43
THE C 1 GENERAL DENSITY THEOREM
FOR GEODESIC FLOWS
MARIO BESSA AND MARIA JOANA TORRES
Abstract. Given a closed Riemannian manifold, we prove the C 1 -general density the-
£T} ■ orem for geodesic flows. More precisely, that C^-generic metrics have dense closed
geodesies.
O'
(N
^; MSC 2000: primary 28D05; 54H20; secondary 37B20.
<^ . keywords: Periodic points, topological dynamics, closing lemma.
CO
1. Introduction: basic definitions and statement of the results
Q.
A Riemannian manifold (M, g) is a C^-manifold with an Euclidean inner product g x
in each T X M which varies smoothly with respect to x G M. So a Riemannian metric is
a smooth section gu — > Symm^ (TM), where Symm^ (TM) is the set of positive bilinear
and symmetric forms in TM. Pick a Riemannian metric on TM and denote by cItm(', ')
the geodesic distance associated to it on TM. Note that since all Riemannian metrics are
^. \ Lipschitz equivalent on compact subsets, the choice of the metric on TM is not important.
The geodesic flow of the metric g is the flow on TM defined by
0* : TM — ► TM
(x,v) .— > (7^(*),7l„(t)),
^ . where, 7^ „ : K — > M denotes the geodesic starting at x with initial velocity v, x G M,
v G T X M. Since the speed of the geodesic is constant, we can consider the flow restricted
to UT(M) := {(x,v) G TM : g x (v,v) = 1}.
Clearly, the orbit of a point (x, v) G UT(M) consists of the tangent vectors to the
geodesic defined by (x, v) and periodic orbits for the geodesic flow correspond to closed
geodesies on M. We write Per^cf) 1 ) C UT(M) for the set of closed geodesies. Recall
that (x,v) G UT(M) belongs to the nonwandering set of 0* denoted by f2(0*), if for
every neighborhood V of (x,v) there exists t n — > 00 such that 0* n (V) fl V 7^ 0. We
say that (x,v) G UT(M) is a <f>\ tj -recurrent point, and we denote this set by i?(0*), if
given any neighborhood V of (x,v), there exists t n such that <j) tn ((x,v)) G V. We have
-R(0*) C 0(0*). Along this paper we are going to consider that M is closed and with
dimension > 2. We observe that the geodesic flow keeps the Liouville volume invariant.
Thus, Poincare's recurrence theorem (see e.g. [8]) asserts that Lebesgue almost every point
is recurrent. Therefore, we conclude that Lebesgue almost every point is nonwandering
and so UT(M) = 0(0*).
Let TZ k (M) denote the set of C k Riemannian metrics in M endowed with the C k
topology, k > 1. We denote by 1Z 3 (M) the enclosure of TZ 3 (M) in ^(M), which is a
Baire space because TZ 3 (M) is clearly C^-dense in TZ l {M) (i.e. TZ 3 (M) = ll l {M)). A
property in Tl}{M) is said to be C 1 - generic if it holds in a residual subset of TV-(M),
that is, the subset of points in which the property is valid contains a C 1 -dense G§. In
1
2 M. BESSA AND M. J. TORRES
particular, by Baire's Category theorem (see [10]) a residual subset of 7V-(M) is C 1 -dense
u\K l (M).
Given a Riemannian metric g, a central question in dynamical systems is to know if
the periodic orbits of 0* are dense in Q^c//). Furthermore, it is a central question in
differential geometry to know when we have closed geodesies. The aim of the present
paper is to prove the general density theorem for C 1 -metrics, i.e. to show that:
Theorem 1. There exists a C 1 -residual subset M ofTZ 1 {M) such that UT(M) = Per(
for any g G &.
ty
Despite the fact that Klingenberg and Takens theorem (see [H]) assures that for a dense
subset of metrics on a compact manifold M we have infinitely many closed geodesies (see
also [H] on more general generic assumptions) , our result guarantees that indeed we have
dense closed geodesies, at least from the C 1 -generic metric standpoint. It is worth noting
that on manifolds with negative curvature the geodesic flow is Anosov and so, by Anosov
closing lemma (see e.g. [8]), the closed geodesies are dense in the manifold without any
need of generic considerations.
