of !; are transverse to an +l' Now denote the sheet
of!; passing through the center 0 '" m of Bo by {3,
LEMMA A. A 25.8
If ' is C 2 -close enough to , then the sheet 'n{3 is close to nm
for any n
(A25.9)
where
0,- To be precise:
IW;mll
e, is defined at (A 25.3).
<
If
1-e
205
APPENDIX 25
Proof:
11'-ll c 2 < 0
implies:
IIYII <
+ ce .
, then
II; y I <
and
Iwi y II
< , and so on. But since 11m II < , inequality (A 25.9) is proved.
REMARK
A25.10
Ily!1 <
IW2n y I ::;
In fact
IIY I <
c and c
if c ? 1 ~ 8
c implies:
? ce/(1- 8) imply:
II; y II <
Ily II
+ ce
< 8c + ce < c .
(Q. E. D.)
Lemma B
Now we consider the sheet Yn of
Yn1
:x:
Yn
:x: 1:
',1,-1
'f'
Yn
'
11,
similar to
e,
by setting:
11: lXnln
11
Z!
lYn1 1n
->
is a diffeomorphism and x
11l xn :
ZI,
Xn
->
Yn1
lXnln
Obviously 3 (0) = 0,
Z!
->
lXnln
ZI,
206
ASSERTION A 25.11
It follows from the very definition of C-systems that 31x : X
n
Xn+ J
is dilating:
(A25.12)
REMARK
A 25.13
f3 n =
'In
f3
C Bn
of the
(See Figure
n Bn
f3 n .
If ' is close
as local coordinates of
f3 n :
the mapping
E
Ixn :
..
FIgure A 25.14
f3 n '
n Bn is a diffeomorphism.
Yn+l
q>'
Xn ~
207
APPENDIX 25
ASSERTION A 25.15
(A25.16)
LEMMA B. A25.17
II ' is C 2 -close enough to ' then there exists a well-defined sheet
o(
'Y'
such that ,n
o is close to
there exists one and only one point Xo ( Xo such that 11;n xoll
<
E for
any n-::: O.
First we need a sublemma.
LEMMA A 25.18
Let R be a union of equi-dimensional Euclidean spaces R n , n
Let T
K + L: R
->
R, T IR n : R n
->
O.
(1) K(O) = 0,
(2) IlL II $. E,
Eilx - y II,
0 > 1,
0 - E > 1,
for any x, y (R. Then, there exists one and only one point x ( Ro such
that' the sequence Tnx is bounded, and
(A 25.19)
IiTnx II $. _ E _
0-1
for any n ~ O.
208
Proof:
->
ilTx- Tyll
(A25.22)
consequently
T-1b1(c) ) T- 2 b 2(c) ) .. ) T-nbn(c) ) ....
->
0 as n
->
00
This
finishes the proof if one observes that c = E/(8 -1) verifies (A 25.22) .
. Proof of Lemma B. A2S.23
According to (A 25.11) and (A 25.15), the mapping ; verifies the conditions of the preceding lemma. It is sufficient to set K
3' L
209
APPENDIX 25
close to the orbit nm for n ::: 0 (in the sense of Lemma 8). If ' is
close enough to '
'!:I'
is the dilating
foliation of ,-1 and the sheet 0 is close to m. Consequently, according to Remark (A2S.10), the sheets mo, (n
t 8 C Bn'
(-00
f3
8
and
center of Bo. The desired homeomorphism k of the Anosov theorem is defined by setting k (m)
= k
APPENDIX 26
INTEGRABLE SYSTEMS
(See Section 19, Chapter 4)
J. Liouville
(A26.1)
aH
aq
= --,
q "" ap
aH ,
p = (Pl" .. , p ), q = (ql'' q ),
n
(A26.2)
are known, then the system is integrable by quadratures.
Many examples of integrable problems of classical mechanics are
known. In all these examples the integrals (A 26.2) can be found. It was
pointed out long ago that, in these examples, the manifolds specified by
the equations F,
A 26.3
f;
constant, i
1, ... , n, define an
independent;
1 Two funchons F(p, q) and G(p, q) are in involution if their Poisson bracket
vanishes idenhcally:
(F, G) = aF aG
ap aq
_ aF aG
aq ap
210
== 0
211
APPENDIX 26
(2) a Jacobian Det lal/af I, which is defined below (A 26.7) does not
vanish identicaIIy.
Then:
(1) M is an n-dimensional torus and the neighborhood of M is the direct product Tn
Rn;
F/I).
1>
0,
aH ,
aI
and the motion on M is quasiperiodic since H = Fl = H(I) and Equations (A 26.1), in action-angle coordinates, are Hamiltonian equations 2
with corresponding Hamiltonian function H(J).
Proof:
NOTATIONS A
26.4
= (p, q)
Xl
, ... , F
x2n
of
a function F(x). The Hamiltonian equations (A 26.1) then take the form:
(A26.5)
x=
1=(O-E)
E-O
I grad H,
where E is the unit matrix of order n. We introduce in R2n the skewscalar product of two vectors x, y:
[x,
Jd
for all x, y,
212
lution.
THE CONSTRUCTION. A 26.6
~;
Let us consider the system (A 26.5) with Hamiltonian F;. Since (F;,
F.) ~ O. all the functions F. are first integrals. and every orbit lies wholJ
fJ
and
f,
~; ~ I
grad F, is tangent to M.
but 3 the velocity field of the system (A 26.5) with Hamiltonian (F.,
F.)
~ O.
,
J
Thus, M is a connected. compact orbit of the group Rn acting smooth
ly and transitively; therefore we proved that M ~ Tn. Besides. M being
specified by the equations F;
and I grad G
IS
an
Hamiltonian vector field with Hamiltonian function - (F, G). We shift the proof to
Appendix 32.
APPENDIX 26
S([, q)
(A26.8)
jq
213
pdq
'lo
p ([, q) ).
The
1> ->
nates:
p
(A26.9)
LEMMA
as
as
oq
af
A26.10
0 ,
O.
1>
->
p, q.
26.11
1>.
214
There-
one~form.
Let us compute the periods of the one-forms d, over the basic cycles
of the torus Mf' According to (A 26.7) we have:
Yi
di
f d(aS)
al.
Yj'
~
d/.
'Yj
dS
~(2771.)
= 2770 ..
d / . } '}
,
Therefore the variables , are angular coordinates on the torus M(l) and
our theorem is proved.
APPENDIX 27
1. Consequent-
ly, the product of the proper values Al and A2 of A is equal to 1. Besides, Al and A2 are roots of the real polynomial Det (A -AE). Therefore,
either Al and A2 are both real, or they are complex conjugate: A2 = Xl .
In the first case, we have:
(A27.1)
Figure A 27.3
215
216
and the roots belong to the unit circle (see Figure A 27.3). The third and
last possible proper value coniiAuration is:
hyperbolic rotation
hyperbolic rotation
with reflection
Figure A27.5
217
APPENDIX 27
EXAMPLE
A27.4
2p, Yz q ,
-+
-+ -
constant. Of course, the fixed point 0 is unstable. From classical theorems of linear algebra it follows that every mapping A of the first type
(Ai of, A2 ; Ai' A2 ( R) is an hyperbolic rotafion, possibly with reflection.
This means that, up to a suitable change of variables, A may be written
EXAMPLE
->
1 Q
AP, X
A27.6
p, q
-+
x
p
Figure A27.7
e- ia,
218
This rotation transforms into an "elliptic rotation" (see Figure A27.7) under a linear change of ,,:ariables. In this case the orbit r'x of x = (p, q)
belongs to an ellipse centered at O. The fixed point 0 is obviously stable. Classical theorems of linear algebra show that every mapping A of
the second type (1'\11 = 1'\21 = 1, '\1';' '\2) is an elliptic rotation.
