Torus Extensions of Anosov Flows

• Torus extensions of Anosov flows are specialized constructions that use toral suspensions and fiberwise bundles to model hyperbolic dynamics and rigidity phenomena.
• They employ techniques like Dehn twists, arithmetic criteria, and surgery methods to analyze topological equivalence, commensurability, and smooth rigidity.
• These models extend to higher dimensions, providing frameworks to study invariant tori, symplectic structures, and complex dynamical behaviors.

“Torus extensions of Anosov flows” does not designate a single standardized construction across the literature. In the most precise usages represented here, it refers to several closely related toral mechanisms: suspensions of hyperbolic automorphisms of $\mathbb T^2$, constant-roof suspension flows on mapping tori of Anosov diffeomorphisms of $\mathbb T^2$, fiberwise Anosov flows on affine torus bundles over lower-dimensional Anosov flows, and torus-boundary cut-and-paste constructions in dimension three. These toral models serve both as canonical examples of Anosov dynamics and as organizing objects for rigidity, surgery, commensurability, and symplectic constructions (Dehornoy, 2013, Barthelmé et al., 2017, Gogolev et al., 2022).

1. Toral suspension models

One of the two “main examples” of $3$-dimensional Anosov flows is the vertical flow on the suspension of a hyperbolic automorphism of the torus. For $A\in SL(2,\mathbb Z)$ with $\operatorname{tr}(A)>2$, the suspension manifold is

$M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$

and the flow is induced by the vector field $\partial/\partial t$ in the interval direction. The stable and unstable directions come from the stable and unstable eigendirections of $A$, transported along the suspension direction, which is why the resulting flow is Anosov (Dehornoy, 2013).

This toral suspension model reappears in several classification results. In the solvable case, a pseudo-Anosov flow in a three-manifold with virtually solvable fundamental group has no singularities and is topologically equivalent to a suspension Anosov flow. In that setting the ambient manifold is, up to finite cover, a torus bundle over $S^1$, so the only genuine pseudo-Anosov dynamics is the suspension model (Barbot et al., 2010).

The same toral construction also underlies the exceptional family in smooth rigidity results for $3$-dimensional volume-preserving Anosov flows. The exceptional case is when the manifold is a mapping torus of an Anosov automorphism of $\mathbb T^2$0 and both flows are constant roof suspension flows; outside that family, $\mathbb T^2$1 conjugacy forces smooth conjugacy (Gogolev et al., 2022).

2. Equivalence, commensurability, and rigidity

The toral suspension family is rigid under topological equivalence but becomes much more flexible under weaker equivalence relations. Two flows are topologically commensurable if they admit finite coverings by topologically equivalent flows. They are topologically almost equivalent if, after removing finitely many periodic orbits from each, the restricted flows are topologically equivalent. They are topologically almost commensurable if, after removing finitely many periodic orbits, the restricted flows are topologically commensurable. Within this framework, all suspensions of automorphisms of the $\mathbb T^2$2-torus and all geodesic flows on unit tangent bundles to hyperbolic $\mathbb T^2$3-orbifolds are pairwise topologically almost commensurable (Dehornoy, 2013).

For toral suspensions themselves, commensurability admits an explicit arithmetic criterion. If $\mathbb T^2$4 and $\mathbb T^2$5 are hyperbolic matrices, then the suspension flows on $\mathbb T^2$6 and $\mathbb T^2$7 are topologically commensurable if and only if there exist positive integers $\mathbb T^2$8 such that

$\mathbb T^2$9

This shows that commensurability is weaker than conjugacy of monodromy, but still nontrivial. The additional allowance of deleting finitely many periodic orbits is what permits toral suspensions to join the same almost-commensurability class as hyperbolic-orbifold geodesic flows (Dehornoy, 2013).

Rigidity results place an opposite emphasis on the same toral family. For $3$0-dimensional $3$1, $3$2, volume-preserving Anosov flows, a $3$3 conjugacy is automatically a $3$4 diffeomorphism unless both flows are constant roof suspensions of Anosov diffeomorphisms of $3$5. In the toral suspension case, periods alone do not determine smooth conjugacy; periodic eigenvalue data remain necessary (Gogolev et al., 2022).

3. Fiberwise torus bundles and higher-dimensional extensions

A more literal notion of torus extension appears in the study of affine torus bundles

$3$6

and flows $3$7 that fiber over a base flow $3$8. Such a flow is fiberwise Anosov if

$3$9

and the vertical bundle $A\in SL(2,\mathbb Z)$0 admits a $A\in SL(2,\mathbb Z)$1-invariant splitting

$A\in SL(2,\mathbb Z)$2

with uniform contraction on $A\in SL(2,\mathbb Z)$3 and uniform expansion on $A\in SL(2,\mathbb Z)$4. When the base flow is Anosov, the total flow is automatically Anosov (Barthelmé et al., 2017).

