Dynamically convex Finsler metrics and $J$-holomorphic embedding of asymptotic cylinders

DYNAMICALLY CONVEX FINSLER METRICS AND
J-HOLOMORPHIC EMBEDDING OF ASYMPTOTIC

CYLINDERS

ADAM HARRIS AND GABRIEL P. PATERNAIN

Abstract. We explore the relationship between contact forms
on S'^ defined by Finsler metrics on §^ and the theory developed
by H. Hofer, K. Wysocki and E. Zehnder in [9l [10]. We show
that a Finsler metric on with curvature K > 1 and with all
geodesic loops of length > tt is dynamically convex and hence it
has either two or infinitely many closed geodesies. We also explain
how to explicitly construct J-holomorphic embeddings of cylinders
asymptotic to Reeb orbits of contact structures arising from Finsler
metrics on §^ with K = 1 thus complementing the results obtained
inlH].

1. Introduction

A contact form A on a closed, connected, oriented manifold of odd
dimension is said to be dynamically convex if the Conley-Zehnder index
of any contractible periodic orbit of the Reeb vector field is at least
three. This notion generalises the case of convex hypersurfaces in
on which the contact form is simply the restriction of the standard
one-form Aq from the ambient space. When the hypersurface is S^, the
Reeb flow induces a fibration by orbits of constant period, namely the
Hopf fibration. If the hypersurface is an ellipsoid Ep^g defined by the
equation p\z\'^ + q\w\'^ = 1 {z, w denoting coordinates of C^) then an
interesting dichotomy arises between the dynamics of the Reeb flow for
I rational or irrational. In the rational case there are still infinitely
many periodic orbits, though not all are of the same minimal period,
while in the irrational case there are just two periodic orbits. This
dichotomy extends to any strictly convex hypersurface, since it was
shown by Hofer, Wysocki and Zehnder [9j that the Reeb vector field
X of any contact form satisfying the dynamic convexity condition on
the three-sphere has either two or infinitely many periodic orbits. A
central role in their theory is played by the existence of finite-energy

2000 Mathematics Subject Classification. 32Q65, 53D10, 58B20.

J-holomorphic embeddings of the cylinder in the "symplectisation"
of a contact manifold, from which a periodic orbit is realised as the
asymptotic limit of an end of the embedded cylinder.

For the standard example associated with the Hopf fibration on S'^
there is a direct correspondence between the asymptotic limit of finite-
energy J-holomorphic embeddings in the symplectisation on §^ x M and
germs of plane algebroid curves centred at the origin of a sufficiently
small ball in C^, via their "link" with the boundary-sphere. It may then
be asked whether a more general correspondence can be established be-
tween plane algebroid curve-germs and J-holomorphic embeddings in
the symplectisation of other contact forms A on S'^. It should be noted
that a similar question has been studied extensively in the context
of pseudoholomorphic curves in a symplectic manifold (cf., e.g., |17j )
though the question there is local rather than asymptotic in nature. In
[8] it was shown that near periodic orbits of "elliptic type" (cf. locally
recurrent orbits, [8]) such that the partial almost complex endomor-
phism j, compatible with the contact structure, is Reeb-invariant (i.e.,
the Lie derivative £xj = 0), finite-energy J-holomorphic embeddings
ip of the cylinder may be represented holomorphically in a suitable
tubular coordinate neighbourhood of the orbit. After transformation
of the cylinder to a punctured disc -D \ {0} via the choice of a com-
plex coordinate z, the periodic component of ip naturally subdivides
D into "quasi-sectors" , the number of these being determined by the
topological degree of ip restricted to the circle | approaches
zero (i.e., as the axial coordinate of the cylinder approaches infinity),
and being referred to as the "charge" of the mapping. In the direc-
tion spanned by a disc A, transversal to the orbit, is represented by
holomorphic functions defined on each of the quasi-sectors of D, which
are hinged together along common edges by the return map a, induced
on A by the Reeb fiow. With respect to (A,j) satisfying the above
criteria and for a given charge n, each finite-energy J- holomorphic
embedding of the cylinder in x M therefore gives rise to a collection
of n holomorphic functions, whose continuity at adjacent boundaries
of their domains is mediated by a (cf. |8], Theorem 1]).

For any contact structure (A, j) such that Cxj = and a is the
identity (e.g., the standard Aq restricted to S'^) it is possible to move
explicitly back and forth between finite-energy J-holomorphic embed-
dings of the cylinder in the tubular neighbourhood of a periodic orbit
and algebraic curve-germs at the origin in of the form (z", F{z)). A
similarly explicit correspondence was obtained in [8J for contact struc-
tures on §^ diffeomorphically equivalent to the restriction of Aq to a
rational ellipsoid (on which a is a non-trivial rational rotation near

2

each of two exceptional orbits). Implicit in these examples is the fact
that the contact form has rotational symmetry along the axis of a suit-
able tubular coordinate neighbourhood of the orbit, thus allowing the
map ip conversely to be determined (up to local diffeomorphic equiv-
alence) by the charge n and holomorphic functions Fk, predefined on
formal sectors Qk, such that Fk+i = a o at the common boundary
of adjacent sectors, < A; < n — 1 (cf. Theorem A below). As we
note in the present article, this crucial symmetry property is also held
(up to local gauge correction) by any contact form for which a is an
irrational rotation (as in the case of the irrational ellipsoids), so that a
ip of charge n is again determined by the holomorphic data F^:

Theorem A. Suppose, for the ensemble (M, A, j) that Cxj = and the
return map a induced locally by the Reeb flow near a locally recurrent
orbit V corresponds to an irrational rotation. For positive integer n let
Qk denote the formal sectors of a disc D G C defined by

k , , k + 1

2-K- < aig{z) < 2tx , < A; < n - 1,

n n

with holomorphic functions w = Fk{z) defined on Qk such that

for each k. Modulo a local gauge correction of the form

X = X-df,

for some smooth function /, defined in a tubular neighbourhood of V
such that X\{f) = 0, these holomorphic data determine a finite-energy
J-holomorphic curve of charge n, asymptotic to V.