The first attempt to obtain the dissipative version of Theorem [T] in the much less
knotty class of homeomorphisms was in [T3] by Palis, Pugh, Shub and Sullivan. However,
the proof in [TH] was not complete as it was observed in [5J. The complete proof is
due to Hurley ([?]) using dissipative arguments. Hurley's proof uses the fact that we
can create a C°-stable periodic sink by a small C°-perturbation and Brouwer's fixed
point theorem guarantees a fixed point for every C°-close homeomorphism. Clearly, this
strategy is meaningless for the volume-preserving setting because sinks simply do not
exist. With respect to the volume-preserving context Daalderop and Fokkink proved
(see [U Proposition 4]) that the general density theorem holds for that systems. In the
present paper we mainly follow the arguments in [5], adapting them to the geodesic
flow framework and combine it with the recent proved Rifford closing lemma (P2j). It
is interesting to remark that the general density theorem was first established in the
C 1 -class by Pugh ([H]) in the late 1960's and much later by Pugh and Robinson for
volume-preserving and symplectic diffeomorphisms as also for Hamiltonians (see [15]).
Actually, the general density theorem is a direct consequence of the combination of the
closing lemma and the stability and persistence of non-degenerated closed orbits given, for
example, from the hyperbolic (and also elliptic in the conservative case) structure. This
stability and persistence holds in the smooth case but sadly the notion of hyperbolicity
(and ellipticity) is no longer valid in the topological context, as it is perceptively observed
by Rifford in [T2]: "The Pugh C 1 - Closing Lemma has strong consequences on the structure
of the flow of generic vector fields... It is worth noticing that our result is not striking
enough to infer relevant properties for generic geodesic flows (for instance, the existence
of an hyperbolic periodic orbit is not stable under C° perturbations on the dynamics)".
Fortunately, we have at hand a range of topological techniques which will allow us to
reach a generic result of undeniable interest.
2. Perturbation results
Let (j) 1 be the geodesic flow of a Riemannian metric g acting on UT(M), the unit
sphere bundle of M. Let tx : UT(M) — > M be the canonical projection. Non-trivial closed
geodesies on M are in one-to-one correspondence to the periodic orbits of 0* From now
on we denote by # 7 the representative of the closed geodesic 7 by picking a single point
in the orbit, say 9 1 = (2, v). Given a closed orbit 7 = (0„(0) : t G [0, a]} of period a we
THE C 1 GENERAL DENSITY THEOREM FOR GEODESIC FLOWS 3
can define the Poincare map "P 9 (£,7) as follows: one can choose a local (2dim(M) — 2)-
hypersurface S in UT(M) containing 9 and transversal to 7 such that there are open
neighborhoods S and S a of 9 in £ and a different iable arrival function 5 : S — > M with
6(9) = a such that the map V g (T,, 7) : S — >• S a given by v t-$ (f) g (v) is a diffeomorphism.
Given a closed geodesic c : R/Z — > M, all iterates c m : R/Z — >• M; c m (t) = c(mt) for a
positive integer m are closed geodesies too.
A closed orbit 7 (or the corresponding closed geodesic c) is called nondegenerate (c.f. [3])
if 1 is not an eigenvalue of the linearized Poincare map P c := D^^VgCE, 7). In that case,
7 is an isolated closed orbit and it o 7 is an isolated closed geodesic.
A Riemannian metric g is called bumpy if all the closed orbits of the geodesic flow
are nondegenerate. Since P c m = P™ this is equivalent to saying that if exp(27riA) is an
eigenvalue of P c , then A is irrational. We state the bumpy metric theorem [TJ [2]:
Theorem 2.1. (Bumpy metric theorem) For 2 < k < 00, the set of bumpy metrics of
class C k is a residual subset oflZ k (M).
Recently, in [TJJ, Rifford was able to overcome a problem that was open for a long
time and showed how to close an orbit of the geodesic flow by a small perturbation of the
metric in the C 1 topology:
Theorem 2.2. (C 1 - closing lemma) Let g be a Riemannian metric on M of class C k with
k > 3 (resp. k = ooj, (x,v) G UT(M) and e > be fixed. Then there exists a metric g of
class C k ~ l (resp. C°°) with \\g — g\\ c i < e such that the geodesic "if xv \ is periodic.
3. Proof of Theorem [Q
Let g £ 7?. 1 (M) and let </>* be the geodesic flow of the Riemannian metric g acting on
UT(M). We define the set of weak-periodic points of (f) 1 by
WPeri^t) : ={( x ' v ) = lim ( x n,v n ):
(x n ,v n ) e Per (</>*), g n e V}(M) and lim \\g n - g\\ c i = 0}.
