Strictly convex ergodic billiards

Wentao Fan and Boris Hasselblatt investigate whether a strictly convex planar billiard can be ergodic. In a mathematical billiard, a particle moves freely inside a table and reflects off its boundary at equal incoming and outgoing angles. Ergodicity means that almost every trajectory explores phase space according to the invariant measure, rather than remaining confined to a smaller invariant region. The authors establish what they describe as the first known strictly convex ergodic billiards: sufficiently asymmetric lemon-shaped tables formed by intersecting two disks. Their boundaries consist of focusing circular arcs that meet at two corners. The result proves a conjecture about lemon billiards whose defining circles have very different radii. The tables are parametrized by the smaller radius \(r\), the larger radius \(R\), and an angle \(\zeta\) describing the intersection geometry. The main theorem applies when \(\zeta\) is a “feasible” parameter, as defined in the authors’ preceding work, between \(0\) and \(\arctan(1/3)\), approximately \(0.322\) radians or \(18.4^\circ\). For each such choice, there is a sufficiently large threshold \(R_{\mathrm{HF}}(r,\zeta)\) such that the billiard map is Bernoulli whenever \(R\) exceeds that threshold. The threshold is at least \(30{,}000r/\sin^2\zeta\). This numerical lower bound comes from restrictions inherited from the earlier hyperbolicity analysis; exceeding it alone does not necessarily satisfy the full threshold requirement. The proof builds on the authors’ earlier establishment of uniform hyperbolicity for a return map to a carefully chosen global section. This section records the next-to-last collision—or the sole collision—in a sequence of visits to the smaller circular arc. Uniform hyperbolicity provides expanding and contracting directions, but it does not, by itself, establish ergodicity. The new work controls singularity curves, which represent trajectories that reach a corner in forward or backward time. These curves can cut stable and unstable manifolds, obstructing the connections needed for a Hopf argument. A central difficulty is that some singularity curves are arch-shaped, have vertical tangencies, and change slope sign. They therefore lack the simple alignment properties found in standard billiard examples. The authors derive quantitative slope bounds and track how these curves transform under subsequent collisions. For incoming arch curves, slopes are below \(-0.495\) or above \(1\). For arches corresponding to at least two consecutive collisions on the larger arc, the negative bound improves to \(-1.49\). A short exceptional segment requires separate treatment. Suitable subsequent collisions align the curves with the invariant cones on the chosen sections. The authors also adapt the Markarian–Del Magno local ergodicity theorem, replacing its Sinai–Chernov condition with a version that accommodates one-sided continuity near singularities. They verify four ingredients: regularity, alignment, unbounded expansion near almost every singularity point, and bounded backward contraction along unstable manifolds. An inductive cutting-and-generating construction shows that every finite-order singularity set consists of finitely many closed arcs whose \(C^2\) regularity extends to their endpoints. These results establish local ergodicity. Finally, connectivity arguments and positive-measure overlaps link the relevant section components. Because the section is global, the conclusion extends to almost all of phase space. The resulting Bernoulli property is stronger than ergodicity. The theorem applies only to a restricted, highly asymmetric parameter regime, not to all lemon billiards. It provides neither quantitative mixing rates nor limit theorems, which remain goals for future research. Crucially, strict convexity here allows corners. Whether a strictly convex ergodic billiard can have a continuously differentiable boundary without corners remains open.

When a Lemon-Shaped Billiard Becomes Ergodic

A particle bouncing around a curved table sounds like a simple mechanical system. Yet predicting its long-term behavior can lead to difficult mathematics. In “Strictly convex ergodic billiards,” Wentao Fan and Boris Hasselblatt establish what they describe as the first known strictly convex ergodic billiards. Their examples are lemon-shaped tables formed by intersecting two disks, with one circle much larger than the other.

The result answers a conjecture about highly asymmetric lemon billiards. It also highlights a crucial distinction: these tables are strictly convex, but their boundaries have corners. Whether a strictly convex billiard with a continuously differentiable boundary can be ergodic remains open.

What does ergodicity mean?