But to how large a class of contact structures on do these corre-
spondences apply? More broadly, to how large a class of structures do
the dynamic convexity results of Hofer, Wysocki and Zehnder apply?
An important class of examples of triples (M, A, j) is given by the con-
tact form and almost complex endomorphism j naturally induced on
the three-manifold corresponding to the unit tangent bundle of a Rie-
mann surface S by the geodesic flow of a metric on S (cf., e.g., [T3]).
If the additional Lie symmetry condition Cxj = is assumed then g
must have Gaussian curvature K = 1 and up to isometry we just have
E = But the class of required structures is much broader if, in-
stead of a Riemannian metric, one considers a Finsler structure on the
two-sphere, i.e., a hypersurface M C TEP' and a surjective submersion
vr : M — i> §^ such that for all p G §^ the fibre vr~^(p) is a smooth, closed
strictly convex curve enclosing the origin in TpEP' (cf. section two).

3

The associated structure equations derived from a canonical framing
of T*M depend on three functions /, J, K over M. When the Finsler
structure corresponds specifically to a Riemannian metric, it follows
that / = and K corresponds to the pullback of Gaussian curvature
from S. The induced almost complex endomorphism j on a Finsler
contact manifold (M, A) satisfies Cxj = precisely when = 1 for
any J, J (cf. Lemma [3. II) .

A famous class of contact structures due to A. Katok [TT] arises from
convex hypersurfaces M C TS^ on which K = 1 and the return map
a associated with the Reeb flow near either of its two distinguished
periodic orbits is a rotation. The Finsler structures connected with
these examples incorporate what are known as Randers metrics, coming
from special perturbations of the norm derived from a Riemannian
metric. When a Finsler structure on S with K = 1 admits a non-trivial
Killing field, one may perturb to a family of non- Randers examples with
the same properties, as is shown in section 3.2, following independent
observations of P. Foulon and W. Ziller [TS]. All the hypersurfaces
in question are doubly covered by S^, such that the contact forms
lift to tight contact structures on the three-sphere. In sections three
and four a precise description of this lifting reveals that the Katok
examples correspond exactly with those induced by the restriction of Aq
to Kp^q C C^. It is then worth noting that the non- Randers structures of
section 3.2 lie genuinely beyond the standard class of examples coming
from convex hypersurfaces of M'^. Returning to the broader notion
of dynamic convexity for contact structures on S^, it is asked in the
final section of this article whether a corresponding criterion can be
found in terms of Finsler structures on Theorem B provides this
criterion for a given Finsler structure, via a lower bound on the length
i of its shortest geodesic loop (we shall say that the Finsler metric
is dynamically convex if its associated contact form is dynamically
convex) :

Theorem B. Let F be a Finsler metric on such that K > 6 > 0. If
i > Ti/\^, then F is dynamically convex.

In particular, by the results in [9J any such Finsler metric has either
two or infinitely many closed geodesies. We note that recently, V.
Bangert and Y. Long pj have shown that any Finsler metric on has
two closed geodesies.

Acknowledgements: The authors would like to express their warm
thanks to Z. Shen for his helpful communications during their research

4

towards this article. We are also very grateful to V. Bangert and H.B.
Rademacher for their comments and remarks.

2. J-HOLOMORPHIC CYLINDERS NEAR LOCALLY RECURRENT
ORBITS OF A CONTACT THREE-MANIFOLD

Let M denote a compact, oriented three- manifold with contact form
A and associated plane-field ^ C TM corresponding to ker(A). Let Xx
denote the Reeb vector field associated with this structure on M, to-
gether with an almost complex structure J acting on ^ such that the
symmetric tensor defined by d\{*, J*) \^ is positive definite. Consider
a periodic orbit of the Reeb flow, denoted V, and a tubular neigh-
bourhood T-p C M. If A represents a disc centred at the origin in
M^, let A C M be an embedded image such that the origin is mapped
to the unique element po of P fl A, with A itself corresponding to a
transverse slice of T-p. The Reeb flow will be assumed moreover to be
Lyapunov-stable near V in the sense that for all p G A', where A' C A
is a sufficiently small disc centred at the origin, there exists a unique
solution 7p : [0, oo) M to the equation

^ = X,(7,(t)), 7p(0)=P,

which depends smoothly on both t and p, and remains inside T-p for all
t > 0. Given p e A', we will define [i) t{p) to be the smallest t >
such that 7p(t) G A , (ii) Tp := 7p((0,r(p)]) and for each connected
open neighbourhood of the origin, C A ,

(in) T{n) := Upef^Fp .

We may now consider a recursively defined system of neighbourhoods
{fifc}, such that Qq := A', while denotes the origin-component of
r(fifc_i) n flk-i- The set floo '■= ^kLo^k was seen in ^ proposition 1
to be conformally equivalent to a disc whenever it corresponds to an
open subset of A'. As in [8J, the Reeb flow will be said to be "locally
recurrent" near a periodic orbit V if it is Lyapunov-stable within a
tubular neighbourhood T-p and for any sufficiently small embedded disc
A, corresponding to a transversal slice through T-p at some point po,
the limit set C A' C A is open. An orbit V itself may also be
referred to as "locally recurrent" in this context. In passing we note
that if the orbit is elliptic of twist type, then by Moser's twist map
theorem it will be locally recurrent. Moreover, the local recurrence
implies that the characteristic multipliers are unimodular.

5

Under the assumption that the Reeb flow is locally recurrent near
P, we now select Q^o x {''^o} as coordinate disc within the initial Mar-
tinet tube (in which A = / ■ {di!} + xdy), where {} denotes the periodic
coordinate and / denotes a function such that f{0,0,'d) = tq and
V/(0, 0, {}) = 0, though these facts are not used here). Without loss of
generality, let {}o be zero and consider the cylinder Qoo x [0, 27i], which
maps to the tube via the obvious identification mod(27r). The cylinder
has V as its axis, x = y = 0, and the Reeb vector field in Martinet
coordinates already looks like -^-^^ when restricted to V. There is no
consequent loss of generality if we "normalise" A by the constant mul-
tiple so that the minimal period is effectively 1. In particular this
will ensure that the notion of "charge" for a J-holomorphic mapping ip
of the punctured disc, as introduced below, is consistent with the stan-
dard definition corresponding to the limit of the integral of ip*X over
the circle = c as c approaches zero (and is specifically an integer).
By analogy with the standard construction of Darboux coordinates,
the next step is to define

