Recognition: 4 theorem links
· Lean TheoremThe intersection dual of geodesic currents
Pith reviewed 2026-05-06 14:03 UTC · model claude-opus-4-7
The pith
A curve function comes from a geodesic current exactly when it is additive on disjoint unions and non-increasing under crossing resolutions.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
A real-valued function on closed curves of a hyperbolic surface comes from intersecting with a geodesic current — a measure on the space of geodesics — exactly when it adds across disjoint components and never increases when you resolve an essential crossing in either of the two ways. The current, when it exists, is unique. This recasts geodesic currents, originally defined as flow-invariant measures, as the additive curve functionals that respect surgery, with no reference to dynamics or measure theory. A group-theoretic version of the same criterion replaces "smoothing" with explicit inequalities on products of elements of the surface group whose axes are parallel or crossing.
What carries the argument
A combinatorial duality between curve functionals and measures on the space of oriented geodesics, built by assigning to each "right-handed box" of geodesics a number defined as a limit (1/2) lim_n [f(bx^n) + f(cx^n) − f(ax^n) − f(dx^n)] using a fixed auxiliary simple curve x; smoothing, additivity, and stability force this to converge, be non-negative, finitely additive, and countably additive, hence extend by Carathéodory to a geodesic current that recovers f.
If this is right
- Any length pseudo-metric on a closed surface of genus ≥ 2 has a dual geodesic current
- intersection-number domination between two currents implies length-spectrum domination for every such metric.
- Lengths on filling embedded graphs with the edge metric are intersection numbers with a half-integral multi-curve
- generalizing Erlandsson's result for simple generating sets.
- Periods of generalized cross-ratios — including the discontinuous ones attached to certain Anosov representations — automatically satisfy smoothing and so define geodesic currents
- unifying constructions of Martone–Zhang and Burger–Iozzi–Parreau–Pozzetti.
- A minimal surface-group action on an R-tree is dual to a measured lamination iff it is irreducible and preserves intersection of axes
- giving a streamlined proof of Skora's theorem and three new equivalent conditions
Where Pith is reading between the lines
- Because the criterion is local at crossings and additive across components
- it should transfer cleanly to any setting where one has a notion of essential crossing and a flow-invariant measure space — surfaces with boundary
- orbifolds
- or train-track-like models — once the right analogue of the auxiliary simple curve x with dense translate-endpoints is supplied.
- The sharp gap between smoothing (exact) and quasi-smoothing (up to additive error) — where quasi-Fuchsian length functions live on the wrong side — suggests that the failure of duality measures
- in a quantitative way
- how far a representation is from being Fuchsian
- the open question on whether the quasi-smoothing constant grows with the quasi-Fuchsian constant looks like a natural rigidity statement.
Load-bearing premise
The proof needs the smoothing inequality together with additivity to force certain limits over long words to actually exist; if smoothing is only approximate, the same recipe is known to fail.
What would settle it
Exhibit a curve functional that is additive on disjoint unions and non-increasing under both resolutions of every essential crossing, yet whose values on closed curves cannot be written as i(μ, C) for any geodesic current μ — or, conversely, a geodesic current whose intersection function violates one of the two axioms at some essential crossing. Either would refute the equivalence.