We observe that thanks to the Poincare recurrence theorem, the geodesic flow is nowan-
dering on UT(M). It follows from the C 1 -closing lemma (Theorem 12.21) that:
Lemma 3.1. If g G K l (M), then UT(M) = WPertyl).
We say that a closed orbit 7 of 0^ is permanent if any g e 7?. 1 (M) and C 1 -arbitrarily
close to g has a (/^-periodic orbit 7 near 7. Let ^(0*) denote the set of all permanent
closed orbits of </>* .
Lemma 3.2. There exists a C l -residual subset M oflZ 1 (M) such that for any g G M we
have Per^g) = ^(0*).
Proof. Along this proof we borrow some argument developed in [B] together with some
elementary fixed point index theory and also the fact that TZ 3 (M) is C 1 -dense in 1Z 1 (M).
It follows from the bumpy metric theorem (Theorem 12.11) that there exists a C 3 -residual
subset e^o of TZ 3 (M), hence C 1 -dense in 7?. 1 (M), such that every metric g in Mq has all
the closed orbits of the geodesic flow nondegenerate.
We claim that exists a C 1 -residual M such that any element in it has all the closed
orbits of the geodesic flow permanent. For that we take a countable base for the topology
{Bi}i(zfq of the unit tangent bundle UT(M). The fixed point index will play a crucial role
along the proof since, in rough terms the existence of non-null index on a set assures a
4 M. BESSA AND M. J. TORRES
fixed point in that set and, moreover, displaying non-null index persists under topological
perturbations.
Now we define, for every i,n G N, the following open subsets of 1Z l (M), which a priori
do not cover the whole set 7?. 1 (M), in the following way:
(1) g G T^ n if g is free of closed geodesies of period less or equal than n in Bi\
(2) g G Xj >n if there exists Bj with diam(Hj) < diam(£>j) such that for some return
time t < n we have ft g (x) 7^ x for any x in the boundary of Bj and ind(ft g , £>,) 7^ 0.
Let ^ n be the C 2 -open and dense subset of metrics without any closed orbit of period
exactly equal to n. We claim that any metric g G Mo fl G n is such that g G J-i jn U Xj.„
for all i,n. In fact, if g has no closed orbit with period < n through £>j, then g G JFi >n ,
otherwise g has a closed orbit 7 with period a < n through B h . But, since g G ^o? 7 must
be stable (hyperbolic, fc-elliptic or completely elliptic), thus isolated from other closed
orbits with period less than a. Therefore, we accomplish (2) and g G X in .
We let the residual on the theorem to be defined by
K*' i,n '— ' -^i,n)
Let us see why it works; take g G M and 7 G Per(ft g ) of period a, then for any £>j
which intersects 7 clearly g G T itn for all n > a. As a consequence, there exists fij
with diam(£?j) < diam(£>j) and such that ind((fi g ,Bj) 7^ and moreover this property is
persistent for small C 1 -perturbations on the original metric. In conclusion 7 is permanent.
□
Let UT(M)* be the set of compact subsets of UT(M) endowed with the Hausdorff
topology.
Lemma 3.3. The map ^3: V} (M) -» UT(M)* defined by <#(g) = Per(ft g ) is lower
semicontinuous on the residual M given by Lemma \3. 2 .
Proof. We must prove that for any g\ G M, and any e > there exists a neighborhood
V of g\ such that ^{ft gi ) C B e (*$(ft g )) for all g G V, or in other words there are no
implosions of the number of closed orbits when we perturb g\. But Lemma [3.21 says that
Per((j)\ L) = 8? {ft qi ) an d the proof is completed by recalling the definition of permanent
closed orbit. □
Proof, (of Theorem H]) Noting that UT(M) = Q{ft g ) we will prove that Q,(ft g ) = Per(ft g )
for some C 1 -generic Riemannian metric in 7?. 1 (M).
From Lemma El the map <£: ^(M) ->■ UT(M)* defined by <p(#) = Per{ft g ) is lower
semicontinuous on ^. It is well-known (see [5]) that the continuity points of ?$\@ is a
residual subset ^1 C M, hence a residual subset of TV-{M). Let us see that if g G M\ (i.e.
g is a continuity point of ^\m) then UT(M) = Per(ft). The non obvious inclusion is
UT(M) C Per{ft). Assume, by contradiction, that exists (x,v) G UT(M) \Per(ft). By
Lemma I3TT1 let {g n } n ^ C iV-(M) and {(x n ,u n )} ne N C Per (ft ) be such that lim g n = g
and lim (x n ,v n ) = (x,v). By Lemma f3T2l there exists {g n }nm such that (x n ,v n ) G &(g n )
and g n is --C 1 -close to g n .