C:={{p,t)\p={x,y)eQo. , 0<t<r(p)},

and a homeomorphism

/i : C ^ fioo X [0, 27r] , h boox{o}= 1 ,

which is smooth for all < t < r(p), coming from solutions of the
ordinary differential equation

It follows that on the interior of C, the standard contact form Aq and
A' := h*X have the same Reeb vector field, corresponding to ^. We
now consider the Cauchy-Riemann system

7r((/i-V),) + Jvr((/i-V)c) = (*),

A'((/i-V)c) = A'((/i-V).) = ac (t) ,

satisfied by some finite-energy J-holomorphic map ip of a punctured
neighbourhood D \ {0} C C into the Martinet tubular neighbourhood
of P (xM). For a sufficiently "thin" neighbourhood of V, the standard
projection (fi,f2,f3) ^ {vi,V2) determines a linear isomorphism /i
between ^' := ker{X') and M^. Hence we define a 2 x 2 matrix-valued
function j{x,y) = fi o J o such that x := {x, y) implies (*) can be
written in the form

y^rjiz) + jx^(z) = .
Let a denote the diffeomorphism of Qoo x {0} defined by the return
map a{p) := 7p(t(p)), hence a{0) = 0. As seen in [8], if CxyJ = 0,

6

then in a neighbourhood of G A, the smooth automorphism ot is
equivalent to a rotation, via a diffeomorphism : A" ^ f/ C f2oo
such that o i o ip^ = jo (the standard multiphcation by i) . More
specifically, let Q'^ denote the simply connected domain inside U which
is stabilised by the Reeb flow. The diffeomorphism a := ip~^ o a o (f
then acts on ip~^{Q'^) C A" as an automorphism such that d(0) =
under the assumption of local recurrence. The additional assumption
CxyJ = implies that a*i = j, hence in particular djo = Jo«; i-e.,
d is a conformal automorphism. Modulo a conformal transformation
identifying ip~^{Q'^) with a disc, a is then equivalent to a rotation.
Now := (^~^(x^) and := (y9~^(x^) implies

^vivX)+jo ■ MvX) = (t*).

As described in [8], each "branch" of

^ := (/lo (v? X 1))~V

is defined smoothly in the interior and continuously up to the bound-
aries of a quasi-sector Qk in D \ {0}, with discontinuities arising at
points zq lying on the smooth arcs, corresponding to 'i9~^(0),that bound
adjacent sectors (in the usual way "±" will be used to denote opposite
sides of the boundary). Discontinuities of the transverse projection of
are therefore described by the relations

lim x(z) := x=^(2;o) a{±'{zQj) = ±'^{zq) .

Hence on each Qk <Z D \ {0} , (f*) defines a holomorphic function
w = Fk{z) which partially describes a branch of \E', such that

lQ.nQ.+i= « ° Fk+i |Q,nQ,+, , < A; < n - 1 ,

where n denotes the charge (or asymptotic degree) of ip in relation to
the periodic orbit V.

Returning now to the particular form of the equations (f) in C note
that any A' with ^ as its Reeb vector field must take the general form

X' = dt + /i(x)(ix + f2i^x)dy .

Such a local presentation of the contact form allows decoupling of (f)
into an inhomogeneous Cauchy-Riemann equation. This property is
preserved under the diffeomorphism ip, however, if it is assumed that
Cx^J = 0, hence in particular the matrix j above is independent of
t. In this case, letting 3ft and 3 signify real and imaginary parts, and
u = t + ia, (t) becomes

g^iu IqJ = -{MFk{z)) ^_ + f2{Fk{z)) ^_ }

7

= -^[(/i + i/2)oF,(^)]-F^(z) ,

keeping in mind that this equation is defined smoothly only on the
interior of each quasi-sector Qk- Define a; := ^(/i + if2){w,w)dw, so
that

\' = dt + 2^{uj) , and
^[if, + if,)oF,iz)].F[(^)dz = F:u; .

Now Cx^^ = implies d*^{uj) = ?R.{uj). In particular, f := (/i,/2)
implies 3?(u;) = (f , *) with respect to the standard inner product on
M^, and hence a*f = f. Similarly '^{uj) = (jof, *), while Cx^J =
implies d*jo = joc^l, so that

a*Q{u;) = (dtjof , *) = Oof, *) = »(^) .
It follows that a*uj = uj, and hence

lQ,nQ,+,= F:{a*u) \Q^nQ^^= (« o F.Yuj Ig^nQ^^^

There now exists a continuous function G{z, z) on D such that

G{z, z)dz |q,:= F*uj , 0<k<n-l,
which is moreover continuously differentiable (cf. [5]), and hence

G{z,z) := / ^^^df^Adft
Jd f^-z

is a twice-continuously differentiable function on D. Moreover, the
inhomogeneous equation satisfied by u now has the form

^[u\q^]=G, 0<k<n-l,
and with a little additional argument we have

Theorem 2.1 ([8J). Let {ip, a) : D\{0} ^ M xR be a J -holomorphic
curve of finite energy and charge n at z = 0, asymptotic to a locally
recurrent periodic orbit V , near which Cx^J = 0- Consider any tubular
neighbourhood ofV in M , diffeomorphic to AxS^ such that {0} x ^
V. There exists a diffeomorphic change of coordinates in A x [0, 27r)
such that on each quasi-sector Qk G D \ {0} the map {ip, a) can be
expressed in the form

(Ffc(z), Hf,{z) - ^Giz, z)) , < A; < n - 1,

ZTTl

where F^ , are holomorphic on and continuous on Qj^, such that

Fk lQfcnQfc+i= « ° Fk+i Ig^nQfe+i '
8

while each corresponds to an analytic branch 0/ 2^ log(p) , ordQ{p) =

n. Moreover, the function G belongs to C^{D) and is bounded by K\z\.
Finally, if a — 1, then each is the restriction of a single function F
holomorphic on D, F{0) — 0.

