Pith. sign in

REVIEW 3 major objections 4 minor 2 references

Ergodic cocycles in hyperbolic and Hadamard spaces

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper proves that ergodic cocycles on non-elementary hyperbolic actions converge almost surely to the Gromov boundary under the asymptotic past–future independence condition.

desk verdict The paper does real work on Mackey ranges and CAT(0) boundaries, but the main convergence proof leans on a hitting lemma that is false for transient random walks. read the letter →

arxiv 2509.06797 v1 pith:ZIR4CYCC submitted 2025-09-08 math.DS math.GR

Corentin Le Bars This is my paper · ORCID
classification math.DSmath.GR MSC 37A3037A5020F6560B1537H99
keywords ergodiccocyclesGromov-hyperbolicspacesHadamardMackeyrangeasymptoticpastandfutureindependenceboundaryconvergencepositivedriftcontractingelements
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proves that for ergodic random dynamical systems on a group acting non-elementarily on a possibly non-proper hyperbolic space, the backward cocycle converges almost surely to a point of the Gromov boundary, provided the system satisfies the asymptotic past and future independence condition (apafi). No integrability or properness is needed. When the cocycle has finite first moment, its drift is strictly positive. The same conclusions hold for Hadamard spaces with independent contracting elements, via curtain-based hyperbolic models. This extends boundary convergence and positive-speed laws from random walks to time-dependent ergodic cocycles.

What carries the argument

The asymptotic past and future independence condition (apafi) is the load-bearing stochastic assumption: past and future Mackey-range boundaries are weakly independent, forming a G-boundary system. The Mackey ranges B±, defined as the spaces of ergodic components of one-sided trajectory spaces under the shift, replace the Poisson–Furstenberg boundary of a random walk. For Hadamard spaces, the curtain-based hyperbolic models X_L convert contracting isometries into loxodromic ones, and the boundary embedding ∂_L: ∂Grom X_L → ∂∞ X transfers hyperbolic convergence back to the visual boundary.

What would settle it

Take the simple random walk on the free group F_2 with a finitely supported non-elementary measure, which is transient and apafi. Fix a loxodromic g and compute the probability that the backward cocycle χ(n,ω) ever equals some fixed positive power of g; transience makes this probability less than 1, contradicting the hitting assumption in Lemma 6.4. This does not by itself disprove the convergence theorem, but it shows the written proof's pivotal support claim fails for a standard example.

Watch

Extended reading notes

Core claim

The central claim is Theorem 1.1: under apafi, for any basepoint o, the backward cocycle χ(n,ω)o converges P-almost surely to a random boundary point ξ+(ω) in the Gromov boundary. The proof constructs past and future Mackey ranges B− and B+ from the cocycle, shows they form a G-boundary system, uses them to produce forward-invariant measures that are Dirac measures, and then exploits weak convexity of shadows to turn boundary contraction into actual convergence in the space. Theorem 1.4 adds that integrability forces the drift λ_X(φ) to be strictly positive. The Hadamard analogues, Theorems 1.2 and 1.5, follow by passing to a hyperbolic model X_L and using a new equivariant homeomorphism bet

Load-bearing premise

The key lemma assumes that apafi guarantees every sample path eventually hits a fixed positive power of any given loxodromic element; transient random walks satisfy apafi but can violate this, and the proof of boundary convergence depends on that hitting event.