In the first case (A 27.1), the fixed point 0 is called an hyperbolic
point and one says that A is hyperbolic at O. In the second case (A 27.2)
the fixed point 0 is called an eIIiptic point and one says that A is eIIiptic at O. Finally, the third case (,\ 2 = 1) is called the parabolic case.
REMARK
A27.8
restricted to lie either on the real axis or on the unit circle (see Figure
A 27.3). Therefore, these roots cannot leave the unit circle, except at
points ,\ = :!::. 1 which correspond to the parabolic case. Finally, we define
the topological index of a vector field at a fixed point. Let us consider a
vector field ';(x) of the plane p, q, with an isolated fixed point ';(0)
This defines a mapping of the unit circle x2
= p2
O.
= 1 onto itself:
If E is small enough, then the topological degree of this mapping does not
depend on E and is called the index of .; at 0, or the index of O.
Now, consider the vector field ';(x) = Ax - x.
If the mapping 1\ is
A27.9
APPENDIX 28
(0,0). Assume that 0 is elliptic, that is, that the differential of A at zero has proper values A1 = e- 1a , A2 ~ e ia . It has been
fixed point 0
B: p, q
->
P, Q,
B (0)
0,
arctg(P/Q)
'
arctg(P'/Q')
then:
I'-I
(A28.1)
'-
The coefficients a, a 1 ,
= O(ls+1)
a+
a1 i+ ai 2 + .. + a/ s +
...
~ 2rrm/n
0([S+1).
719
220
THEOREM A 28.2
~eneric
I,
Similar theorems are obtained concerning the stability of equilibrium positions and elliptic periodic solutions of Hamiltonian systems with two degrees of freedom (see Arnold [7]).
J. Moser [1]
The fixed point of an elliptic canonical mapping A of the plane is stable provided that:
(1) a f, 217
j!- ,
217 :
(2) a 1 f, 0 ;
J. Moser [1].
REMARK A 28.4
If a
Civita [1].
APPENDIX 29
PARAMETRIC RESONANCES
(See Section 20, Chapter 4)
The analysis of the stability of the fixed point (0,0) of a linear mapping of the plane is due to Poincare and Lyapounov. Only in recent times
(1950), were these results extended by M. G. Krein [11. [2] to systems with
many degrees of freedom. Krein's investigations have been enlarged by
Jacoubovich [11. Gelfand and Lidskii [41. and so on. J. Moser [3] published
a report of Krein's theorem.
Let A be a linear symplectic mapping 1 of the canonical space R2n.
We say that A is stable if the sequence An is bvunded. We say that A is
parametrically stable if every symplectic mapping, close to A, is
st~ble.
X, and I/X
product and 1 =
(~=~), E
[e-,." 1 = (1
, ) is the inner
[Ae-. A."l
[e-. ."l
221
and
A iA
= 1.
222
Prool:
It is sufficient to prove that the characteristic polynomial of' A is real
= Det(A-..\E) =
Det(-lA,-1 1 + ..\p)
~nd
1..\1 = 1. Quadruples are formed by ..\,X,..\-I, and X-I (see Figure A 29.3) .
1.3
I
I
\ I
\I
~.l
\I'
,.
'~'I
;\3
\
\
Flgure A 29.3
223
APPENDIX 29
COROLLAR Y
A 29.4
If the proper values are simple and lie on the unit circle IAI = 1, then
A is parametricalIy stable, because if all the proper values are simple and
lie on IAI
= 1,
then:
to A, lie on IAI = 1. In fact, assume the contrary, then A' would have'
two proper values A and X-I close to a unique isolated proper value of A.
(see Figure A 29.3).
Let us now assume definitively that :i 1 are not proper values of A.
Krein classified the proper values belonging to the unit circle IAI = 1:
they split into positive and negative proper values. First assume that all
the proper values are simple; we prove the following lemma:
LEMMA A29.5
1, or [~1' ~2] = O.
Proof:
Since A ~1
[A~I' A~2]
Al A2 [~1' ~2]
[~1' ~2]
(Q. E. D.)
COROLLAR y A
Let
(J
29.6
1)]
1)
of
(J
IS
nonvanishing.
Assertion (1) is a direct consequence of AIA3 ~. 1, A/\3 ~ 1: in view
224
= 0,
= 0 for
every
~3' Therefore [~1' 71] = 0 for every 71, which is impossible. Consequent-
A29.7
~ of the
A~
and
of a are non-
f. 0 on
a.
REMARK A
29.8
The sign of a proper value lias a simple geometrical meaning. The plane
a
Therefore, oile may speak of positive (or negative) rotations. The restriction of A to a is an elliptic rotation through an angle a, 0
two proper values with opposite signs can leave the unit circle after they
have collided, so forming a "quadruple" with their conjugates (see Figure
A 29.3).
To be p~ecise, let A(t) be a symplectic mapping -.yliich depends continuouslyon a parameter t, and the proper values of which are different
from :. 1 if It I <
A are simple
collide for t
THEOREM
an~
T.
O.
A 29.9
If all the proper values that collide have identical sign, then they re-
225
APPENDIX 29
main on the unit circle after the collision and the mappinl1 A remains sta-
(A 2 - E) Ae-
0 and
del1enerate for It
I<T
this form is positive definite for t < O. In fact, every vector 1/ is equal
to the sum of its projections 1/k into the invariant planes Uk corresponding to the proper val,ues Ak
Xk
[A1/. 1/]
=I
[A1/k' 1/1]
k,l
because
t) .-
So, the form [A (t ) e-, e-] is positive definite for t < 0 and nondel1enerate
226
0 and, consequent.
An is symplectic.
=
< E.
constant, that is
(Q. E. D.)
RE:'IARK A 29.11
>'k lie
1.\1
1,
>.;
1. Besides, the
Xk
APPENDIX 30
n=
F:
n ...
RI,
f:
CI)
II =
n ...
RI,
CI):
8 1 ... RI ,
A30.I
n:
, ~ =
CI) ( [ )
1I
E F ([, )
+ Et([, )
and
(A30.3)
= E
---.L
21T
If
CI) ( [ )
,;,
0 in
n,
21TF (j, ) d .
= ]
(0), satisfy 1 :
227
:S
l/f .
228
Proof:
P ~ P(I, ) ~ 1+ Eg(J, ),
g(I,) ~
F(P)-F(P,) d;
cu (P)
2TT
(ii - F)d
0,
and therefore g( + 2TT) ~ g(). Now, our system (A30.S) may be written:
(A 30.7)
Let P(t) be the solution of (A 30.7) with initial data P(O) ~ J(O)
(A30.8)
= /(O):
IP(t)-J(t)1
< Cl'
IP(t}-J(t}!
< C2
for every t.
Inequalities (A30.9) and (A 30.10) conclude the proof. They prove also
that the motion decomposes into the averaged motion and fast small oscillations (see Figure A30.U).
229
APPENDIX 30
I (t )
"-="_- J(t)
t=o
Figure A30.1l
APPENDIX 31
SURFACES OF SECTION
(See Section 21. 9, Chapter 4)
Let H(p, q) be the Hamiltonian function of a system with n degrees of
freedom (therefore the phase space is 2n-dimensional). Let L: H = h,
q1 =
------;...... y
,I
,
I
----__\r-"
Figure A 31.1
point x of
a point
APPENDIX 31
231
THEOREM 1
A 31.2
(A 31.3)
PdQ =
where PdQ
PdQ,
Ay
Proof:
Consider the orbits emanating from y in the (2n + 1) -dimensional space
I(p, q, t)l. The curves y and Ay of the space I(p, q)l are the projections of
two closed curves y'and Ay'of !(p, q,Ol which are formed respectively,
q
p
Figure A 31.4
Poinc~re-Cartan
theorem:
232
L pdq- Hdt
(A 31.5)
~,
where pdq
pdq- Hdt ,
Ay'
then:
.! Hdt
~y'
o.
