Isometryinvariant geodesics
and the fundamental group
Abstract.
We prove that on closed Riemannian manifolds with infinite abelian, but not cyclic, fundamental group, any isometry that is homotopic to the identity possesses infinitely many invariant geodesics. We conjecture that the result remains true if the fundamental group is infinite cyclic. We also formulate a generalization of the isometryinvariant geodesics problem, and a generalization of the celebrated Weinstein conjecture: on a closed contact manifold with a selected contact form, any strict contactomorphism that is contactisotopic to the identity possesses an invariant Reeb orbit.
Key words and phrases:
Isometryinvariant geodesics, closed geodesics, Morse theory2000 Mathematics Subject Classification:
58E10, 53C221. Introduction
1.1. Background and main result
Since the work of Hadamard [Had98] and Poincaré [Poi05], the problem of the existence of closed geodesics has occupied a central place in Riemannian geometry. In the presence of symmetry, described by an isometry of the Riemannian manifold , the analogous problem consists in searching for geodesics on which the isometry acts as a nontrivial translation: namely, geodesics that, after being suitably reparametrized with constant positive speed, satisfy for all . These geodesics are simply called isometryinvariant or, more specifically, invariant. Their study was initiated by Grove [Gro73, Gro74] in the 1970s, and since then several existence and multiplicity results have been established (see [Maz13] for some history of the problem).
In a beautiful short paper of 1984 [BH84], Bangert and Hingston proved that on any closed Riemannian manifold with infinite abelian fundamental group, the growth rate of closed geodesics is at least the one of prime numbers (in particular, there are infinitely many of them). This lower bound is obtained by a clever use of a simple nondivisibility argument: on the space of closed curves whose homotopy class is a prime multiple of a generator of infinite order of the fundamental group, a closed geodesic obtained by means of a suitable minimax scheme is not iterated. In the current paper, we investigate the generalization of Bangert and Hingston’s result to the case of geodesics invariant by an isometry homotopic to the identity. The situation, here, appears way more complex: the nondivisibility argument mentioned above does not seem to hold in the isometryinvariant setting. However, a subtle study of the local properties of isometryinvariant geodesics still allows to recover a multiplicity result in case the fundamental group is not cyclic.
Let be a closed connected Riemannian manifold whose fundamental group is isomorphic to , for some nontrivial abelian group . Every isometry of homotopic to the identity admits infinitely many invariant geodesics.
It seems reasonable to expect that the result still hold in case the fundamental group is isomorphic to . The seemingly technical difficulty we come across when we relax the noncyclicity assumption is that the available minimax scheme that produces isometryinvariant geodesics is not onedimensional anymore: as in the closed geodesics case, we would need to perform minimax on a suitable family of nontrivial spheres, for some , inside the space of isometryinvariant paths. We refer the reader to Remark 3.2 for a more precise discussion of this issue.
1.2. Invariant Reeb orbits
The geodesic flow of can be seen as the Reeb flow on the unit cotangent bundle equipped with the contact form given by the restriction of the Liouville form. An isometry of lifts to a contactomorphism of that preserves the contact form. In this way, the problem of isometryinvariant geodesics can be seen as a special instance of the problem of invariant Reeb orbits, which goes as follows. Let be a closed contact manifold, equipped with a contact form such that and a contactomorphism that preserves . We denote by the corresponding Reeb vector field, which is defined by and . We call invariant Reeb orbit a curve such that and, for some timeshift , satisfies . We wish to stress the importance of the requirement that the timeshift be nonzero: with this definition, invariant Reeb orbits are precisely closed Reeb orbits. Therefore, the existence problem for invariant Reeb orbits contains the corresponding problem for closed Reeb orbits. It seems natural to formulate the following generalization of the celebrated Weinstein conjecture (see e.g. [HZ94, page 120]).
In a closed contact manifold with a selected contact form, every contactomorphism that preserves the form and is contactisotopic to the identity admits a invariant Reeb orbit. ∎
The notion of invariant Reeb orbit is related to the one of translated point, introduced by Sandon in [San12]. Given a contactomorphism of the above contact manifold that does not necessarily preserve the contact form , a translated point of is a point that belongs to the same Reeb orbit of its image and satisfies . In the same paper, Sandon conjectured that any such that is contactisotopic to the identity possesses at least as many translated points as the minimal number of critical points of a smooth realvalued function on . When preserves the contact form , its invariant Reeb orbits are precisely the Reeb orbits of translated points that are closed or not contained in the set of fixed points of . If we further assume that does not have fixed points (which is true for a generic perturbation of by the Reeb flow) our Conjecture 1.2 becomes a special case of Sandon’s one. However, we wish to stress that the two conjectures have quite a different flavor: indeed, ours contains the Weinstein conjecture, whereas this is not the case for Sandon’s one.
