gpt-6.2-sol high: bw-holonomy-note
Reviewer record, not the publication verdict. The response includes unaccepted proposals and source limits. Missing antecedent proofs are not automatically defects in the paper. The initial request for Lemmas 8.4–8.5 was our inventory error; the journal ends at Lemma 8.3. The curated report records final dispositions.
Completed visible assessment, preserved as a readable check record. It is a model assessment, not a formal certificate. This document contains no private reasoning stream.
The reproducible obstruction is to the 2005 note's auxiliary Lemma 1.1 in both time directions. Its failure invalidates the printed projective iteration; the derivative Cauchy formula has a separate defect. These findings do not refute the local holonomy theorem or either principal ergodicity theorem. A separate linear comparison proves the compact coherent local holonomy result, but the journal's uniform fake-foliation interface and its cited historical inputs remain conditional. The historical cutoff is October 11, 2005, the journal received date printed on p. 489. The dated preliminary note contains the needed statement before that date on its face; the later bibliographic entry [BW06b] cannot by itself establish which proof was available by the cutoff. No post-cutoff result was used to close an obligation. Model training cannot exclude later knowledge; the audit relies on supplied dated evidence and the explicit calculations above, not on external inspection.
H1
Checked statement: The June 11, 2005 note's Theorem 0.2: a C^{1+δ} partially hyperbolic, dynamically coherent diffeomorphism of a compact manifold, with ν^θ < γ hatγ for some θ in (0,δ) that is a Hölder exponent for E^c, has C^{1+β} stable holonomy between center transversals for some β>0. Also checked the local uniformity intended in the proof and its first approximation step.
Returned assessment: repairable_gap
The note assumes genuine center-stable and center-unstable foliations and uses their intersections as center leaves. The stable holonomy is inside a common center-stable leaf, between local center plaques; there is no volume-preservation or accessibility hypothesis. The required bunching is one-sided: ν is the forward stable contraction bound, γ the forward center conorm bound, and hatγ^{-1} the forward center norm bound. The image on p. 1 permits γ hatγ=1. Neither an unstable bunching assumption nor a C^1 stable distribution is needed for this holonomy conclusion. Theorem 0.2 states a common positive exponent; the proof additionally seeks estimates uniform over short stable legs and local plaques, not bounds independent of the length of arbitrary holonomy chains. I checked the local approximation hypotheses and the time direction of the C^0 estimate. The theorem is not refuted by the subsequent failure of the projective estimate. A separate local-branch construction and proof are given in H3.
Premises used
- The note's derivative inequalities on printed p. 1 are for unit vectors; the center, stable and unstable derivative bounds there are non-strict, but the domination and bunching inequalities are strict.
- Compactness gives uniform slack in ν<γ and ν^θ<γ hatγ. Thus one can increase the stable upper bound and decrease the center lower bound slightly while preserving both inequalities.
- A local center plaque written as a graph has uniformly θ-Hölder derivative when its tangent distribution is θ-Hölder and the graph angle is uniformly controlled.
Missing inputs
- The original proof does not give a branchwise construction justifying the asserted unique extension of π from a local plaque to its entire forward image.
Scope changes
- The conclusion audited is local holonomy regularity for the compact, dynamically coherent system of the note. Transport to the journal's noncompact auxiliary model is a separate obligation in H3.
- The note's Proposition references and lemma numbers were not identified with journal results of the same number. The note's Theorem 0.3 and corollaries were not used to prove Theorem 0.2.
Located concern
June 11, 2005 note, printed pp. 4–5, properties 1–3 of π and the sentence opening p. 5
A uniform local projection does not by itself justify the asserted extension and consistency on forward images of a whole center plaque.
The forward image of a center plaque need not stay in one projection chart. Property 3 is a compatibility assertion, not a consequence of choosing transverse disks separately for different local plaques. On a long image, the local target sheet must be specified; membership in the same global center leaf does not specify that sheet.
Define the projection pointwise on the sheet selected by the forward image of the genuine stable holonomy. Use a single smooth local family of disks transverse to E^c, and prove that the nearby intersections agree on overlaps in the original plaque parameter. H3 gives a separate construction of this kind.
