Ekeland 1989
Ekeland 1989
Ekeland 1989
Contents
I. Symplectic Capacities
* The results were obtained while the first author held a Distinguished Visiting Professorship at
Rutgers and the second author was a member of the Institute for Advanced Study
** Research partially supported by NSF Grant DMS-8603149, the Alfred Sloan Foundation, and a
Rutgers Trustees Research Fellowship grant
356 I. E k e l a n d and H. Hofer
~ f * co < a.
S
In the next section we will give an alternate construction in symplectic vector-
spaces V utilizing Hamiltonian systems. This allows to define a capacity for any
358 I. Ekeland and H. Hofer
Acknowledgement. We would llke to thank ProfessorM. Gromov for very stimulating discussions.
(x,y) = ( x o , Y o ) + 2 n . r l k l ( x k , Yk).
(4) S + ~ h i ( E - @ E ~ (~ [0, + ~ ) e ) 4= 4
for all t ~ [ 0 , 1]. I f x e E - 9 E ~ we have a(x) <-_0 so that h,(x) = x by condition (3).
Hence
1 - IIht(x)II = 0
to be solved for x e f2. This is equivalent to
x + T~(x)=O, x~O
with
Tt(x ) = t(P ~ + e -'~- (x) P - )k(x) - (~ + 1 - IIh,(x) I])e
x=y+~e, yeE- @ E~
We have just seen that
(7) x + T t ( x ) 4=O for all t e [0,1] and xec3P.
F o r t = 0, we have
x+ To(x)=x-(~+ 1 --II xll)e
=y-(1 -Ijxll)e.
Since the m a p (t,x)---, Tt(x ) is compact, we can apply Leray-Schauder-Degree
theory. Using (7) we see that
deg(I + T1, O, 0) = deg(I + To,f2,0 ).
Symplectic Topology and Hamiltonian Dynamics 361
deg(I + T o, Q, 0) = d e g ( I - e, O, 0)
= deg(I, Q, e)
zl.
This completes the proof. []
h(S +) = x o + e S +
We m a y assume that H is zero in a n e i g h b o r h o o d Of Xo. Since H ( x o ) = O, H ' ( x o )
= 0 and H " (x o) = 0 we see that
cu =< infcbH(ho(S+)) + 89
N o w introduce three locally Lipschitz continuous maps qo1, q~2 , ~o3: E ~ [0, 1]
such that
~1 (X) = 0 if cku(x ) (i [cH - go, CH + gO]
Cpl(X)= 1 if q~u(X)E [On -- 89 + 89
(pz(X)= 1 if II ~;~(x)II < 1
(12)
~o2(x)= 11q,~(x)l1-1 if ]]~h(x)tl .> 1
2
~3(x) = 1 if IIXI] __<(sup 11ho(S +) ]l) + - -
go
d
~(x,t) __>6o~
If we can show that h e F then we have obtained the contradiction c~ > cn. This
would prove that cn is a critical level. We define a m a p d: R x E ~ N by
a(t,x) = ~o~(x.t) ~o~(x.t) q~ (x.t).
Symplectic Topology and Hamiltonian Dynamics 363
1
We put r = and define
~0
9+ ( s , x ) = i d ( a , x ) d a
s
+ (x) = i + (0, x)
This gives the desired decompositions of h. See for example [18] for more
details. []
Proposition 3. Let H E ~ and let s--+ 7is be a smooth homotopy of the identity in
De(C"). Define H ~ E ~ by H~ = H ~ ~s. Then the map
S ~ COs
is constant
Proof The critical levels of q~s:= ~n, are independent of s, since q~Ho~(q)-a(x))
= q~n(x) for q~ exact symplectic and since q~-l(x) is a solution for the Hamiltonian
vectorfield associated to H o q) iff x is a solution for the Hamiltonian vectorfield
associated to H. Further, there is an estimate of the form
That for example cn, ~ co is clear since lB.(x) - H(x) l < e, for every x e C" where
e,--* 0. F r o m this it follows immediately that I c n . - cnl < e,.
Next we need
364 I. Ekeland and H. Hofer
Proof. Since for large Izlwe have H(z) = ~Izl2 we see that H~(z) = ~Izl2 for large Izl,
too. Since ~ > 7z and ti~rt N again the Palais-Smale condition holds, 1-17]. I f x is a
1-periodic solution of 2 = X n ( x ) then xz defined by
xa(t) = 2 - 2 x ( t )
In other words ifX c ~ is the set of critical levels ofq~ n, then X~:= 2 - 2 z ~ is the set
of critical levels of ~uz- Arguing as in Proposition 3 we see that the m a p 2 ~ 2 2 c~ is
continuous. If cu is an isolated critical level the m a p 2 ~ 2 2 cz has to be constant.