Conjecture 1.2 is known to be true in the geodesics setting: in a closed Riemannian manifold, every isometry that is homotopic to the identity possesses an invariant geodesic, as Grove proved in [Gro73, Theorem 3.7]. If the isometry is not homotopic to the identity but is fixed point free, an invariant geodesic can be found by a simple minimization procedure [Gro73, Proposition 2.7]. However, the topology of the Riemannian manifold may force the existence of fixed points (e.g. if the Euler characteristic is nonzero), and the hypothesis that the isometry be homotopic to the identity cannot be completely relaxed, as the example of the isometry on the flat torus shows.
More generally, Conjecture 1.2 holds when the contact manifold is a unit cotangent bundle (equipped with the Liouville contact form) of a closed manifold whose fundamental group is either finite or has infinitely many conjugacy classes^{1}^{1}1There is no known example of a finitely generated infinite group with finitely many conjugacy classes.. This statement is not explicitly proved in the literature, but can be inferred by combining the arguments in [AM13] and [Kan10]. The author is grateful to Peter Albers and Will Merry for pointing this out to him.
1.3. Organization of the paper
In Section 2 we discuss the features of the variational principle for isometryinvariant geodesics: in Sections 2.1, 2.2, and 2.3 we give the background on the Morse theory of the energy function, while in Section 2.4 we provide a crucial technical result (Lemma 2.4) on minimax isometryinvariant geodesics. In Section 3 we prove multiplicity results: in Section 3.1 we establish Theorem 1.1 in the easier case in which the rank of the fundamental group is at least two, and finally in Section 3.2 we carry out the proof of Theorem 1.1 in the general case.
2. Preliminaries
2.1. The energy function
Let us recall the variational setting for the study of isometryinvariant geodesics from [Gro73]. We consider a closed connected Riemannian manifold of dimension larger than one, equipped with an isometry . We denote by the space of curves such that for all . This space, equipped with a suitable infinitedimensional Riemannian metric, is a complete Hilbert manifold. The real line acts on by timeshift: given an invariant curve , a real number acts on it by . This action is not free on the subspace of periodic curves, the isotropy groups being generated by the minimal period of the curve if it is nonzero, or the whole if the curve is stationary.
We will be interested in the energy function , which is given by
This is a smooth invariant function that satisfies the PalaisSmale condition. Its critical points may be of two kinds: “genuine” critical points with positive critical value, which are invariant geodesics of parametrized with constant positive speed, and “spurious” critical points with critical value zero, given by those curves that are constantly equal to some fixed point of the map . Genuine critical points come in critical orbits , which are homeomorphic to either a circle if is a periodic curve, or the real line if is an open curve.
The critical orbits of open invariant geodesics are nonisolated ones, as it was proved by Grove in [Gro74, Theorem 2.4]. This important fact implies that, when studying the multiplicity of isometryinvariant geodesics, one can always assume that all such geodesics are periodic ones. Any periodic invariant geodesic gives rise to an infinite sequence of critical orbits of the energy: indeed, the geodesic can be parametrized with some with minimal period , so that ; for every positive integer , the curve given by is also a critical point of , but it is equivalent to from the geometric point of view.
2.2. Topology of the space of invariant paths
The easiest way to find critical points of a bounded from below function satisfying the PalaisSmale condition is by minimizing it on any connected component of its domain. For the energy function , such a procedure may end up giving a critical point with zero critical value, which is simply a fixed point of the isometry . However, this happens only in finitely many connected components of the space of invariant paths.
The critical points of the energy function with zero critical value are contained in finitely many connected components of .
Proof.