Conversely, given an ensemble (M, A, J), Cx^J — 0, with locally re-
current periodic orbit V corresponding to the asymptotic limit of some
J-holomorphically embedded cylinder of finite energy, we may ask for
the essential holomorphic data which determine such J-holomorphic
curves in general near V. Our main result in this direction is the fol-
lowing:

Theorem A. Suppose, for the ensemble (M, A, J) above, that the re-
turn map a induced locally by the Reeb flow near a recurrent orbit V
corresponds to an irrational rotation. For positive integer n let
denote the formal sectors of a disc D G C defined by

k k -\- 1
27r- < arg(2;) < 27r < A; < n - 1 ,

n n

with holomorphic functions w = F^^z) defined on Qk such that

for each k. Modulo a local gauge correction of the form

\ = X-df ,

for a smooth function f , defined in a tubular neighbourhood ofV such
that Xx{f ) = 0, these holomorphic data determine a finite-energy J-
holomorphic curve of charge n, asymptotic to V .

Proof. Returning to the general form of A with respect to the coordinate
tube C above, we have

X'^dt + /i(x)dx + /2(x)dy ^dt + n(u;) ,

noting that Q;*3?(a;) = 3?(a;), in the case of a an irrational rotation
(hence in particular the orbit of any point w under a is dense in a
circle of radius \w\). Now

^{u) = ri{p)du + C{p)dp = ri{\w\){xdy - ydx) + df ,

where w — X -\- \y , p—\w\ , v — axg{w) , and

l\{r)dr.
Jo

9

Letting ?)(p^) = ?7(p), and after making the gauge correction A : =
\ — df := dt + 3fJ(a)), we have

LJ = fj{ww) ■ iwdw ,

and hence

<l>(|w|^) := / fi{r)dr
Jo

imphes lj = It should be noted at once that dX = dX , Xx{f) =

together imply that Xx remains the Reeb vector field of the contact
form defined by A. Recall that

indicates that there exists a single smooth function \F\ on D, corre-
sponding to Fk on each Qk, hence we may define

G{z, z) ■= id^{\F\^) , and G{z, z) := 2tx\ ■ \^{\Ff) = -27r$(|Fn .

Under the assumption that V is already the asymptotic limit of some
finite-energy mapping of the punctured disc (or cylinder), we recall
moreover from the discussion of [S], that the return time r(p) for each
p e fl'^ is constant (i.e., normalised to value 1). Note that the ro-
tational symmetry of the contact form ultimately implies that G is a
real-valued function, compatible with the formal specification of simple
sectors Qa: as domains of the holomorphic functions F^, and moreover
that the t-component of a J-holomorphic mapping derived from these
data can be defined by t = 2^ arg(2;"). Now from Theorem 1, we
may write

w = t + ia=-i^(log(z")+27r<|.(|Fp)),
2tti

from which it follows that the associated a-component must be

aiz) = ^\ogi\zn-H\F\') .
Ztx

One or two remarks should be made concerning the almost complex
structure with respect to which the mapping \E', determined sector-wise
by the Fk{z) and the corresponding analytic branches of 2^(log(2;"') +
27r$(|Fp)), may be said to represent a J-holomorphic mapping of
charge n of the cylinder into 7p x R, for a tubular neighbourhood
T-p. It is straightforward to see that in a sufficiently thin tube, the
standard projection mapping defines an isomorphism between and
the contact planes of both A and A at any given point which we will
denote by /3 : ^ — >• ^ , i.e., [3{y) = v — A(v)X , where X denotes the

common Reeb vector field of A and A. The almost complex structure

10

J := (3 o J o f3 ^ is then automatically induced on ker(A). Moreover,
given v' G ker(A), we have

dX{V, J-V) = dA(/3(v), J ■ /3(v)) = dX{(3{v), f3{J ■ v))

= dX{v - A(v)X, J • V - A(J • v)X) = rfA(v, J ■ v) .

Hence the quadratic form dX{*, J*) \^ is also positive definite.

It remains now to check that the pseudoholomorphic curve defined
with respect to A , J in 7p x M is of finite energy. Following [9], let ^
denote the space of smooth functions /i : R — [0, 1] such that h' > 0,
and define extensions Xh of the contact form from 7p to 7^ x M such
that Xh{p,a) := h{a) ■ X{p). The "energy" of \Ef is then defined as

E(^) := sup /

dXv,

D\{0}

Clearly,

Now

where the formula

^*dXh = / ^*A;, - lim / ^*A;,

D\{0} >^|z|=l ^~'°J|z|=e

= / h{a)'^*X-\im [ /i(a)^*A .

J\z\=l ^^^J\z\=e

71

^l!*X = -r]{\F\)diy{F) + —d^Tgiz) ,

Ztt

V = aig{w) = tan ""^ | — i

w + w
implies

dviF) = 2m— ■ F'(z)dz) = m—^dz) .
dw F

(Here we have adopted a harmless abuse of notation, in the sense that

F'

the ratios define a single continuous function on D \ {0} when a is
a rotation). Note, inside any \z\ = e sufficiently small, that 2; = is
the unique zero of the smooth function and hence

w — w

^*A-n

\z\=e J\z\=e

where the latter quantity approaches zero as e goes to zero, if it is
recalled that //(p) is a smooth function such that ri{0) = . Now
|z| — > implies a +00, and thus

E{^) = snp( [ h{-(^{\F{z)\^))-^*X~n-h{+oo)) ,
:f \J\z\=i J
11

which is clearly finite.

□

In the next section we will consider a class of examples of contact
structures to which the above theorem may be applied, but it should
first be estalished that holomorphic data of the sort specified are in
plentiful supply. For arbitrary consider f{z) holomorphic on D C C,
such that ordolf) > n, and an arbitrary irrational rotation of the form
a = e^'^"^ , < c < 1. Let D be divided into formal sectors as in
theorem 1, with

Foiz) := , Fk+iiz) = a ■ Fk{z) , < A; < n - 1,

noting that Fn{z) = Fq{z). For convenience we may define z""'^ =
^-ncLog(z)^ whcrc Log{z) denotes an analytic branch of the complex
logarithm defined on C minus the positive real axis.

3. Examples

3.1. Canonical coframing. Let S be a closed oriented connected
surface. A smooth Finsler structure on E is a smooth hypersurface
M C TE for which the canonical projection vr : M E is a surjec-
tive submersion having the property that for each x G E, the vr-fibre
TT'^{x) = M nT^-E is a smooth, closed, strictly convex curve enclosing
the origin 0^ G T^^E.

