On the cohomological equation of magnetic flows
Nurlan S. Dairbekov, Gabriel P. Paternain

TL;DR
This paper investigates the cohomological equation for magnetic flows on closed manifolds, establishing conditions under which solutions exist, specifically that the right-hand side must be zero and the 1-form must be closed.
Contribution
It extends previous results by proving the cohomological equation has solutions only when the forcing term is zero and the 1-form is closed, without assuming Anosov flow conditions.
Findings
Solutions exist only if h=0 and θ is closed
The result generalizes prior work to non-Anosov flows
Provides conditions for the solvability of the cohomological equation
Abstract
We consider a magnetic flow without conjugate points on a closed manifold with generating vector field . Let and let be a smooth 1-form on . We show that the cohomological equation \[\G(u)=h\circ \pi+\theta\] has a solution only if and is closed. This result was proved in \cite{DP2} under the assumption that the flow of is Anosov.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMathematical Dynamics and Fractals · Geometric Analysis and Curvature Flows · Geometry and complex manifolds
On the cohomological equation of magnetic flows
Nurlan S. Dairbekov
Kazakh British Technical University, Tole bi 59, 050000 Almaty, Kazakhstan
and
Gabriel P. Paternain
Department of Pure Mathematics and Mathematical Statistics, University of Cambridge, Cambridge CB3 0WB, England
Abstract.
We consider a magnetic flow without conjugate points on a closed manifold with generating vector field . Let and let be a smooth 1-form on . We show that the cohomological equation
[TABLE]
has a solution only if and is closed. This result was proved in [10] under the assumption that the flow of is Anosov.
1. Introduction
In the present paper, which is a sequel to [10], we study the cohomological equation for a certain class of second order differential equations on the tangent bundle of a closed connected manifold with canonical projection ( for , ).
The dynamical systems that we will consider are magnetic flows. In their most general form they are determined by a pair , where is a Finsler metric on and is a closed 2-form. The Legendre transform associated with the Lagrangian is a diffeomorphism and defines a symplectic form on , where is the Liouville 1-form on . The magnetic flow of the pair is the Hamiltonian flow of with respect to the symplectic form .
Contrary to what happens for geodesic flows, a magnetic flow can change behaviour drastically as we change energy levels. One can obtain the behaviour in all energy levels by restricting the flow to the unit sphere bundle and changing by , where . Henceforth, we shall only consider acting on and we will denote by the infinitesimal generator of . A curve given by will be called a magnetic geodesic.
The cohomological equation (also known in the literature as the kinetic equation) is simply
[TABLE]
where are functions on . The function is said to be a coboundary. The importance of the cohomological equation in dynamical systems is well known; it arises for example in the study of invariant measures, conjugacy problems, reparametrizations, rigidity questions and inverse problems.
Suppose now that and we try to solve for . There are some obvious obstructions for doing so. Let be the space of all invariant distributions, i.e., is an element of the dual space of such that for every smooth function . Hence, if admits a smooth solution we must have for all . In particular,
[TABLE]
for every -invariant Borel probability measure . Remarkably, in some cases invariant distributions are the only obstructions. For Anosov flows, the smooth version of the Livšic theorem ([17]) says that there exists such that where if and only if has zero integral along every closed orbit of the flow. More recently, L. Flaminio and G. Forni [12] proved the following:
Theorem 1.1**.**
Let be a closed oriented hyperbolic surface and let be the area form. If for all then there exists such that
[TABLE]
Flaminio and Forni proved this theorem for the usual horocycle flow , but it is easy to see that and are conjugated by the map , where is the vector obtained by rotating by following the orientation of the surface. Using representation theory they were able to determine completely the space in terms of the spectrum of the Laplacian of and the genus of . We will return to their results below.
In the present paper we are concerned with the following geometrical aspect of the cohomological equation related to the fibration . Assume there exists such that and suppose that for every , is a polynomial of degree in . What does it imply about ? Must also be a polynomial in ? Of special interest to us is the case , i.e., we suppose that , where and is a smooth 1-form which we view as a function . The case is also of great interest, but with the exception of Proposition 4.3 we will not consider it here. Even for cohomological equations of this special kind appear in problems related to spectral rigidity and regularity of the Anosov splitting [9, 10]. Note that if and is exact, then because .
The literature on this topic is abundant, see [14, 15, 19, 8, 11, 23, 9, 10]. However in all these references the results require either non-positive curvature (and transitivity) or uniform hyperbolicity of . Here we show that these conditions can be weakened to just no conjugate points.
Theorem A. Suppose does not have conjugate points. Let and let be a smooth 1-form on . Then the cohomological equation
[TABLE]
has a solution only if and is closed.
