gpt-6.2-astra max: bw-modern-holonomy
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.
There is a false auxiliary lemma in the dated June 2005 note and a consequential failure of its printed holonomy proof. There is no counterexample here to a principal holonomy or ergodicity theorem. At the historical cutoff the supplied proof does not justify the low-regularity journal input; the C² route is separate. For modern use, conditional on Saghin's 2023 uniform fake-holonomy result and the BW graph construction before (vii), the missing strong-inside-weak regularity input can be supplied at the original bunching strength. Brown's extra-bunched argument is not the reason for that full-strength conclusion. The independent linear-map argument proposed in H3 also works locally for the compact coherent note theorem. Neither modern substitution nor new argument is retrospective validation, and repairing this input is not an audit of the whole density proof.
M1
Checked statement: The all-pairs forward and backward estimates, including the small-initial-distance clauses, in Lemma 1.1 of the June 11, 2005 preliminary note; their role in the proof of that note's Theorem 0.2 and the relevance to the journal's distinct Theorem 0.3.
Returned assessment: claim_false
Both proposed systems work against the note. Write c = γ hatγ. On p. 6 the points being quantified over need not have the same footpoint, and the proof explicitly includes base distance in the maximum defining distance. In a flat product with constant one-dimensional center, the projective center fiber is a singleton and the distances in question are just the base distances. Nor is there a hypothesis that the base points lie in one center plaque. With constant rates every inverse c-product in either estimate is c^(−n); changing its base index cannot save the estimate.
First witness: A = [[2,1],[1,1]] is symmetric unimodular, with positive eigenvalues λ = (3−√5)/2 and λ^(−1). For A × id on the flat three-torus the derivative on the three orthogonal summands has norms λ, 1, λ^(−1). The constant splittings integrate, including the two weak splittings. Use μ = hatμ = 3/10, ν = hatν = 2/5, γ = hatγ = 9/10, δ = 3/4 and θ = 1/2. Then 3/10 < λ < 2/5, 9/10 < 1 < 10/9, and the reciprocal inequalities give the unstable bounds. The order of the three rate intervals is strict. Also 2/5 < (81/100)^2 and (4/9)^2 < 3/10. Thus even both journal bunching inequalities and both journal exponent inequalities hold. The map is smooth, volume preserving, and its center bundle is θ-Hölder. For ANY proposed lemma radius r take 0 < η < r below a flat injectivity radius. Use base points 0 and ηλ^n e_s (and equal circle coordinate), with their center lines. At time −k the distance is ηλ^(n−k) for 0 ≤ k ≤ n, with no reduction modulo the lattice affecting this calculation. Even the extra initial threshold holds, since λ < c = 81/100. Part 2 requires η ≤ √η (√λ/c)^n. Here √λ/c = 50(√5−1)/81 < 1. This fails for large n at the fixed η. Using e_u and positive times gives the same test of part 1. The circle coordinate is constant along su-paths, so this witness is not essentially accessible.
Second witness: let L = (3+√5)/2, and take A × A² on the flat four-torus. Assign both contracting directions to the stable bundle (norms L^(−1), L^(−2)), the first expanding direction to the center (norm L), and the second expanding direction to the unstable (norm L²). Again the foliations are affine and the map smooth. With ν = 1/2, hatν = 1/5, γ = 0.99 L, hatγ = 0.99/L, one has L^(−1) < 1/2 < γ < L < L/0.99 < 5 < L². For θ = 1/2 < δ = 3/4, both √ν and √hatν are less than c = 0.99². The constant center bundle has the required exponent. Now take a displacement s L^(−2n) in the second expanding direction, with 0 < s < r inside the injectivity radius. The distances through time n are s L^(−2(n−j)), 0 ≤ j ≤ n. Part 1 requires s ≤ s^θ (L^(−2θ)/c)^n, impossible since L^(−1) < c. The second contracting direction gives exactly this contradiction in backward time for part 2. Since L^(−2) < c these pairs also meet the respective extra initial thresholds. The second witness need not satisfy journal (8); indeed with its stated hatν and hatγ one cannot have hatν/hatγ < √hatμ with hatμ < L^(−2). That is NOT a hypothesis of note Lemma 1.1 or note Theorem 0.2. Both witnesses have smooth actual center holonomies.