Besides. we have:
pdq
Ay'
pdq.
Ay
In view of ql
Thus. finally.
,pdq-Hdt = jPd Q,
pdq-Hdt
Ay'
1.
PdQ.
Ay
31.6
be a closed convex curve of the plane E2. Suppose that a material point
N
Figure A31.7
233
APPENDIX 31
M moves inside r and collides with r according to the law "the angle of
The
states of M, immediately before aAd immediately after a reflectim, are determined by the angle of incidence a, 0 :::; a :::; 217, and the point of incidence. The point of incidence A is defined by the algebraic length q2 of
the arc OA of r (0 is an arbitrary origin). In other words, the set of the
states of M, immediately before and immediately after a reflection, form a
torus
r2
r2
info another
Proof:
MN is the distance
= Ivl
sin a,
P2
= Ivl
cos a.
v2
=""2
In the four-dimensional space I(P l , P2 ' ql' q2)~' consider the surface ~
whose equation is:
1, M ( r) :
234
From one reflection to the next, the motion defines a mapping :\: L
->
L.
The coordinates P2' q2 are local coordinates of L (a 1= 0) which is a Surface of section. In view of Theorem (A 31.2), the mapping t\ is canonical
and therefore preserves the two-form:
dP 2 /\ dq2 = sin a' dq2 /\ da .
(Q. E. D.)
An elementary proof of this theorem, due to G. D. Birkhoff [1], requires ex~
tensive computations.
APPENDIX 32
FUNCTIO~S
TilE GENERATI:\G
OF
"APPI~GS
CANONICAL
Let x ~ (p, q), (p ~ (Pi' ... , Pn)' q ~ (ql' ... , qn))' be a point of the
.canonical space R2n. The differentiable mapping:
is
call~d
(A 32.1)
pdq
1.
pdq,
Ay
[(a)
ff
dp 1\ dq
ff
dp 1\ dq
[(Aa) .
Aa
235
236
P(p, q),
A(x)
Q(p, q),
=
0).
Therefore:
~:) ~
O.
Then, ~(x) can be regarded as a function of P, q, defined in the neighborhood of the point P, q:
(A 32.4)
A(P, q)
(P, q)
p(P, q),
Q(P, q).
DEFINITION A 32.5
LEMMA
Q,
aA
aq
= p.
A32.7
in the
nei~hborhood
= P(p,
q), Q
= Q(p, q),
mappin~
A.
237
APPENDIX 32
mapping
where
S
S(p, q; E) ,
By definition, the infinitesimal canonical mapping Se is a class of equivalent families Se: two families Se and SE' are equivalent if ISE- Se' I
O(E 2 ),
DEFINITION
A32.9
The function S(p, q) on the phase-space is called the generating function of the infinitesimal mapping SE (or Hamiltonian function).
Of course, S is defined up to a constant. Now we prove that the function S is a geometric object: S neither depends on the canonical coordinates p, q, nor on the choice of a representant SE in the class of equivalence: it is a mapping S: R 2n ... Rl. In fact, let y be a curve of R2n
1 ThiS is a way to memorize (A 32.6).
238
joining x and y:
ay = y - x. We set
Ye = Se Y ,
and we denote by a(E) the strip form-
aoe
Figure A32.10
Let us set:
(A 32.11)
![a(E)]
11
dp/\dq.
a(e)
According to (A 32.2), this integral does not depend on the canonical coordinates and, according to (A 32.1), does not depend on the curve Y, but
only on x and y.
LEMMA A32.12
The generating function S. of the infinitesimal canonical mapping Se
is given by:
(A32.13)
S(y) - S(x) =
![a(E)]1
dE
'
e=O
ox
= (op,oq).
E(S(y) -S(x)) = E
f~
yap
dp +
dq
aq
(oqdp-opdq) + 9(E2)
(A32.1S)
239
APPENDIX 32
(Q. E. D.)
A32.16
(A 32.17)
S (x) + constant.
Proof;
Let YE and a (E) be the curve and the surface of Lemma (A 32.12).
The curve y' = Ay joins the points Ax and Ay. Besides, the curves
TE ,y', 0
~ E' ~ E,
(A 32.18)
From (A 32.13) follows:
(A32.19)
S(y)-S(x) = ~ J[a(E)),
dE
T(Ay)-T(Ax) = ~[[r(E)].
dE
(Q. E. D.)
COROLLARY A32.20
Let '\ and CE be infinitesimal canonical mappings with corresponding generating functions Band C, and let ,\ be a finite canonical mapping . Then, the infinitesimal canonical mapping:
(A 32.21)
240
(A 32.22)
B '(x) =
c (x)
+ B (x) - C (A -1 X) + constant.
In fact, (A 32.8) implies that the generating function of the product of two
infinitesimal mappings is the sum of their generating functions, and also
that the generating function of the inverse mapping C;-1 is - C. Relation
(A32.22) is easily derived from these remarks and from Lemma (A 32.16).
C. Lie Commutators and Poisson Brackets
Given two infinitesimal canonical mappings Ae and Be' there exists
one and only one infinitesimal canonical mapping Ce such that:
(A32.23)
AaBbA_aB_b
A 32.24.
of
Ce
(A32.2S)
'lie
[V A, V Bl.
gradient.
Proof:
ay
y - x. We consider the
aa 1
Y - Y1 + ''',
a2
\y~. -a<E<O,
oa2
Y1 - Y2 + "',
a3
BeY2'
0 < E < b,
aa3
Y2 - Y3 + ''',
a4
.'\ Y3'
0< E < a,
aa4
Y3- Y4+"',
~egments
241
APPENDIX 32
Figure A 32.26
(A 32.27)
+ [(7 y) + [(7)
[(~)
0 .
(A32.28)
1(ul )
[(u2 )
[(u 3 )
- a[A(Yl)-A(x l )] + 0(a 2 ) ,
b[B(Y2)-B(x2 )] + 0(b 2 ),
[(u 4 )
a[A(Y3)-A(x 3 )] + 0(a2 ) ,
[(us)
[(7 )
Finally, from (A 32.8) itO follows that the vector fields corresponding to Ae
and Be are 1 V A and 1 V B. Consequently, up to terms of order 0 (a 2 + b2 ),
we Have:
A(Y3)-A(Yl) = (VA, Y3- Yl) = (VA, Y3- Y2) + (VA, Y2- Yl).
=
{V A, 1 V B)b - (V A, 1 VA)a =
= {VB, VA]b -
242
Similarly:
- [V A, V B]a .
Thus, we deduce from (A 32.28):
(A 32.30)
Comparison of (A 32.27), (A 32.29), and (A 32.30) yields (A 32.25).
(Q. E. D.)
APPENDIX 33
GLOBAL
CANO~ICAL
\lAPPINGS
Let
p)1
n
->
->
and q
= q(x)
(n.
of the point x
/Pl' ... ,
= p(x)
->
(A33.1)
pdq =
Y
pdq
Ay
for anyone-cycle y (even nonhomologous to zero). In conformity to Appendix 32, the mapping A is locally determined by a generating function
Pq + A (P, q), provided that
Det (~: )
0,
(A33.2)
p
=P
aA ,
aq
aA
= q+ - ,
ap
A (P, q) =
(P, q)
243
244
Let us set:
A(x) = A(P(x), q(x)) ,
where
x
LEMMA
(p(x), q(x)) (
n.
