# Burns–Wilkinson: scoped audit and repair checks

Igor Rivin · Math Referee · 21 November 2026

**Completed scoped audit.** Six independent source-only assessments and two
focused Astra 6.2 challenges have completed, all without truncation. The record distinguishes
checked defects, proposed repairs, conditional imported results, and unchecked
parts. This is an AI-assisted source audit, not a human certification or a
complete verification of every cited theorem. Publication authorized on 21 November 2026.

The target is Burns and Wilkinson, *On the ergodicity of partially hyperbolic
systems*, Annals of Mathematics 171 (2010), 451–489,
[publisher record and paper](https://annals.math.princeton.edu/2010/171-1/p09),
together with the separate June 2005 preliminary note *A note on stable holonomy
between centers*. The journal references the note under a slightly different
title and a 2006 date. Those versions are not silently identified.

## 1. Present conclusion

There is a demonstrably false auxiliary estimate in the supporting note, and
there are local formulation and proof defects in the published Annals paper.
Selected later papers do reuse affected inputs or arguments. This campaign has
**not found a counterexample to the principal ergodicity theorems**. The
available evidence supports a substantially narrower claim than saying that
the main theorems and their descendants are false.

The two regularity branches matter. The C2 branch uses a corrected
Pugh–Shub–Wilkinson input. The C1+delta branch uses the separate BW note and
stronger bunching assumptions. A defect in the latter input does not remove the
former route. Local density arguments are shared and need their own repairs.

It would also be wrong to say that the holonomy problem went unnoticed.
[Wilkinson's publication list](https://www.math.uchicago.edu/~wilkinso/papers/papers.html)
acknowledges errors in the note and directs readers to Brown. Brown and Saghin
provide later repair routes with different stated hypotheses and scope; matching
those routes to the BW construction was a specific audit obligation. The
focused check now supports the Saghin substitution at the original bunching
strength, conditional on that imported result and the graph construction.

## 2. Confirmed local defects

### Supporting-note Lemma 1.1(2)

The iterated estimate fails even for a smooth linear product with constant
center distribution. Let

$$
A=\begin{pmatrix}2&1\\1&1\end{pmatrix},\quad
f=A\times\mathrm{id}_{S^1},\quad
\lambda=(3-\sqrt5)/2.
$$

Use the flat product metric, take
$\mu=\hat\mu=3/10$, $\nu=\hat\nu=2/5$,
$\gamma=\hat\gamma=9/10$, $\theta=1/2$, and regularity exponent
$\delta=3/4$. These satisfy the strict rate and bunching inequalities.
For any proposed lemma radius choose a smaller fixed $\eta>0$. The two
base points $0$ and $\eta\lambda^n e_s$, with identical center directions,
remain within that radius for all backward times from zero through $n$.
The asserted estimate would give

$$
\eta\le\sqrt\eta
\left(\frac{50(\sqrt5-1)}{81}\right)^n\longrightarrow0,
$$

an impossibility. The coordinator checked the argument and exact rational
inequalities; both source-only providers independently found counterexamples
to this estimate. Astra additionally checks that both time directions, including
the lemma's extra small-initial-distance clauses, fail. The [exact-check bundle](/papers/human-burns-wilkinson-ergodicity/files/burns-wilkinson-checks.zip) contains a portable
Python script and its recorded output. It checks the scalar inequalities; the
general geometric argument is the one given here.

This example is not essentially accessible and its actual center holonomies
are smooth. It refutes the auxiliary estimate, **not** the note's regularity
conclusion or the ergodicity theorem. The note's later convergence steps cannot
simply cite the estimate as printed. The two initial reviewers propose different
replacement arguments. Astra has now checked Sol's actual-linear-map argument
for the compact coherent note theorem; the coordinator accepts the local
argument with the scope and premises described in Section 5 below. Fable's
different proposed repair has not had a separate focused adjudication.