The approximants are not justified on their asserted common domain as written. This is a construction gap, not a counterexample to the theorem.
Located concern
June 11, 2005 note, printed p. 5, Step 1 estimate after (7)
The displayed cocycle indices are wrong for a pullback from time n to time zero, and the bound needs control of the intermediate center arcs.
The image prints γ_{−n}(x) and then Cγ_{−n}(x)ν_{−n}(x), while the next sentence switches to ν_n. The inverse factors appropriate to the reference segment are γ_{−n}(x_n)=1/γ_n(x); the terminal stable error is a forward product. Even with corrected indices, a pointwise conorm at x_i does not by itself bound lengths of nearby center arcs.
Use a slightly enlarged forward stable bound a and a slightly decreased center conorm bound g with a<g. Pull back a short terminal center arc; a first-exit induction in the reference tube bounds its length at time k by C a_k(x) times the product of a/g from time k through n−1. H3 spells out this estimate.
The C^0 step is repairable by domination, independently of the false projective lemma.
H2
Checked statement: The June 11, 2005 note's Lemmas 1.1 and 1.2, their time indices and smallness clauses, the backward induction, and the projective Cauchy comparison at the start of printed p. 8.
Returned assessment: claim_false
Both all-pairs conclusions of Lemma 1.1 are false, even under the standing hypotheses of Theorem 0.2 and even when the entire orbit segment stays inside the required small neighborhood. The counterexample below also meets both additional small-initial-distance thresholds. On a single projective fiber, a derivative condition-number estimate is compatible with the note's intended argument. But the lemma compares points over different base points, and its distance includes base distance. There is no bound on base expansion by the center condition number. Even a correct one-step bound K d^θ does not iterate with exponent θ; literal iteration would give exponent θ^n. Lemma 1.2 considers a much narrower class of pairs generated by the projections along a stable leg. The counterexample does not refute that restricted conclusion. Its printed induction and the resulting Cauchy proof are nevertheless invalid. H3 gives a new deduction of the restricted estimates with buffered rates, without using Lemma 1.1.
Premises used
- Write c=γ hatγ and c_n(x)=the product of c(x_i) for 0≤i<n. The inverse condition-number factors over x_n down to x_k use c_{k−n}(x_n)=c_k(x)/c_n(x). Over x down to x_{−n}, the product in the denominator is c_n(x_{−n}), not c_n(x_n).
- In the p. 6 metric convention, parallel projective directions in the flat example have zero fiber distance; the distance on the projective bundle for those pairs is exactly base distance.
- The image on printed p. 6 has a θ power in the lemma's statements but no θ power in the last displayed inequality of the proof. The images on pp. 7–8 likewise show the linear-in-distance uses, not missing extraction superscripts.
Missing inputs
- A valid estimate for the projected stable pairs, including both their base-point separation during pullback and their derivative or angular separation, is required in place of the all-pairs lemma.
Scope changes
- The false statement is the supporting note's Lemma 1.1, not the journal's stable/unstable distance lemma.
- The corrected estimates described here use buffered rates and stable-pair geometry. They are not assertions of the original all-pairs lemma with better indexing.
Located concern
June 11, 2005 note, printed p. 6, Lemma 1.1, parts 1 and 2 and the two smallness clauses
False auxiliary lemma in both time directions.
Let A=[[2,1],[1,1]], let λ>1 be its larger eigenvalue, and let f=A×A² on the flat four-torus. Let E^s be the two contracting eigendirections, E^c the expanding eigendirection in the first factor, and E^u the expanding eigendirection in the second. These constant distributions give dynamical coherence by affine foliations. The map is smooth; take δ=3/4, θ=1/2, ν=1/2, hatν=1/5, γ=0.99λ and hatγ=0.99/λ. All the note's rate inequalities hold, E^c is θ-Hölder, and ν^θ<c=0.99². For any proposed radius r>0 fix 0<s<r small enough for flat distances. Put one base point at the origin and the other at displacement sλ^{−2n} on a unit expanding eigendirection of the second factor. Both projective directions are the constant center line. The successive distances at times j=0,…,n are sλ^{−2(n−j)}, so all the premises hold; also sλ^{−2n}≤r c^n for large n. The claimed bound at time n is s≤s^θ(λ^{−2θ}/c)^n, which fails. For part 2 use the contracting eigendirection of the second factor and run backward. It gives exactly the same distances, threshold and contradiction.