Given such a structure it is possible to define a canonical coframing
{ui,U2,uj3) on M that satisfies the following structural equations (see
[21 Chapter 4]):

(1) dui = -UJ2 A tus,

(2) duj2 = -uJs A {iOi - IUJ2),

(3) dus = —{Kui — JuJ^) A UJ2.

where /, K and J are smooth functions on M. The function / is called
the main scalar of the structure and it vanishes if and only if M is the
unit circle bundle of a Riemannian metric. When / = 0, i.e. when the
Finsler structure is Riemannian, K is the vr-pullback of the Gaussian
curvature.

Let Xi, X2 and X3 be the vector fields on M that are dual to the
coframing {001,002, 003). The form ooi is the canonical contact form of M
whose Reeb vector field is the geodesic vector field Xi.

As a consequence of ([IH2D the framing (Xi, X2, X^) satisfies the com-
mutation relations:

(4) [X3, X,] = X2, [X2, X3] = Xi + 1X2 + JX3, [Xi, X2] = KXs.

12

Note that the coframing also defines a natural almost complex struc-
ture J on ,^ = kertui. Indeed, we may set:

J(xX2 + yXs) = yX2 - xX^.

If we let 7] = XX2 + yX^ G kercji then using ([1]) we see that:

dui{r], J?7) = +

and thus J is compatible with the contact structure.

Lemma 3.1. Cxil = z/ and only if K = 1.

Proof. Let (pt be the fiow of Xi. Note that £xiJ = if and only if dcpt
is an isometry of the inner product in kercui, given du!i{*,S*). (Recall
that (pt preserves duJi.)

Let 1] = XX2 + yX^ G kercji and write:

d(f)t{^) = x{t)X2 + y(t)X3.

Thus the fiow (pt is an isometry of the inner product (ia;i(*, J*) if and
only if

equivalently if and only if

(5) XX + yy = 0.

Write

T] = x{t)d(P.t{X2) + y{t)d(l).t{Xi)
and differentiate with respect to t to obtain

= XX2 + x[Xi, X2] + 2/X3 + y[Xi, X3].

Using the structure equations (jl]) and regrouping we have:

= (x - y)X2 + {y + xK)Xs,

hence

x = y,
y + Kx = 0.

If we use the last two equations in we see that (pt is an isometry if
and only if

xy{l -K) =

and thus £xiJ = if and only ii K = 1. □

13

Remark 3.1. Note that the proof above shows that in general, if we
let Tj e kercui and write

d(t)t{v) = x{t)X2 + y{t)x^

then

x = y,
y + Kx = 0.
We will use this fact later on.

We now recall some global consequences of i^' = 1 as explained by R.
Bryant in [B] . The first thing to observe is that the structure equations
imply:

(6) 0*u;i=cji,

(7) (pt^2 = COS tuj2 + sin t

(8) (pt^3 = — smtuj2 + cos t UJ3

Suppose that S is geodesically complete and connected. Then it
can be shown that S is diffeomorphic to S"^ and there exists a unique
orientation reversing isometry A : S — >■ S such that dA\M = (pn (we
call A a quasi- antipodal map). Moreover for any point p G E every
unit speed geodesic leaving p passes through A{p) at distance vr and S
has diameter vr. According to [6], Proposition 4] we have the following
dichotomy:

(1) A"^ is the identity on S in which case all geodesies are closed
with the same minimal period 27r;

(2) A"^ has exactly two fixed points, say a and A{a). Moreover there
exists a positive definite inner product on T^-E that is preserved
by d{A'^){a) : T^-S T„T, and there is an angle 9 e (0, 27r)
such that d{A'^){a) is counterclockwise rotation by 9 in this
inner product.

Here we will be mostly interested in case (2). We note that this case
has two possible subcases. Suppose that 9/2tt is rational and write
9 = 2'n'p/q where < p < g with p and q coprime. Then A^"? is the
identity and thus (f)2nq = identity, that is, every orbit of is closed with
period 27rg, although some orbits may have smaller minimal period.

When 9/27T is irrational, then the iterates of A'^ are dense in a circle
of isometries of the Finsler surface S. This circle of isometries fixes a
and A{a) and S is rotationally symmetric about a. Hence the geodesic
flow of S is completely integrable with a "Clairaut" first integral. The
surface is also symmetric with respect to A about a circle (the equator)

14

E. The unit tangent vectors to E determine two closed orbits 7±. The
equator divides S into two disks D„ and DA{a) which contain a and
A{(t) respectively. {A maps D^j to -Da(o-) and fixes E setwise.)

Let S be the subset of M given by those pairs (x, v) where x & E and
V points inside the region D^. The set S (diffeomorphic to x (0, vr))
is a section of the geodesic fiow with return map 027r|5 = We
see that there are no other closed orbits besides 7±.

If 7ro7^ is the closed geodesic which travel around E counterclockwise
(as seen from a) then we find the lengths of vr o 7^ to be:

:=£(7ro7+) = 27r-rf(p,A2(p)),

£_ := ^{'K o 7_) = 27r + d{A^{p),p),

where p is any point in E. Since 9/2ti is irrational we see that 1±/2t[
are irrational and in view of ([7j) and ([8]) we conclude that 7± are elliptic
orbits of 0. It is also clear that they are locally recurrent.

3.2. Examples with only two closed geodesies. Summarizing the
discussion above, Theorem A can be applied to a Finsler metric on
with K = 1 and 6/2tt irrational. Examples of such metrics are given
by the well known Katok examples pjj analyzed by W. Ziller in [TS] .
The fact that these metrics have i^' = 1 is proved by Z. Shen [TB] (see
also [HI Section 5] for a discussion of these examples).

However these are not the only examples and a Katok type construc-
tion also gives a larger class as we now explain.

Suppose F is a Finsler metric on with K = 1 and A"^ = Id (i.e. all
the geodesies are closed and with the same minimal period). Suppose
in addition that F admits a nontrivial Killing field V. We may suppose
without loss of generality that the fiow ft of V is such that /2,r = Id-
Define a 1-parameter family of Finsler metrics {e small) by giving
its co-metric G* in T*^^ as follows:

(9) G*{x,p) = F*{x,p) + ep{V{x)).

An unpublished result of P. Foulon asserts that also has K = 1 (this
can be checked along the lines of the calculations in [18]). The Katok
examples arise when F is the standard Riemannian metric with K = 1
in §2.