### Journal Proposition 2.4, printed p. 462

The definition of a regular family on p. 459 includes nesting and positive
ambient volume. The fiber-ratio hypotheses in Proposition 2.4 do not imply
either condition. Alternating equal disjoint rectangles in a vertical foliation
violates nesting while satisfying the ratios. A shrinking vertical interval
over a singleton transversal base satisfies the inequalities but has zero
ambient volume.

Adding nesting and positive-volume assumptions makes the printed integration
argument prove the intended successive-volume bound. Both initial providers
find that the concrete BW applications supply the needed extra conditions.
Astra checks the BW applications and the corrected abstract integration proof.
For ABW, it identifies a specific one-step nesting condition not present in its
supplied excerpts. The coordinator then checked the additional primary pages:
Proposition 5.4(3)(b), printed p. 950, gives the moving-chart inverse unstable
contraction, while (5.8) and the following display, p. 957, give
$\tau\hat\nu\prec\sigma<1$. Hence the inverse image of the terminal
radius $\tau_{n+1}$ is contained in the preceding radius $\tau_n$,
on the controlled local plaques. Lemma 5.8 on p. 957 supplies the shrinking
working-neighborhood control. These statements discharge the identified
nesting condition on their geometric premises; their entire upstream
nonuniform graph construction has not been re-proved.
This is a local overstatement, not a failure of the underlying integration
estimate.

### Journal Lemma 8.2, printed p. 485

The proof uses same-index julienne membership where Proposition 5.3 provides
fixed index shifts. It also infers a positive lower stable-fiber length from
diameter upper bounds, which do not alone supply that lower bound.

A checked local completion uses stable-coordinate projection in a fixed local
chart. The endpoints remain O(sigma_n) close; tangent continuity makes the
weak-plaque contribution o(sigma_n), while the stable base-boundary
displacement has a positive multiple of sigma_n as its projection. Local
product structure identifies fiber boundaries with base boundaries. The boundary
lower bound gives an inradius, not merely a lower diameter: an escaping path
would cross that boundary. Fixed rank shifts absorb the resulting constants.
This is a new justification,
not a claim that the missing lower estimate was already proved by the displayed
triangle inequalities. Astra and the coordinator accept this local completion
for BW on its uniform plaque premises. It also works for ABW in one fixed
time-zero adapted chart, without asserting a uniform comparison of all future
adapted metrics. The HT transfer remains conditional on its precise history
product and base-inclusion inputs.

## 3. Additional findings from the independent panel

Both providers identify a metric convention needed for the weak cs/cu bounds
in Proposition 3.1. Norm estimates on Es, Ec and Eu separately do not imply the
same bounds on their sums for a strongly nonorthogonal metric. This can be
exhibited by the same product map using nearly parallel stable and center unit
vectors in the chosen inner product. An orthogonal product metric on the
invariant splitting, followed by a sufficiently close smooth approximation,
is a remedy using the strict compact margins. Astra confirms this normalization
without changing the bunching hypotheses, smooth measure class or density
points. The overstrong metric-dependent clauses must not be used for every
metric satisfying only the three separate summand bounds.

The fake-Jacobian application of Lemma 4.4 also needs an explicit orbit-indexed
version: a fake distribution depends on its orbit chart and need not define a
single function on the original manifold. Uniform Holder bounds on the chart
functions give the same summable log-distortion calculation. This is distinct
from genuine unstable Jacobians, which do define one function on the manifold.

In Lemma 6.2, genuine stable holonomy cannot be treated as membership in the
same fake center-stable leaf. The checked local completion constructs the terminal
local-product intersection and pulls it back, checking the working domains and
rank shifts. Other recorded discrepancies are orbit subscripts, cocycle base
points, and endpoint strict inequalities. A typographical discrepancy is not
counted as an independent failure of a main theorem.

Several panel items are **coverage limits**, not established defects. Examples
include unprovided graph-transform proofs, full hypotheses of classical
absolute-continuity results, and the precise Pinsker input. The initial density
packet itself mistakenly requested Lemmas 8.4 and 8.5; Section 8 ends at
Lemma 8.3. That was our target-inventory error and has been corrected.