Premises used
- The note's p. 1 rate bounds are for unit vectors; p. 2 assumes a compact dynamically coherent partially hyperbolic system, C^{1+δ}, and the one-sided inequality ν^θ < γ hatγ with θ a center-bundle exponent. It assumes neither essential accessibility nor nonconstant center directions.
- The projective Lipschitz estimate for a single linear isomorphism on a single fiber involves its condition number. It is not an estimate for expansion of the base of the projective bundle.
- The limiting arguments above test the radius quantifier; they do not infer a counterexample from one numerical trial.
Missing inputs
- The actual bibliographic [BW06b] cited on journal pp. 455 and 467 is not supplied as a second manuscript. Identity of its proof with the dated June 2005 preliminary note, or a different proof available by the received-date cutoff, is not established here.
Scope changes
- This is the supporting note's Lemma 1.1, not the journal's Lemma 1.1.
- The historical cutoff is 2005-10-11. The June 11, 2005 date and title are on the note itself. The journal citation key and its reported 2006 bibliographic version are kept distinct from that note.
- I checked the note's proof steps for these dependencies, not the proofs of the external C² regularity theorems.
Located concern
June 2005 note, printed p. 6, proof of Lemma 1.1
The one-step calculation and the proposed induction do not give the printed iterated estimate.
The penultimate bound has a factor d(v,w)^θ, not d(v,w); the last bound on the page is linear. Iterating a Hölder bound of exponent θ < 1 would itself put θ^n, not θ, on the initial distance. The orbit-segment counterexamples defeat the statement even with all intermediate distances small.
Discard the all-pairs iteration. Control base orbits and compare directions or linear maps only for the pairs generated in the holonomy construction.
Auxiliary statement false in both time directions, not merely an omitted proof.
Located concern
June 2005 note, printed p. 7 (Lemma 1.2), p. 8 (end of Step 2 and Lemma 1.3), p. 9 (Step 3), pp. 9–11 (Step 4)
There are direct and indirect uses of the invalid estimate, and bounded distortion alone would not complete the derivative proof.
The p. 7 backward induction actually uses linear propagation of the terminal discrepancy. The p. 8 comparison of two successive projectivized approximants pulls back a discrepancy between different base points; a same-fiber estimate does not suffice. A pullback from time n to time k also calls for the inverse product over [k,n), based at time n. The log sum in Lemma 1.3 is valid given the claimed exponentially decaying discrepancies, but its input has not been proved. On p. 9 the pullback derivatives for h_n and h_{n+1} are evaluated on different projected orbits/directions, so their difference does not factor with the displayed common denominator. A bound on each distortion ratio is not convergence of their difference. Step 4 uses Lemma 1.1 again, forwards and backwards, as well as the earlier convergence estimates.
Use a comparison with a summable error for actual derivatives, together with convergence of the underlying maps on the selected target sheets. M3 tests one such alternative in the compact coherent setting.
The printed proof of Theorem 0.2 is defective. The counterexamples do not themselves falsify Lemma 1.2 for its special projected pairs, or Theorem 0.2. Step 1 (C⁰ approximation, pp. 4–5) does not depend on Lemma 1.1; its sheet and pullback estimates should be justified independently.
Located concern
BW, Annals 2010, printed pp. 454–456 and 465–467
The affected journal edge is the low-regularity holonomy input, not a logical implication from a false lemma to a false ergodicity statement.
The June note's Theorem 0.2 is about C^{1+β} stable holonomy in a coherent system. Journal Theorem 0.3 does not assume coherence; it imposes volume preservation, (6)–(8), and essential accessibility, and concludes ergodicity and K. Journal p. 455 describes the Lipschitz input and p. 467 uses [BW06b] on the extended fiber dynamics for the 1 < k < 2 case. The k ≥ 2 paragraph on p. 467 instead uses bundle regularity along weak leaves under ordinary bunching; it refers the weak-leaf smoothness difficulty to [PSW97, PSW00].