Given we define the vertical subspace at as , where is the canonical projection. We say that the orbit of does not have conjugate points if for all ,
[TABLE]
The energy level is said to have no conjugate points if for all , the orbit of does not have conjugate points. Since magnetic flows are optical, the main result in [5] says that has no conjugate points if and only if the asymptotic Maslov index of the Liouville measure .
Note that if is Anosov, then there are no conjugate points [25, 21, 5] and Theorem A was proved in [10] using a non-negative version of the Livšic theorem [18, 27] and an integral version of the Pestov identity. Here we will use the same integral Pestov identity and show how we can do without hyperbolicity.
Theorem A seems to be new even for the particular case of the geodesic flow of a Riemannian metric ( and is Riemannian). There are several interesting examples of geodesic flows without conjugate which are not Anosov and have regions of positive curvature. The example of W. Ballmann, M. Brin and K. Burns in [1] is of this kind and has the additional feature of having non-continuous Green subbundles (see Section 2). Another interesting class of magnetic flows without conjugate points is given by compact quotients of complex hyperbolic space . If we let be the Kähler structure with holomorphic sectional curvature on , then the magnetic flow of the pair has no conjugate points and is an algebraic unipotent flow. For we obtain the flow in Theorem 1.1. The proof that these magnetic flows have no conjugate points is fairly simple and is given in the appendix, where we also collect other facts about them which are probably well known to experts, but not readily available in the literature.
It is very likely that in Theorem A one can replace “closed” by “exact”, but we do not know how to prove this in general. Let us explain what are the complications and at the same time indicate very general conditions under which we can claim that must be exact.
Let be the space of all -invariant Borel probability measures. Clearly . To any element we can associate its asymptotic cycle by setting
[TABLE]
where and is any closed 1-form in the class . Invariance of ensures that is well defined.
Suppose now that is surjective. Then it is immediate to see that if with closed, then must in fact be exact. We know of no example of a magnetic flow without conjugate points for which is not surjective. We indicate now several conditions that imply the surjectivity of .
If , then is surjective. This follows from the fact that is a compact convex set containing the origin in its interior, which in turn follows from the fact that every non-trivial homotopy class contains a closed geodesic, see [22, Chapter 1].
If is Anosov, the closed orbits are dense and it is not hard to see that is also surjective (cf. [26]).
But one can say more. Let be the universal covering of . Consider the exponential map of the energy level , given by , where , and . It is unknown if the abscence of conjugate points in implies that is a diffeomorphism for all (see [6, p. 907]). But suppose it does and to simplify matters suppose also that is a Riemannian metric. Then, is diffeomorphic to and the lift of to is exact. Write . Then we can associate to the magnetic system a critical value in the sense of Mañé [3, 24]:
[TABLE]
(Note that as ranges over the form ranges over all primitives of , because any two primitives differ by a closed 1-form which must be exact since is simply connected.)
We will say that a magnetic flow is Mañé critical if . If is a diffeomorphism, the same proof of Theorem D in [7] shows that . If , then assuming a technical condition on which seems to hold always, it is known that every non-trivial homotopy class contains a closed magnetic geodesic [24] and again is surjective.
We are left with the question: suppose has no conjugate points and is Mañé critical, is surjective? We mentioned before that compact quotients of complex hyperbolic space with the Kähler structure with holomorphic sectional curvature have no conjugate points. It turns out that they are also Mañé critical (for this is proved in [4, Example 6.2] and for the proof is similar). We do not know other examples of Mañé critical magnetic flows without conjugate points.
For we will check using Flaminio and Forni’s explicit computation of the invariant distributions that is surjective (see Section 4). We suspect that the same is true for any , but finding the invariant distributions and understanding the cohomological equation using representation theory and harmonic analysis seems quite a laborious task. A direct argument using the Pestov integral identity obtained in [10] will allow us to show exactness in this case. More precisely we will show:
Theorem B. Let be a smooth compact quotient of complex hyperbolic space . Let be the Kähler structure with holomorphic sectional curvature and let be the vector field generated by the magnetic flow of the pair . Let and let be a smooth 1-form on . Then the cohomological equation
[TABLE]
has a solution if and only if and is exact.
We note that if Theorem 1.1 extends to compact quotients of for , then Theorem B shows that must be surjective for .
1.1. Convex Hamiltonians
Theorem A above can be used to prove a fairly general result for an arbitrary convex superlinear Hamiltonian ( close and connected). The Hamiltonian is said to be convex if is everywhere positive definite. The Hamiltonian is superlinear if for all ,
[TABLE]
Let be the canonical projection. Given an arbitrary smooth closed 2-form on , we consider endowed with the symplectic structure where is the Liouville 1-form. Given a convex superlinear Hamiltonian we let be the Hamiltonian vector field of with respect to . We denote by the flow of . Let be a regular value of and set .