## 4. Selected downstream dependencies

This is a check of the following mathematical uses, not an exhaustive citation
graph or a complete audit of any later paper.

| Later source | Source-supported dependence | Scope that must be preserved |
| --- | --- | --- |
| Rodriguez Hertz–Rodriguez Hertz–Ures (2008), Proposition B.7, printed p. 374 | Cites the BW holonomy note for a lower-regularity strong-subfoliation assertion | Its main Theorem B is C2, and the next paragraph gives a separate Pugh–Shub route for C2 with one-dimensional center. |
| Avila–Bochi–Wilkinson (2009), Proposition 5.2(1), p. 946 | Repeats the fiber-ratio regularity criterion, explicitly citing BW Section 2.3 | Check nesting and positive measure in the actual applications; its analytical Theorem B is C2. |
| Avila–Bochi–Wilkinson, Lemma 5.21, pp. 965–966 | Reuses the same-index membership and lower stable-length argument | A fixed-chart repair must work with its adapted metric; C1 in the genericity title is not the differentiability hypothesis of this analytical theorem. |
| Avila–Santamaria–Viana (2013), Section 2.1, p. 16 | Explicitly invokes BW Theorem 0.3 for the blanket strongly bunched C1+alpha extension | Keep that extension separate from its C2 statements. |
| Hammerlindl–Tyler (2025), Proposition 3.1(vii), pp. 14–15 | Imports both regularity clauses through the BW construction | Its main Theorem 1.1 is C2; the additional lower-regularity clause is not the only route to it. |
| Hammerlindl–Tyler, Lemma 9.4, p. 28, Appendix B, pp. 32–33 | Adapts Lemma 8.2 and repeats the unsupported lower-length inference | Unlike BW's paragraph it does invoke a holonomy-control result. The repair must respect projection from the orbit space and local fiber geometry. |

No conclusion about the authors' awareness or the falsity of the later main
theorems follows from these edges. Other descendants mentioned in the owner's
report remain outside the source checks completed here.

## 5. Historical proof and modern repair

The historical external-input cutoff is the journal's received date,
11 October 2005. The June 2005 preliminary note is before that cutoff. The
bibliography's 2006 label does not establish that no earlier note existed.
Brown and Saghin are later work and must remain labeled modern replacements.

Brown's retrieved manuscript expressly uses stronger bunching. Its arXiv side
stamp and internal date differ; this record preserves the actual frozen file
without claiming all of its content existed in 2016. Saghin's 2023 preprint
contains uniform C1 fake holonomy as Corollary 1.1 and explains the noncompact
uniform setting in Section 3.7. The corresponding journal text labels the
result Corollary 1.5. We have the complete preprint PDF and page images; the
published PDF could not be downloaded locally and was read only as web text.

### Modern replacement of the fake-holonomy input

The focused check supports the substitution at the original BW bunching
strength, without Brown's extra numerical assumptions. Its premises are
Saghin's uniform fake-holonomy and nearby-pair derivative estimates, and BW's
coherent uniform graph construction **before** Proposition 3.1(vii). It does
not assume the strong C1 subfoliation that it is meant to supply.

The rate dictionary is
$\lambda_s=\nu$, $\lambda_u=\hat\nu^{-1}$,
$\lambda_c^-=\gamma$, $\lambda_c^+=\hat\gamma^{-1}$.
The stable and unstable theta-bunching tests are exactly BW's two tests.
Its auxiliary rate inequalities force $0<\theta<1$; $\theta<\delta$
then gives C1+theta regularity. This avoids an unstated change of exponent.

Fix one cutoff extension of the fiber maps and one associated graph family.
The nonlinear error vanishes to first order at zero. Compactness and the strict
margins give uniform derivative Holder, inverse derivative, cone and bunching
bounds. To check a holonomy away from the zero section, translate the same
global fiber dynamics along that point's full orbit. This retains all those
bounds and the same graph family; constructing unrelated new fake leaves at
each point would not suffice. This is the uniform noncompact setting described
in Saghin Section 3.7.