Finally, using a genericity argument we conclude that in all cases
2 2 C2 = e H
c(~'(s)) = ~c(S)
where S is a bounded subset.
Proof Assume first a = 1. Then we find a s m o o t h arc s --* ~s in Dr ~) such that
qq (S) = ~(S).
By Proposition 3 we obtain
CHo ~,~, = c u for H ~ ~ (S)
and therefore
e(~'(s)) = e(Cq (s)) = e(S).
Next assume a > 0 is different from one. Then a - ~ T J e D ( C ") and by the
previous step
c(~-~ 'e(s)) = e(S).
Next we employ Proposition 4 in order to obtain
c(:~-~ 7"(s)) = ~- ~ (7'(s)).
Symplectic Topology and Hamiltonian Dynamics 365
Summing up we find
c(~'(s)) -- ~c(S)
as required. []
Next we define for any subset S of C" a number c(S) ~ [0, + oo] by
(14) c(S) = sup {c(T)l T is bounded, T = S }.
As a corollary of Proposition 5 we find
Corollary 1. For any subset S of C" and any diffeomorphism ~P of C" such that ~P* ~o
= eco for some ct > 0 we have
(15) c ( ~ ( s ) ) = ~c(S).
Proof The first part is evident by Proposition 5. For the second part we just note
that if S ~ T, then Y ( T ) c i f ( S ) . []
In view of (15) we can now define for any subset S of a symplectic vectorspace
(V, co~) a number c,(S) by
(16) c~(S): = c(7J(S))
where 7J: ( V, co~) ~ (C", me, ) is a linear symplectic diffeomorphism. Clearly c,(S) is
well defined. We shall show that c is the desired capacity. From our definition it is
clear that we may assume in the following that (V, co~)=(C",co). From our
previous discussion we know that (A2) and (A3) hold. So it remains to prove (A1)
(Normalization). In fact, we shall show even more, namely the properties claimed in
Theorem 1.
m a p g: ~ ~ N such that
g(s) = b for s < ro
g(s)= (k + ~ ) ~ for s large
g(s) >= (k + 1) ~s 2 for s _-> r o
O< g'(s) <= (2k + 1)~s for s > r o.
Moreover, let (o: ~ ~ N be a smooth m a p such that
0 if z s B a
(p (e) if z e A~
(18) H(z) =
b i f z ~ B~f~ ,[z[ =< r o
g(Izl) if [zl > ro.
Define
c(Ba)b, k = inf{ cn [ H e ~ ( B a ) ~ , k } .
N o w we construct a particular sequence (Hm), where Hm is as in (18) and (o = cp,,
and g = gin' F o r that we choose the fl~' and fl~' in such a way that
(20) go'~(s)6Z s if g0,,(s)e [fl~', b - fiT]-
N o w given any H e Y ( B a ) b , k we can pick such an H,, such that H,, => H.
(Perhaps we have to modify the gin, but positive critical levels do not depend on gin.)
Symplectic Topology and Hamiltonian Dynamics 367
Hence
cn~ ~ c(B a)b,k.
Using the choice of b and the definition of H " we can argue as in [17] to find
critical points x,, such that
9 u~
' (Xm) = 0, C,m = 4~H,,, (x')
x'([0, 1])= A~, ~m~(fln~,flT).
,~.(2") __>cl2"1
provided m is large enough, since e" ~ 0. Consequently the length can be estimated
by the action.
As in [17] changing perhaps the parameterization of the x " we can use the
Ascoli-Arzela-Theorem to find lb ~ 7/, lb > 1, and Pb ~ [A] such that
(21) c (B a)b, k = Ib A (P)
or c(Ba)b,k = lbA(Pb) - - b.
The m a p b~c(Ba)b, k is nonincreasing. Hence the m a p b--*lbA(Pb) must be
nonincreasing in both cases in (21) since it takes values in a discrete set 22 Since
C(Ba)b,k = lbA(Pb).