For a modern substitution at the journal interface use M2. Do not use the June note's statement about compact coherent dynamics as if it were already the required uniform fake-foliation theorem.
Neither journal Theorem 0.1 (C²) nor Theorem 0.3 is refuted. The June proof cannot certify the low-regularity input historically. No failure of the separate C² input is established by this test.
M2
Checked statement: Conditional modern substitution: Saghin's 2023 fake-holonomy result, used with the BW uniform graph construction but without Proposition 3.1(vii), supplies the local strong-inside-weak regularity and bounded Jacobians at the original journal Theorem 0.3 rates.
Returned assessment: no_issue_found
Yes, for this particular proof input, conditional on the stated uniform results. The following patch uses local holonomies along short legs; it does not posit a bound for arbitrarily long holonomy chains.
Rates. Use the BW metric and set λ_s = ν, λ_u = hatν^(−1), λ_c^- = γ, λ_c^+ = hatγ^(−1). These have the order and unit-vector bounds of Saghin Definition 1.1. His stable ratio test is ν^θ < γ hatγ, and the unstable test is (hatν^(−1))^θ > hatγ^(−1)/γ. These are exactly journal (7), pointwise, with uniform strict margins by compactness. For the inverse map at f(p) the stable and unstable bounds are hatν(p) and ν(p)^(−1), and the center lower and upper bounds are hatγ(p) and γ(p)^(−1); one must shift the evaluation point when reversing time. There is no loss at the exponent restriction θ ∈ (0,1]. Indeed μ < ν < 1 and hatμ < hatν < 1 under the unit-vector rate conventions, and γ hatγ ≤ 1. If θ ≥ 1, (8) gives ν hatν ≤ ν hatν/(γ hatγ) < (μ hatμ)^θ ≤ μ hatμ < ν hatν. Thus 0 < θ < 1, θ < δ, and the dynamics are C^{1+θ}. Also (7) implies ordinary bunching (5). The journal uses (8) for a center Hölder exponent; Saghin's corollary does not require an extra θ-pinching or a θ-Hölder hypothesis on the center bundle.
Which fake foliations. Fix ONE globalization F_r as on journal pp. 465–466. In the fiber over p write L_p = T_p f. In uniform exponential charts one can take F_p(v) = L_p v + χ(v/r)(exp_{f(p)}^(−1) f exp_p(v) − L_p v), with smooth cutoff, equal to the chart dynamics on the r-ball and linear outside 2r. The term in parentheses vanishes to first order at zero. With α = min(δ,1), its value and derivative on the 2r-ball are O(r^{1+α}) and O(r^α); its derivative has a uniform Hölder bound of exponent at least θ. Differentiating the cutoff gives uniform fiber derivative Hölder bounds (the derivative of the error has size O(r^α) and θ-seminorm O(r^{α−θ})). Take r small. These are diffeomorphisms of whole fibers: composition with L_p^(−1) gives identity plus a perturbation of Lipschitz constant < 1. There is a common co-Lipschitz constant m > 0, by the lower singular value of L_p and small derivative error. In particular the inverse derivatives are bounded by 1/m, with θ-seminorm bounded by m^(−2−θ) times a forward derivative seminorm. No assertion that the values of the linear maps on the whole fibers are bounded is needed. Compact rate margins, the narrow cones of the graph construction, and this C¹ closeness give the derivative bounds on every fake summand, with rates taken at p independently of v. Rates can be buffered slightly within the strict inequalities. In particular b_s = sup ν^θ/(γ hatγ) and b_u = sup hatν^θ/(γ hatγ) are still < 1. The cone-scale versions follow by taking the cones thin and the comparison scale small. All angles and local graph sizes can be taken uniform.