We only have to consider the case in which possesses some fixed points, otherwise the statement is trivially true. We denote by the Riemannian ball in that is centered at and has radius , and by the injectivity radius of our closed Riemannian manifold. For each there exists such that all points lie at distance less than from their image . For all such points we consider the unique geodesic parametrized with constant speed such that joins with and has length equal to their distance. Notice that, if , its associated curve is stationary at . Since is compact, we can find finitely many points such that the union of the balls covers . Consider the map given by . This map is clearly continuous, and thus sends each Riemannian ball into some connected component of . Since the stationary invariant curves (which are the fixed points of ) are contained in the image of , they are contained in finitely many connected components of . ∎
As it was first remarked by Grove [Gro73, Lemma 3.6], the homotopy type of the space of invariant curves depends only on the homotopy class of within the space of continuous map of into itself. In particular, if is homotopic to the identity, is homotopy equivalent to the free loop space . Given a continuous homotopy , with and , an explicit homotopy equivalence can be constructed as
(1) 
A homotopy inverse is the map given by
(2) 
Actually, in order for to lie in , all the curves of the form must have regularity . One can achieve this by an arbitrarily small perturbation of the homotopy , for instance as follows. Choose an integer that is large enough so that, for all and with , the Riemannian distance between and is less than the injectivity radius of . Let be the (unique) continuous curve such that, for each , its restriction is the shortest geodesic parametrized with constant speed joining and . We can now reset the homotopy to be . ∎
From now on, let us assume that the fundamental group is abelian. This implies that there is a bijective map that sends any homotopy class of loops to the corresponding connected component. By taking the composition with the map induced by the homotopy equivalence (1) we obtain a bijective map
(3) 
Consider the evaluation map that sends every loop to its base point . Since the fundamental group of is abelian, this map induces a surjective homomorphism of the fundamental group of any connected component of the free loop space onto the fundamental group of . By precomposing with the isomorphism induced by the homotopy equivalence (2), we also get a surjective homomorphism of the fundamental group of any connected component of onto the one of . More precisely, for all and with , the evaluation map given by induces a surjective homomorphism
(4) 
2.3. Local properties of isometryinvariant geodesics
Let be an invariant geodesic that is periodic with minimal period (for the remaining of section 2 we do not need to assume the isometry to be homotopic to the identity). The local Morsetheoretic properties of the sequence of critical orbits were thoroughly investigated by Grove and Tanaka [GT78, Tan82]. Here, we shall recall only the ones that will be needed for the proof of our multiplicity results in section 3 (for a quick reference, see [Maz13, Sections 2.2 and 3.1]).
We recall that the Morse index is the (finite) dimension of the negative eigenspace of the Hessian of at the critical point. The behavior of the index along our sequence of critical orbits is described by the following lemma. {lem}[GroveTanaka] The sequence converges to a nonnegative real number as . Moreover, if this limit is zero, then for all . ∎
In order to discuss further properties, it is more convenient (and formally equivalent) to see each critical orbit of our sequence as the same critical orbit living in different spaces. For all we introduce the space of invariant paths
and the average energy given by
In order to go forth and back from this new setting to the former one, we can employ the diffeomorphism given by , which relates the energies as . The curve , which is a critical point of , corresponds to the curve , which is a critical point of . The next lemma tells us that, if we want to understand the behavior of the function near the critical point for all , it is enough to do it for finitely many values of . {lem}[GroveTanaka] There exists finitely many spaces , integers that are multiples of the period , and a partition with the following properties:

each space contains , is contained in
as a Hilbert submanifold of each of the spaces involved in the intersection, and is invariant by the gradient flow of for all ;

we have for all ;

the null space of the Hessian of at is the same as the null space of the Hessian of restricted function at for all . ∎
2.4. Bangert’s homotopy and isometryinvariant geodesics
A sufficiently iterated closed geodesic cannot be detected by a 1dimensional minimax scheme. This fundamental remark was first made by Bangert [Ban80], and played a crucial role in the proof of many existence results in Riemannian geometry (see e.g. [BTZ81, BK83, BH84, Ban93, AMP13]). A detailed account that employs our notation can be found in [Maz13, Section 3.2].
[Bangert] Let be a continuous path. For each , let denote the inclusion. The path can be homotoped, fixing its endpoints, to a new path such that
where is a quantity independent of . Moreover, there is a continuous function such that for all . ∎
By means of Bangert’s Lemma and the results of section 2.3, we obtain the following important application in the context of isometryinvariant geodesics.