Editorial extensions

If this is right

  • Almost-sure convergence to the Gromov boundary holds for every ergodic apafi cocycle on a non-elementary hyperbolic action, covering non-proper settings such as arc graphs, extension graphs, and R-trees.
  • For integrable cocycles the drift is positive, so sample paths escape to infinity at a positive linear rate; in Hadamard spaces they track a geodesic ray with sublinear error.
  • In Hadamard spaces with contracting elements, the hitting measure is supported on curtain-visibility points in B_L, so any two distinct limit points subtend angle π.
  • The map from the Gromov boundary of each hyperbolic model to the visual boundary is an Isom(X)-equivariant homeomorphism onto B_L, giving a dictionary between model directions and true boundary points.
  • The Mackey-range actions are coarsely metrically ergodic, extending strong approximate transitivity from Poisson boundaries to ergodic cocycles.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the boundary embedding for curtain models extends to other wall-based hyperbolic models that detect strongly contracting directions, the same argument should give convergence and positive drift for actions on injective spaces or hierarchically hyperbolic spaces.
  • The apafi condition is plausibly close to necessary: without weak past–future independence, past-oriented stationary measures need not be diffuse, and the support argument for attracting fixed points breaks down; constructing a non-apafi ergodic cocycle with oscillating sample paths would delineate the boundary of the theorem.
  • The Busemann-cocycle proof of positive drift is set up for second-moment arguments; adding moment assumptions on displacement should yield central limit and iterated-logarithm rates for apafi cocycles.
  • Because convergence requires no integrability, apafi alone rules out a zero-drift non-convergent regime; checking whether apafi also forbids sublinear but unbounded displacement would sharpen the dichotomy between convergence and escape rate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper studies discrete random dynamical systems generated by an ergodic measure-preserving transformation and a measurable map f from the probability space to a countable group G acting by isometries on a (possibly non-proper) Gromov-hyperbolic or CAT(0) space. Under the 'asymptotic past and future independence' (apafi) condition of Bader–Furman, it claims almost sure convergence of the backward cocycle to the Gromov or visual boundary, positive drift for integrable cocycles, and analogous statements for Hadamard spaces via hyperbolic models introduced by Petyt–Spriano–Zalloum. The paper also develops ergodic properties of Mackey ranges, including strong approximate transitivity and coarse metric ergodicity, and proves an equivariant embedding of the boundary of the hyperbolic models into the visual boundary of the CAT(0) space.

Significance. If the main theorems were established, they would give a genuinely general boundary convergence theorem for ergodic cocycles without moment assumptions, recovering the Maher–Tiozzo random walk theorem and extending it to CAT(0) spaces. The paper has several strengths: it is clearly written, it properly attributes external results, it contains no fitted parameters or hidden normalizations, and it contributes original technical tools, notably the adaptation of BCFS24 to almost geodesic hyperbolic spaces, the proof of Theorem 1.3, and the coarse metric ergodicity result for Mackey ranges. However, the proof of the central convergence theorem relies on a false hitting-time assertion in Lemma 6.4. This is a load-bearing gap: it invalidates Corollary 6.5, the proof of Theorem 6.2, and consequently the main hyperbolic and CAT(0) convergence and drift theorems as stated.

major comments (3)
  1. [§6.2, Lemma 6.4] In the proof of Lemma 6.4, the paper asserts that by apafi and Proposition 5.6, for almost every ω there exists n such that χ(n,ω)=g^k for some k≥k0. Proposition 5.6 only shows that supp(f_*P) generates G as a semigroup; it gives no information about individual sample paths realizing a specified word at some time. In the random walk case, which the paper explicitly includes, χ(n,ω) is the product of i.i.d. increments. For a transient admissible random walk on a free group, for instance simple random walk on F_2 with support {a,b,a^{-1},b^{-1}}, the event that the walk ever equals a fixed element a^k has probability strictly less than 1. Hence the set ⋃_n Ω_n defined in the proof is not conull, and the assertion is false.
  2. [§6.2, Corollary 6.5 and Theorem 6.2] Corollary 6.5 is deduced directly from Lemma 6.4 and asserts that the support of μ+_{p+(ω)} is the full limit set Λ(G). Since Lemma 6.4 is unsupported, Corollary 6.5 is not established. The proof of Theorem 6.2 then explicitly uses Corollary 6.5 to choose two points ξ,η in the support of μ+_{p+(T^{-n}ω)} and applies the shadow argument of Proposition 6.6. Without a valid support statement, the contraction in Proposition 6.3 cannot be converted into convergence of the cocycle in X. The error therefore propagates to Theorem 1.1 and, through Theorem 7.1, to Theorems 1.2, 1.4, and 1.5.
  3. [§1.1 and §7] The paper states that Theorem 1.1 recovers the Maher–Tiozzo random walk convergence theorem. That recovery is not justified by the present proof. For random walks the support of the stationary measure is known to be the entire limit set, but the argument supplied here, via Lemma 6.4, is invalid. A correct proof would need a different mechanism to show that all loxodromic fixed points lie in the support under apafi. The Hadamard-space results inherit the same gap because Theorem 7.1 applies Theorem 6.2 to the hyperbolic model; no independent support argument is given there.
minor comments (4)
  1. [§8.1, proof of Theorem 8.1] The proof refers twice to 'Proposition 8.4', but the stated result is Corollary 8.4. Please correct the cross-reference.
  2. [§6.2, Lemma 6.4] There is a typo in the citation to [Arn98]: 'Corolloary' should be 'Corollary'.
  3. [§6.2, proof of Theorem 6.2] The phrase 'freedom to chose a pair of points ξ,η in the support' is imprecise. For the argument one needs, for a sequence of times, points in the support of μ+_{p+(T^{-n}ω)} that can be chosen measurably and stay away from the limit point. This is secondary to the support issue but should be clarified in a revision.
  4. [§2.4] The reduction from separable to non-separable spaces via [GST20, Remark 4] is sketched in one sentence. Since the paper claims to drop separability, a fuller explanation of the equivariant quasi-isometric embedding Y↪X would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained modulo properly attributed external theorems; the flagged Lemma 6.4 issue is a proof gap, not a circular reduction.

