Burns–Wilkinson: tracing a proof gap and its repairs

A false supporting estimate, local proof gaps and modern repair routes. A scoped audit traces selected downstream uses without claiming the principal ergodicity theorems are false.

Igor Rivin

How much of a theorem survives when a supporting estimate is false?

Keith Burns and Amie Wilkinson's On the ergodicity of partially hyperbolic systems is a useful case study. Our scoped audit identifies an explicit counterexample to an estimate in a companion note, local defects in the published argument, and selected later uses of those inputs. It also examines repairs. We have not found a counterexample to the principal ergodicity theorems.

The correction history belongs at the beginning of this story. Wilkinson's publication list already acknowledges errors in the companion note and directs readers to work of Aaron Brown. Our contribution here is a source-based account of what fails, which assumptions the available repairs require, and how specific proof steps are used later. We make no claim that the holonomy problem went unnoticed.

The theorem and the geometric ingredient

Ergodicity says that an invariant measurable set has either zero or full measure. For a volume-preserving dynamical system, it rules out a decomposition into two invariant regions of positive volume.

A partially hyperbolic system has directions that contract, directions that expand, and intermediate center directions. The Burns–Wilkinson theorem connects ergodicity to accessibility along stable and unstable leaves, together with inequalities controlling how these directions interact. The paper's $C^2$ and $C^{1+\delta}$ versions use different regularity inputs.

One such input concerns holonomy: slide a point along a stable leaf from one center surface to another. To compare volumes and density points, the proof needs control of this sliding map. The June 2005 supporting note attempts to establish that control under lower differentiability assumptions.

A false estimate in a smooth example

The problem already appears for a linear map of the two-dimensional torus, with an untouched circle as its center direction:

$$A=\begin{pmatrix}2&1\\1&1\end{pmatrix},\qquad f=A\times\mathrm{id}_{S^1}.$$

Its contracting eigenvalue is

$$\lambda=\frac{3-\sqrt5}{2}.$$

Choose two points separated by $\eta\lambda^n$ in that contracting direction. Under $n$ backward iterates their separation becomes $\eta$. All intermediate separations can be kept inside any prescribed small neighborhood.

With admissible constant rate bounds, the note's Lemma 1.1 would imply

$$\eta\le\sqrt\eta \left(\frac{50(\sqrt5-1)}{81}\right)^n.$$

The factor in parentheses is less than one. For a fixed positive $\eta$, the right side tends to zero. The estimate is therefore false. The report verifies the rate assumptions explicitly and supplies a small exact-arithmetic script.

There is an instructive distinction here: the actual holonomies in this example are smooth. The system also fails the essential-accessibility hypothesis of the ergodicity theorem. The example refutes the auxiliary estimate, not the regularity conclusion it was intended to prove, and not the main ergodicity theorem.

Two local gaps, two different repairs

The published Annals paper contains a regularity criterion, Proposition 2.4, that needs additional assumptions. Its volume ratios do not by themselves make a family nested or give it positive ambient volume. Equal disjoint rectangles can alternate while preserving those ratios; a family over a zero-measure transversal base can have positive fibers and zero ambient volume.

Adding the missing conditions makes the integration argument work. The applications we checked supply those conditions. This is a useful correction to an abstract statement, not a failure of the underlying integration method.

Lemma 8.2 has a different issue. Upper bounds on distances do not establish the lower bound on a stable fiber that its proof needs. A repair projects the geometry onto the stable coordinate in one fixed chart. The base boundary has displacement proportional to the scale, while the competing weak-direction contribution is smaller. Local product structure then turns the boundary estimate into a genuine inner-radius bound. The report states the plaque and scale assumptions used in this completion.

The repair must match the hypotheses

Brown's retrieved manuscript uses stronger bunching assumptions. That gives a repair route, but it cannot simply be substituted into every original statement without comparing the assumptions.

Radu Saghin's later work supplies a different route through uniform holonomy estimates. Our focused application check supports using that result at the original Burns–Wilkinson numerical bunching strength. It requires the same uniform fake-foliation construction throughout, control of inverse derivatives, and a passage from holonomy estimates to the volume comparisons used by the density argument. Those steps are recorded in the report, along with the imported results on which they depend.

The panel also produced a separate replacement argument for the compact, dynamically coherent setting of the note. It works with actual derivative operators and a convergent telescoping sum. That proof has its own scope; it does not automatically settle the uniform fake-foliation setting.

These are modern and newly checked replacements. They do not make an invalid step in the original proof valid retroactively.

What carries forward to later papers?

We traced selected uses in work of Rodriguez Hertz–Rodriguez Hertz–Ures, Avila–Bochi–Wilkinson, Avila–Santamaria–Viana, and Hammerlindl–Tyler. Some cite the lower-regularity input; others repeat local density arguments. Those are different kinds of dependence.

The differentiability assumptions matter. Several later main results are $C^2$, even when a nearby discussion includes a lower-regularity extension. A defect in that extension does not automatically invalidate the $C^2$ result. Likewise, a repair in a fixed manifold chart must be checked again before using it in an orbit-space construction.

The Hammerlindl–Tyler transfer is a useful next investigation. Its full history-product, base-inclusion and measure arguments remain outside the completed checks. The present report documents the relevant edges without claiming that the later theorems are false.

Explore the evidence

Six source-only assessments from Sol and Fable were followed by two focused Astra checks. All eight returned complete assessments. The public record separates those model responses from the source-adjudicated conclusions and lists the remaining unchecked inputs.

The aim is to identify precisely what must change, what can be retained, and what still needs to be checked.

Cite this spotlight

If this spotlight informs your work, please cite it.

Igor Rivin. Burns–Wilkinson: tracing a proof gap and its repairs. Math Referee, .

Download .bib
View BibTeX
@misc{rivin2026spotlight-burns-wilkinson-proof-gap-and-repairs,
  author = {Rivin, Igor},
  title = {{Burns--Wilkinson: tracing a proof gap and its repairs}},
  year = {2026},
  month = {November},
  date = {2026-11-21},
  howpublished = {Math Referee},
  url = {https://mathreferee.dimensionreducers.ai/spotlight/burns-wilkinson-proof-gap-and-repairs/},
  note = {Paper Spotlight}
}

Take a closer look at your paper.

Explore the public reports, then join the waitlist for our invitation-only research preview. Joining is free; reviews are paid.

Join the preview waitlist