VII.2. The Hofer–Zehnder capacity c
0
265
for a real number a 0, and so is equal to x.t/ D e
atJ
x.0/ D .cos at/x.0/ C
.sin at/J x.0/. We see that all the solutions are periodic! In the case a D 0, the
solution is constant D x.0/, in the case a>0the solution has the minimal period
T D 2=a. From 0 f
0
<, it follows that a<2and hence T>1.We
have verified that the function H does, indeed, belong to the set H
a
of admissible
functions.
We have demonstrated the estimate c
0
.B.1/; !
0
/ by means of a single
example. Since the inclusion map B.1/ ,! Z.1/ is a symplectic embedding, we
obtain, in view of Lemma VII.11, the following estimate.
Corollary VII.14. c
0
.B.1// c
0
.Z.1//.
In order to verify the normalization axiom (A3), it remains to prove the estimate
c
0
.Z.1// . This requires an existence proof, namely, the proof of the following
theorem. If proved, Theorem VII.10 is also proved and c
0
is a symplectic capacity,
as claimed.
Theorem VII.15. If H 2 H.Z.1// satisfies m.H / > , then the Hamiltonian
vector field X
H
on Z.1/ possesses a non-constant periodic solution x.t/ 2 Z.1/
of period T D 1.
It follows that c
0
.Z.1// , so that Theorem VII.10 is proved.
It remains to find a global periodic solution of period T D 1, for a Hamiltonian
vector field X
H
defined on the open cylinder Z.1/ where the Hamiltonian function
H belongs to the set H.Z.1/; !
0
/ and satisfies m.H / > .
Our existence proof will be based on a variational principle for the action func-
tional of classical mechanics and we start with some preparations.
Since the Hamiltonian function H W Z.1/ ! R belongs to H.Z.1/; !
0
/ it satis-
fies H 0 on an open set U and H m.H / is constant outside of a compact set K
that contains U . We have already proved that c
0
is a symplectic invariant, because
the monotonicity axiom (A1) holds true for c
0
. Therefore, we may assume that U
is an open neighborhood of 0 2 R
2n
by means of a symplectic diffeomorphism of
R
2n
having compact support in the cylinder Z.1/.
We prefer to work with a Hamiltonian function defined on the whole space R
2n
and not merely on the cylinder Z.1/ and we therefore extend the function H to
a smooth function
x
H W R
2n
! R. In principle, this is not a problem, we could,
for instance, extend H by the constant m.H /. For technical reasons, this is not
good enough and we shall choose a more clever extension which, far away from
the origin, looks like a specific quadratic form.
Since the set K Z.1/ is compact, there exists an ellipsoid E D E
N
Z.1/
containing the compact set K E, so that the Hamiltonian function H also belongs
to the set H.E; !
0
/. Indeed, we choose
E Dfz 2 R
2n
j q.z/<1g