As before we say that the orbit of does not have conjugate points if for all ,
[TABLE]
where now . The energy level is said to have no conjugate points if for all , the orbit of does not have conjugate points.
Theorem C. Suppose does not have conjugate points and let be a smooth 1-form on . Then the cohomological equation
[TABLE]
has a solution only if is closed.
2. Preliminaries
2.1. Green subbundles
If has no conjugate points, one can construct the so called Green subbundles [6, Proposition A] given by:
[TABLE]
[TABLE]
These subbundles are Lagrangian, they never intersect the vertical subspace and, crucial for us, they are contained in . Moreover, they vary measurably with and they contain the vector field .
Assume now that has no conjugate points and let be one of Green subbundles. Using the splitting
[TABLE]
where is the horizontal subspace, we can represent as the graph of a linear map . The correspondence is measurable and [6, Proposition 1.7].
2.2. Semibasic tensor fields
For the reader’s convenience we recall various definitions and notations from [10]. Henceforth is a closed -dimensional manifold and is a Finsler metric on .
Let be the natural projection, and let denote the bundle of semibasic tensors of degree , where is the bundle of tensors of degree over . Sections of the bundles are called semibasic tensor fields and the space of all smooth sections is denoted by . For such a field , the coordinate representation
[TABLE]
holds in the domain of a standard local coordinate system on associated with a local coordinate system in . Under a change of a local coordinate system, the components of a semibasic tensor field are transformed by the same formula as those of an ordinary tensor field on .
Every “ordinary” tensor field on defines a semibasic tensor field by the rule , so that the space of tensor fields on can be treated as embedded in the space of semibasic tensor fields.
Let be the fundamental tensor,
[TABLE]
and let be the contravariant fundamental tensor,
[TABLE]
In the usual way, the fundamental tensor defines the inner product on , and we put .
Let
[TABLE]
be the spray induced by . Here are the geodesic coefficients [29, (5.7)],
[TABLE]
Let
[TABLE]
be the decomposition of into horizontal and vertical vectors. Here
[TABLE]
with
[TABLE]
and
[TABLE]
Let
[TABLE]
be the Chern connection,
[TABLE]
where
[TABLE]
are the connection forms. Recall that
[TABLE]
Given a function , we put
[TABLE]
and, given a semibasic vector field , put
[TABLE]
We have
[TABLE]
and
[TABLE]
In the usual way, we extend these formulas to higher order tensors:
[TABLE]
and
[TABLE]
We define the operators
[TABLE]
by
[TABLE]
and
[TABLE]
For convenience, we also define and by
[TABLE]
2.3. Modified horizontal derivative for magnetic flows
The form , regarded as an antisymmetric tensor field , gives rise to a corresponding semibasic tensor field. We define the Lorentz force by
[TABLE]
We also define
[TABLE]
Note that is skew symmetric with respect to :
[TABLE]
Straightforward calculations show that
[TABLE]
If , then by (4)
[TABLE]
Suppose that for a smooth function we have
[TABLE]
Extend to a positively homogeneous function (of degree [math]) on , denoting the extension by again.
For , define
[TABLE]
Then on we have
[TABLE]
where is the positively homogeneous extension of to of degree .
Given , put
[TABLE]
Finally, given , define
[TABLE]
2.4. An integral identity
A crucial element in our proofs is the following integral version of the Pestov identity proved in [10]:
[TABLE]
We note the following points:
- (1)
vanishes if and only if is the pull back of a function on ; 2. (2)
and are respectively the Riemann curvature operator and the Landsberg tensor from Finsler geometry; is the Lorentz force associated with the magnetic field; 3. (3)
is the dimension of .
We may regard the identity as a kind of “dynamical Weitzenböck formula”. We will also need the following lemma [10, Lemma 4.4]:
Lemma 2.1**.**
Let be such that , where is independent of while depends linearly on . Then
[TABLE]
In [10] we dealt with the left hand side of (5) assuming that is Anosov. We will show in the next section, how to bypass hyperbolicity just using the Green subbundles and a further calculation to show that in Theorem A must be closed.
3. Proof of Theorem A
Define
[TABLE]
Then the following holds:
[TABLE]
Suppose and extend to a positively homogeneous function of degree zero on (still denoted by ). Then . From (5) and Lemma 2.1 we infer that
[TABLE]
[TABLE]
For each , let be the orthogonal projection onto with respect to the fundamental tensor at .
Theorem 3.1**.**
Assume has no conjugate points and let be a semibasic vector field with for all . Let