SL_2(R)-developments and Signature Asymptotics for Planar Paths with Bounded Variation

Horatio Boedihardjo; Xi Geng

REVIEW 2 major objections 7 minor 17 references

SL_2(R)-developments and Signature Asymptotics for Planar Paths with Bounded Variation

T0 review · 2 major / 7 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read A planar path's length is determined by the asymptotic growth of its iterated integrals, for every strongly tree-reduced path of finite length.

desk verdict A genuine advance on the planar signature asymptotics conjecture, with one load-bearing cited estimate that a referee should verify against the source. read the letter →

arxiv 2009.13082 v2 pith:S5ZDYNOU submitted 2020-09-28 math.CA

Horatio Boedihardjo , Xi Geng This is my paper · ORCID

classification math.CA MSC 26A4560L1060L20

keywords signaturetransformiteratedintegralspathlengthtree-reducedSL2(R)developmentangledynamicsboundedvariationroughpaths

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 the length of a planar path of finite length can be recovered from the growth rate of its normalized signature, provided the path is strongly tree-reduced, meaning it has no local tree-like pieces. This confirms a conjecture implicit in earlier work and explicit in the recent literature, which previously had been established only for paths with continuous derivatives. The result matters because it separates the topological property of being tree-reduced from any regularity assumptions on the path. The proof works by lifting the path to the group SL2(R) and showing that a certain angle dynamics tracks the path's tangent angle closely at large scales, yielding an effective lower bound on the signature norm.

What carries the argument

The key object is the Cartan development of the path into the special linear group SL2(R), which satisfies a linear matrix ODE. In polar coordinates the dynamics decouples into a radial equation dρ/dt = λρ cos(α−2φ) and an angular equation dφ/dt = λ sin(α−2φ), where α is the reversed tangent angle. The angular equation is independent of the radial component and acts as a strong mean-reverting force pushing 2φ toward α when the parameter λ is large. The proof's core is a set of lemmas showing how fast 2φ enters and stays near the interval where α spends most of its time, quantified in terms of λ and the measure of exceptional sets.

What would settle it

Compute, for a strongly tree-reduced planar bounded-variation path with a singular cusp (parameter c in (0,1) as in Section 6), the normalized signature norms under the projective tensor norm for large n; if the n-th root limit comes out strictly below the path length, the theorem would be false. Alternatively, numerically test the imported lower bound on a path whose angle undergoes dense rapid oscillations to see whether the growth rate of the Cartan development can fall short of the signature asymptotics.

Watch

Extended reading notes

Core claim

The central claim is that for a path γ parametrized at unit speed in the plane, if every interior point has a neighbourhood on which the angular path stays in an interval of length at most π except on a negligibly small set, then the signature asymptotics formula holds: the limit of the n-th root of the norm of n! times the n-fold iterated integral equals the length of γ. This condition, called strong tree-reducedness, captures the absence of local tree-like pieces without imposing differentiability. The proof establishes the lower bound needed for the formula by analysing the Cartan development of the path into SL2(R).

Load-bearing premise

The argument imports a lower bound from the authors' earlier work asserting that the signature asymptotics functional is at least the normalized exponential growth rate of the Cartan development matrix; if that estimate carries hidden regularity conditions beyond the bounded-variation setting, the lower-bound half of the proof would not reach the length.

Editorial extensions

If this is right

For every strongly tree-reduced planar path of finite length, the length is determined by the asymptotic growth of its normalized signature, so the signature inversion problem has a sharper lower bound for this class.
The result includes all C1 paths and piecewise C1 paths with intersection angles strictly below π, as well as paths with regular cusps, extending the previously known C1 case.
Combined with the uniqueness theorem for rough path signatures, strong tree-reducedness implies the path is tree-reduced in the classical sense, confirming the heuristic connection between the two notions.
The decoupling of the angle dynamics from the radial dynamics reduces a global analytic estimate to the behaviour of a single scalar ODE, a structure that may simplify other asymptotic signature questions.
The effective signature lower bound obtained here is the type of estimate needed for convergence of algorithms that invert the signature to reconstruct a path.