full rationale

The paper's central claims (Theorem 1.1 convergence in hyperbolic spaces, Theorem 1.4 positive drift, and their CAT(0) analogues) are not normalized into their inputs. The apafi condition is an explicit hypothesis in the sense of Bader–Furman, not a conclusion of the convergence theorem, and it is not defined in terms of the target result. The boundary maps and uniqueness statements are imported from external sources—[BCFS24], [BF14], [BF25], [PSZ24]—with proper attribution, and none of these works are by the present author, so there is no self-citation chain carrying the argument. The transfer to Hadamard spaces is not circular: the hyperbolic models of Petyt–Spriano–Zalloum are cited, but the needed boundary homeomorphism (Theorem 1.3) is proved in Section 3 from those ingredients rather than assumed. The positive drift result is obtained by a separate Kingman/Birkhoff ergodic argument with no fitted parameter being relabeled as a prediction. The only substantive concern raised by a careful reading is Lemma 6.4: it asserts that under apafi, for almost every omega there is n with chi(n,omega)=g^k for a fixed loxodromic g, and the cited Proposition 5.6 only shows the support generates G as a semigroup. That could be a genuine gap or even a false assertion for transient random walks, but it is not circularity: the lemma is not an input assumption, its conclusion is not equivalent to the convergence theorem, and failure of the lemma does not mean the paper's conclusion was built into its premises. No pattern from the enumerated circularity kinds is present.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the apafi hypothesis, the non-elementary action assumption, and substantial imported theorems from BCFS24, BF25, PSZ24. No free parameters or invented entities are introduced; the Mackey ranges and hyperbolic models are standard or borrowed constructions.

assumptions (5)
  • domain assumption The RDS satisfies the asymptotic past and future independence condition (apafi) of [BF14]
    Used in every main theorem; it is the key mixing-like hypothesis that makes the past and future boundaries independent.
  • domain assumption The G-action on X is non-elementary, i.e. contains two independent loxodromic elements (hyperbolic) or two independent contracting elements (CAT(0))
    Needed for the richness of boundary dynamics, in particular for the classification of boundary maps used in Theorem 6.1.
  • domain assumption The hyperbolic model (X,d_L) of [PSZ24] is δ-hyperbolic and α-almost geodesic, and the boundary embedding ∂L: B_L → ∂Grom X_L holds; in the non-proper case this is Theorem 1.3 proven in Section 3
    The CAT(0) arguments transfer to the hyperbolic case only through these models; the author proves the required boundary statement but inherits curtain machinery from PSZ24.
  • domain assumption The classification of boundary maps in Theorem 6.1 ([BCFS24, Theorem 3.1]) and the boundary system properties of [BF25] apply to the Mackey ranges built here
    The proofs of Section 6 and Section 5 import these external results as black boxes.
  • standard math Doob's martingale convergence, Birkhoff's and Kingman's ergodic theorems, and Rohlin disintegration hold in the standard Borel spaces used
    Used throughout Sections 4-8 without proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ergodic cocycles in hyperbolic and Hadamard spaces." pith.science (2026). https://pith.science/paper/ZIR4CYCC