A33.4
n.
Proof:
Let
y be a
closed curve of
(A33.S)
n.
f<Q-q)dP+(P-P)d q =
y
o.
(A33.6)
pdq =
PdQ:
Thus, we obtain:
fd
[P(Q-:- q)].
d[P'(Q-q)] = O.
Y
(Q. E. D.)
B. A Topological Lemma
APPENDIX 33
LEMMA
245
A33.8
The tori T and AT have at least 2n common points (counted with their
order 01 multiplicity) provided that the equation 01 AT
(A33.9)
IS:
p = p(q),
I(x) =
(A33.10)
p(x)dq(x) ,
Kij
where x
(p(x), q(x))
Function I(x) is well-defia.ed since the integral (A33.10) does not depend
on the path. In fact, let y be a closed curve of AT. then:
pdq =
1.
pdq = 0,
A-'y
0 on T. The
tersection of AT and T.
,
= ~o
(Q. E. D.)
bl' b,
= i-th
246
COROLLARY
A33.11
p = p'(q), p =
(A33.12)
p"(q)(I~:'1
<
00, I~'I
<
00)
In fact, if dp 1\ dq
:=
-+
p- p'(q), q
REMARK
If n
p"(q).-p'(q).
A33.13
1 (mappings of annuli), Lemma (A 33.8) still holds without con-
dition (A33.9). The proof makes use of Jordan's theorem and does not extend for n > 1. Whether T and AT intersect, for n> 1, if condition (A33.9)
is not fulfilled, is an open question.
in the plane many-body problem, are such that the ellipses do not intersect.
Then, whatever
be, there exist initial phases 2 1., g. such that the axis
"
T.
A 33.14
If we drop condition (A 33.9), Lemma (A 33.8) cannot hold without assuming that i\ is a diffeomorphism, because (even for n
1) regular and
247
APPENDIX 33
C. Fixed Points
A:
p., q -+ p, q <+
(p)
Assum~ that; on the torus p = Po' all the frequencie~ are commensurate:
(A33.16)
and that the twisting is nondegenerate:
(A33.17)
THEOREM
Det (
aw ) ~. 0
ap Po
A33.18
x.
Proof:
In view of (A33.1S), (A33.16), and (A 33.17) the mapping AN can be
written under the form:
(A33.19) A": p, q
-+
a(p)
0,
Det
(aa)
ap
Po
N[w(p)-w(po)]
BN : (p, q
-+
p+
f3 1(p, q),
q + a(p) +
f3 2 (p, q))
(P, Q)
~ 0
2<tS
= q,
that is:
(A33.21)
From the implicit function theorem we deduce:
(1) equation (A 33.21) determines a torus T, close to p
wh~ch
Po and
(A 33.22)
I I
00
The mapping BN , being globally canonical, is given by a generating function of the form Pq + B (P, q). In view of Lemma (A 33.4), the function
B(x) = B(P(x), q(x)) is s-ingle valued in
a.
of B (x) to the torus T. This is a smooth function on an n-dimensionaltorus, thus B has at least 2" critical points (compare to Lemma A 33.S).
Let us prove that these critical points belong to the intersection of T
and BNT. Formula (A33.3) yields:
dB
(p, q)
->
(P(x), Q(x)).
0 on our torus
p.
(Q. E. D.)
REMARK
A 33.23
o need
not verify dp 1\ dq
JhJapping proves:
;r
APPENDIX 34
A. Newton s Method
Newton's approximation for the construction. of a zero x of a function
f, with a prescribed approximation, consists in replacing the curve y =
f(x) by its tangent at ~xo' f(x o))' where Xo is an approximation of x. If
Ix-xol < E . the substitution gives an error of order E2. Hence, the linearized equation
f(x o ) + f'(xo)(x ~ x o) = 0
conver~
gent approximation:
(A 34.2)
I' x n+ l-xl<Clx-x
n
n
n- 112
I
Hence, the deviation of the n tl :approximatlOn from the solution will be:
I
Ix -
x n 1 .....JE
249
2n-1
250
Figure A 34.1
This method extends easily to equations in Banach spaces 1. But, in analysis, one is most often concerned with poly-Banachic spaces. and the operator ['(xO)-l maps a space into another one. This situation is illustrated
by'the following lemma, which offers a characteristic example, and where
(A 34.4) plays the role of (A 34.2).
LEMM.A
A34.3 2
Let L be an operator which maps the functions (z), which are analytic in each complex domain G, into the functions Lf(z) which are analytic
in the domain 3 G-o and such that, for every 0
(A344)
< 0 <00 :
See KantoroVlch [ I
J.
2 This lemma was.already known by H. Cartan [1]. whose' paper [I] is one of
the first appearances of Newton's approximatlOns m theoretical analysis
3 Here and further,
251
APPENDIX 34
Prool:
Let
~
2v + 1
and u~
2
~ 3/2
= u1 ,
M2
... , U~ s+1
Ms3/2,
hence
M
= 02v+l
s
Then, if
we have:
Now, let
G 1 = G,
Then,
IIIG < Ml
IIIG s < M
G2=Gl-ol.'Gs+l~Gs-oS:
implies
ILsII Gs <
implies:
inG-o',
(Q. E. D.)
B. Small Denominators
~
k~O
'
252
(A34.5)
where g is an unknown 271 periodic function. This equation admits a "formal" solution:
(A 34.6)
g(q) =
gk' e,(k,q),
ei(k,c. _ 1
k~O
<
<
M always holds
gue measure zero), the function g (q), defined by (A 34.6), is analytic and,
for Ilmql
Here 00
< p-o,
we have: Ig(q)1
< Mo-II, II
2n+4, if 0<0<00.
n (K)
(A34.8)
lei(k,c. - 11
< K~II, II
> KN-l/
<
N. Let us de-
II
= n+2.
le,(k,c.>o) - 11
n (K) c
> K
Ik I-II,
no(K)
LEMMA A34.10
Almost every (in the Lebesgue measure sense) point Wo belongs to
n (K)
[3],
APPENDIX 34
253
Proof:
Let
lk,d = Iwollei(k,Cil) - 11
Then, clearly:
only on.
n.
mea~ (lk, d
n m :s.
Iwol.
Let:
w, Iw-wol
< dl .
since
!k I-v <
for v
00,
n+2 .
Therefore:
meas
n n (K)
o.
k"'O
(Q. E. D.)
Now, let
~ fk
f(q)
ei(k,q)
be an analytic function.
LEMMA
A34.11
(A) Iffor IImql < p we have If(q)1 < M, then Ifkl < Me-plkl.
(B) If Ifk(q)! < Me-P !kl, then for 11m ql < p-o (where 0 < 0 < 0 0 ):
< P we have
If(q)1
<
< P -0,
0<0 < 0 0 :
~ f n . ei(k,q) .
ikl>N
254
s~ift
fk =
to
f e-i(k, q) dq
. ip.
Proofs of (B) and (C) consist in mere summations of geometrical series.
Lemma (A34.7) follows at once from Lemmas (A34.10) and (A 34.11):
take Wo ( Do(K) and take into account (A34.6), (A34.9), (A) and (B) from
Lemma (A34.11), and the elementary inequality:
<
e-lkI5'lklv
C(v)o-V
<
K/C(v). We
A34.12
0 for Ik
>
<
<
N if [k = 0 for
Ikl > N.
c. Sketch of the Proof
Now, let us recall notations of Theorem (21.11) (see Chapter 4). The
set
~ Bn
is
invari~
255
APPENDIX 34
C(BA)C- 1
where B'
C B A C- 1 A-I.