Since c (B a)b, k --* C(Ba) as b, k ~ + oo and since X is discrete, we find P e [A ] and
k e N, k > 1, with
c(Ba) = k A (P)
as required. So far we assumed that 22 is discrete. If S is not discrete we take a
generic A being C~-close to A. This can be done due to results in [20]. N o w being
of restricted contact type is a C l - o p e n condition. So we can a p p r o x i m a t e A by
368 I. Ekeland and H. Hofer
z] e R so that our arguments hold for A. We can pick k'e N, k" > 1, a n d / 5 e [zt] such
that
c(Bs) = kA(P).
If A ~ A we find c(Bs) ~ c(Ba). In fact, we have if z~ is close enough to A,
Ba_ ~ ~ B 5 c Ba~
(The convergence A --* A is understood in the following sense: z~ is the graph of a
section of the n o r m a l bundle of A and the section converges in C ~ to the
zerosection.) (*) implies using the monotonicity of c,
e-~c(Ba) <c(B5) <-_e%(Ba).
Moreover, since length (/3) __<cA (/3) if z] --+ A for some constant c > 0 (since
A e R), we can use the Ascoli-Arzela-Theorem, as in [17] for a suitable parametri-
zation of the/~ to find k e N , k >__ l, and PE[A] such that
c(Ba) = kA (P).
By our definition of c(S) we immediately see that
c(A) < c(n~)
We like to show equality. In order to do so we go through our construction
which gives (18) again. Define 7a: C" --, N as follows. Pick for a > 1, za: R --+ N such
that
1
ra (s) = a for s < - -
a
1 1
z'a(s)<0 for--<s< ---
a 2a
1
% (s)= 0 for s >
= 2a"
N o w define 7a by
I a for zeBal_~ I1
ya(Z)= Za(e) forzEa~,---<e<0
a
0 for zCB a.
Hence the positive critical levels of ~ba, and q~a are the same. By a genericity
argument the m a p (0, + o o ) ~ (0, + oo): s ~ ca+sy, has to be constant since
we infer
co >= CH+~, a = c a > e(Ba).
This implies
e(A) > c(Ba)
proving our desired result since we already k n o w that c(A) < c(Ba). []
Finally we have to show the normalization property
= soe(t ).
Hence
e A x ) = ~so~ - q,(So)
__< ~so~
< zc(1 + 2c5)2.
Summing up this implies that
c(B e) <= c11 < ~z(1 + 23) 2.
Since 6 > 0 was arbitrary we obtain
(25) c(Be) < ~.
On the other hand, we have by monotonicity
(26) c(B2"(1)) < c(Be) < zc.
7", (z) = lt 7' (tz) for t ~ (0, 1) and 7"0 (z) = 7" (0)z. Then t --* 7", is a continuous arc in
the space of symplectic embeddings of B(1) into C". We can smoothen the arc and
take a Hamiltonian H: U ~ ~, where U c [0, 1] • C ~ is open in [0, 1] x C", which
generates the smoothened arc. Then we restrict H to a suitable smaller set and
extend the restriction smoothly to a Hamiltonian of c o m p a c t s u p p o r t . If y is the
time 1-map for the associated Hamiltonian system then f = f o 7"o is the desired
global symplectic diffeomorphism. Using this simple remark we can use the
capacity c to prove a celebrated result of G r o m o v [-12].
Theorem 2. Assume 7":BZ"(r) q S(r'): = BZ(r ') • C n- 1 is a symplectic embedding.
Then r <_ r'.
Remark. Note that the cylinder 2~(r)= { z ~ C " J p ~ + p ~ < r 2} has an infinite
capacity for every r > 0 as a consequence of the remark before Section III. For
cS > 0 sufficiently small consider the symplectic linear m a p defined by
which implies
g)Z~r2 = 52 r2 c(B(1) ) = c (3 B(r) ) < e( S (r') ) = 7r(r') 2.
Hence for every 6e(0,1) we find 62r z < (r') 2 proving that
r'(r'. []
This result is remarkable since Z(r') has infinite volume and yet, for r < r', does
not have leave r o o m enough to fit a symplectic copy of B(r) which has finite
volume. One can also use T h e o r e m 2 as a starting point for constructing symplectic
capacities.
F o r F c C" define
(1) 5(F) = inf{Trr2[ 3 T e D ( C " ) with T (F) c S(r.) }
c ( F ) = sup {ztrZl 3 T e D ( C " ) with T(B(r) c F}.
Here r E [0, or) w { + ~ }. F r o m the definition of c and ? it is clear that
First we study linear maps which preserve a capacity c. We will then deduce that
nonlinear C l - m a p s which preserve a symplectic capacity are either symplectic or
antisymplectic. C~ will be an easy consequence. Define a set E c C" to be a
b o u n d e d ellipsoid if there is a positive definite quadratic form q such that
E = { z e C " j q ( z ) < 1}.