We now note that Bryant [5] has produced several families of
Finsler metrics with K = 1 and A"^ = Id. Among them there are sub-
families with rotational symmetry which are not of Randers type. For
example Theorem 10 in |;4] gives a 1-parameter family of rotationally
invariant projectively fiat Finsler metrics with K = 1.

15

To complete our construction we note that Ziller in [TH] observed
that for a Finsler metric defined by with e irrational, the only
closed geodesies of are those invariant under ft. Hence there will
be only two closed geodesies for e irrational.

We remark that the results in [3j show that the Katok examples are
the only examples (up to isometry, of course) of Randers metrics with
K = 1. A Randers metric is a Finsler metric of the form \/gx{v, v) +
6x{v), where g is a Riemannian metric and 6* is a 1-form.

4. Lifting Finsler metrics to contact structures on

Let F be a Finsler metric on The Lagrangian gives rise to
a Legendre transform ip : TS^ \ {0} ^ T*§2 \ {q} and if we let A be
the Liouville 1-form on T*S^, it is well known that i*pX restricted to
the unit sphere bundle M of F is precisely the contact form uji from
the previous section.

On T*S^ we also have a Finsler co-metric F* such that F = F* o ip
and a corresponding unit co-sphere bundle M*. We consider on §^
the canonical metric with curvature 1. It has an associated Legendre
tranform : T§^ —>■ T*§^. In what follows | ■ | denotes the norm of the
canonical metric in both TS^ and T*S^.

Observe that there exists a unique smooth function g : T*§^ \ {0} — *
such that F*{x,p) = g{x,p)\p\. {g is homogeneous of degree zero
in p.)

Let r : Mq M* be given by

We identify with §U(2) and with the matrices in §U(2) of the

as above then A~^xA G S^. This also gives a natural embedding of §^
in M^. Below we will often write (t, z) to indicate a point in §^ instead
of the corresponding matrix in SU(2).
Consider the matrices:

r{x,p) = {x,p/g{x,p)).

form

where t G M, 2 G C and + \z

1. If A G §U(2) = §3 and x G §Ms

Now let G :

Mo be the following map:

G{A) = {A'^jA,A-^kA).

16

It is easy to see that the unit vectors A ^jA, A ^kA in are orthog-
onal. Let A e §11(2) be written as:

A

Wi W2
-W2 Wl

where (^1,^2) G and + 1^2^ = 1. The map G may also be
written as

(10) G{A) = [{2'^{wiW2),wl + wl), {-2^{wiW2),i{wl - wj))].

If Ao is the canonical contact form on §^ (Ao(a;) = {ix, *)) and cj^
is the contact form of Mq, then it is not hard to check that (and it is
explicitly done in |[7i, Proposition 1.1]):

(11) G*o;? = 2Ao.

We now claim (compare with [71 Proposition 1.1]):
Lemma 4.1. Let / := 1/^ o 4 : TS^ \ {0} M+. Then
G*i;r* (A|mO = 2(/oG)Ao.

Proof. Let r : T*S^ be the canonical projection and note that

r*\x,p){0 =Pl9{d'r{dr{i))) = -A(^,p)(c?r(0)

9

and that

t,{\/g) = u;l,/{goQ.
Combining these two equalities with flTT]) the lemma follows.

□

The lemma is saying that the geodesic flow of a Finsler metric on §^
is (up to a double covering) smoothly conjugate to the Reeb flow of a
(tight) contact form on §^ of the form h Aq where h = 2{f oG) and / is
related to the Finsler metric as described in the lemma. Conversely, if
we have a contact form hXo with h{—A) = h{A), then h will give rise
to functions / and g as above. In general, for an arbitrary h invariant
under the antipodal map, the hypersurface M* of T*§^ determined by
g does not need to be fibrewise strictly convex, but it will be clearly
starshaped. Note that in the proof of the lemma we did not really need
M* to come from a Finsler metric. If M* is just starshaped, then A|m*
is also a contact form.

Summarizing, the lemma gives this: there is a 1-1 correspondence
between starshaped hypersurfaces of T*§^ and smooth positive func-
tions h on §^ with h{A) = h{—A). If in addition M* is fibrewise strictly
convex we obtain a Finsler metric.

17

5. Ellipsoids and the Katok examples

In contact geometry there is a well studied class of examples given
by the ellipsoids

Ep.g := {(wi, W2) e : p\wi\^ + g|w2p = 1},

where p and q are positive real numbers. The restriction of Aq to Ep^g
determines a Reeb flow whose dynamics is very simple: the flow is just
(f)t{wi,W2) = (wie*^*, ^26*''*). There are two periodic orbits correspond-
ing to Wi = and W2 = 0. These are the only periodic orbits if p/q is
irrational, whereas Ep^g is foliated by periodic orbits if p/q is rational
(but not all will have the same minimal period).

On the other hand, in Finsler geometry there are the well known
Katok examples, which in particular provide examples of Finsler met-
rics with only two closed geodesies. The Finsler co-metric of the Katok
examples in geodesic polar coordinates (r, 0) G (0, vr) x [0, 27r] is

;i2) F*{r,(f),pr,p^) = \IpI + -^pI + ep^,

V sm r ^

where e G (—1,1). We will show in this section that the ellipsoids
and the Katok examples are related precisely by the correspondence
described in the previous section.
Let /I : §3 ^ M+ be the function

(13) h{wi,W2) = -. — 7:r—\ — T^-

p\wi\'^ + q\w2\

Also let v9 : — > Ep^g be given by ip{wi,W2) = ^h{wi, W2) (ifi, ^2)- It
is easy to check that

(p*Xo = hXo

and hence we can think of the Reeb flow of the ellipsoids as being
defined on §^ with contact form h Aq where h is given by f|T3l) . Clearly
h{A) = h{—A). To find / we write G{A) = (x, v) G Mq where x = (t, z)
and V = {b,ri). The expressions for t,z,b,ri are given by ( ITOi) . Using
them we derive

-2 z-'^v

^ 2 '

2 z + ir]

Wn = .

^ 2

Using that (1^1,^2) G S'^ we obtain

1

f{x,v)

{p ~ ~ "i-vl + 25

18

We now introduce geodesic polar coordinates (r, 0) on §^ such that

t = cos r,
z = sinre*"^.

Thus {b,rj) = {t,z). In these coordinates the Legendre transform io is
simply:

Pr = r, P(f, = 0sin^r.