read the original abstract
Geodesic currents on closed hyperbolic surfaces are measures on the unit tangent bundle invariant under geodesic flow and orientation reversal. Every geodesic current induces a dual function on curves via the geometric intersection pairing. It is natural to ask which curve functions are dual to geodesic currents, that is, which arise as intersection functionals of a geodesic current. In this paper we give a purely axiomatic and combinatorial characterization of curve functionals dual to geodesic currents. This yields a new definition of geodesic currents as curve functionals or, equivalently, as functions on surface groups, without reference to measures or flows. More precisely, we show that a function on curves arises as the geometric intersection pairing with a geodesic current if and only if it is additive under disjoint union and satisfies a simple \emph{smoothing} property: it is non-increasing under surgery of essential crossings. As applications, we obtain new axiomatic characterizations of measured laminations and hyperbolic length functions, and new descriptions of small surface group actions on real trees, including a concise proof of a classical theorem of Skora. We also provide a unified framework for dual geodesic currents arising from metric structures and generalized cross-ratios, including those associated with certain Anosov representations. Our approach subsumes all previously known constructions of dual geodesic currents and yields broad new families of examples.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proves that for a closed orientable surface S of genus ≥ 2, a real-valued functional on (unoriented) multi-curves f : C(S) → R arises as f(C) = i(μ_f, C) for a unique geodesic current μ_f if and only if f is additive on disjoint unions and non-increasing under smoothing of essential crossings (Theorem A). An equivalent group-theoretic version (Theorem A') replaces unoriented smoothing by class invariance, stability, and oriented connected/disconnected smoothing inequalities on pairs (a,b) ∈ π_1(S). The construction of μ_f proceeds by defining a pre-measure on a semi-ring of 'right-handed boxes' indexed by quadruples in π_1(S), whose values are double limits of the form lim_{n,m} f([bx^n]) + f([cx^n]) − f([ax^n]) − f([dx^n]); existence of these limits is established via axis-convexity bounds derived from a sequence of explicit smoothing reductions (§4.6, Cor. 4.32, Lemma 5.15). Carathéodory extension then produces the geodesic current. Applications include axiomatic characterizations of measured laminations and hyperbolic Liouville currents (Theorem D), a new conceptual proof of Skora's theorem and three new equivalent characterizations of S-trees dual to measured laminations (Theorem E), dual currents for arbitrary length pseudo-metrics on S (Corollary B) and for embedded filling graphs (Corollary C), and a unified treatment of currents from generalized cross-ratios (Theorem F).
Significance. If correct, the paper provides the natural converse to Bonahon's intersection construction: a clean, intrinsic, group-theoretic characterization of which curve functionals on a closed surface of genus ≥ 2 arise as intersection with a geodesic current. The two formulations (Theorem A on unoriented multi-curves, Theorem A' on π_1(S)) together give a flexible and falsifiable criterion — additivity plus a non-increase-under-smoothing inequality — that subsumes essentially all previously known constructions of dual currents (Bonahon Liouville, Otal/CFF/HP/BL/DLR/Con for Riemannian, NPC, and singular metrics; Erlandsson for simple word lengths; Martone–Zhang and Burger–Iozzi–Parreau–Pozzetti for cross-ratios from Anosov representations) and produces new ones (arbitrary length pseudo-metrics in Cor. B, embedded graphs in Cor. C, a clean characterization of measured laminations and hyperbolic Liouville currents in Theorem D, a streamlined proof and refinement of Skora's theorem in Theorem E). The applications are concrete and falsifiable: max-smoothing characterizes laminations; the parallelogram identity (8.7) characterizes hyperbolic length functions among all curve functionals. The resu
major comments (3)
- [§4.6, Eq. (4.10) and Cor. 4.32, Eq. (4.14)] The trailing Y-exponent is displayed as Y^{n-2s} in both equations, but the input has independent exponents Y^m and x^n. Reading the proof of Eq. (4.10), the chain ends with the line containing [Ax^s aY^s][Y^{m-2s}][Y^s x^s][x^{n-2s}], so the displayed Y^{n-2s} should be Y^{m-2s}. This matters because Eq. (4.14) is the load-bearing input for the lower bound on g_-(n,m) in the proof of Lemma 5.15: with the displayed (incorrect) exponent, subtracting nf([x])+mf([y^{-1}]) leaves a term (n-2s-m)f([y^{-1}]) that is not bounded below as m grows; with the corrected m-2s, the bound is const - 2s(f([x])+f([y^{-1}])) as required. Please correct the display and confirm the rest of Section 5.2 reads off the corrected exponent.
- [§5, Prop. 5.12 (invariance under p_i ↦ x p_i)] The constant s in Proposition 4.24 / Corollary 4.32 producing the splitting configuration depends on the four group elements (a,b,x,y). In the proof of Prop. 5.12 invariance under Eq. (5.4a), the argument applies Lemma 5.15 to the translated quadruple, and then Proposition 5.16 to recombine. It would be helpful to state explicitly that the s for (xa, xb, c, d) and the s for (a,b,c,d) can be taken uniformly large (since both are eventually-axis-convex with possibly different thresholds, and the limit argument only needs sufficiently large n, m). One sentence near the end of the proof of Prop. 5.12 confirming this would close a small but non-trivial gap that a careful reader will pause on.