B'A
Now, observe that this equation, for each fixed p, has precisely the form
of Equation (A34.5) (with w
= w(p),
f = -B,
n' c n
ried out as in Lemma (A 34.3); Inequalities (A 34.13) are proved with the
help of Lemma (A 34.7): they hold "far from resonances," that is for w(p)
( n(K).
5 This process was already used by Bogolubov. Mltropolski [1]. One could do
without it by replacing the frequencIes cu(p) with constants cu ( Q K in the small
denominators.
256
arbi~
IF I h M"
(A 34.14)
provided that 0 <
LEMMA
<
0,
M<
IF I :s.
MO- v 2
oV I
A34.15
IS
ana-
<
n M.
6 That is that these constants depend on the d,mens,ons of the domains dealing
with the proposition a, on the number of depivatives, etc., but neither on the functions nor on the domains.
257
APPENDIX 34
Proof:
From the Cauchy formula
dkf
k!
F()d(
277i . Y (_z)k+l
we deduce:
.E kf
\ dz k
:s.
1
k! MB- k
1
k'
o.
(Q.~. D.)
Here:
F (0, M) =
sup
f.G.z (G-O
dkf I
,
Idz
k
where sup ranges over all the bounded domains G, all the analytic
tions f,
If I
func~
Then. if
F(B, M) ~ M.
f.J
Now, let us make clearer how the global canonical mappings S, close
to the identity. are related to their generating functions Pq + S (P, q). Let
Tn
I,
LEMMA
A34.16
< M.
258
(A34.17)
as ,
q+-
ap
S (p, q)
P (p, q),
Q(p, q) ;
S: [0] - 20
->
[0] - 8 ,
I(P LEMMA
p) - as (p, q)
ap
I <
q) I <
I(Q- q) + as(p,
aq
n
M2
n'
M2.
A34.18
LEMMA
M,
(A 34.17),
IS(p, q) -
(p,q)
M2
A34.19-
ST of the corresponding
by an analytic gen"
->
([0
is 'given by Ip I < y', 11m q I < p', 0 < y', p' < 1). Let a-1ly - xi
< lA<r)-A(x)1 < aly-xl, and S be an analytic function in [OJ, S the canonical diffeomorphism corresponding to S by virtue of Lemma (A 34,16).
259
APPENDIX 34
LEMMA
A 34.20
The formula T
of [n'] - 38 into
to'] - 28.
Let us now begin the construction of the invariant tori of the mapping
B A (see Theorem 21.11. Chapter 4).
To understand the estimates performed in Sections E - F. it is useful
to keep in mind that the positive numbers {3. y. 8. M and p. K. 0 satisfy:
0,< M 8
and that vi'
Ci
are absolute
(3
constants~ v
p. K. 0- 1
<1
> 1 > c.
A 34.21
!\: P. q
-+
P. q + w( q)
r. f
ii(p)
= (21T)-n
w 1(p)
aii
ap
= w(p) + -
Figu~
7 In thiS lemma, the constant 50' which enters mto the defmihon of
also on a.
A 34.22):
, depends
260
Figure A 34.22
Of course, BA = B'A 1
LEMMA CONCERNING THE VARIATION OF THE FREQUENCIES. A34.23
Suppose that:
O-l\dp\
hold in the domain \p - p*\ < y, \Im q \ < p. Then the global canonical
mapping B' possesses a generating function Pq + B '(P, q) which satisHes 8 :
where
y-o, \Imq\
<
p-o.
Proof:
Application of Lemma (A34.19) to the mappings Band AA1l: p, q
-->
F. Fundamental Lemma
We now make use of the estimates of the small denominators (Section B)
to obtain inequallties of the form \B 1\ ~ M2, after a suitable change of
coordinates C.
8 The constant 50' m the relation
eventually depends on
e.
261
APPENDIX 34
A34.24
FUNDAMENTAL CONSTRUCTION.
A: p, q
-+
p, q + wI (p) ,
~ B k(P)ei(k, q)
B(P) +
k,t 0
We put: -
C (P, q)
Ck(P)ei(k,q) ,
O<lkl~N
(A 34.25)
Bk(P)
e-i(k,Ull (P)) _
where N is a positive integer. We denote by C the global canonical mapping, whose generating function is Pq + C (P, q), and we consider the
global canonical mapping Bl
C B A C- 1 A-I. Of course:
BIA
(see Figure A 34.26).
C(BA)C- 1
B ""'C
A
B1 .
B
cl
----
Bl
F,gure A34.26
FUNDAMENTAL LEMMA
We suppose that
wI (p),
in
A34.27
the domain
q)1 <
M,
p, the functions
IB(p)1 < Ii
262
(2) in the domain, IP - p*1 < y -5, 11m q I < p -{3, the generating function Pq + B 1 (P, q) of B1 satisfies:
(A34.28)
provided that [for (1) and (2) l:
Proof:
From Idwl < OldPI and y < C4 N-(n+2), C 4 < K/O, it follows that
all the points w 1 (p) of the domain Ip-p*1 < y ,:erify Iw-w*1 < KN-(n+2).
Consequently, according to the last remark of Section 8", the assertion of
Lemma (A 34.7) holds"for the trigonometric sum C. Let 50 ~ 5 0(n, K) be
t~e constant of Lem~a (A 34.7). If C 3 < 00' then we have 5 < 50; therefore Lemma (A 34.7) implies that IC I < M5- vI for jIm q I < p -8. This is
Or < p.
9 That is. tpey depend only on the dimension n and the constants K,
e.
The fun-
damental lemma states that there exist constants vI' v2 (large enough) and C l'
C 2 , C 3 , C 4 (small en0':1gh) such that (1) and (2) hold.
263
APPENDIX 34
(A34.30)
But from (A34.2S) follows:
(A34.31)
If 11m q I < P - (3 < P - 8, then assertion (C) of Lemma (A 34.11) implies:
IRNB
(A34.32)
Since 181
A 34.33
-+
p, q + w(p) ,
-+
p, q + w 1 (p) ,
B'
"1'
264
B1A1
C(BA) C- 1
c ._ _ _ _ _ _ ---.:}
Figure A 34.34
THE INDUCTIVE LEMMA
A 34.35
w(p*)
w*
O(K) .
P:
is analytic and
IC I <
M o-Vt ;
main IP - p*1 < y, 11m q I < p, and in this domain the function B1 is analytic and
provided that:
1/1,1/2,1/3' 1/4
solute constants, that is depend only on the dimension n and on the constants
eo'
K, but not on B, w,
e,
M, and so on.
265
APPENDIX 34
Proof:
Proof of the inductive lemma follows at once from the two preceding
lemmas. The only two novelties are:
P: :
Ip: I <
ities:
IdWI
dp
(see Section E) and from M8- v
<
<
_ dw
dp
belongs to
<
M8- v
Ip - p:1 < Y1
I
<
< 8 v2- v
C4 y, the domain
~ ow
M<
M28- V4
(Q. E. D.)
H. Proof of Theorem (21.11)
(see Chapter 4)
CONSTRUCTION A34.36
< Ns <
00',
Ns
-+
H2
we have:
B2A2
C1(R1 A1)C 1- 1 .
266
000
As
Figure A 34.37
This
construction also determines the points p*,
ws(p*)
s
s
Ts(w*) denotes the torus p
T:
be the torus p
P:,
Now
p:
lim
s~""
A : T*
""
= w*.