In the following we shall consider C" exclusively as a 2n-dimensional real vector-
space. If we speak a b o u t a linear subspace of C" we mean a linear subspace of the
~-vectorspace C". Similarly for linear maps.
Theorem 3. Assume T: C" -~ C" is a linear map such that for every bounded ellipsoid
E we have
c(E) = e(T(E)).
Then T is symplectic or antisymplectic:
T*~o = ( o or T*~ = -co.
The p r o o f depends on a simple lemma. We have to fix some notation. A 2-
dimensional subspace W of C" is said to be null iff oJI W = 0. This is clearly
SymplecticTopologyand HamiltonianDynamics 373
equivalent to W c W% where W '~ = { z e C " l z ~ W} is called the ~o-orthogonal of
W. Here z~• W means co(z, w) = 0 for every w e W. The co-orthogonal is of course
defined for every subspace of C".
L e m m a 3. Let ~2 be a bounded open subset of C" and W a linear subspace of
codimension 2. Consider the cylinder (2 + W. Then
c(~+W)= +~if W' ~
0 < c(Q + W ) < + ~ if W ~ is not null.
Proof We assume without loss of generality that ~2 contains the origin. Using a
linear symplectic change of coordinates we can bring ourselves to the situation
where:
W = [ z e C " j R e ( z l ) = Re(z2) = 0} in the first case
W = { z e C " l z 1 = 0} in the second case.
In the first case we take e > 0 so small that
(Re(z1))2 + (Re(z2))2 <=~2 =~ ze~2 + W.
Now observe that for any N > 0 the ellipsoid
{zl~Re(za)Z+-~sIm(zl)2+~Re(z2)Z+~Im(z2) 2
+ ~ ~lzk 1 62< 1 }
k=3
is contained in ~2 + W. This ellipsoid, say EN, is symplectically equivalent to the
o < c(S) = c ( ~ e ( s ) ) = o.
~u(w ~)
= ((~*)- 1 (W))C~ '
the constant dk defined by (3) simply by d. I f d > 0 we define ~ = ~ 4~ and ifd < 0
( ( ~ , ) - 1 = ~. Hence
Corollary 2. I f T: C ~ -~ C" is a linear map and there exists an open nonempty subset
S such that c(TJ(S)) > O, then 7t is invertible. []
Next we go over to the nonlinear case
Theorem 4. Let 7~: C" ~ C" be a C 1 -map such that for every bounded ellipsoid S we
have
c ( ~ ( s ) ) = c(S).
Theorem 5. Let (t])k) be a sequence o f continuous maps of the open unit ball B(1) into
C n converging uniformly to ~. Assume all the t~ k preserve the capacity of bounded
ellipsoids
c ( ~ k ( s ) ) = e(S) Vk ~
which gives
~(s) = c(S)
: c(~(s))
__<e ( ~ ( s ) )
= e(g)
< ~ + ~(~(s)).
Since e > 0 was arbitrary we infer that
(4) ~(S) < ~(~(S))
for every bounded ellipsoid S with closure in B(1). Hence for t e (0,1)
If now t+0, ~cb(t.) converges uniformly to ~b'(0) and we conclude by the same
a r g u m e n t as above
limo 6 ( ~ ( t S ) ) ~ ~(~'
Consequently
(6) ~(s) < ~(~'(0)s).
for every bounded ellipsoid S with closure in B(1). F r o m Corollary 2 we see that
q~'(0) is invertible. Next we have to show the reversed inequality in (6). Since h is
differentiable at 0 we find a continuous increasing m a p s: (0, 1) -* (0, oo ) such that
s(s) --* 0 as s -* 0 and assuming 4(0) = 0
If(x)- ~'(O)xl < e(Ixl)Ixl.
Given 5e(0, 1) we find k(6) such that for k > k(5)
I,P~(x)- 4~'(O)xl < ~(Ixl)Ixl + ~.
Pick a linear ellipsoid S with closure in B(1) and let te(0,1). Pick 7 > 0; then, if z is
small enough (1 + 7)z < 1. Consider the equation
(7) t q)k((1 + 7)ZX) + (1 -- t ) ~ ' ( 0 ) ( 1 + ?)ZX) = eb'(O)(zZ)
for given z ~ S and solvable for some z e [0, 1] and x e OS. Pick a constant d > 0
such that
I'~'(O)xl-->_ dl xl.