Use the bi-infinite sequence of Euclidean fibers over p_i = f^i p. Coherence, global graph leaves, subordination of the strong leaves, and uniform Hölder tangencies of some positive exponent are inputs from the construction before (vii). They give the coherent model in Saghin §3.7, not coherence of f on M. The origin is not a restriction on the holonomies. Given any z in a fiber, put z_i on its full F_r-orbit and use G_i(w) = F_{p_i}(z_i+w) − z_{i+1}. Then G_i(0) = 0, the derivative, inverse and bunching bounds are unchanged, and the error from L_{p_i} in values is bounded by twice the original sup error. The translated invariant graphs are the graphs for this conjugate sequence; one does not cut off the maps afresh along z or ask that exp_{p_i}(z_i) follow the original orbit in the local balls. In these coordinates the smooth comparison in §3.7 is the linear cu-plane with parallel u-planes (cs and s in reverse time). In particular one is not approximating a distant leaf by a plane at an unrelated origin. Even for a long cu piece there is height control here: for a cu graph of stable height bounded by H, its forward graph transform has height at most (sup ν)H + C sup|G_i−L_{p_i}|, by projection to the stable linear summand. Starting from horizontal graphs through zero in the past gives the uniform bound C sup|G_i−L_{p_i}|/(1−sup ν) on the limiting graph through the marked orbit. The inverse construction gives the cs bound. This and the small slopes explain the particularly simple flat comparison; bounded angles alone need not bound heights on large disks. Constants are independent of the orbit, fiber, recentering and iteration time. By using the same graph construction and extension as BW, followed by the SAME exponential projection, one gets the fake plaques allowed in Proposition 3.1. The remark on journal p. 468 rules out identifying fake leaves just by their name for different extensions.
Regularity obtained. Corollary 1.1 gives uniform C¹ holonomy along the fake strong leaves between the fake center plaques, in the appropriate weak leaf. Taking the leg in the opposite order gives the inverse map with the same local bounds. Pushforward by the uniformly smooth charts preserves uniform local bi-Lipschitz bounds. For clarity about continuity, a bound on derivative norms alone is not a common modulus of continuity. One can use the short-leg estimate in Saghin's proof: for nearby u-related points a,b, the derivative between centers at a differs from P_{a,b} by at most C d_u(a,b)^θ, where P is projection from the center at a to the center at b parallel to the su-space at a in fiber coordinates. See the Lipschitz-jet estimate and its differential version, preprint pp. 13–14, with the uniform-fiber proof on p. 16. The estimate is relative to a varying projection, not a claim that the center field is θ-Hölder. The transfer of that estimate uses the same proof inputs: the flat smooth comparison just described, uniform transversality, and the recurrence of Lemma 2.2 (pp. 6–7), whose coefficient on the previous constant is < 1 by cone-scale bunching and whose additive term is bounded by the derivative Hölder and angle bounds. I use the jet estimate of the imported proof, not a deduction of differentiability from convergence of bounded derivatives. There is a stable version by time reversal.
In particular the needed joint continuity can be seen in the fibers. For x,y on a short fake u-leg, transport P_{x_{−n},y_{−n}} to time zero by the forward center derivative on the y-orbit and the inverse center derivative on the x-orbit. By holonomy equivariance and the short-leg estimate this continuous finite product differs from Dh^u_{x,y}(x) by at most C d_u(x,y)^θ b_u^n: the leg in the past contracts by the product of hatν, and the multiplying center norms cost the product of 1/(γ hatγ) over the SAME time interval. For s, pull P_{x_n,y_n} back from the future, with bound C d_s(x,y)^θ b_s^n. For each fixed n these products have uniform continuity moduli within the fibers, by the uniform derivative bounds and Hölder tangencies, including for the oblique projections. Thus the limits are continuous in the nearby pair with uniform control. This explains the use of uniform C¹ in the patch without reading the 2023 corollary as the stronger wording of a different edition.