""
~ T* ,
""
A""q
q +w
Finally, we put:
os --
C 1-1 C2-1
...
C s-1
-1
11m
Us
s~""
""
We now prove that the above limits exist and that D:\""
B -\ D on T (w * ).
(Q. E. D.)
267
APPENDIX 34
A 34.38
CONVERGENCE,
In view of dw/dp
0-11 dp I
2(),
Let us put:
oY,s '
f3 s
Y '/.o(n+2)
s
-Y,(n+2)
Ys
'
then we have:
f3 s+ 1 = f3 s3/2 '
N s+l
o.
Ys+l
y3/2
s'
Ns3 / 2
Denote by a a
Let C, 0 < C < 1/10, be an absolute constant, that is, C depends only
on n, K, 00' a and on the constants
tive lemma. If 0 < C,' then,
\!k'
induc~
obviou~:;ly:
(1)
(A34.39)
(2)
(A 34.40)
(3) for any s
(A 34.41)
268
eo
we introduced above;
(4)
(A 34.42)
Our number 8 is nowoselected to satisfy: 0 < 8 < C. Hence, the numbers
N s' on which our construction depends, are determined.
Pq+ B 1 (P, q} which verifies IB11 < M1 = 8 1a in Ip-p~1 < Y1' 11m ql <
P1 = Y:zp. According to Lemma (A 34.18) this inequality holds if E, under
by A
p;,
P1 - (31' In fact, (A34.39~ implies P2 > 0; (A 34.40) and the third part
of the inductive lemma show that "2 satisfies the inequalities:
Finally, the second part of the inductive lemma, (A34.42), and a>
a >
1J 1
21J 4
+ 1,
+2, imply:
M 1 e- iV'I1
In other words:
Let G s be the domain Ip-p!1 < Y s ' IImql < P s ' Then,thediffeomor
1 map G 1 into G and, in the C 1.norm, we have:
phisms Cs
s+
s
APPENDIX 34
269
E = identity,
(A 34.43)
in G s+I' The point
p:
...
00
'.
From the estimate (A 34.43), follows at once the convergence of the ns
on
the torus
r*
00
nG
s>1
The inequalities
on the torus
IB sl < 8~
imply;
r*.
00
wsl < 8 s
pings As:
Thus'
'\
00'
on
r*
00
(Q. E. D.)
BIBLIOGRAPHY
Abramov. L. M.
'[l] Concerning a note of Genis, R Z Math. 8 (1963) p. 439.
[2] Metric Automorphisms with Quasi-Discrete Spectrum, lsvestia Math.
Series 26 (1962) pp. 513-531.
Anosov. [). V.
[1] Roughness of Geodesic Flows on Compact Riemannian Manifolds
of Negative Curvature, Dokl. Akad. Nauk. 145 (1962) p. 707-709.
[Sov. Math. Dokl. 3 No.4 (1962) pp. 1068-1069.]
[2] Ergodic Properties of Geodesic Flows on Closed Riemannian Manifolds of Negative Curvature, DokI. Akad. Nauk. 151 (1963) pp.
1250-1253. [SOy. Math. Dokl. 4 No.4 (1963) pp. 1153-1156.]
[3] A,feraging in Systems of Ordinary Differential Equations with Rapidly Oscillating Solutions, lsvestia, Math Series 24 (1960) pp. 721742.
Anzai. H.
[1] Ergodic Sltew Product Transformations on the Torus, Osaka Math.
J. (1951) pp. 83-99.
Arnold. V. I.
[1] Sur la geomtHrie des groupes de difeomorphismes et ses applications en hydrodynamique des fluides parfaits, Ann. lnst. Fourier
(1966).
[2] Remarks on Rotation Numbers, Sibirski Math. Zh. 2 No.6 (1961)
pp. 807-813.
[3] Some Remarks on Flows of Line Elements and Frames, Dokl. Akad.
Nauk. 138 No.2 (1961) pp. 255-257. [SOy. Math. Dokl. 2 (1961) pp.
562-564.]
271
272
[4] Small Denominators and Problems of Stability of Motion in Classical and Celestial Mechanics, Usp. Math. Nauk. 18 No.6 (1963) pp.
91-196. [Russian Math. Surveys 18 No.6 (1963) pp. 85-193.]
[5] Proof of a Theorem of A. N. Kolmogorov on the Invariance of QuasiPeriodic Motions under Small Perturbations of the Hamiltonian,
Usp. Math. Nauk. 18 No.5 (1963) pp. 13-40. [Russian Math. Surveys 18 No: 5 (1963) pp. 9-36.]
Arnold. V. I.
[7] On the Stability of Positions of Equilibrium of a Hamiltonian Sys. tern of Ordinary Differential Equations in the General Elliptic Case,
Dokl. Akad. Nauk. 137 No.2 (1961) pp. 255-257. [SOy. Math Dokl.
2 pp. 247-279.]
.
[8] On the Generation of a Quasi-Periodic Motion from a Set of Periodic
Motions, Dokl. Akad. Nauk. 138 No.1 (1961) pp. 13-15. [SOy. Math.
Dokl. No.2 pp. 501-503.]
[9] On the Behavior of an Adiabatic Invariant under a Slow Periodic
Change of the Hamiltonian, Dokl. Akad. Nauk. 142 No.4 (1962) pp.
758-761. [SOy. Math. Dokl. No.3, pp. 136-139.]
[10] On the Classical Theory of Perturbations and the Problem of Stability of Planetary Systems, Dokl. Akad. Nauk. 145 No.3 (1962)
pp. 487-490. [SOy. Math. Dokl. No.3 (1962) pp. 1008~101l]
[ll] Small Denominators I, on the -Mapping of a Circle into Itself, Izvestia Akad. Nauk. Math. Series 25, 1 (1961) pp. 21-86. [Transl.
Amer. Math. Soc., Series 2,46 (1965) pp. 213-284.]
[12] Conditions for the Applicability, and Estimate of the Error, of an
Averaging Method for Systems which Pass through States of Resonance in the Course of Their Evolution, Dokl. Akad. Nauk. 161
No.1 (1965) pp. 9-12. [SOy. Math. Dokl. 6 No.2 (1965) pp. 331337.]
[13] Instability of Dynamical Systems with Many Degrees of Freedom,
Dokl. Akad. Nauk. 156 No.1 (1964) pp. 9~12. [SOy. Math. Dokl. 5
No.3 (1964) pp. 581-585.]
[14] Stability and Instability in Classical Mechanics, Second Summer
Math. School, 1964 2, Kiev (1965) p. 110.
.
BIBLIOGRAPHY
273
J. R. and Hanson O. L.
[1] On the Isomorphism Problem for Bernouilli Schemes, Bull. Am.
Math. Soc. 69 No.2 (1963) pp. 221-223.
Blum,
n.
274
a la sur-
Eul('r. L.
[1] Theoria Motus Corporum Solidorum Seu Rigidorum (1765).
Ge Ifand. I. M. and Shapiro-Piatct.lski. I. I.
[1] A Theorem of Poincare, Dok!. Akad. Nauk. 127 No.3 (1959) pp.
490 493. [Math. Review (1960) No. 6460.]
0
.\. I.
[3] All Example of a Toroidal Magnetic Field Not Having Magnetic
Surfaces, Dok!. Akad. Nauk. L-rJ No.1 (1962) pp. 81083. [SOy.
Phys. Dok!. 7 pp. 223 724.]
0
Gourcvitch. B. M.