For a closed $A\in SL(2,\mathbb Z)$5-dimensional base manifold carrying an Anosov flow, these torus extensions satisfy a sharp dichotomy. If

$A\in SL(2,\mathbb Z)$6

is an affine torus bundle and $A\in SL(2,\mathbb Z)$7 is a fiberwise Anosov flow over $A\in SL(2,\mathbb Z)$8, then either $A\in SL(2,\mathbb Z)$9 is topologically orbit equivalent to a suspension flow of an Anosov automorphism of $\operatorname{tr}(A)>2$0 and $\operatorname{tr}(A)>2$1 is also a suspension of an Anosov diffeomorphism, or the stable and unstable dimensions of $\operatorname{tr}(A)>2$2 satisfy

$\operatorname{tr}(A)>2$3

The mechanism behind this dichotomy is a theorem saying that if the base has periodic orbits freely homotopic to inverses of each other, then $\operatorname{tr}(A)>2$4 must be even, $\operatorname{tr}(A)>2$5, and the fiberwise stable and unstable distributions have equal dimensions (Barthelmé et al., 2017).

This framework has both suspension and non-suspension examples. A suspension of $\operatorname{tr}(A)>2$6, where $\operatorname{tr}(A)>2$7 and $\operatorname{tr}(A)>2$8 are hyperbolic toral automorphisms, fibers over the suspension of $\operatorname{tr}(A)>2$9 and realizes the suspension branch. By contrast, Tomter’s mixed-type algebraic Anosov flow gives a fiberwise affine Anosov flow on an affine $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$0-bundle over a hyperbolic surface geodesic flow, with

$M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$1

showing that the balanced branch is nonempty (Barthelmé et al., 2017).

4. Invariant tori, toroidal manifolds, and ambient topology

The existence of torus-based hyperbolic dynamics imposes strong constraints on the ambient $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$2-manifold. An embedded $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$3-torus $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$4 is called an Anosov torus if there exists a diffeomorphism $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$5 such that $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$6 and the induced action on $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$7 is hyperbolic. A closed orientable irreducible $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$8-manifold admits an Anosov torus if and only if it is one of three types: the $M_A:=\mathbb T^2\times[0,1]/_{(x,1)\sim (Ax,0)},$9-torus, the mapping torus of $\partial/\partial t$0, or the mapping torus of a hyperbolic automorphism of the $\partial/\partial t$1-torus (Hertz et al., 2010).

This classification is not itself a classification of Anosov flows, but it identifies the ambient topological types in which toral hyperbolic behavior can occur as an invariant torus. In particular, the mapping torus of a hyperbolic toral automorphism is the canonical torus-bundle ambient space for a suspension Anosov flow. The same paper proves a boundary version: if a compact orientable irreducible $\partial/\partial t$2-manifold with incompressible torus boundary admits an Anosov torus, then it is $\partial/\partial t$3 (Hertz et al., 2010).

Toroidal $\partial/\partial t$4-manifolds also enter the broader pseudo-Anosov theory. In a Seifert fibered manifold, a pseudo-Anosov flow is, up to finite covers, topologically equivalent to a geodesic flow. In a solv manifold, it is topologically equivalent to a suspension Anosov flow. For periodic Seifert pieces inside a JSJ decomposition, the flow has a standard form built from finitely many Birkhoff annuli. This gives a torus-decomposition perspective in which suspension flows and geodesic flows appear as the rigid toroidal prototypes (Barbot et al., 2010).

5. Transverse tori, Dehn twists, and symplectic realizations

A transverse torus to an Anosov flow carries two distinguished one-dimensional foliations, induced by the weak stable and weak unstable bundles. For two transverse $\partial/\partial t$5 foliations $\partial/\partial t$6 on $\partial/\partial t$7, there exists a loop

$\partial/\partial t$8

with $\partial/\partial t$9, such that $A$0 remains transverse to $A$1 for all $A$2, and the point-tracks $A$3 are non-null homotopic. More precisely, if $A$4 and $A$5 have no parallel compact leaves, every class in $A$6 occurs; if they have parallel compact leaves in class $A$7, the admissible classes are exactly the cyclic subgroup $A$8. Applied to a non-transitive Anosov flow on an oriented closed $A$9-manifold, this yields a general Dehn-twist construction: for any finite family of transverse tori with no return and any admissible torus directions, the composition of sufficiently large time map with the corresponding Dehn twists is an absolutely partially hyperbolic diffeomorphism, robustly dynamically coherent and plaque expansive (Bonatti et al., 2016).

In Mitsumatsu’s Liouville construction, a transverse torus $S^1$0 can also be asked to produce a Lagrangian in $S^1$1. Ruscelli defines $S^1$2 to be pre-Lagrangian if, for some supporting Anosov-Liouville pair $S^1$3 and some function $S^1$4, the graph

$S^1$5

is Lagrangian in $S^1$6. A necessary condition is that the maps

$S^1$7

associated to the induced weak-stable and weak-unstable foliations be trivial. A sufficient condition is that these two foliations have no parallel compact leaves. Suspension fibers satisfy the positive criterion, while the distinguished transverse torus in the Franks–Williams example violates the obstruction and is not pre-Lagrangian (Ruscelli, 24 Aug 2025).