Reading between the lines

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

The angle-tracking mechanism suggests the length-recovery phenomenon is fundamentally about the absence of π-turns rather than regularity; a testable extrapolation is that the formula should hold under the weaker local turning-angle condition that the paper poses as an open question, at least when the angle switches are not dense.
The proof relies on the monotonicity of the sine function in the one-dimensional angular ODE; in higher dimensions the angular process lives on a sphere and loses this monotonicity, so a genuinely different mechanism would be needed for paths in R^3 or beyond.
A concrete numerical experiment on the singular cusp family with parameter c in (0,1) could probe how quickly 2φ^λ tracks α as λ grows; the comparison-lemma proof predicts an exponential-in-λ entry into the good region, which is measurable.
If the full length conjecture holds for all tree-reduced bounded-variation paths, then the asymptotic signature would act as a complete metric invariant of the reduced path class; the present theorem is evidence toward that stronger invariant view.

Share X Bluesky LinkedIn Reddit HN

Formalized claims in Lean

Claim #1: The central claim is that for a path γ parametrized at unit speed in the plane, if every interior point has a neighbourhood on which the angular path stays in an interval of length at most π except on a negligibly small set, then the signature asymptotics formula holds: the limit of the n-th root of the norm of n! times the n-fold iterated integral equals the length of γ. This condition, called st

/-- @claim 1 The central claim is that for a path γ parametrized at unit speed in the plane, if every interior point has a neighbourhood on which the angular path stays in an interval of length at most π except on a negligibly small set, then the signature asymptotics formula holds: the limit of the n-th root of the norm of n! times the n-fold iterated integral equals the length of γ. This condition, called st -/
def central_claim : Prop :=

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

This paper delivers a real advance. Theorem 2.1 extends the Hambly-Lyons / Chang-Lyons-Ni signature asymptotic formula from C1 unit-speed paths to all bounded-variation planar paths satisfying strong tree-reducedness. That condition is local and allows regular cusps, and the proof only needs measurability of the angle. The SL2(R) development with the decoupled radial/angular ODE system is new and the angle-tracking lemmas in Section 4 are clean. The local-to-global patching in Section 5 is dense but the parameter order (lambda to infinity, eta to 0, M to infinity, epsilon to 0, delta to 0) is handled carefully. The C1 case is reproved in the appendix, which is a good sanity check.

The main soft spot is Proposition 3.1, imported from the authors' earlier paper [3] without proof. It is the only bridge from the signature functional to the Cartan development, so the whole lower-bound argument rests on it. This is not a defect in itself—mathematicians cite theorems—but the stress-test note is right that if the version stated here is not exactly what [3] proves, the lower-bound argument fails. My reading is that the proposition is tailored to BV paths and [3] treats pure rough paths, so BV should be a special case. Still, a referee should check that the statement matches.

Second, Remark 2.1 admits that strong tree-reducedness is only shown to imply tree-reducedness as a consequence of Theorem 2.1 plus the uniqueness theorem. That is not circular, but it means the paper does not independently characterize its class. Tree-reduced paths that are not strongly tree-reduced remain outside the theorem, and the authors are honest about this gap.

Third, Section 6 on singular cusps treats a model example rather than a general condition. It is illustrative, not a general theorem.

Overall, the central estimate appears coherent, the auxiliary C1 proof checks out, and the external citation is the only serious dependency. I would send this to a competent referee; it deserves publication if the referee confirms Proposition 3.1 applies verbatim to the BV setting.

Referee Report

2 major / 7 minor

Summary. The paper proves the signature asymptotics formula (1.1) for planar paths of finite length parametrized at unit speed, under a 'strongly tree-reduced' condition (Definition 2.4) that allows the angle function to be merely measurable. The method lifts the path to SL2(R) via the Cartan development, obtains a decoupled radial/angular ODE system (Lemmas 3.1-3.2), and then analyzes the angular dynamics at microscopic scales. The global case (Theorem 5.1) is handled by Lusin's theorem, a finite partition with mesh inversely proportional to λ, and the local estimates Lemmas 4.1-4.3 and 5.1. The local case (Theorem 2.1) is reduced to the global one by a covering lemma (Lemma 5.2) and a consistency argument for initial conditions (Lemma 5.3). An extension to a class of singular cusps is given in Section 6, and a self-contained proof of the C1 case is relegated to the appendix.