[1] The Entropy of Horocycle Flows. Dok!. Akad. Nauk. I % No.4
(1961) pp. 768 770. [SOy. Math. Dok!. 2 (1961)pp. 124-130.
0
BIBLIOGRAPHY
275
Grant, A.
[1] Surfaces of Negative Curvature and Permanent Regional Transitivity, Duke Math. J. No.5 (1939) pp. 207-229.
Guirsanov, I. V.
[1] On the Spectra of Dynamical Systems Which Arise from Stationary
Gaussian Process, DokI. Akad. Nauk. 119 No.5 (1958) pp. 851853.
Gysin, W.
[1] Zur Homologietheorie der Abbildungen und Faserungen von Mannigfaltigkeiten, Comment. Math. Helv. 11 (1941) pp. 61-122.
Hadamard. J.
[1] Les surfa~es a courbures opposees et leurs lignes geodesiques,
J. Math. pures appl. (1898) pp. 27-73.
Hajian. A. B.
[1] On Ergodic Measure Preserving Transformations Defined on an
Infinite Measure Space, Proc. Am. Math. Soc. 16 (1965) pp. 45-48.
Halmos. P. R.
[1] Lectures on Ergodic Theory, Chelsea, New York (1958).
[2] Measure Theory, New York (1951).
[3] Introduction to Hilbert Spaces, Chelsea, New York (1 q57).
Hedlund. G.
[1] The Dynamics of Geodesic Flows, BuIl. Am. Math. Soc. 45 (1939)
pp. 241-246.
Helgason. S.
[1] Differential Geometry and Symmetric Spaces, Academic Press,
New York (1962).
Henon. M. and Heiles. C.
[1] The Applicability of the Third Integral of Motion: Some Numerical
Experiments, Astron. J. 6q No.1 (1964) pp. 73-79.
Hopf, E.
[1] Ergodentheorie, Springer, Berlin (1937).
lusvinski i. S. A.
[1] On Metrical Automorphisms with Simple Spectrum, DokI. Akad.
Nauk. 172 No.5 (1967).
276
Jacoubovitch, V. I.
[1] Questions of the Stability of Solutions of a System of Linear Differential Equations of Canonical Form with Periodic Coefficients,
Mat. Sbornik. 37,79 (1955) pp. 21-68. [Transl. Am. Math. Soc. 2
10 (1958) pp. 125-175.]
Kagan, V. F.
[1] Foundations of the Theory of Surfaces, Osnovy Teorii Poverchnoster I, Moscow (1947).
Kantorovitch, L. B.
[1] Functional Analysis and Applied Mathematics, Usp. Math. Nauk.
3 No.6 (1948) pp. 89-185.
Kasuga, T.
[1] On the Adiabatic Theorem for the Hamiltonian System of Differential Equations in the Classical Mechanics, I. II. III., Proc. Japan.
Acad. 37 No.7 (1961) pp. 366-382.
Katok, A. B.
[1] Entropy and Approximation of Dynamical Systems by Periodical
Mappings, Funkzionalnyi Analys i ego Prilojenija, Moscow, I,
No.1 (1967) pp. 75-85.
[2] On Dynamical Systems with Integral Invariants on the Torus,
Funkz. Anal. i ego Prilojenija, Moscow I No.3 (1967).
Katok, A. B. and Stepin, A. M.
[1] On the Approximations of Ergodic Dynamical Systems by Periodical Mappings, Dokl. Akad. Nauk. 171 No.6 (1966) pp. 1268-1271.
Kolmogorov, A. N.
[1] Sur les systemes dynamiques avec un invariant integral la surface du tore, Dok!. Akad. Nauk. 93 No.5 (1953) pp. 763-766.
[2] A New Metric Invariant of Transitive Systems and AutomorphisJ.T1s
of Lebesgue Spaces, Dokl. Akad. Nauk. 119 (1958) pp. 861-864.
[Math. Review 21 No.2035a.]
[3] Foundations of Probability Theory, Chelsea, New York (1956).
[4] On the Entropy per Time Unit as a Metric Ihvariant of Automor- .
phisms, Dokl. Akad. Nauk. 124 (1959) pp. 754-755.
[5] La theorie genera Ie des systemes dynamiques et la mecanique
classique, Amsterdam Congress 1(1954) pp. 315-333. [Math. Review 20 No. 4066.]
[6] On the Conservation of Quasi-Periodic Motions for a Small Change
in the Hamiltonian Function, Dokl. Akad. Nauk. 98 No.4 (1954)
277
BIBLIOGRAPHY
Levi-Civita, T.
[1] Sopra alcuni criteri di instabilita, Ann. Mat. pura appl. 3,5 (1901)
pp. 221-307.
278
Margoulis, A.
[1] Sur quelques questions Liees la theorie des C-systemes d'Anosov
(Russian) Int. Congress, Moscow VIII (1966).
Melnikov. V. K.
[1] On the Stability of a Center for Time Periodic Perturbations, Trudy
Moskovskogo Math. Obschestva 12 No.3 (1963) pp. 3-53. [Math. Review No. 5981 (1964).]
[2] On Some Case of C~nservation of Conditionally Periodic' Motions
under a Small Change of the Hamiltonian Function, Dokl. Akad.
Nauk. 165 (1965) pp. 1245-1248.
Meshalkin, L. D.
[I] A Case of Isomorphism of Bernouilli Schemes, Dokl.
128 (1959) pp. 41-44.
A~ad.
Nauk.
Milnor, J.
[I] Morse Theory, Ann. Math. Studies, Princeton (1961).
Mosel', J.
[1] On Invariant Curves of Area-Preserving Mappings of ail Annulus,
Nachr. Akad. Wiss. Gottingen No.1 (1962).
[2] On Invariant Surfaces and Almost Periodic Solutions for Ordinary
Differential Equations, Am. Math. Soc. Notices, 12 No.1 p. 124.
[3] New Aspect in the Theory of Stability of Hamiltonian Systems,
Commun. Pure Appl. Math. II (1958) pp. 81-114.
[4] A New Technique for the Construction of Solutions of Non-Linear
Differential Equations, Proc. Nat!. A cad. Sci. U. S. 47 (1961)'pp .
I
. 1824-1831.
[5] . On Theory of QuaSi-Periodic Motion, Siam Review (1966).
[6] A Rapidly Convergent Iteration Method and Non-Linear Partial Differential Equations 'I, II, Annali Della Scuola Norma/e Superiore di
Pisa, Serie III 20 Fasc. II, III (1966) pp. 265-315; pp. 499-535.
[7] Convergent Series Expansions for Quasi-Periodic Motions, Math.
Annalen, 169 (1967) pp. 136-176.
[8] On a Theorem of V. Anosov (in press).
Moser, J. and Jeffrey>, W. II.
[9] QuaSi-Periodic Solutions for the Three':'Body Problem, Astron. J.
71 No.7 (1966) pp. 568-578.
Nemytskii, V. and Stepanov, V. V.
[I] Qualitative theory of Differential Equations, Princeton (1960).
BIBLIOGRAPHY
279
A. P.
[1] Etude des oscillations d'urt satellite dans Ie plan d'une orbite elliptique, Kosmicheskie Issledovania 2 No.5 (1964) pp. 657-666.
Peixoto, M. H.
[1] On Structural Stability, Ann. Math., Series 2, t. 69 (1959) pp. 199-
222.
Poincare If.
[1] Oeuvres completes, 1 pp. 3-221.
[2] Les methodes nouvelles de la mecanique celeste, 3. Gauthier-Villars,
Paris (1899).
[3] Sur un theor~me de geometrie, Rendiconti circolo Mathematica di
Palermo, t. 33 (191.2) pp. 375-407.