Passage to subfoliation charts. Work in one small weak plaque W. Write S for its strong subfoliation, K for its center tangent planes, and H for T S. For two uniformly transverse C¹ disks T,T' and a local S-leg x to y between them, factor the holonomy germ through the center disks C_x and C_y. At a crossing point, sliding from T to C_x along S has derivative the projection from T_x T to K(x) parallel to H(x). Indeed the two disks pass through x; in graph coordinates transversality bounds the displacement and the length of the short connecting plaque arc by O(distance to x). Its component normal to H(x) is little-o of this distance, by continuity of the plaque tangents. Linearizing the two disks gives the stated projection. The same argument applies from C_y to T' at y. Thus the general derivative is the center-to-center derivative composed on either side with these projections. It is jointly continuous, bounded, and boundedly invertible for disks with uniform transverse angle. This argument does not require smooth holonomy ALONG the centers.
For example, in a weak graph chart take the disks T_t given by fixing the coordinate in a fixed strong plane, and send (x,t), x in T_0, to S(x) ∩ T_t. On smaller uniform boxes this is a foliation homeomorphism. The x-partial derivative is the holonomy derivative just computed. The t-partial is the inverse of coordinate projection restricted to H at the image, since each strong plaque is a C¹ graph. Both partials are continuous. Transversality, the transverse inverse bound and the identity in the t-coordinate bound the derivative and its inverse. These are uniform C¹ strong-foliation boxes INSIDE W. They give both the Lipschitz transport of center radii and the measure consequence: transverse Jacobians for controlled transversals are bounded above and away from zero (by the corresponding powers of the norm and inverse bounds). In the boxes, induced weak-leaf volume is uniformly comparable to transverse volume times strong-plaque volume, by change of variables and Fubini. In particular this works for fake unstable fibers in a fake cu-leaf, not just for a single holonomy through the marked base point. Smooth exponential projection and restriction to smaller local balls do not destroy these bounds.
Premises used
- Modern input assumed: Saghin, arXiv:2307.01321v1 (July 2023), Corollary 1.1, printed p. 2: uniform C¹ fake unstable holonomy between fake centers for C^{1+θ}, θ-unstable bunched partially hyperbolic diffeomorphisms, 0 < θ ≤ 1, and the stable counterpart. I use it in the uniform disjoint-Euclidean-fiber sense specified in §3.7, p. 16. For full C¹ charts the uniform C¹ control is used with the nearby-pair derivative control of that proof (§§3.4–3.7); a mere bound on the Lipschitz constant suffices for less. The genuine coherent Theorem 1.1 on p. 2 or the linear cocycle Theorem 1.2 on p. 3 alone is not the fake-holonomy result.
- Construction assumed: BW journal pp. 464–468, Proposition 3.1 through (vi) and its extended invariant graph families, with the stated uniformity. In particular the strong and center graph foliations subfoliate the appropriate weak graph foliations topologically with continuously varying tangent planes. Neither a C¹ strong-foliation box nor bounded transverse Jacobians inside the weak leaf are assumed. Saghin's reference [4] on p. 16 is used for these graphs, not as a proof of their bunched holonomy regularity. This avoids circularity.
- Uniformity needed is on controlled local boxes, in each component of the orbit model with constants independent of the component and of p. No horizontal differentiability on TM with the discrete base is used.
- The remaining density argument, when invoked as a consumer, is a separate premise. BW p. 473 explicitly takes the uniform subfoliations on the disjoint orbit balls and distinguishes true from fake saturation. BW p. 472 uses Theorem 5.1/Corollary 5.2, volume preservation, Birkhoff averages and essential accessibility, and for K the Brin–Pesin input that Pinsker sets are bi-essentially saturated. Fake Jacobians do not prove that last input or ambient absolute continuity of the true strong foliations.
Missing inputs
- No extra numerical bunching assumption is missing for the Saghin application on the stated premises. The complete correctness of the imported graph-transform foundations and Saghin's proofs of the holonomy/jet results has not been independently certified in this application audit.
- The full remaining journal density-sequence proof, including its volume-comparison applications and the separate genuine-foliation measure inputs, is not supplied page by page here. A verdict on that entire proof does not follow from the patch.