In a mathematical billiard, a particle travels freely inside a table. Whenever it reaches the boundary, it reflects with equal incoming and outgoing angles. The geometry of the boundary determines how successive trajectories separate, reconnect, or remain trapped in organized patterns.

To describe the dynamics, mathematicians use phase space, which records both collision position and direction. Ergodicity means that almost every trajectory explores this space according to the system’s invariant measure, rather than remaining confined to a smaller invariant region of positive measure. It makes a statement about typical long-term behavior, not every individual trajectory.

Fan and Hasselblatt prove something stronger: their billiard map has the Bernoulli property, a powerful form of statistical randomness. The theorem therefore establishes more than ergodicity alone, although it does not provide quantitative mixing rates or limit theorems.

A precise recipe for asymmetry

The tables are described by three parameters: the smaller circle’s radius r, the larger circle’s radius R, and an angle zeta specifying the intersection geometry. Their boundaries consist of two focusing circular arcs meeting at two corners. Strict convexity here excludes straight boundary segments; it does not exclude corners.

The theorem applies when zeta is a “feasible” parameter, in the technical sense defined in the authors’ preceding work, between zero and arctan(1/3). That upper endpoint is approximately 0.322 radians, or 18.4 degrees. For each eligible choice of r and zeta, there is a sufficiently large threshold R_HF(r, zeta). Whenever R exceeds that threshold, the billiard map is Bernoulli.

The threshold is at least 30,000 times r divided by the square of sin(zeta). This striking scale reflects restrictions inherited from the earlier hyperbolicity analysis. It is not a stand-alone recipe: exceeding that numerical lower bound does not necessarily mean the full threshold requirement has been met.

Why hyperbolicity was not enough

The new proof builds on the authors’ earlier establishment of uniform hyperbolicity for a return map on a carefully chosen global section. This section records the next-to-last collision—or the sole collision—in a sequence of visits to the smaller circular arc.

Uniform hyperbolicity supplies expanding and contracting directions. Nearby states separate along one direction and approach each other along another. But those stretching and contracting mechanisms do not, by themselves, prove that almost all trajectories belong to one ergodic component. The proof must also show that the relevant stable and unstable manifolds connect sufficiently well.

Here, corners create the main obstacle. Trajectories that reach a corner, either forward or backward in time, produce singularity curves in phase space. These curves can cut stable and unstable manifolds, disrupting the connections needed for a Hopf argument.

Controlling troublesome arches

Some singularity curves in these lemon billiards are arch-shaped. They have vertical tangencies and change slope sign, so they lack the straightforward alignment found in standard billiard examples. The authors must track their geometry quantitatively rather than simply assume that they point in favorable directions.

They establish slope bounds for incoming arch curves: slopes are below −0.495 or above 1. For arches associated with at least two consecutive collisions on the larger arc, the negative bound improves to −1.49. A short exceptional segment needs separate treatment. By following suitable subsequent collisions, the authors show how these curves become aligned with the invariant cones on the chosen sections.

From local control to a global result

The authors also adapt an existing local ergodicity theorem, replacing one of its conditions with a version that accommodates one-sided continuity near singularities. They verify four essential ingredients:

• Regularity of the singularity sets.
• Alignment with the relevant invariant directions.
• Unbounded expansion near almost every singularity point.
• Bounded backward contraction along unstable manifolds.

An inductive cutting-and-generating construction shows that each finite-order singularity set consists of finitely many closed arcs, with twice continuously differentiable regularity extending to their endpoints. Together, these controls establish local ergodicity.

Connectivity arguments and positive-measure overlaps then link the relevant section components. Because the section is global, the conclusion extends to almost all of phase space, yielding the Bernoulli property.

A breakthrough with clear boundaries

The theorem does not cover every lemon billiard. It establishes a restricted, highly asymmetric family and leaves quantitative statistical questions for future work. Its achievement is nevertheless substantial: strict convexity can coexist with ergodicity, even when every smooth boundary piece is focusing. The remaining challenge is to understand how far this phenomenon extends—and whether corners can eventually be removed.