Polya, G. and Szego, G.
[1] Aufgaben und Lehrsaetze ausder Analysis, 1, 2nd ed. Springer, Berlin (1954).
Rohlin, V. A.
[1] In General a Measure-Pr~serving Transformation Is Not Mixing,
Dokl. Akad. Nauk. 13 (1949) pp. 329-340.
.
. [2] .On Endomorphisms of Compact Commutative Groups, Izvestia Math.
Nauk. 13 (1949) pp. 329-340.
[3] On the Fundamental Ideas of Measure Theory, Mat. Sbomik (N. S.)
25,67 (1949) Pi>. 107-150. [Transl. Am. Math. Soc. 1,10 (1962) pp.
1-54.]
[4] Exact Endomorphisms of Lebesgue Spaces, Izvestia Akad. Nauk.
25 (1961) pp. 499-530. [Transl. Amer. Math. Soc. 2, 39(1964)pp.1-36.l
Rohlin, V. A. and Sinai, Ya;
[5] Construction afld Properties of Invariant Measurable Partitions,
Dokl. Akad. Nauk. 141 No.6 (1961) pp. 1038--1041.
[Sov. Math. Dokl. 2 No.6 (1961) pp. 1611-1614.]
.
Schwartzman, S.
[1] Asymptot'ic Cycles, Ann. Math. 66 No.2 (1957) pp. 270-284.
Siegel,
[1]
[2]
[3]
C. L.
Iterations of Analytical Function, Ann. Math. 43 (1942) pp.607-612.
Vorlesungen iiber Himmelsmechanik, Springer, Berlin (1956).
Uber die Exisfenz einer Normalform analytisclter Hamiltonsche
Differentialgleichungen in der Nahe einer GleichgewichtslOsung,
Math. Annalen. 128 (1954) pp. 144-170.
280
Sinai. Ya.
[1] The Central Limit Theorem for Geodesic Flows on Manifolds of
Constant Negative Curvature, Dok!. Akad. Nauk. 133 (1960) pp.
1303-1306. [Sov. Math. Dok!. I No.4 (1961) pp. 983-987.]
[2] Properties of Spectra of Ergodic Dynamical Systems, Dokl. Akad.
Nauk. 150 (1963) pp. 1235-1237. [SOy. Math. Dok!. 4 No.3 (1963)
pp. 875-877.]
[3] Some Remarks 'on the Spectral Properties of Ergodic Dynamical
Systems, Usp. Mat. Nauk. 18 No.5 (1963) pp. 41-54. [Russian Math.
.
SurVeys 18 No.5 (1963) pp. 37-51.]
[4] On the Foundations of the Ergodic Hypothesis for a Dynamical
System of Statistical Mechanics, Dokl. Akad. Nauk. 153 No.6
(1963). [SOy. Math. Dokl. 4 No.6 (1963) pp. 1818-1822.]
.
[5] Vestnik Moscovskovo Gosudrastvennovo Universitata. Series Math .
. No.5 (1962).
[6] Dynamical Systems with Countably Lebesgue Spectra, Izvestia
Math. Nauk. 25 (1961) pp. 899-924. [Transl. Amer Math. Soc.,
Series 2, 39 (1961) pp. 83-110.]
[7] On the Concept of Entropy of a Dynamical System, Dokl. Akad.
Nauk. 124 (1959) pp. 768-771. [Math. Review 21 No.2036a.]
[8] Letter to the Editor, Usp. Math. Nauk. 20 No.4 (124) (1965) p. 232.
[9] A Weak Isomorphism of Transformations with an Invariant Measure,
Dok!. Akad. Nauk. 147 (1962) pp. 797-800. [SOy. Math. Dok!. 3
(1962) pp. 1725-1729.]
[10] Geodesic Flows on Compact Surfaces of Negative Curvature, Dokl.
Akad. Nauk. 136 (1961) p. 549. [SOy. Math: Dok!. 2 No.1 (1961)
pp. 106-109.]
[ll] Dynamical Systems with Countably Multiple Lebesgue Spectra II.
Izvestia Math. Nauk. 30 No.1 (1966) pp. 15-68.
Sitnikov, K.
[1] The Existence of Oscillatory Motions in the Three-Body Problem,
Dokl. Akad. Nauk. 133 No.2 (1960) pp. 303-306. [SOy. Phys. Dokl.
5 (1961) pp. 647-650.]
Slater, N. B.
[1] Distribution Problems and Physical Applications, Compositio
Mathematica, 16 fasc. 1.2 (1963) pp. 17&-183.
Smale, S.
[1] DynamiCal Systems and the Topological Conjugacy Problem for
Diffeomorphisms, Proc. Int. Congress Math. (1962) pp. 490-496.
BIBLIOGRAPHY
281
[2] Structurally Stable Systems Are Not Dense, Am. ]. Math. 88 (1966)
pp. 491-496.
[3] Differentiable Dynamical Systems, Rull. Am. Math. Soc. n (1967)
pp. 747-817.
Slepin, A. M.
[1] O~ th~ Approximation of Dynamical Systems by Periodic Mappings
and Its Spectrum, Funkz. Analys i ego Prilojenija, Moscos 1 No.2
(1967).
Weyl, H.
[I] Sur une application de la theorie des nombres fI la mecanique statistique, Enseignement Math. 16 (1914) pp. 455-467.
[2] Uber die Gleichveirteiligung von Zahlen mod. 1, Math. Annalen 77
(1916) pp. 313-352.
[3] Selecta Hermann Weyl, Bassel-Stuttgart (1956) pp. 11-147.
[4] Mean Motion I, Am. ]. Math. 60 (1938) pp. 889-896.
[5] Mean Motion II, Am. ]. Math. 61 (1939) pp. 143-148.
Winlner, A.
[I] Upon a Statistical Method in the Theory of Diophantine Approximations, Am. ]. Math. 55 (1933) pp. 309-331.
INDEX
A- Entropy, 48
Abstract dynamical system, 7
284
INDEX
Discrete spectrum
theorem, 27
Geodesic flow
of the torus, 3, 12, 18, 94, 117
Dynamical system, 1
Elliptic point, 218
Endomorphism, 124, 143
Entropy, 24, 35
of an automorphism, 40, 163
of a classical system, 46
of a flow, 45
of partition, 36
of a partition with respect to an
automorphism, 38
of a partition a with respect to a
partition (3, 37, 158
Equipartition, 129, 135
Ergodic system, 12, 26, 134
classical, 147
Ergodicity, 16, 20, 24
EuleroPoinsot motion, 119
Fibre bundle, 57
Foliation, 59, 185
absolutely continuous, 75
co.ntracting, 73
dilating, 73
ergodic, 73
Generating function, 99, 235, 243
Generator, 41, 163
Generic elliptic type, 219
Generic property, 12
Geodesic flow, 3, 18, 151
of the ellipsoid, 3, 94, 96
of the Lie groups, 3, 120
of the manifolds of negative curo
vature, 34, 60, 76, 136
285
INDEX
Spectral invariant, 22
Spectral measure, 24
Spectral multiplicity, 24
Stability (theory of), 85
Stability of fixed point, 219
Stable system, 81
286
INDEX
Structural stability, 64
of C-system, 64
problem (see Smale theorem)
Surface of section, 98, 230
Swing, 81
Symplectic
manifold,S
mapping, 212, 215, 221
(see also Linear symplectic
mapping)
System of evolution, 101, 108
Torus
translation, I, 5, 11, 16, 18, 21,
115, 130, 132, 135
Transition chain, 111
Transition torus, 110
Unsta~le
system, 53