Significance. If correct, Theorem 2.1 constitutes a substantial advance: it replaces the C1 regularity assumption of earlier works with a purely geometric non-degeneracy condition and confirms the Hambly-Lyons/Chang-Lyons-Ni conjecture for a large class of bounded-variation planar paths. The technique of reducing signature asymptotics to a decoupled angular ODE on SL2(R) is elegant and likely to be influential. The appendix provides a clean self-contained proof for the C1 case, and the main global/local proof is mostly self-contained after the quoted Proposition 3.1. The paper is careful in separating the tree-reducedness condition from regularity assumptions, which is methodologically valuable.

major comments (2)

[§3.1, Proposition 3.1 and §3.3, Lemma 3.3] Proposition 3.1 is the only bridge from the signature functional L1(γ) to the growth of the Cartan development Γ^λ, and it is imported from [3] without proof or verification that its hypotheses are satisfied for the measurable-β unit-speed BV paths considered here. The proof of Theorem 2.1 depends on it via Lemma 3.3; if the result in [3] carries additional conditions (for example, a pure-rough-path or p-variation assumption, or yields only a limsup bound), the lower-bound argument in Sections 5.1 and 5.2 would not reach the length. The authors should either reproduce a proof of Proposition 3.1 for the BV case or state the precise theorem from [3] and check its hypotheses under Definition 2.4. In addition, the strict inequality in Proposition 3.1 (and hence in Lemma 3.3) appears to be a typo: if the right-hand limit equals L, L1(γ) > L would contradict the universal upper bound L1(γ) ≤ Length(γ); these inequalities should be non-strict.
[§5.1.3, limiting order (5.4) and estimates (5.11), (5.15)] The final passage to the limit is not fully rigorous as written. The estimates (5.11) and (5.15) contain the quantity cos(2ε + 2Mη/(ε sinε)) as well as terms of order L/M and η. In order to conclude that the liminf of Iλ is at least L, the parameters must be chosen so that ηM/ε → 0 and all remainder terms vanish. The statement 'λ→∞, η→0+, M→∞, ε→0+, δ→0+' in (5.4) is ambiguous as an iterated limit order; a naive reading in which η→0 before M→∞ leaves the term ηM/ε uncontrolled. The authors should specify a diagonal choice (e.g., fix δ and ε, then choose M large, then choose η small depending on M, then choose λ large) and show that the final lower bound is attained. This is a rigor issue in a load-bearing step and should be clarified.

minor comments (7)

[§2, Definition 2.3] In the definition of regular cusp, 'a closed subset F ⊆ L' should read 'F ⊆ [0,L]'.
[§3.2 and Appendix, Corollary A.1] The angular path is said to take values in R^2 in two places; it should be R-valued. (See 'β : [0,L]→R^2' in Section 3.2 and in Corollary A.1.)
[§5.1.3, definition of K_n] In the definition of K_n, the integrand is written as cos(2ψ^λ_t − α_t); this should be cos(2φ^λ_t − α_t).
[§6, equation (6.2)] The intermediate equality λ sin(2ψ − (θ−π)) = λ sin(2ψ − θ) is incorrect as written; the first expression should be λ sin((θ−π) − 2ψ). The final equality is correct, so this is a presentation error.
[§5.2.3, estimate (5.15)] The derivation of (5.15) should explicitly account for the exceptional sets (F'_i)^c and justify the factor 2L/M; as it stands the reader has to reconstruct the argument from the global-case estimates.
[§5.1.3 and §5.2.3, final inequalities] The conclusion 'lim_{λ→∞} ∫ cos(α−2φ) dt > L' (and the similar claim in Theorem 5.1) should read '≥ L', since the integral is bounded above by L; the strict inequality would conflict with the universal upper bound.
[Introduction] The word 'tree-redcuedness' appears in the Introduction and should be 'tree-reducedness'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the proof combines the cited parameter-free lower estimate of Proposition 3.1 with a new, self-contained SL2(R) angle-dynamics estimate.

full rationale

The derivation chain is not circular. Theorem 2.1 asks for the lower bound L1(γ) >= Length(γ); the upper bound is the usual triangle inequality. The lower bound is reached in two independent parts: (i) Proposition 3.1, quoted from the authors' prior paper [3], gives L1(γ) >= lim log||Γ^λ_L||/(λ||Φ||) for bounded-variation paths; (ii) Sections 4-5 prove, from the strong-tree-reducedness hypothesis on the measurable angle path β, that the associated SL2(R) angle dynamics satisfies lim ∫ cos(α_t - 2φ_t) dt >= L. Part (ii) is new and self-contained; it does not assume the target formula or the tree-reducedness conclusion. Proposition 3.1 is a parameter-free theorem with stated hypotheses (finite-dimensional normed spaces, bounded-variation path, linear map and representation) that do not include the signature-asymptotics equality for strongly tree-reduced planar paths, so under the review rules it counts as independent support rather than circular input. The only other self-citation, the uniqueness theorem [2], appears in Remark 2.1 as a post-theorem observation and is not used anywhere in the proof of Theorem 2.1. No fitted parameter is renamed as a prediction, and no definition smuggles the conclusion into the hypothesis. Therefore the paper exhibits no significant circularity.