Discard the all-pairs statement. Retain the same-fiber angular bound, or use a restricted comparison along genuinely stable-related orbits and their controlled projected approximants. No change of cocycle indices alone repairs the stated lemma.
This refutes Lemma 1.1, not the note's holonomy theorem or the journal's ergodicity theorem.
Located concern
June 11, 2005 note, printed pp. 6–8, proof of Lemma 1.1, proof of Lemma 1.2 and the Step 2 Cauchy calculation
The proof replaces a one-step Hölder bound by a linear one, then uses that stronger bound in induction and in the Cauchy comparison. The inverse products are also based at the wrong time in several displays.
The sum in the middle of the p. 6 proof factors against d^θ, not d. In general the intermediate direction (x,ξ) also need not be in P(E^c(x)) in an ambient trivialization; a bundle trivialization requires controlling its transverse variation. On p. 7 the multiplier from time n down to k is c_{k−n}(x_n), not c_{k−n}(x); even with that correction the displayed linear induction does not follow from the lemma's stated power. On p. 8 the compared points at time n lie over the outputs of two different approximants, so the same-fiber bound cannot justify pulling their difference back. If the two distances to the reference orbit at time n are O(a_n^θ), one needs a propagated bound of the form O(a_n^θ/c_n), or a summable substitute, not an application of a bound with another θ power.
Use the linear-map comparison in H3 at each starting time k. It gives an error O(a_k^θ) relative to near-identity identifications for 0≤k≤n, and a geometric error bound for convergence at k=0. Projectivization of matrices with uniformly bounded norms and conorms preserves those local bounds. The unused k<0 part of the printed quantifier need not be used in the induction; on the compact projective bundle it follows from bounded diameter after increasing C, if the backward iterates are defined.
The proof of Lemma 1.2 and the printed Step 2 Cauchy deduction do not survive, although neither is disproved merely by the all-pairs counterexample.
H3
Checked statement: Completion of the note's C^{1+β} holonomy proof, and the use of a holonomy result in the 2010 journal Theorem 0.3 and Proposition 3.1(vii), including separate checks on the C^2 and C^{1+δ} branches.
Returned assessment: repairable_gap
Original proof: the log-distortion sum in Lemma 1.3 is a valid calculation conditional on an actual bound d(v_k,v'_k)≤C a_k^θ with sup a<1. It then gives absolute summability of the log differences; finitely many small n can be absorbed into the constant if the approximants have uniform bounds. It does not, by itself, give a Cauchy estimate for successive derivatives. Step 3 uses an invalid common denominator, and Step 4 again uses the false projective lemma. Neither the note's theorem nor the journal theorem is refuted by those failures.
New repair, separate from the supplied proof, for local center holonomy in the compact coherent setting. Choose continuous rates a,g with ν<a<g<γ and η=sup a^θ/(g hatγ)<1. These choices follow from compactness and strict bunching. Use short genuine stable pairs x,y=h(x), and write x_i=f^i x, y_i=f^i y, and a_{k:n}=∏{i=k}^{n−1}a(x_i). At nearby points of the ambient manifold use parallel transport followed by orthogonal projection onto the target center space as the identification J. On the diagonal J is identity, and locally it varies θ-Hölder. In a smooth chart this follows from smooth transport and the θ-Hölder center projections. Small stable arcs satisfy length at time i ≤ℓ a{0:i}, where ℓ is an upper bound for their initial lengths: on a fixed small tube the stable derivative is bounded by a(x_i), by continuity and the strict margin. The same margins bound the center conorm below by g(x_i) in that tube.