- [§7, Prop. 7.8 and Theorem A] Stability is used as a hypothesis in Theorem 6.9 and is then derived from additivity + power-smoothing + oriented disconnected smoothing in Prop. 7.8. The route 'Theorem A ⇐ Theorem 6.9 ⇐ stability ⇐ Prop. 7.8' is correct, but the statement of Theorem A in §1 does not list stability among its hypotheses, deferring entirely to 'smoothing+additivity'. Readers translating this to non-additive functionals or to alternative settings (e.g., quasi-smoothing) will need to know that this derivation uses additivity essentially. Consider adding a one-line remark after Theorem A pointing to Prop. 7.8 and noting that it does not survive the relaxation to quasi-smoothing (Remark 7.9).
minor comments (10)
- [Abstract / §1] The abstract states the characterization for 'closed hyperbolic surfaces' but the standing hypothesis (Table 1) is genus ≥ 2 closed orientable. A brief sentence in §1 noting that surfaces with boundary or punctures are not treated, and pointing to where the closed orientable hypothesis is used (e.g., Lemma 4.34 on density of axis endpoints, the use of π_1(S) torsion-freeness in Lemma 4.22), would orient the reader.
- [Definition 2.1] 'Most of the multi-curves we consider will be oriented' — please make this explicit in Notation: from §2.1 onward, C(S) means oriented unless qualified, but Theorem A as stated uses unoriented C(S). Some inline reminders would reduce friction.
- [§4.2, Definition 4.6] The figures (4.2a–e) are central to identifying the four smoothing types. Please label the components of the result in each subfigure (e.g., which arcs are [a], [b], [aB] after smoothing); the current labels assume the reader is tracking the convention from MGT25.
- [§5.1, Definition 5.1] Condition (3) ('for sufficiently large n, …R-cross') is asymmetric in style with (1) and (2). Stating explicitly that for boxes of type 1 the crossing is essential for all n (as in Remark 5.6) immediately after the definition, rather than several pages later, would aid readability.
- [§5.2, Eq. (5.5)] The factor of 1/2 in the definition of \hat μ_f is essential (cf. Appendix B) but its origin is not explained when first introduced. A line of motivation — that it produces flip-invariant intersection numbers via i = ⃗i + σ_*⃗i — at the point of definition would help.
- [§6.1, Proposition 6.5] The proof uses 'V_0 := W(B_0, ε)' to handle the boundary of B_0, with V_0^∘ ⊂ \bar B_0 \setminus B_0 (open in subspace topology). The set inclusion \bar B_0 \setminus B_0 ⊂ V_0^∘ in the proof would benefit from one extra sentence: namely that the upper-right half-edge is what fails to be covered by the B_i (which are half-open in the (s,t)-decreasing sense).
- [§9.2, Theorem 9.11] Subcase 2.2 in the proof has a long sub-argument deriving a contradiction; consider extracting it as a separate lemma since it is the most intricate case-analysis in the section.
- [Notation table] G is used both for an embedded graph (§8.3) and (implicitly) for the inverse of g elsewhere. Please add 'G' as graph to the notation table or rename the graph G → Γ_G to avoid clash.
- [§3.4, Remark 3.29] Helpful clarification, but the example g(n,m) = -mn deserves a sentence noting that this g is in fact bounded above on N×N at 0, but unbounded below — confirming that the boundedness hypothesis in Lemma 3.27 is used essentially.
- [Bibliography] Please double-check the cross-references to [DRMG25] (used in Prop. 3.19, Remark 8.12, Lemma 8.14) — author/year format suggests this is in preparation; if so, mark accordingly.
Simulated Author's Rebuttal
We thank the referee for the careful and constructive report, and in particular for identifying a load-bearing typographical error in Eq. (4.10) / Eq. (4.14) and for highlighting two expository improvements that genuinely sharpen the manuscript. We accept all three major comments. The revisions required are localized: (i) correcting the trailing exponent Y^{n-2s} to Y^{m-2s} in Eqs. (4.10) and (4.14), and re-confirming that the surrounding arguments in Section 5.2 (notably the lower bound on g_-(n,m) in the proof of Lemma 5.15) read off the corrected exponent; (ii) adding one sentence to the proof of Prop. 5.12 stating that the splitting thresholds from Cor. 4.32 for the original and translated quadruples may be taken uniformly large, which is all the limit argument needs; and (iii) inserting a one-line remark after Theorem A pointing to Prop. 7.8 and Remark 7.9, alerting readers that stability is derived using additivity essentially and that the derivation fails under quasi-smoothing. None of these touch the structure of the proof of Theorem A or the applications. We are grateful for the precision of the report.