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

The paper introduces no free parameters fitted to data and no new entities. The proof rests on the standing strong-tree-reducedness hypothesis, on standard measure-theoretic and ODE tools, and on the authors' earlier lower estimate (Proposition 3.1 from [3]) for the signature in terms of Cartan development growth.

assumptions (5)

domain assumption Proposition 3.1 (from [3]): L_1(γ) ≥ lim_{λ→∞} log||Γ^λ_L||/λ for the SL2(R) Cartan development.
Imported from Boedihardjo, Geng and Souris (Adv. Math. 2020), the bridge from signature asymptotics to the ODE dynamics; cited, not proved here.
domain assumption Strong tree-reducedness (Definition 2.4) is the standing hypothesis of Theorem 2.1.
The theorem is conditional on this non-degeneracy condition; Remark 2.1 notes its equivalence to tree-reducedness is only established retroactively via the theorem.
standard math Lusin's theorem: measurable functions are continuous on large compact sets.
Used in Section 5.1.1 to find a compact set F2 on which α is uniformly continuous.
standard math ODE existence and uniqueness, and Gronwall's inequality.
Used throughout Sections 3 to 6 to define and compare angle dynamics solutions.
standard math Compactness of [0,L] and existence of finite subcovers.
Used in Lemma 5.2 to replace the local regular-cusp assumption with a finite covering.

how reviews work

0 comments

Cite this review

Pith. "Pith review of SL_2(R)-developments and Signature Asymptotics for Planar Paths with Bounded Variation." pith.science (2026). https://pith.science/paper/S5ZDYNOU

@misc{pith2026200913082,
  author       = {Pith},
  title        = {Pith review of: SL_2(R)-developments and Signature Asymptotics for Planar Paths with Bounded Variation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S5ZDYNOU}},
  note         = {Machine review of arXiv:2009.13082}
}

abstract

The signature transform, defined by the formal tensor series of global iterated path integrals, is a homomorphism between the path space and the tensor algebra that has been studied in geometry, control theory, number theory as well as stochastic analysis. An elegant isometry conjecture states that the length of a bounded variation path $\gamma$ can be recovered from the asymptotics of its normalised signature: $\text{Length}(\gamma)=\lim_{n\rightarrow\infty}\big\Vert n!\int_{0<t_{1}<\cdots<t_{n}<T}d\gamma_{t_{1}}\otimes\cdots\otimes d\gamma_{t_{n}}\big\Vert^{\frac{1}{n}}$. This property depends on a key topological non-degeneracy notion known as tree-reducedness (namely, with no tree-like pieces). Existing arguments have relied crucially on $\gamma$ having a continuous derivative under the unit speed parametrisation. In this article, we prove the above isometry conjecture for planar paths by assuming only local bounds on the angle of $\gamma'$ (which ensures the absence of tree-like pieces). Our technique is based on lifting the path onto the special linear group ${\rm SL}_{2}(\mathbb{R})$ and analysing the behaviour of the associated angle dynamics at a microscopic level.

Figures

Figures reproduced from arXiv: 2009.13082 by the authors.

**Figure 2.** The covering structure when k = 4. In what follows, we always work with a fixed covering structure given by Lemma 5.2. 5.2.2 Step two: consistency of initial conditions From Lemma 5.2, we know that α|[ui−1,vi] is a regular cusp. In particular, we know by assumption that αt ∈ [ai , ai + π] for a.a. t ∈ [ui−1, vi ] with some ai ∈ R. The main issue here is that, we cannot directly apply the results from Section 5.1, si… view at source ↗

**Figure 3.** Regular cusp, Singular cusp and tree-like cusp. [PITH_FULL_IMAGE:figures/full_fig_p025_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages

[3]

Boedihardjo, X

H. Boedihardjo, X. Geng and N.P. Souris, Path developments and tail asymptotics of signature for pure rough paths,Adv. Math.364 (2020)

work page 2020
[1]

Boedihardjo and X

H. Boedihardjo and X. Geng, A non-vanishing property for the signature of a path,C. R. Math. Acad. Sci. Paris357 (2): 120-129, 2019

work page 2019
[2]

Boedihardjo, X

H. Boedihardjo, X. Geng, T. Lyons and D. Yang, The signature of a rough path: unique- ness, Adv. Math.293 (2016): 720-737

work page 2016
[4]

Chen, Iterated integrals of diﬀerential forms and loop space homology, Ann

K.T. Chen, Iterated integrals of diﬀerential forms and loop space homology, Ann. of Math. 97 (2) (1973): 217-246

work page 1973
[5]