Uniform C1 holonomy in both directions gives upper and lower derivative bounds.
For joint continuity, more than a bound on the derivative norm is needed.
Saghin's pp. 13–14 give a short-leg derivative estimate relative to projection
between the two center spaces; p. 16 supplies the uniform-fiber setting. By
equivariance, transporting those projections from a contracting end of the
orbit approximates the holonomy derivative with a uniformly geometric tail.
The finite products are continuous; hence so is their uniform limit. The
coordinator inspected these primary page images as well as the application
argument.

Factor holonomy on a general controlled transversal through center disks at
its two crossing points. The crossing maps have derivatives given by linear
projection along the continuous strong tangent field. Combining these with
the center-to-center derivative gives continuous, uniformly invertible
transverse derivatives. In local weak-leaf graph coordinates, the strong
coordinate supplies the other derivative of the foliation box. This yields
uniform C1 strong-foliation boxes inside the weak leaf and two-sided volume
comparison by change of variables and Fubini. These are the inputs needed for
center-radius control and fake unstable disintegration in the BW density proof.

This closes the **application check** on the stated imported premises. It does
not independently reprove every part of Saghin or the graph-transform theory,
prove genuine strong-foliation absolute continuity, or settle the Pinsker
argument. It is a modern replacement, not a retrospective validation of the
June 2005 proof or a blanket audit of every descendant.

### A separate new proof for the compact coherent note theorem

Sol proposed an independent construction, and Astra checked it without using
Brown, Saghin or Proposition 3.1(vii). It uses the note's compact coherent
geometry and its theta-Holder center bundle. The coordinator checked the
linear-map telescoping identity, summability, branch selection and scope.
This gives a locally checked replacement argument, not a claim of mathematical
priority or human certification.

Buffer the rates so that $\nu<a<\min(1,g)$, $g<\gamma$,
$\rho=\sup a/g<1$, and $b=\sup a^\theta/(g\hat\gamma)<1$.
For a short stable leg from x to y=h(x), mark at time n the target center sheet
through f^n(y). Intersect it with a short smooth transverse disk through f^n(x).
The marked sheet resolves recurrence ambiguity without differentiating h.
The local intersection maps are C1. Pulling back their center errors gives
$O(\ell a_{0:i}\rho^{n-i})$ at time i; a first-exit argument keeps the
arcs in the tubes where the buffered conorm applies.

Write $A_i$ and $B_i$ for center derivatives on the source orbit and
the approximating orbit, and $J_i$ for smooth transport followed by center
projection. With forward product convention, put

$$
R_i=B_i^{-1}J_{i+1}A_i-J_i.
$$

If $P_n$ is the terminal intersection map, the chain rule and exact
telescoping give

$$
B_{0:n}^{-1}DP_nA_{0:n}-J_0
=\sum_{i=0}^{n-1}B_{0:i}^{-1}R_iA_{0:i}
+B_{0:n}^{-1}(DP_n-J_n)A_{0:n}.
$$

The terms are bounded by $C\ell^\theta b^i$ and the final error by
$C\ell^\theta b^n$. A summable bound alone would not establish convergence
because the summands depend on n: for every fixed i the approximating orbit
also converges uniformly to the true y-orbit. Truncate the series and then use
its uniform geometric tail. Thus the actual derivatives converge uniformly,
as do the maps, proving C1 holonomy. Short-leg closeness to projection supplies
invertibility. This avoids the gap between convergence of projective directions
and convergence of actual derivatives.

The finite transported derivative products have Holder constants at most
$CD^m$, while their tails are $O(b^m)$. Balancing these bounds gives a
positive Holder exponent, for example
$\theta\log(1/b)/(\log D+\log(1/b))$. No original explicit exponent is
claimed. Longer fixed local legs are covered by finite compositions; there is
no uniform bound asserted for unbounded holonomy length. This compact coherent
argument does not itself establish the separate uniform fake-model transfer.