SymplecticTopology and Hamiltonian Dynamics 377
Theorem 6. Assume q)k: B(1) --* C" is a sequence of symplectic embeddings converging
unifrmly to a continuous map q): B(1) --* C", which is differentiable at 0 with derivative
cP' (O). Then q)'(0)sSp(C").
Proof Given a subset S of B(1) with el(S) ~ B(1) we must have C(~k(S)) = c(S) for
any capacity as the consequence of the extension after restriction principle. From
Theorem 5 we conclude that q~'(0) is symplectic or antisymplectic. If n is odd a
antisymplectic is orientation reversing. Since q~k preserves the orientation the same
has to be true for q)'(0). Hence ifn is odd q~'(0) can be only symplectic. So assume n
is even. Then Id c x I~ k --1. Id c x q~ uniformly on B 2(n+ 1)(3) for some small fie(0, 1).
Since Idc x 4~k is symplectic we conclude from the previous argument that
Idc x (b'(0) is symplectic which implies that ~'(0) is symplectic.
Finally we obtain as a simple corollary of Theorem 6 a celebrated result by
Eliashberg and Gromov.
378 I. Ekeland and H. Hofer
Bibliography
1. Amann, H., Zehnder, E.: Nontrivial solutions for a class of nonresonance problems and applications
to nonlinear differential equations. Ann. Sc. Norm. Super Pisa 7, 593-603 (1980)
2. Arnold, V.I.: Sur une propri6t6 topologique des applications canoniques de la m+canique classique.
C.R. Acad. Sciences Paris 261, 3719-3722 (1965)
3. Arnold, V.I.: Mathematical methods of classical mechanics. Berlin Heidelberg New York: Springer
1978
4. Bennequin, D.: Probl~mes ellptiques, surfaces de Riemann et structures symplectiques. S~m.
Bourbaki Fevr 1986, 1-23
5. Chaperon, M.: Quellques questions de g6ombtrie symplectique (d'apras autres Poincar6, Arnold,
Conley-Zehnder), S2m Bourbaki, n ~ 610. Ast~risque 105-106, 231-249 (1983)
6. Clarke, F.: Periodic solutions to Hamiltonian inclusions. J. Differ. Equations 40, 1-6 (1981)
7. Clarke, F., Ekeland, I.: Hamiltonian trajectories having prescribed minimal period, Commun. Pure
Appl. Math. 33, 103-116 (1980)
8. Conley, C., Zehnder, E.: The Birkhoff-Lewis Fixed point theorem and a conjecture by V.I. Arnold.
Invent Math. 73, 33-49 (1983)
9. Ekeland, I., Hofer, H.: Convex Hamiltonian energysurfaces and their periodic trajectories. Commun
Math. Phys. 113, 419-469 (1987)
10. Ekeland, I., Hofer, H.: Two symplectic fixed point theorems. To appear
11. Floer, A.: Morse-theory for Lagrangian intersection. To appear in J. Differ. Geom.
12. Gromov, M.: Pseudo-holomorphic curves in symplectic manifolds. Invent. Math. 82, 307-347 (1985)
13. Gromov, M.: Partial differential relations. Berlin Heidelberg New York: Springer, 1986
14. Gromov, M.: Symplectic geometry hard and soft. ICM Talk, Berkeley, 1986. To appear
15. Herman, M.: Une remarque, manuscript, Paris, 1983
16. Hofer, H.: Lagrangian embeddings and critical point theory. Ann. Just. Henri Poincar~, 2, 407-462
(1985)
17. Hofer, H., Zehnder, E.: Periodic solutions on hypersurfaces and a result by C. Viterbo. Invent.
Math. 90, 1-9 (1987)
18. Hofer, H.: On strongly indefinite functionals with application. TAMS 275 (11), 185-214
19. MacDuff, D.: Examples of symplectic structures. Invent. Math. 89, 13-36 (1987)
20. Takens, F.: Hamiltonian systems: generic properties of closed orbits and local perturbations. Math.
Ann. 188, 304-312 (1970)
21. Viterbo, C.: A proof of the Weinstein conjecture in ~2,. Ann. Just. Henri Poincar~, Analyse
nonlin6aire 4, 337-357 (1987)
22. Weinstein, A.: On the hypotheses of Rabinowitz's periodic orbit theorems. J. Differ. Equations 33,
353-358 (1979)
23. Weinstein, A.: Lagrangian submanifolds and Hamiltonian systems. Ann. Math. 98, 377-410 (1973)