We can now compute g = l/(/ o £q ^) in (r, (l),Pr,p<i,) -coordinates:

g{r, (f),pr,p^) = 2q + {p — q)y r'^ cos^ r + sin^ r (1 +

sin^ r

= 2q + {p — q)]^ pf. cos^ r + sin r + 2p^
This is the expression of g on Mq, that is, when in addition

Pl + ^ = l.
sm r

Thus if we simplify it further we obtain

g{r,(j),pr,p^) = 2q + {p - q){l + p^) = p + q + {p - q) p<p.

But in view of (fT2l) the value of F* on Mq is just 1 + eptp. Note that
by homogeneity a Finsler metric is completely determined by its value
on Mq. Hence if we choose p and q such that p + q = 1, the ellipsoids
induce, under the correspondence described in the previous section,
exactly the Katok examples with e = p — q.

Remark 5.1. In [9l Theorem 1.1] the authors show that a strictly
convex hypersurface S* C carries either 2 or infinitely many periodic
orbits. The ellipsoids, of course, provide examples of this dichotomy
and the authors point out that it is not true that the first alternative
holds only for the irrational ellipsoids. They also remark in [9], Page
200] that M. Herman constructed examples of hypersurfaces S which
are C°°-close to the ellipsoid, admit only two periodic orbits but have a
transitive flow on S. Apparently these examples of Herman are unpub-
lished, but we remark here that it is easy to construct such examples
using the results above combined with Katok's main result in [TT]. In-
deed, Katok shows in [11] that given any r, one can approximate
in the topology by a Finsler metric F with ergodic geodesic flow
and only two closed geodesies. We have shown above that under the
lifting procedure described in Lemma 14.11 lifts to an ellipsoid, and
hence F will give rise to a smooth hypersurface S which is C'-close to

19

an ellipsoid and such that the flow on S has only two closed orbits and
is transitive.

Note that the examples described in Subsection 13.21 give rise to a
new 1-parameter family of hypersurfaces in exhibiting the same
dynamics and dichotomy as the ellipsoids.

Finally we note that if under the lifting described in Lemma 14.11 a
Finsler metric gives rise to a strictly convex hypersurface, then it will
carry either 2 or infinitely many closed geodesies. Recently, V. Bangert
and Y. Long have shown that any Finsler metric on §^ has two closed
geodesies. In the next section we will describe geometric conditions on
the Finsler metric that ensure dynamical convexity so the main results
in [9] can be applied.

6. Dynamically convex Finsler metrics

We first recall the definition of dynamically convex contact form A
on a closed, connected and oriented manifold M with 7r2(M) = 0.

The Reeb vector field X is transversal to the contact structure ^ =
ker A so that naturally we have a splitting T^M = MX(x) © C,x- Let
0j be the flow of X. Clearly dcpti^x) : ^M^) symplectic with

respect to d\. A contractible periodic solution x with period T has an
integer- valued index ^{x,T) which we briefly recall (cf. [D]).

Let D G C he the closed unit disk. Choose a smooth map a :
D ^ M such that a{e'^^^*^'^) = x{t). Choose a symplectic trivialization
ijj : a*^ —y D X M."^ of the symplectic bundle a*^ with symplectic form
a*dX. Here is endowed with the standard symplectic form. We can
use this trivialization to define a symplectic arc

$: [0,T]-.5p(l)

by setting

Any such symplectic arc has a Conley-Zehnder index yu($), whose def-
inition we recall below, and we define

/i(x,T)=/i($).

It can be checked that T) is well defined and does not depend on
the choices made.

There are several possible ways to define yu($). We shall present the
definition that is most appropriate to our purposes. Our reference for
what follows is [9l Section 3].

Let $ : [0,T] — >• Sp{l) be a smooth arc with $(0) = / and set
A{t) := — J$(t)<l'(t)^^. Then A{t) is a smooth path of symmetric

20

matrices and $ solves the linear differential equation:

$ = JA^, $(0) = /

for t G [0,T]. Given r G M, suppose v (not identically zero) solves the
first order differential equation

(14) - Jv{t) - A{t)v{t) = Tv{t)

and satisfies the periodic boundary condition w(0) = v(T). Then v is
an eigenvector with eigenvalue r of the operator

La{v) = -Jv - A{t)v

defined on the space of T-periodic if^-maps into M^. Since v never
vanishes we may choose a smooth angle (p{t) such that

2ni,fiit) _ ^(^) f ^ \o r]

~ \vit)\' ^U'^J-

Define the winding number A{t,A) G Z by

Air, A) :=^(T)-^(0).

One can see that A(r, A) depends only on the eigenvalue r and not on
the eigenf unction v. For every integer A; G Z there are precisely two
eigenvalues (counting multiplicities) ti and T2 such that k = A{ri, A) =
A{t2, A). Hence we may label the eigenvalues r of La by their winding
numbers. Indeed, set A{Tk,A) = [|] for G Z and < tj for k < j.
With this labeling we define the Conley-Zehnder index of $ as

:= max{A; : < 0}.

A contact form A on M is said to be dynamically convex if ii{x, T) > 3
for every contractible periodic solution of the Reeb vector field X. In
[9] it is shown that if a contact form A on comes from a strictly
convex hypersurface in M^, then it is dynamically convex.

Recall from the previous sections that a Finsler metric on §^ defines
a contact form in §0(3). We shall say that the Finsler metric is dy-
namically convex if its associated contact form is dynamically convex.

Given a Finsler metric on S^, let £ be the length of the shortest
geodesic loop.

Theorem B. Let F be a Finsler metric on such that K > 6 > 0. If
£ > tt/v^, then F is dynamically convex. In particular by the results
in [9] any such Finsler metric has either two or infinitely many closed
geodesies.

21

Proof. By a simple rescaling argument we may assume that K > 1 and
that i > 7c.

Consider a closed geodesic 7 with length T. The vector fields X2
and X3 provide a trivialization of ^ without the need of extending the
bundle to D. Given 77 G ^, write f] = 0X2 + 6X3. Then

V ^ (b, a)

gives a symplectic trivialization between dui) and with the canon-
ical symplectic form. (Note that we need to swap a and b since dui =
—U2 A UJ3.) Recall that if we write d(l)t(rf) = XX2 + I/X3, then x = y
and X + Kx = 0. Thus, using the trivialization, we see that $ satisfies
the linear differential equation:

where

J=( J and A=( J ^

Take v an eigenvector of with eigenvalue r and set w{t)
Using (HM it is straightforward to check that w{t) satisfies

w = J{A + tI)w — w{w, J{A + tI)w).