The toral suspension model also has a distinctive symplectic package. For the Anosov Liouville domain associated to the suspension of a linear Anosov diffeomorphism on the torus, there are no closed exact Lagrangian submanifolds that are orientable, projective planes, or Klein bottles. At the same time, the domain contains weakly exact Lagrangian torus fibers, and the exact Lagrangian cylinders over periodic orbits form the object set of the orbit category inside the wrapped Fukaya category. This sharply separates the torus-suspension domain from the hyperbolic geodesic-flow case, where exact Lagrangian tori do exist (Cieliebak et al., 2022).

6. Surgery, gluing, and toroidal proliferation

Starting from the suspension flow $S^1$8 of a hyperbolic toral automorphism $S^1$9, one may perform Dehn–Goodman–Fried surgeries along periodic orbits and study the resulting class

$3$0

If all surgery coefficients have the same sign, the resulting flow is $3$1-covered and twisted according to that sign. The mixed-sign case is subtler. Given any flow $3$2, there exists $3$3 such that every nontrivial surgery along an $3$4-dense periodic orbit produces an $3$5-covered twisted flow. Conversely, there exist periodic orbits $3$6 such that every flow obtained by surgeries with distinct signs on $3$7 and $3$8 is non-$3$9-covered. In this way torus suspensions generate both twisted $\mathbb T^2$00-covered and branching non-$\mathbb T^2$01-covered descendants, often on hyperbolic $\mathbb T^2$02-manifolds (Bonatti et al., 2020).

A newer toroidal construction uses bicontact plugs with quasi-transverse torus boundary. If two strongly adapted bicontact plugs have compatible quasi-transverse periodic boundary tori, then after perturbation near the boundary there is a gluing diffeomorphism $\mathbb T^2$03 such that the glued manifold again carries a strongly adapted bicontact structure; if the resulting manifold is closed, it carries an Anosov flow. This framework produces closed toroidal manifolds carrying many pairwise non-orbit-equivalent transitive Anosov flows, including manifolds obtained by gluing two copies of the figure-eight knot complement and manifolds with one hyperbolic and one Seifert fibered JSJ piece (Pinsky et al., 23 Mar 2026).

Surface-bundle constructions provide a different large-scale analogue. For every $\mathbb T^2$04, there exists a finite index subgroup

$\mathbb T^2$05

such that every element of $\mathbb T^2$06 has a representative $\mathbb T^2$07 for which the mapping torus

$\mathbb T^2$08

carries a transitive Anosov flow. For almost every element of $\mathbb T^2$09, $\mathbb T^2$10 is hyperbolic. This is not a torus extension in the narrow sense, but it extends the toroidal surgery viewpoint to a large class of fibered hyperbolic $\mathbb T^2$11-manifolds (Béguin et al., 6 Mar 2026).

7. Higher-dimensional analogues and open directions

In higher-dimensional holomorphic rigidity, torus-based models persist in a more algebraic form. A topologically transitive transversely holomorphic Anosov flow on a smooth compact $\mathbb T^2$12-manifold is either $\mathbb T^2$13-orbit equivalent to the suspension of a hyperbolic automorphism of a complex torus, or, up to finite covers, $\mathbb T^2$14-orbit equivalent to the geodesic flow of a compact hyperbolic manifold. Here the toral branch is a suspension over a complex $\mathbb T^2$15-torus rather than over $\mathbb T^2$16 (Abouanass, 10 May 2025).

A related $\mathbb T^2$17-dimensional theorem concerns transversely holomorphic partially hyperbolic flows with compact subcenter foliation and trivial holonomy. In that setting there exists a smooth compact connected $\mathbb T^2$18-manifold $\mathbb T^2$19, a smooth fiber bundle map

$\mathbb T^2$20

and a smooth transversely holomorphic Anosov flow $\mathbb T^2$21 on $\mathbb T^2$22 such that

$\mathbb T^2$23

If the quotient flow is topologically transitive, it satisfies the same dichotomy as in dimension five: complex-torus suspension or, up to finite covers, hyperbolic geodesic flow. This gives a higher-dimensional bundle-extension analogue of the toral suspension paradigm (Abouanass, 29 Jan 2026).

The broadest unresolved question in the background remains whether toral models are universal after suitable weakening of equivalence. Fried asked whether every transitive Anosov flow admits a genus-one Birkhoff section. A positive answer would imply that every transitive Anosov flow is topologically almost equivalent to the suspension of some automorphism of the torus. In the same circle of ideas, Ghys conjectured that any two transitive Anosov flows are topologically almost commensurable. What is proved at present is the toral–geodesic statement for suspensions of automorphisms of $\mathbb T^2$24 and geodesic flows on hyperbolic $\mathbb T^2$25-orbifolds, not a classification of all Anosov flows by torus suspensions (Dehornoy, 2013).