For the approximants, choose a smooth plane field complementary to E^c locally (one can smooth the continuous orthogonal projection in finitely many charts, patch by a partition of unity and take its spectral subspace near 1). Use small exponential disks in this complementary field. Near x_n project along these disks onto the local center sheet selected by y_n. Uniform graph slopes and transversality give a unique intersection z_n and a center arc from y_n to z_n of length at most Cℓ a_{0:n}. The intersections define a map P_n on the source sheet: for nearby parameters the branch is unchanged by continuity of h and uniqueness in the local graph chart. This glues in the original plaque parameter, without requiring f^n of the plaque to be embedded in a single box. Differentiating the intersection equation gives D P_n=J(x_n,z_n)+O((ℓ a_{0:n})^θ) on E^c(x_n): the base derivative of the disk parametrization differs from identity in transported coordinates by O(d(x_n,z_n)), its vertical derivative spans an almost unchanged complementary plane, and solving for the target component is a uniformly bounded oblique projection. The source and target center planes differ by O(d(x_n,z_n)^θ), so replacing that oblique projection by J on the source center plane has the stated error.
Write z_i=f^{i−n}z_n. Pulling back the terminal arc gives length at time k at most Cℓ a_{0:n}/∏{i=k}^{n−1}g(x_i), hence at most Cℓ a{0:k}. To justify the tube in this induction, pull back only up to the first exit of the arc from the tube at time k. Up to such an exit the inverse derivative bound is 1/g(x_k). The resulting length bound is smaller than the distance from y_k to the tube boundary when ℓ is chosen small, ruling out exit. This also gives uniform convergence z_0→y, since sup(a/g)<1, and convergence z_i→y_i for each fixed i.
For the derivative, let A_i=Df|E^c(x_i), B_i=Df|E^c(z_i), and J_i=J(x_i,z_i). Subscripts 0:i denote ordered products carrying time 0 to time i. The one-step defect R_i=B_i^{-1}J_{i+1}A_i−J_i is O((ℓ a_{0:i})^θ): compare Df in smooth coordinates at the two points and use the estimates on J_i and J_{i+1}; inverses have uniform bounds. The exact telescoping formula for the derivative of the pulled-back projection is J_0+∑{i=0}^{n−1} B{0:i}^{−1}R_i A_{0:i}+B_{0:n}^{−1}(D P_n−J_n)A_{0:n}. Each summand is bounded by Cℓ^θ η^i and the terminal term by Cℓ^θ η^n, because the product of the two norm bounds at time i is at most 1/(g(x_i)hatγ(x_i)). For each fixed i, its term converges uniformly to the corresponding term with z_j replaced by y_j. Uniform summability proves convergence of the actual derivatives, with no choice of projective sign. The same calculation starting at k bounds the derivative's difference from J_k by Cℓ^θ a_{0:k}^θ; thus it also provides the restricted comparison needed in H2. The limiting derivative is uniformly close to J(x,y) for short stable legs, and is invertible. Uniform convergence of maps and derivatives in local center charts, using integration along source chart segments, proves C^1 regularity and uniform local derivative bounds.
For the positive Hölder exponent, the limiting derivative on genuine short stable pairs is the uniform limit of K_m(x,y)=(Df^m|E^c(y))^{-1}J(x_m,y_m)Df^m|E^c(x), with tail O(η^m) by the same telescoping sum. In smooth bundle coordinates the finite products have θ-Hölder constants bounded by C L^m for some L>1. Indeed, replace one factor at a time in the product: the factors and one-step inverses are uniformly bounded, each has a uniform θ-Hölder bound in the pair at its own time, and the distance of two pairs at time i is bounded by Lip(f)^i times their initial distance. These comparisons can be made in a fixed Euclidean bundle representation using finitely many chart frames multiplied by a partition of unity, extending each map by projection onto its source space. Balancing C L^m times distance^θ against Cη^m gives a positive Hölder exponent on the space of short stable pairs. Restricting to y=h(x) preserves that exponent because h is locally Lipschitz in center charts by the derivative bound already proved. All choices are uniform over sufficiently short legs. This repair does not assume the note's projective or scalar Cauchy estimates and does not use a later holonomy theorem.