Scope changes
- The verdict is positive for the CONDITIONAL MODERN APPLICATION, not for the original [BW06b] proof. It uses the 2023 preprint's Corollary 1.1, not a visual inspection of a 2025 journal Corollary 1.5.
- The full-strength input needed in BW is local regularity inside weak leaves, not C¹ regularity of the strong distribution on the ambient manifold. Volume preservation and essential accessibility enter the ergodicity consumer, not the construction of these boxes.
- I checked the BW construction/interface on the supplied journal pages, the Saghin formulation and the features of its proof relevant to the uniform-fiber transfer, and the Brown hypothesis comparison. I did not reprove all the imported theorems.
Located concern
BW journal pp. 465–468, 473; Saghin 2023 preprint pp. 2, 13–16
Several weaker transfers would not establish the stated conclusion.
Local fake plaques in M are invariant only while iterates remain in the construction balls. Regularity proved only at the zero section, or for some different fake family, would miss general holonomies in the orbit boxes. Also an upper Lipschitz bound for one direction is not a lower Jacobian bound. Even uniform bounds on C¹ norms, if interpreted only as sup norms of first derivatives, do not by themselves imply joint continuity of derivatives as target plaques vary.
Use the full recentered invariant fiber dynamics, the short-leg bounds in both orders, and the transverse chart argument above. If one imported ONLY derivative norm bounds, the crossing-disk argument and the reverse leg would still give bi-Lipschitz and volume control, but the joint continuity step for literal C¹ foliation charts would need a common C¹ modulus or the short-leg argument.
With the specified 2023 result and uniform construction, the low-regularity clause of (vii) needed here has a modern replacement. It gives no fake-to-true saturation implication and no assertion about arbitrary locally chosen plaques.
Located concern
Brown, supplied manuscript, printed pp. 1–3, 14–15
Brown's argument is not an interchangeable citation at the full original rates.
In Brown's buffered log-rate setup, (1) controls a bundle exponent barθ < β, (3) requires the center log spread divided by −κ_n to be < θ, and (2) additionally requires sup exp(βκ_n − κ_n θ − βγ_n) < 1. Here γ_n is the LOWER center log rate. In BW rates the extra stable inequality corresponds to (ν/γ)^β < ν^θ with slack; reverse for the unstable side. Brown p. 3 expressly distinguishes (2) from the BW test. Theorem 4.1, pp. 14–15, uses an exponent below barθ with the extra comparisons (13), written there non-strictly. Care is needed with its notation: p. 15 displays γ hatγ^(−1) on the right of the max bound, whereas the inverse spread of the p. 14 center interval [γ,hatγ^(−1)] is γ hatγ. The log-spread test in §2 avoids relying on that display to identify the hypotheses. For instance, at the level of rate data take ν = hatν = exp(−1), μ = hatμ = exp(−1.05), γ = hatγ = exp(−0.30), δ = β = 0.75. BW permits θ = 0.65 (0.65 > 0.60 and 0.70 > 1.05·0.65). At those rates the log-spread test requires an exponent above 0.60 while (2) requires it below 0.75·0.70. This tests non-implication for the rate data, not non-ergodicity of any map.
Use Brown as a later result with its own extra hypotheses for its globalized-holonomy/ergodicity application; use Saghin, not Brown's extra-bunched argument, for the substitution at journal (6)–(8).
No historical validation and no claim that Brown refuted BW. Brown's first page carries the arXiv v1 August 21, 2016 side stamp and an internal October 19, 2021 date; no assertion about when each line was written or about the text of the 2022 journal edition is required.
M3
Checked statement: The new branchwise projection and telescoping proof proposed in Sol H3, restricted to local stable holonomy under the compact dynamically coherent hypotheses of the June 2005 note's Theorem 0.2.