Two classical sources were obtained after the paid requests were frozen and
are coordinator-only supplements. Brin–Pesin Proposition 5.1 is indeed on
printed p. 203 and is introduced under a smooth invariant measure. Its proof
cites Sinai's partition theorems. The extension to all the measure and rate
conventions used in BW remains a named external obligation here, not a new
counterexample. Pugh–Shub Theorem 2.1, pp. 4–5, gives measurewise C1 strong
foliations with a jointly continuous positive Jacobian. In one transversal
dimension that conclusion yields C1 holonomy, but applying it on the relevant
weak leaves still requires its geometric hypotheses. Full independent proofs
of these classical inputs have not been audited.

## 6. Provenance, cost and remaining work

The owner's consolidated report is preserved but was withheld from the six
initial requests. Those requests used primary journal page images, the complete
October 2006 text as labeled context, the separate holonomy note, and the PSW
correction as appropriate. Sol 6.2 high and Fable 5.2 high each reviewed geometry,
density, and the note. This is a two-provider independent panel, not a
three-provider panel. Later papers and repair literature entered only the
focused challenges. Fable's replies arrived after those challenges were frozen.

All eight assessments are structured and uncapped. Conservative usage totals
**USD 9.516245**; the uncached batch-tier estimate is **USD 7.766056**. The initial
six cost USD 5.799921 under the conservative accounting, and the two focused
checks USD 3.716324. The child allowance was delegated from the original follow-up
hold, not additional to it. The campaign ceiling remains **USD 25**. These figures
are token-based accounting, not invoices. No inference request remains pending
and no further calls are needed to complete this scoped record.

The remaining mathematical limits are explicit: complete imported proofs of
graph transforms and strong-foliation measure results; the full Pinsker
application; historical note-version collation; HHU's weak-leaf use of its
alternate theorem; the full ASV extension; ABW's upstream nonuniform
construction; and HT's history-product/base-inclusion and measure arguments.
They are not additional demonstrated false statements. Neither a complete proof of either principal theorem nor an exhaustive
downstream audit is claimed.

An invalid auxiliary estimate does not refute its consumer theorem.

The neutral reading map indexes 37 numbered statements, 10 sections and 32
bibliography entries. Its 83 nodes include separate regularity branches and
imported inputs. The 32 source-attributed edges cover the selected argument,
not every mathematical dependence. The panel decision graph, repair records,
reviewer responses and cost/coverage record remain separate from that reading map.

## 7. Sources and downloads

The downloadable main paper is the complete October 2006 author edition
(50 pages). The audit's printed page references and primary proof images are
from the published Annals edition (2010, pp. 451–489). The extracted source
index combines that journal text with the separate June 2005 note. It is a
quotation/indexing aid, not native TeX; mathematical page images are authoritative.
The differing title/date in the journal's reference to the note is recorded,
without asserting that every historical version has been collated.

- [Main paper: author edition](/papers/human-burns-wilkinson-ergodicity/files/paper.pdf).
- [June 2005 supporting note](/papers/human-burns-wilkinson-ergodicity/files/holonomy-note-2005.pdf).
- [Source editions and hashes](/papers/human-burns-wilkinson-ergodicity/files/source-edition.json).
- [Download this report](/papers/human-burns-wilkinson-ergodicity/files/burns-wilkinson-scoped-review.pdf).
- [Exact-check bundle](/papers/human-burns-wilkinson-ergodicity/files/burns-wilkinson-checks.zip).
- [Coverage and costs](/papers/human-burns-wilkinson-ergodicity/files/coverage-and-cost.json).
- [Repair dispositions](/papers/human-burns-wilkinson-ergodicity/files/repair-dispositions.json).

Later-source checks used the following primary texts:
[HHU (2008)](https://www.fing.edu.uy/~jana/www/publications_files/2008inventiones.pdf),
[ABW (2009)](https://numdam.org/item/10.24033/asens.2113.pdf),
[ASV (2013)](https://numdam.org/item/AST_2013__358__13_0.pdf), and
[Hammerlindl–Tyler (2025)](https://arxiv.org/abs/2505.05620).
The repair comparison used
[Brown's retrieved manuscript](https://arxiv.org/abs/1608.05886) and
[Saghin's 2023 preprint](https://arxiv.org/abs/2307.01321).
The full source record distinguishes the versions actually inspected from
published metadata and retains their hashes.