But 2TiipJw = w and thus 27iip = {{A + tI)w, w). Since K > 1, we see
that {{A + tI)w, w) > 1 + r for all t G [0, T] and hence

A(r,A)=^(T)-^(0)>^(l + r).
Since A{t3,A) = 1 we deduce

(15) £(l + r3)<l.

Let us combine this inequality with the lower bound on i. If 7 is a
simple closed curve in S^, the orbit t x(t) = {'y(t),'j(t)) will not be
contractible in M. If 7 is not a simple closed geodesic, then it must
have length T > 2i > 2tt. Inequality f[T^ shows that < and by
the definition of /x we see that fj,{x, T) > 3 as desired.

□

Remark 6.1. The proof of the theorem indicates that one cannot
expect to get dynamical convexity just assuming positive curvature.
In fact if 7 is a closed geodesic with length T such that K = 1 along
it, then the proof of the theorem shows that

(16) l = ^il+Ts).

Ztx

22

Now consider a convex surface of revolution whose equator has K = 1.
It is clear that one can make the length T of the equator as short as
desired at the expense of increasing the curvature elsewhere. But using
(fT6|) we see that the Conley-Zehnder index of the equator iterated twice
(to get a contractible curve in the unit sphere bundle) is < 2 if T is
small enough since T3 > in this case. A concrete example of this is
given by the ellipsoid in M^:

The equator z = has curvature K = 1 and length 27ra, so for a = 1/2
we get that the equator iterated twice has t^, = and fi = 1. Note that
in this case 1 < K < 16. This example shows that Theorem B is in
fact sharp.

Hence there are positively curved Riemannian metrics on such
that under the lifting procedure described in Lemma 14.11 they do not
give rise to convex hyper surf aces in M^.

The results of Hofer, Wysocki and Zhender in [9] treat a lot more
than just the dichotomy "two or infinitely many" closed orbits. Their
results state that the Reeb flow has a closed disk as a surface of section
whose boundary is an unknotted closed orbit P with fi = 3. Moreover,
any other closed orbit has to be linked with P.

If the Finsler metric is reversible, then one can obtain a lower bound
for i from an upper bound on curvature. Indeed if < K < a, then
i > In j ^Ja. For Riemannian metrics this is a classical result (cf. [121
Theorem 3.4.8]) and for reversible Finsler metrics the proof is quite the
same (cf. [l3l Theorem 4]).

Thus Theorem B implies:

Corollary 6.1. A strictly 1 /^-pinched positively curved reversible Finsler
metric is dynamically convex.

The corollary seems new even for the case of Riemannian metrics.
Given a Finsler metric F, the reversibility r of F is (cf. |14j):

r := max {F{x, —v)} > 1.

Clearly r = 1 iff F is reversible. In Theorem 4] Rademacher gives a
lower bound for the length L of the shortest closed geodesic as follows.
If < < 1, then

L > 7r(l + 1/r).

The same bound holds for i [15j and hence we may also conclude
that a Finsler metric with (l — j^)^ < K < 1 is dynamically convex.

23

References

[1] V. Bangert, Y. Long, The existence of two closed geodesies on every Finsler
2- sphere, preprint.

[2] D. Bao, S.S. Chern, Z. Shen, An introduction to Riemann- Finsler geometry,
Graduate Texts in Mathematics, 200. Springer- Verlag, New York, 2000.

[3] D. Bao, David, C. Robles, Z. Shen, Zermelo navigation on Riemannian mani-
folds, J. Differential Geom. 66 (2004) 377-435.

[4] R. Bryant, Projectively flat Finsler 2-spheres of constant curvature, Selecta
Math. 3 (1997) 161-204.

[5] R. Bryant, Some remarks on Finsler manifolds with constant flag curvature,
Houston J. Math. 28 (2002) 221-262.

[6] R. Bryant, Geodesically reversible Finsler 2-spheres of constant curvature,
preprint, math.DG/0407514,

[7] G. Contreras, F. Ohveira, densely the 2-sphere has an elliptic closed geo-
desic, Ergod. Th. and Dynam. Sys. 24 (2004) 1395-1423.

[8] A. Harris, K. Wysocfci, Branch structure of J -holomorphic curves near peri-
odic orbits of a contact manifold. Trans. Am. Math. Soc. to appear (cf. also
math .CV/0701496)

[9] H. Hofer, K. Wysocfci, E. Zehnder, The dynamics of three-dimensional strictly
convex energy surfaces, Ann. of Math. 148 (1998) 197-289.

[10] H. Hofer, K. Wysocki, E. Zehnder, Finite energy foliations of tight three-
spheres and Hamiltonian dynamics, Ann. of Math. 157 (2003) 125-255.

[11] A. Katok, Ergodic properties of degenerate integrable systems. Math. USSR
Izv. 7 (1973) 535-572.

[12] W. Klingenberg, Riemannian geometry, De Gruyter Studies Math.l, de
Gruyter, Berhn-New York, 1995.

[13] G.P. Paternain, Geodesic flows, Progress in Mathematics, 180 Birkauser 1999.

[14] H.B. Rademacher, A sphere theorem for non-reversible Finsler metrics. Math.
Ann. 328 (2004) 373-387.

[15] H.B. Rademacher, Personal communication.

[16] Z. Shen, Two-dimensional Finsler metrics with constant curvature, Manusc.

Math. 109 (2002) 349-366.
[17] J.C. Sikorav, Singularities of J -holomorphic curves. Math. Z. 226 (1997) 359-

373.

[18] W. Ziller, Geometry of the Katok examples, Ergod. Th. and Dynam. Sys. 3
(1983) 135-157.

School of Mathematics, Statistics and Computer Sciences, Univer-
sity OF New England, Armidale NSW 2351, Australia
E-mail address: adamh@turing.une.edu.au

Department of Pure Mathematics and Mathematical Statistics, Uni-
versity OF Cambridge, Cambridge CB3 OWB, England
E-mail address: g . p . paternainSdpmms . cam .ac.uk

24