read point-by-point responses
-
Referee: [§4.6, Eq. (4.10) and Cor. 4.32, Eq. (4.14)] The trailing Y-exponent is displayed as Y^{n-2s} but should be Y^{m-2s}, since the chain ends with [Ax^s aY^s][Y^{m-2s}][Y^s x^s][x^{n-2s}]. This matters because Eq. (4.14) is the load-bearing input for the lower bound on g_-(n,m) in Lemma 5.15: with the displayed exponent the bound fails as m grows; with m-2s it gives const - 2s(f([x])+f([y^{-1}])) as required.
Authors: The referee is correct: this is a typographical error in the trailing exponent. Tracking the chain of reductions in the proof of Eq. (4.10), the final line indeed reads [Ax^s aY^s][Y^{m-2s}][Y^s x^s][x^{n-2s}], so the displayed exponent should be Y^{m-2s}, and likewise in Eq. (4.14) of Corollary 4.32. We thank the referee for catching this and for spelling out precisely why it is load-bearing: in the proof of Lemma 5.15, applying f to Eq. (4.14) and subtracting nf([x])+mf([y^{-1}]) yields the lower bound g_-(n,m) >= -f([ab]) - f([ax^s AY^s]) - f([Bx^s bY^s]) - 2s(f([x])+f([y^{-1}])), which is bounded below uniformly in n,m as needed. With the (incorrect) Y^{n-2s} the residual term (n-2s-m)f([y^{-1}]) is unbounded below as m grows, breaking the boundedness hypothesis of Lemma 3.27. We will correct the displayed exponent in both Eq. (4.10) and Eq. (4.14), and we have re-checked the proof of Lemma 5.15 (and the rest of Section 5.2) to confirm that the surrounding text already uses the corrected exponent implicitly; only the displayed equations require revision. revision: yes
-
Referee: [§5, Prop. 5.12 (invariance under p_i ↦ x p_i)] The constant s in Proposition 4.24 / Corollary 4.32 producing the splitting configuration depends on (a,b,x,y). In the proof of invariance under Eq. (5.4a), the argument applies Lemma 5.15 to the translated quadruple, and then Proposition 5.16 to recombine. It would be helpful to state explicitly that the s for (xa, xb, c, d) and the s for (a,b,c,d) can be taken uniformly large.
Authors: We agree this point deserves to be made explicit. The constants s = s(a,b,x,y) and s' = s(xa,xb,c,d) produced by Proposition 4.24 / Corollary 4.32 are each finite, and the eventual axis-convexity in Lemma 5.15 only requires n,m to exceed the relevant threshold. Taking s_max := max{s(a,b,x,y), s(c,d,x,y), s(xa,xb,x,y), s(xc,xd,x,y)} (all finite), both quadruples are eventually-axis-convex past s_max, and the double limits in the chain of equalities in the proof of Prop. 5.12 may be evaluated using a single uniform threshold. We will add a sentence near the end of the proof of Prop. 5.12 stating: "The constants produced by Corollary 4.32 for the translated quadruples (xa,xb,c,d) and (a,b,c,d) are each finite, so the limits in Lemma 5.15 may be taken past a common threshold; this is all that is required for the equality of double limits via Proposition 5.16." revision: yes
-
Referee: [§7, Prop. 7.8 and Theorem A] Stability is used as a hypothesis in Theorem 6.9 and is then derived from additivity + power-smoothing + oriented disconnected smoothing in Prop. 7.8. The route is correct, but the statement of Theorem A in §1 does not list stability among its hypotheses. Readers translating to non-additive functionals or quasi-smoothing will need to know the derivation uses additivity essentially. Consider adding a one-line remark after Theorem A pointing to Prop. 7.8 and noting it does not survive the relaxation to quasi-smoothing (Remark 7.9).