Let be an invariant geodesic that is periodic with minimal period . For all integers large enough, every path such that and for all can be homotoped, with fixed endpoints, inside the sublevel set .
Proof.
If , the statement is easy to prove: according to Lemma 2.3, the Morse index becomes large if is large, and thus we obtain the desired homotopy by pushing our path in a suitable negative direction of the Hessian of near . Let us now focus on the other, hard, case: . We work in the setting of section 2.3: we see as the curve with average energy
Consider the partition given by Lemma 2.3. We choose any and the corresponding space . We recall that all are periodic curves, where is some multiple of . We fix smaller than half the injectivity radius of , and we define the following neighborhood of in
Consider the sublevel set . We denote by the pathconnected components of . Notice that they are finitely many: is bounded from above by the rank of the local homology , which is finite. For all pair of distinct values we fix a continuous path
(5) 
such that , is contained in , whereas is contained in .
Let be the inclusion, where . We claim that, for all and sufficiently large, can be homotoped, with fixed endpoints, to a path that is contained in the sublevel set . This can be seen as follows. We denote by the largest integer such that , and by the inclusion. By Lemma 2.4, the path can be homotoped, with fixed endpoints, to a path such that
(6) 
Moreover, we have for a suitable continuous function . We define the continuous path to be
This path is clearly homotopic, with fixed endpoints, to : indeed, it suffices to extend in the obvious way the homotopy from to . The average energy of the curves along this path can be estimated as
where for the inequality we have used the fact that . By (6) and since we further obtain
The second and third summand in the previous line go to zero as , whereas both and are bounded from above by , for some . Therefore provided is large enough.
In order to conclude the proof of the lemma, all is left to do is to establish the following claim: let be any continuous path with and for all ; then, there is a homotopy, with fixed endpoints, from to a new path such that for all , and the restriction , for some , coincides with one of our standard paths (5).
Let us prove this claim. Consider a characteristic manifold for the function at . Namely, is a finitedimensional submanifold diffeomorphic to a ball, containing the critical point in its interior, whose tangent space is the nullspace of the Hessian of at , and such that the gradient of is tangent to along . Since , Lemma 2.3 tells us that the Morse index of at is zero. According to the generalized Morse Lemma (see e.g. [GM69, page 501] or [Cha93, page 72]), has a tubular neighborhood and a diffeomorphism (where is a Hilbert ball) that maps the zero section to and such that
In particular, we have a continuous deformation retraction given by
Notice that the deformation decreases the value of (away from the characteristic manifold ). Let be such that the portion of the path is contained inside . We take a continuous function supported inside and such that , and we replace by the homotopic path . This shows that we can assume our given path to satisfy for all . In particular and for some . Let us choose a continuous path joining and , and another continuous path joining and . We concatenate , , and in order to obtain the continuous path , i.e.
Since both paths and lie inside the small open set , there exists a homotopy such that , , and all the paths have the same endpoints. One such map can be constructed as follows. Assume for the sake of simplicity that and are piecewisesmooth invariant curves for all . We can always achieve this by small homotopies of the paths and with fixed endpoints, where denotes the inclusion as before. We set to be the shortest geodesic parametrized with constant speed joining the points and . Since
such a geodesic is unique. Moreover, since the isometry maps geodesics to geodesics, we have . Therefore, is well defined as a homotopy of paths in . Finally, we extend to a continuous path of the form by setting outside . This proves our claim, and concludes the proof of the lemma. ∎
3. Multiplicity results
3.1. Fundamental groups of rank at least two
Throughout this section we will consider a closed connected Riemannian manifold whose fundamental group is abelian, equipped with an isometry that is homotopic to the identity. If the rank of the fundamental group is larger than one, we can easily find infinitely many closed geodesics on the manifold: it suffices to pick two linearly independent elements of infinite order , and to consider the closed geodesic of minimal length among all the curves freely homotopic to a representative of^{2}^{2}2Here and in the following, we employ the additive notation for the abelian fundamental group. ; all curves and , for , are geometrically distinct. The situation is analogous for the case of invariant geodesics, even though we must be slightly more clever in the selection of the elements of the fundamental group employed to carry over the minimizations.
Let be a closed connected Riemannian manifold whose fundamental group is abelian of rank at least two. Every isometry homotopic to the identity has infinitely many invariant geodesics.
Proof.