Insertion algorithm for inverting the signature of a path

J. Chang and T. Lyons, Insertion algorithm for inverting the signature of a path, arXiv:1907.08423, 2019

work page Pith review arXiv 1907
[6]

Chang, T

J. Chang, T. Lyons and H. Ni, Super-multiplicativity and a lower bound for the decay of the signature of a path of ﬁnite length,C. R. Acad. Sci. Paris 356(1) (2018): 720–724

work page 2018
[7]

Chevyrev and T

I. Chevyrev and T. Lyons, Characteristic functions of measures on geometric rough paths, Ann. Probab.44 (6) (2016): 4049–4082

work page 2016
[8]

Dyson, The radiation theories of Tomonaga, Schwinger and Feynman, Phys

F.J. Dyson, The radiation theories of Tomonaga, Schwinger and Feynman, Phys. Rev. 75 (1949) 486–502

work page 1949

Show all 17 references

[9]

Folland, Real analysis: modern techniques and their applications, 2nd edition, John Wiley & Sons, 1999

G.B. Folland, Real analysis: modern techniques and their applications, 2nd edition, John Wiley & Sons, 1999

1999
[10]

Geng, Reconstruction for the signature of a rough path,Proc

X. Geng, Reconstruction for the signature of a rough path,Proc. Lond. Math. Soc.114 (3): 495-526, 2017

2017
[11]

Hambly and T

B. Hambly and T. Lyons, Uniqueness for the signature of a path of bounded variation and the reduced path group, Ann. of Math. 171 (1) (2010): 109–167

2010
[12]

Lyons and Z

T. Lyons and Z. Qian,System control and rough paths, Clarendon Press, Oxford, 2002

2002
[13]

Lyons and N

T. Lyons and N. Sidorova, On the radius of convergence of the logarithmic signature, Illinois J. Math.50 (4) (2006): 763–790

2006
[14]

Lyons and W

T. Lyons and W. Xu, Hyperbolic development and the inversion of signature, J. Funct. Anal. 272 (7) (2015): 2933–2955

2015
[15]

Lyons and W

T. Lyons and W. Xu, Inverting the signature of a path,J. Eur. Math. Soc.20 (7) (2018): 1655–1687

2018
[16]

Reutenauer,Free Lie algebras, Clarendon Press, Oxford, 1993

C. Reutenauer,Free Lie algebras, Clarendon Press, Oxford, 1993

1993
[17]

Ryan,Introduction to tensor products of Banach spaces, Springer, 2002

R.A. Ryan,Introduction to tensor products of Banach spaces, Springer, 2002. 34

2002

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.

[1] [3]

Boedihardjo, X

H. Boedihardjo, X. Geng and N.P. Souris, Path developments and tail asymptotics of signature for pure rough paths,Adv. Math.364 (2020)

work page 2020

[2] [1]

Boedihardjo and X

H. Boedihardjo and X. Geng, A non-vanishing property for the signature of a path,C. R. Math. Acad. Sci. Paris357 (2): 120-129, 2019

work page 2019

[3] [2]

Boedihardjo, X

H. Boedihardjo, X. Geng, T. Lyons and D. Yang, The signature of a rough path: unique- ness, Adv. Math.293 (2016): 720-737

work page 2016

[4] [4]

Chen, Iterated integrals of diﬀerential forms and loop space homology, Ann

K.T. Chen, Iterated integrals of diﬀerential forms and loop space homology, Ann. of Math. 97 (2) (1973): 217-246

work page 1973

[5] [5]

Insertion algorithm for inverting the signature of a path

J. Chang and T. Lyons, Insertion algorithm for inverting the signature of a path, arXiv:1907.08423, 2019

work page Pith review arXiv 1907

[6] [6]

Chang, T

J. Chang, T. Lyons and H. Ni, Super-multiplicativity and a lower bound for the decay of the signature of a path of ﬁnite length,C. R. Acad. Sci. Paris 356(1) (2018): 720–724

work page 2018

[7] [7]

Chevyrev and T

I. Chevyrev and T. Lyons, Characteristic functions of measures on geometric rough paths, Ann. Probab.44 (6) (2016): 4049–4082

work page 2016

[8] [8]

Dyson, The radiation theories of Tomonaga, Schwinger and Feynman, Phys

F.J. Dyson, The radiation theories of Tomonaga, Schwinger and Feynman, Phys. Rev. 75 (1949) 486–502

work page 1949