Journal interface: on pp. 455–456, (6) is read as a bound on unit vectors, as required for the positive lower and upper rates; (8) and (7) have different jobs. The domination inequalities (8) are invoked for the Hölder exponent, while (7) supplies the two one-sided bunching margins. Inverting the dynamics exchanges stable and unstable rates, with evaluation shifted to the preimage, and converts the stable calculation to the unstable one. With the usual 0<θ<1 convention and 0<ν,hatν<1, strong bunching also implies ordinary bunching. The missing step is not the algebra of the rates: it is checking the uniform exponent and local geometry on the fiber model of p. 467. The genuine compact-manifold theorem cannot be applied verbatim to that model.
Premises used
- No external regularity theorem is used in the new local repair; it starts with the invariant coherent foliations, the Hölder center bundle and the rate bounds in the note's hypotheses.
- For the journal's C^2 branch, the relevant conditional input is [PSW97], Theorem B, read with [PSW00]: in the compact C^2 dynamically coherent partially hyperbolic setting with ν<γ hatγ, E^s is uniformly C^1 on center-stable leaves and center holonomy is C^1; the inverse version treats E^u. This formulation is quoted in the 2005 note p. 2. available_from=1997 for the paper, 2000 for the correction; date_evidence=2010 journal p. 489 and note p. 11 for the original; date_basis=publication year; version=Duke Math. J. 86 (1997), 517–546, together with 105 (2000), 105–106. The cited texts and the applicability of their proof to the auxiliary leaf spaces were not supplied.
- For the journal's C^{1+δ} exponent step, the conditional input cited on p. 456 is [PSW97], Theorem A: the needed specialization is θ-Hölder regularity of E^c for a C^{1+δ} partially hyperbolic map with θ<δ and the two pointwise domination ratios of (8) relative to the lower rates of (6). The journal states this specialization; the exact hypotheses of the imported theorem and its noncompact uniform extension have not been verified. available_from=1997, with correction available_from=2000; date_evidence=2010 journal p. 489; date_basis=publication year; version=the Duke paper and correction just identified.
- The holonomy input identified as [BW06b] on journal pp. 455 and 467 is distinct as a source record from the supplied June 2005 note. For the note: available_from=2005-06-11 as internal version date, independent circulation date unverified; date_evidence=note p. 1; date_basis=dated preliminary manuscript; version=A note on stable holonomy between centers. For the cited version: available_from=2006 on the supplied record; date_evidence=2010 journal p. 488; date_basis=bibliographic preprint year; version=A note on unstable holonomy between centers, text not supplied. Equivalence of the proofs and earlier availability of the cited version are not established.
Missing inputs
- The actual [BW06b] version and dated evidence for any version containing the needed proof available by October 11, 2005.
- The supplied-text gaps in the historical inputs: the precise [PSW97] Theorems A and B and [PSW00], and a verification of their hypotheses on the auxiliary construction where they are used.
- For the C^{1+δ} branch on p. 467: a uniform θ-Hölder estimate for the fake center distributions on the disjoint tangent fibers, uniform bounds on the nonlinear maps and their inverses and on local product charts, and a check that the local stable and unstable holonomy estimates give exactly the uniform subfoliation regularity in Proposition 3.1(vii). The two inequalities (8) having strict slack makes this plausible but does not itself prove it.
- The journal images containing the full statement of Proposition 3.1 and the downstream julienne proofs were not supplied. References to those downstream uses identify interfaces from author-version context, not checked journal wording.
Scope changes
- The replacement proof is independent and local. It proves the compact coherent holonomy estimate under the note's standing geometric hypotheses, not the existence of the journal's fake foliations or their uniform graph-transform bounds.
- Holonomy regularity inside fake weak leaves is not absolute continuity of the genuine strong foliations in the ambient manifold. Neither allows replacing genuine saturation by fake saturation.
- The journal's ergodicity and Kolmogorov conclusions, and the detailed downstream julienne proof, are not independently audited in this packet.
Located concern
June 11, 2005 note, printed pp. 8–10, Lemma 1.3 and Step 3
Bounded scalar distortion is used as if successive derivative norms had an identical pullback factor. Projective convergence also leaves a sign ambiguity unless an oriented comparison is supplied.