Returned assessment: no_issue_found
I find the proposed method sufficient for the local theorem. The following details matter to that verdict. Use ordinary Hölder exponents (at most one), a small source center disk, and a continuous genuine stable holonomy y = h(x) into a marked target center disk, with connecting stable arcs of length ≤ ℓ. Such short product boxes are available from coherence, transversality and compactness, without differentiability of h. Use products a_{i:j} = ∏{i≤t<j} a(x_t), and similarly for other rate functions. Choose ν < a < min(1,g), g < γ with ρ = sup(a/g) < 1 and b = sup a^θ/(g hatγ) < 1. Uniform slack in domination and bunching permits these small changes even though the note's derivative inequalities need not be strict. In a tube of a fixed radius about x_i the stable derivative is ≤ a(x_i) and the center conorm ≥ g(x_i). Starting with small ℓ, the stable arcs have length ≤ ℓ a{0:i} at time i by induction, so stay in the required tubes.
Terminal projections. Fix a smooth plane bundle uniformly complementary to the centers (smooth approximation of a continuous complementary bundle suffices). Use short exponential disks in that bundle. At time n intersect the disk based at x_n with the center sheet marked by y_n = f^n h(x), obtaining z_n. Uniform local center graphs and the transverse angle give uniqueness on that sheet and a center arc from y_n to z_n of length ≤ Cℓ a_{0:n}. The important continuity is INTRINSIC: h maps between local plaques and f^n restricted to the target leaf is leafwise continuous. For each fixed n and each parameter x, nearby parameters therefore select the same local target graph, even if other sheets of the same global leaf return to the box. On that neighborhood the intersection is a C¹ function P_n of the source point by the implicit function theorem, with no differentiation of h. The sheet-marked definitions agree on overlaps in the original source parameter. There is no need for a tubular neighborhood of an entire recurrent center leaf, a uniform size of those parameter neighborhoods in n, or an integrable smooth complementary bundle.
Let J(u,v) be short smooth transport followed by orthogonal projection from E^c(u) to E^c(v). In uniform coordinates it is θ-Hölder in the pair near the diagonal. The disk parameter of the intersection has size O(ℓ a_{0:n}). The base derivative of the smooth disk parametrization is transport of the identity plus O(ℓ a_{0:n}); eliminating the vertical parameter projects onto the target tangent plane parallel to the disk tangent. This oblique projection is uniformly bounded. On the transported source center plane it differs from orthogonal projection by O(d(x_n,z_n)^θ), since the two center planes are that close. Consequently DP_n on the source tangent differs from J(x_n,z_n) by O((ℓ a_{0:n})^θ). A projective angle estimate alone would not give this assertion about actual linear maps.
Put z_i = f^{i−n} z_n for this approximant. The terminal center arc pulls back to one of length ≤ Cℓ a_{0:n}/g_{i:n} ≤ Cℓ a_{0:i} ρ^{n−i} at time i. To justify use of the conorm in this estimate, induct backwards and stop a preimage arc from y_i at its first exit from the tube about x_i. Up to that exit its length is at most the next arc length divided by g(x_i). This bound plus d(x_i,y_i) is less than the tube radius for ℓ sufficiently small, excluding an exit. Thus d(x_i,z_i) ≤ C'ℓ a_{0:i}, and for each fixed i, z_i tends to y_i uniformly, with closeness on the target center sheet. In particular the maps x ↦ z_0 converge to h on a common source disk.
Derivatives. Write A_i = Df|E^c(x_i), B_i = Df|E^c(z_i), J_i = J(x_i,z_i), and products of linear maps in forward composition order. R_i = B_i^(−1) J_{i+1} A_i − J_i has norm ≤ C(ℓ a_{0:i})^θ: the expression vanishes on the diagonal, and in smooth coordinates the derivative and inverse factors and center identifications have the required local Hölder bounds. The chain rule gives B_{0:n}^(−1) DP_n A_{0:n} for the derivative at time zero. Subtracting J_0 gives EXACTLY the sum over 0 ≤ i < n of B_{0:i}^(−1) R_i A_{0:i}, plus B_{0:n}^(−1)(DP_n−J_n)A_{0:n}. The multiplying norms cost at most the product of 1/(g(x_j) hatγ(x_j)) over j < i or j < n, respectively. Hence the bounds on terms are Cℓ^θ b^i and Cℓ^θ b^n. The summands are n-dependent, so a summable majorant ALONE would not prove convergence. For fixed i, uniform z_j → y_j for j ≤ i+1 does give convergence of the summand to the analogous true-pair summand. Compare changing target spaces in local frames or a fixed ambient bundle representation. Truncation followed by the geometric tail estimate now proves uniform convergence of the actual derivatives. Their limit is O(ℓ^θ) from J(x,y), hence invertible on sufficiently short legs. In common local center graph coordinates, uniform convergence of maps and derivatives implies C¹ convergence to h by integration on source coordinate segments. This avoids both the all-pairs estimate and any choice of sign from projectivized derivatives.