THE C 1 GENERAL DENSITY THEOREM FOR GEODESIC FLOWS 5
Since V}{M) endowed with the C 1 -topology is a Baire space we get that M is dense
in 7Z}(M). Therefore, there exist {g n } n <=N C & and {(x n , v n )}neN C Per((pi ) such that
d TM {(x n ,v n ), (x n ,v n )) < i and g n is ^-C^-close to <j n .
We conclude that limg^ = g and lim (x n ,v n ) = (x,v). Then, since g is a continuity
point of *#\m we have that lim*p(g n ) = *P(g), i.e., lim Per{ffL ) = Per^cj) 1 ). Finally, we
n— >-oo n— >oo
observe that (x, v) G Per((j)i ), or equivalently, (x,f) G Per((j) g ) which is a contradiction
□
with the assumption that (x, v) G UT(M) \ Per((p g
Acknowledgements
MB was partially supported by National Funds through FCT - "Fundagao para a
Ciencia e a Tecnologia", project PEst-OE/MAT/UI0212/2011.
MJT was partially financed by FEDER Funds through "Programa Operacional Factores
de Competitividade - COMPETE" and by Portuguese Funds through FCT - "Fundacao
para a Ciencia e a Tecnologia", within the Project PEst-C/MAT/UI0013/2011.
References
[1] R. Abraham, Bumpy metrics, Global Analysis, Proc. Sympos. Pure Math., XIV (S.S. Chern and
S. Smale, eds.), (1970), 1-3.
[2] D.V. Anosov, On generic properties of closed geodesies, Math. USSR, Izv., 21, (1983), 1-29.
[3] G. Contreras-Barandiaran and G.P. Paternain, Genericity of geodesic flows with positive topolog-
ical entropy on S 2 , J. Differential Geom., 61 (1), (2002), 1-49.
[4] F. Daalderop and R. Fokkink, Chaotic homeomorphisms are generic, Topology Appl. , 102, (2000),
297-302.
[5] M. K. Fort, Category theorems, Fund. Math., 42, (1955), 276-288.
[6] E. M. Coven, J. Madden and Z. Nitccki, A note on generic properties of continuous maps, Ergodic
theory and dynamical systems, II, 97-101, Progr. Math., 21, Birkhauser, Boston, Mass., 1982.
[7] M. Hurley, On proofs of the C° general density theorem, Proc. Amer. Math. Soc, 124 (4), (1996),
1305-1309.
[8] A. Katok and B. Hassclblatt, Introduction to the Modern Theory of Dynamical Systems, Cam-
bridge University Press, 1995.
[9] W. Klingenberg and F. Takens, Generic properties of geodesic flows, Math. Annal, 197 (1972), 4,
323-334.
[10] K. Kuratowski, Topology, vol. 1, Academic Press, 1966.
[11] H-B Rademacher, On a generic property of geodesic flows, Math. Ann. 298 (1), 1994, 101-116.
[12] L. Rifford, Closing geodesies in CI topology, J. Differential Geom., 91 (3), (2012), 361-382.
[13] J. Palis, C. Pugh, M. Shub and D. Sullivan, Genericity theorems in topological dynamics, Dynam-
ical systems — Warwick 1974, Lecture Notes in Math., 468, 241-250, Springer, Berlin, 1975.
[14] C. Pugh, An improved closing lemma and a general density theorem, American Jr. Math., 89 (4),
(1967), 1010-1021.
[15] C. Pugh and C. Robinson, The C 1 closing lemma, including Hamiltonians, Ergod. Th. & Dynam.
Sys., 3 (1983), 261-313.
Departamento de Matematica, Universidade da Beira Interior, Rua Marques d'Avila
e bolama, 6201-001 covilha, portugal.
E-mail address: bessa@ubi.pt
CMAT, Departamento de Matematica e APLicAgoES, Universidade do Minho, Campus de
Gualtar, 4700-057 Braga, Portugal
E-mail address: jtorres@mattL.uminho.pt