The chain rule for D h_n at time n evaluates the pullback derivative at π_n(x_n), in the direction Dπ_n v_n. For D h_{n+1} the corresponding point and direction at time n come from f^{-1}π_{n+1}f. The two products in the denominator are therefore different, even when each separately satisfies Lemma 1.3. Printed p. 9 instead factors out one product and writes fπf^{-1} in the inner comparison. The projective property 2 on p. 4 alone does not estimate the norm of a terminal derivative or fix its sign. Even identical projectivizations and norms do not distinguish I from −I.
Compare actual linear maps with the telescoping identity above. Alternatively, a scalar-series proof would need convergence of the perturbed directions at every fixed time, a uniform summable bound for their log discrepancies, a terminal norm factor tending to one, and an independent oriented derivative comparison.
Neither bounded distortion alone nor the displayed difference formula proves C^1 convergence as written.
Located concern
June 11, 2005 note, printed pp. 9–11, Step 4
The Hölder proof uses the false cross-fiber estimate twice and omits justifications for the distance scale and cocycle comparisons.
The backward comparison and the estimates of the forward projective orbits of p and p′ use Lemma 1.1. Comparing products on these two orbits with a uniform multiplicative constant requires a distortion argument; continuity alone does not supply a summable modulus. Also ρ can equal 1 under the allowed rates, so the stated scale selection need not exist without a strict buffer. The chosen distance interval [ρ_n,ρ_{n−1}) does not give the hypothesis d≤ρ_n used in the next estimate. The second equality for β′ on p. 11 drops a factor θ when the β on p. 9 is substituted.
The finite-product approximation with geometric tails above gives some uniform positive Hölder exponent, without the asserted numerical formula. In an argument retaining the original scales, enforce c<1 by slack, shift the scale index, and prove the orbit-product comparisons before summing.
The printed C^{1+β} completion is not valid as it stands. The separate local repair obtains a positive exponent, not the printed explicit one.
Located concern
2010 journal, printed pp. 455–456 and 467, Theorem 0.3 and regularity of the fake subfoliations
The C^{1+δ} citation is not a verified transfer from the supplied compact coherent note to the uniform fake-foliation statement. The C^2 branch has a different prerequisite.
The note concerns genuine invariant foliations of a compact dynamically coherent map. Journal p. 467 applies holonomy regularity to the globally extended fiber dynamics before pushing the foliations into local manifold balls. The union of fibers is not a compact manifold with the original coherent foliations. Uniform derivative estimates in center directions do not on their own certify every chart assertion in 'C^1-subfoliates.' The C^2 argument on p. 467 has the correct strict-margin calculation at a=1 and hence for a>1 close enough to 1, but explicitly defers the issue that the weak leaves are not a C^k manifold of the assumed class to the regularity references.
Check the graph-transform outputs and the rate inequalities uniformly on the extended fibers, including the θ-Hölder center estimate and uniform local geometry. A uniform local center-holonomy argument can then run entirely inside the extended model. For a C^1 chart conclusion, use continuously varying local stable plaque coordinates and show that holonomy derivatives in center directions vary continuously with the nearby plaque parameter; uniform Hölder derivative bounds plus continuous dependence of the maps give this latter continuity by difference quotients in uniform charts. Check separately the leaf regularity theorem and its corrected version in the C^2 branch.
The compact note, even with a repaired proof, does not by itself verify the journal's C^{1+δ} regularity input. The C^2 branch does not depend on the false lemma. Neither branch is shown false.
Outside the audit
- The contents of the [BW06b] preprint cited by the journal and its version history relative to the supplied note.
- The full external proofs in [PSW97]/[PSW00], the complete uniform graph-transform construction, and the precise journal statement pages not in the packet.
- The proof of the journal's julienne density, genuine absolute-continuity, ergodicity and Pinsker conclusions beyond identifying their dependencies. No line-by-line audit of the complete author extraction is claimed.
- The note's additional examples, corollaries and separate claimed perturbative exponents outside the assigned Theorem 0.2 proof.
Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.