Positive Hölder exponent. The true-pair sum just obtained telescopes up to time m to K_m = (Df^m|E^c(y))^(−1) J(x_m,y_m) Df^m|E^c(x), with error O(b^m) for all short stable pairs. The finite products have θ-Hölder constants ≤ C D^m in pair distance for some D > 1: extend the factors by center projection in smooth ambient bundle coordinates, replace one factor at a time, and use the one-step uniform bounds and Lip(f)^i for the base points at time i. A finite smooth bundle embedding into a trivial bundle, or equivalent finite chart identifications, allows these comparisons without following a single chart for m iterates. The linear-in-m factor from the replacements is absorbed into D^m. For pair-distance t balance b^m against D^m t^θ; this gives, for example, exponent θ log(1/b)/(log D + log(1/b)) > 0. Along y = h(x) the pair parameter is locally Lipschitz in center coordinates by the already proved derivative bound. Center graph coordinate changes are uniformly C^{1+θ} since their tangent planes are θ-Hölder. The limit derivative on the source center disk is therefore Hölder with a uniform positive exponent. General local holonomies along a longer chosen leg can be composed from short ones; no uniform derivative bound for unbounded leg length is implied.
Premises used
- The note's compact coherent foliations give local product homeomorphisms, C¹ plaques tangent to the specified distributions, and uniform local graph geometry. The θ-Hölder hypothesis is on E^c itself, not just on the derivative of f. These are essential to this particular proof.
- The rates in the proof are buffered and multiplied along the reference x-orbit. The expanding factor on the source uses the upper center bound at x_i, and the inverse on the approximant uses the buffered conorm in the tube at that SAME time.
- Only elementary local graph, chain-rule, compactness and convergence arguments in addition to this geometry are used here. Neither Saghin nor Brown nor the journal's Proposition 3.1(vii) is a premise of this check.
Missing inputs
Scope changes
- This is adjudication of a NEW proof, not evidence that the 2005 proof contained it.
- It uses genuine compact coherent geometry and the stated center-bundle exponent. It does not by itself establish uniform graph-transform hypotheses on BW's noncompact fake model, replace the application check in M2, or prove ergodicity.
Located concern
sol-note-proposal H3, branch selection and restricted estimates, contrasted with June 2005 note pp. 4–7
Two scope qualifications are needed; neither leaves a gap for the compact local conclusion.
Ambient proximity alone would not specify a target sheet, and a bound using a_{0:k} with a > ν is not literally a bound with ν_k for fixed unbuffered rates. Also, closeness of projectivizations is not closeness of derivatives.
Mark the sheet by the intrinsic point f^n h(x), use the actual derivative of the disk intersection and retain the buffered rates all the way through the summable series, as above.
This supplies a separate proof of the local C^{1+β} conclusion, not a verification of the printed Lemma 1.1, the exact estimates of Lemma 1.2, or the explicit exponent calculation on note pp. 9–11.
Outside the audit
- The text/version history of the separate bibliographic [BW06b] and the full journal submission as of 2005-10-11.
- Full proofs of the external graph-transform, PSW/correction, genuine strong-foliation absolute-continuity and Pinsker inputs.
- Independent line-by-line validity of Saghin's entire preprint and Brown's full proof. The 2025 Saghin and 2022 Brown journal PDFs were not used as primary texts.
- BW's remaining julienne/density lemmas and other downstream works mentioned in the coordinator record, outside M1–M3.
Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.