The fundamental group of our Riemannian manifold has the form , for some abelian group . Fix two generators of . For all , we denote by the connected component of the path space corresponding to under the bijection (3). Notice in particular that distinct ’s give rise to distinct ’s. We denote by a global minimum of the energy function on the connected component . By Lemma 2.2, for all but finitely many ’s the curve is not stationary, and thus it is a genuine invariant geodesic. We claim that the family contains infinitely many (geometrically distinct) invariant geodesics. Assume by contradiction that our claim does not hold. In particular, there exists an infinite subset , a curve that is periodic with minimal period , and a function such that
We proceed iteratively for increasing values of : there exists an infinite subset , a curve that is periodic with minimal period , and a function such that
(7) 
By the pigeonhole principle, there exist integers such that (and thus ). We take two integers with . Now, consider the homotopy equivalence defined in (2). Notice that the effect of applying to an invariant curve is to concatenate with a path that depends only on and the homotopy . By equation (7), we have
(8) 
and analogously
(9) 
The loop represents an element of the fundamental group of the form
for some and . Equation (8) implies that and . This contradicts equation (9), which implies and . ∎
3.2. Fundamental groups of rank at least one
Theorem 3.1 holds under the weaker assumption that the fundamental group is infinite abelian but not cyclic. This is the statement of our main result, Theorem 1.1, which we restate here for the reader’s convenience.
Theorem 1.1. Let be a closed connected Riemannian manifold whose fundamental group is isomorphic to , for some nontrivial abelian group . Every isometry of homotopic to the identity admits infinitely many invariant geodesics.
The situation here is much harder than in the previous subsection: since the rank of the fundamental group may be one, in general we cannot expect to find infinitely many invariant geodesics that are minima of the energy function in some connected component of the space of invariant curves. Instead, as in the special case of closed geodesics [BH84], we need to consider minimax critical orbits of the energy.
Let be a generator of the subgroup . For all , we denote by the connected component of the path space corresponding to under the bijection (3). In particular and are different connected components if . Moreover, Lemma 2.2 guarantees that, for all but finitely many , does not contain stationary curves, and therefore the critical points of are genuine invariant geodesics. We denote by a global minimum of the energy on the connected component . Since we are looking for infinitely many invariant geodesics, we assume that all the critical orbits of the energy are isolated: in particular is a strict local minimum of .
If the family corresponds to infinitely many geometrically distinct invariant geodesics, we are done. Therefore, let us focus on the other case: there exists an infinite subset , a function , and a periodic invariant geodesic with minimal period such that
Since the ’s belong to pairwise distinct connected components of , the function must be injective. By applying the homotopy equivalence defined in (2), we obtain
(10) 
This implies that the homotopy class does not belong to the subgroup of the fundamental group . Up to replacing by for some , we can assume that is not a point of selfintersection of , that is, we have if and only if is a multiple of the period .
Fix a nonzero and, for all , a corresponding such that , where is the surjective homomorphism in (4). We consider the minimax value
(11) 
Standard critical point theory (see e.g. [Kli78, Cha93, Maz12]) implies that is a critical value of the energy . We claim that
Indeed, assume by contradiction that . Since is a strict minimum of the energy , there must be a achieving the minimax (11) and taking values inside the critical orbit . Since is not a point of selfintersection of , the homotopy class must be a multiple of . This is a contradiction, since is not contained in the subgroup .
For our purpose, it will be useful to single out a particularly simple achieving the above minimax.
There exists that achieves the minimax in (11) and such that lies in the sublevel set except at finitely many in which it goes through some point in .
Proof.
We start by considering an arbitrary achieving the minimax. By pushing down with the antigradient flow of , we can assume that touches the level set only along critical orbits of . Moreover, by homotopying fixing the base point , we can always assume that all the curves are piecewise smooth (indeed, even broken geodesics). Let be a small enough real number. More specifically, we want to be smaller than half the injectivity radius of , and such that for any pair of critical points belonging to distinct critical orbits there exists (actually ) such that . Fix large enough so that
For all such that the path intersects the minimax level set, we denote by [resp. ] the minimal [resp. the maximal] such that . Notice that, by our choice of and , the critical points and of the energy belong to the same critical orbit. For the other values of we may simply set . Let be the real number with smallest absolute value such that , i.e.
We define the continuous loop