@misc{pith2026250906797,
  author       = {Pith},
  title        = {Pith review of: Ergodic cocycles in hyperbolic and Hadamard spaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZIR4CYCC}},
  note         = {Machine review of arXiv:2509.06797}
}
read the original abstract

We consider discrete random dynamical systems induced by a non-elementary group action on a non-proper hyperbolic space. We prove that if the system is ergodic and satisfies the ``asymptotic past and future independence condition'' as defined by Bader and Furman, the associated ergodic cocycle converges to the Gromov boundary almost surely. If the cocycle has finite first moment, we show that its drift is positive. Using hyperbolic models introduced by Petyt-Spriano-Zalloum, we prove analogous statements for groups acting on Hadamard spaces with a pair of contracting elements.

Figures

Figures reproduced from arXiv: 2509.06797 by the authors.

Figure 1
Figure 1. Illustration of Lemma 3.3. are quasi-geodesic hyperbolic space with hyperbolicity constants depending only on L, and Isom(X) acts by isometries on (X, dL) [PSZ24, Theorem 3.9]. Moreover, a semisimple isometry of (X, d) is contracting if and only if there exists L ∈ N such that the corresponding isometry on (X, dL) is loxodromic [PSZ24, Theorem 4.9]. 3.3. Boundaries. Definition 3.5. We say that a geodesic ray γ : [0,… view at source ↗
Figure 2
Figure 2. Proof of Proposition 3.8. 3.3, for all m, m′ ≥ pn, d(γm(t n m), γm′(t n m′)) ≤ 4L + 3. Without loss of generality, we can assume that t n m ≤ t n m′. Note that because curtains are thick, and since the projections do not increase distances (Proposition [BH99, Lemma II.2.4]), d(o, γm(t n m)) ≥ d(o, hn), hence t n m ≥ n. By convexity of the CAT(0) metric [BH99, Proposition II.2.2], we have d(γm(t n m), γm′(t n m)) ≤ 4… view at source ↗
Figure 3
Figure 3. A discrete RDS seen as a bundle Ω × X over Ω. Definition 4.3. A group random element generator (greg) (Ω, P, T, G, f) is the invertible measurable map defined by φ: Z × Ω → G (t, ω) 7→ φ(t, ω), with φ(1, ω) = f(ω) and for all t, s ∈ Z, ω ∈ Ω, φ(t + s, ω) = φ(t, T sω)φ(s, ω). Note that in this case, by the cocycle relation (4), we have that Id = φ(0, ω) = φ(1 − 1, ω) = φ(1, T −1ω)φ(−1, ω) ⇒ φ(−1, ω) = φ(1, T −1ω) −1 … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

2 extracted references · 1 canonical work pages

  1. [231]

    arXiv:2208.04861.↑5 [Zbi24] Stefanie Zbinden,Hyperbolic spaces that detect all strongly-contracting direc- tions, 2024

    MR2164775↑8, 9, 10 [Yan25] Wenyuan Yang,Conformal dynamics at infinity for groups with contracting elements, 2025. arXiv:2208.04861.↑5 [Zbi24] Stefanie Zbinden,Hyperbolic spaces that detect all strongly-contracting direc- tions, 2024. arXiv:2404.12162.↑5 [Zim78] Robert J. Zimmer,Amenable ergodic group actions and an application to Pois- son boundaries of ...

    arXiv 2025

  2. [2016]

    Patterson-Sullivan theory for groups with a strongly contracting element

    MR3560700↑51 [BS00] M. Bonk and O. Schramm,Embeddings of Gromov hyperbolic spaces, Geom. Funct. Anal.10(2000), no. 2, 266–306. MR1771428↑8, 10, 11 [Cho23] Inhyeok Choi,Central limit theorem and geodesic tracking on hyperbolic spaces and Teichmüller spaces, Adv. Math.431(2023), Paper No. 109236, 68. MR4627690↑2 ERGODIC COCYCLES IN HYPERBOLIC AND HADAMARD S...

    work page Pith review arXiv 2000

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.