Authors: We thank the referee for this suggestion, which we adopt. Theorem A indeed implicitly uses stability via Prop. 7.8, whose proof essentially uses additivity (combining Eqs. (7.4)–(7.6)), and Remark 7.9 already records that the analogous derivation fails for C-quasi-smoothing. To make this transparent for readers working in non-additive or quasi-smoothing settings, we will add a brief remark immediately after the statement of Theorem A: "Stability (f(C^n) = n f(C)) is implicit in the hypotheses of Theorem A: it is derived from additivity, power-smoothing, and oriented disconnected smoothing in Proposition 7.8. The proof uses additivity essentially and does not survive the relaxation to quasi-smoothing; see Remark 7.9." This points the careful reader directly to the relevant derivation and its limitations. revision: yes
Circularity Check
No significant circularity: the duality theorem is proved against external benchmarks (Bonahon's intersection, Otal's uniqueness, Carathéodory extension), with self-citations supplying independently checkable lemmas.
full rationale
This is a pure-math characterization paper. The central claim (Theorem A: a curve functional is dual to a geodesic current iff additive + smoothing) is verified against externally fixed objects: (a) Bonahon's intersection pairing on Curr(S) [Bon86], (b) Otal's marked-length-spectrum rigidity [Ota90] used to get uniqueness, and (c) the Carathéodory extension theorem [Bog07]. None of these are self-citations. The construction of μ_f from f is genuinely independent of f's representation: μ_f(B) is defined by an explicit limit (Eq. 5.5) of f-values on long curves [bx^n], [cx^n], [ax^n], [dx^n], and then one proves i(μ_f, C) = f(C) (Theorem 6.9) — the input is f's values on closed curves, the output is a measure, and recovery is a non-trivial computation, not a tautology. Self-citations to MGT21 (Theorem 2.14, smoothing→continuous extension), MGT25 (smoothing trichotomy Prop. 4.10, Lemma 4.5), and Thu00 (max-smoothing for laminations) are used as tools (lemmas about crossings/smoothings) and are independently verifiable from those papers' own arguments. They are not invoked as a "uniqueness theorem" forcing the conclusion of Theorem A; rather, Theorem A's necessity direction is checked separately (Appendix C) and its sufficiency is built measure-theoretically. The skeptic's concern about the chain of smoothings (Eq. 4.10/4.14, Cor. 4.32, Lemma 5.15) underwriting boundedness of g_-(n,m) is a *correctness/robustness* concern about a specific bound, not a circularity concern: the bound is derived from the axioms (smoothing, convex union, stability, homogeneity) on f via explicit reductions in π_1(S), not by importing the conclusion. Even if a transcription issue exists (e.g., the n-2s vs m-2s exponent in Eq. 4.14), that is an error-checking issue, not a circular derivation. There are no fitted parameters, no "predictions" that reduce to inputs, no ansatz smuggled via citation, and no renaming of a known result as a new one (Luo's criterion and Skora's theorem are explicitly recovered as corollaries, with credit). Score: 1.
Axiom & Free-Parameter Ledger
axioms (7)
- standard math Carathéodory extension theorem and standard measure-theoretic preliminaries
- domain assumption Bonahon's bilinear continuous extension of intersection to geodesic currents
- domain assumption Otal's theorem: flip-invariant geodesic currents are determined by their marked length spectrum
- domain assumption MGT21 Theorem A (continuous extension of quasi-convex functionals to currents)
- domain assumption Parry's characterization of R-tree translation length functions
- domain assumption Bonahon's algebraic characterization of Liouville currents (Bon88, Theorem 13)
- domain assumption MGT25 smoothing trichotomy (Propositions 4.9–4.10)
Lean theorems connected to this paper
-
IndisputableMonolith/Cost.leanCost.Jcost_symm; Cost.dalembert_identity unclearTheorem A: A curve functional f:C(S)→R is dual to a geodesic current μ_f ... if and only if it satisfies the smoothing property f(crossing) ≥ max{f(smoothing_1), f(smoothing_2)} ... and is additive on connected components.
-
IndisputableMonolith/Cost/FunctionalEquation.leanJlog_as_cosh; ode_cosh_uniqueness_contdiff echoesTheorem D (Hyperbolic metric case): λ(a)λ(b) = λ(ab) + λ(aB), where λ(g) := 2 cosh(f(g)/2). This identity comes from the hyperbolic parallelogram identity (Lemma A.2).
-
IndisputableMonolith/Cost.leanCost.Jcost_submult unclearStability: f(a^n) = n f(a). Power smoothing: f([a^{n+m}]) ≥ f([a^m][a^n]).
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.